Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Simulation modeling to evaluate cost-benefit of multi-level screening strategies involving behavioral components to improve compliance: the example of diabetic retinopathy
(USC Thesis Other)
Simulation modeling to evaluate cost-benefit of multi-level screening strategies involving behavioral components to improve compliance: the example of diabetic retinopathy
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
DISSERTATION SIMULATION MODELING TO EVALUATE COST- BENEFIT OF MULTI-LEVEL SCREENING STRATEGIES INVOLVING BEHAVIORAL COMPONENTS TO IMPROVE COMPLIANCE: THE EXAMPLE OF DIABETIC RETINOPATHY Irene Vidyanti Committee Chair: Dr. Shinyi Wu Dissertation committee members: Dr. Carl Kesselman, Dr. Chih-Ping Chou Epstein Department of Industrial and Systems Engineering, University of Southern California 2 ABSTRACT Recent health care reform brings forth the importance of preventive strategies such as screening. However, concerns about the growing population and rising healthcare spending necessitate health plans and public health policymakers to consider and determine cost-beneficial population-based screening strategies. Screening strategies often vary on compliance and, thus, result in suboptimal cost-benefit for the population. To achieve maximum cost benefit, especially for screenings that often suffer low compliance, such as Diabetic Retinopathy screening, policymakers need to consider multi-level (e.g. both patient- and provider-level) interventions to improve screening compliance. To determine which of such interventions are cost-beneficial, many different screening strategies need to be considered. Yet, it is costly, impractical, and time-consuming to do clinical trials to test all the different screening strategies. Simulation models provide a more cost- and time-effective way to help determine cost-beneficial screening strategies. Simulation models have been used extensively to address cost-benefit of screening. However, there are shortcomings to existing models. First, they do not have a structure that enables the evaluation of strategies that include policies affecting compliance, even though screening compliance is often low. Thus, strategies that include policies targeting the improvement of compliance may be necessary to achieve maximum cost-benefit. Additionally, even policies not specifically targeting compliance may affect compliance, and models evaluating such policies without considering their impact on compliance will under- or over-estimate the policy impact. Second, current simulation models do not have a structure that can evaluate multi-level strategies (e.g. those targeting patient, providers, and clinics) even though they are more likely to have sustained or powerful effect than those targeting only the individual-level. This research develops a generic conceptual model for screening services that addresses the two shortcomings and then constructs the model for the case of Diabetic Retinopathy screening to illustrate the model. 3 The first shortcoming is addressed by including compliance as a mediating variable in the model, rather than a fixed input variable as in current methods. Compliance is influenced by patient characteristics (demographics, disease severity, self-care, health belief) and screening strategy used, and in turn compliance influences disease progression and healthcare utilization and, thus, cost and benefit of the screening strategy. The second shortcoming is addressed by using hierarchical simulation with nested design where policy effects are manifested not universally but through a hierarchical structure. This enables evaluation of the impact of policies at a higher aggregate level (e.g. policies that target providers) on individual (patient) outcomes. The multi- level design has the benefit of taking into account behavior at several levels and enabling the incorporation of economy of scale (e.g. how clinic size may affect cost-benefit of certain screening strategies) in the model. The developed model is then compared with models without compliance as a mediating variable and models without the hierarchical structure to illustrate its advantages in terms of policy impact assessment. In addition, the models were used to evaluate cost-benefit of different screening strategies using multiple perspectives. As the results indicate differing decisions made by policymakers viewing through these different perspectives, a case for context-based decision making is made. 4 CONTENTS Abstract .................................................................................................................................................................................................. 2 Chapter 1. Introduction ................................................................................................................................................................. 7 1.1. Background ............................................................................................................................................................................ 7 1.2. Limitations of current simulation models ............................................................................................................. 9 1.2.1. The importance of having a model structure that enables the evaluation of strategies that include policies affecting compliance ........................................................................................................................ 10 1.2.2. The importance of having a simulation model structure that can evaluate multi-level strategies ................................................................................................................................................................................... 13 1.3. Purpose of study ................................................................................................................................................................ 15 1.4. Structure of dissertation ............................................................................................................................................... 16 Chapter 2. Literature Review.................................................................................................................................................... 19 2.1. Compliance ........................................................................................................................................................................... 19 2.2. Multi-level Screening Strategies ............................................................................................................................... 21 2.3. DR & DR screening ........................................................................................................................................................... 23 2.3.1. Diabetes and DR ....................................................................................................................................................... 23 2.3.2. DR screening strategies ........................................................................................................................................ 25 2.4. DR screening modeling ............................................................................................................................................. 28 Chapter 3. Research questions and approach .................................................................................................................. 31 3.1. Research questions .......................................................................................................................................................... 31 3.2. Approach ............................................................................................................................................................................... 32 Chapter 4. Generic modeling framework ........................................................................................................................... 34 4.1. Screening strategy and screening strategy component definition ......................................................... 34 4.2. Definition of cost-benefit .............................................................................................................................................. 37 4.3. Conceptual model framework .................................................................................................................................... 40 4.3.1. Model 1 (general structure of existing simulation models) .............................................................. 42 4.3.2. Model 2 .......................................................................................................................................................................... 45 4.3.3. Model 3 .......................................................................................................................................................................... 50 4.4. Modeling schema .............................................................................................................................................................. 52 4.4.1. High-level modeling schema for models 1, 2, 3 ........................................................................................ 52 4.4.2. Detailed generic modeling schema ................................................................................................................. 54 4.5. Form of compliance model .......................................................................................................................................... 56 5 4.6. Model comparison ............................................................................................................................................................ 58 4.6.1. The extent to which each model can evaluate different mechanisms of influence of an intervention ............................................................................................................................................................................. 59 4.6.2. The estimation of policy impact assessment using each model ...................................................... 59 4.6.3. The resultant most cost-beneficial screening strategy determined by each model ............. 60 Chapter 5. Application of generic model to DR screening example: Model profile for DR screening. 61 5.1. Model purpose .................................................................................................................................................................... 61 5.2. Model overview ................................................................................................................................................................. 62 5.3. Assumption overview ..................................................................................................................................................... 64 5.4. Component overview ...................................................................................................................................................... 67 5.4.1. Cost model ................................................................................................................................................................... 68 5.4.2. Benefit model ............................................................................................................................................................. 69 5.4.3. DR screening module ............................................................................................................................................. 70 5.4.4. Compliance module ................................................................................................................................................ 74 5.5. Parameter overview ........................................................................................................................................................ 79 5.5.1. Input parameters (population cohort, screening strategy) ............................................................... 79 5.5.2. Length of simulation period and number of simulation iterations ............................................... 83 5.5.3. DR progression parameters ............................................................................................................................... 85 5.5.4. Cost-benefit parameters ....................................................................................................................................... 87 5.5.5. Compliance module parameters ...................................................................................................................... 93 5.5.6. Implementation Module Parameters ............................................................................................................ 95 5.5.7. Limitations .................................................................................................................................................................. 95 5.6. Output overview ................................................................................................................................................................ 97 Chapter 6. model verification & validation ....................................................................................................................... 99 6.1. Preliminary model verification ................................................................................................................................. 99 6.2. Models 2 and 3 verification ...................................................................................................................................... 105 6.3. Preliminary model validation .................................................................................................................................. 107 6.3.1. Disease progression module validation .................................................................................................... 107 6.3.2. Compliance module validation ...................................................................................................................... 108 6.3.3. Cost module validation ...................................................................................................................................... 108 Chapter 7. Results ........................................................................................................................................................................ 110 7.1. Results of models 1-3................................................................................................................................................... 110 7.1.1 Results of Model 1 ................................................................................................................................................. 111 6 7.1.2. Results of Model 2 ..................................................................................................................................................... 114 7.1.3. Results of Model 3 ..................................................................................................................................................... 117 7.2. Sensitivity analysis ........................................................................................................................................................ 121 7.3. Model comparison ......................................................................................................................................................... 127 7.3.1. Intrinsic differences between the three models and expected effects on outcomes ......... 127 7.3.2. Effects of model differences in precision of outcome estimates .................................................. 130 7.4. Policy analysis and context-based decision making .................................................................................... 138 7.4.1. Incremental analysis: Choosing from the pool of screening strategies .................................... 138 7.4.2. Comparative analysis: Comparison to status quo ................................................................................ 150 7.4.3. The case for context-based decision making .......................................................................................... 154 Chapter 8. Discussion ................................................................................................................................................................ 156 Chapter 9. Conclusions .............................................................................................................................................................. 162 Acknowledgments ....................................................................................................................................................................... 164 Bibliography ................................................................................................................................................................................... 165 Appendix A: Parameterizing the compliance module - Determinants of compliance ............................. 177 A.1. Framework for analysis ............................................................................................................................................. 177 A.2. Methods .............................................................................................................................................................................. 180 A.2.1. Data source .............................................................................................................................................................. 180 A.2.2. Dependent variables ........................................................................................................................................... 181 A.2.3. Independent variables ....................................................................................................................................... 181 A.2.4. Analysis ...................................................................................................................................................................... 183 A.3. Results ................................................................................................................................................................................. 183 A.4. Discussion.......................................................................................................................................................................... 186 Appendix B: Sensitivity analysis results……………………………………………………………………………………188 Appendix C: Sample codes……………………………………………………………………………………………………….203 C.1. Code for DR progression submodule ................................................................................................................... 203 C.2. Code for the main function for one iteration of the DR screening simulation…………………….203 C.3. Code for the compliance module…………………………………………………………………………………….214 7 CHAPTER 1. INTRODUCTION 1.1. BACKGROUND More than one in four people in the United States reported at least some difficulty getting medical care for themselves or their family members, and there is a dramatic increase in reported difficulties on the ability to see medical specialists, including ophthalmologists for DR screening. In the Commonwealth Fund 1998 International Health Policy Survey, more than half of US respondents citing access difficulties name insufficient money or insurance to pay for care as a source of access problems (1). In 2010, the Patient Protection and Affordable Care Act (PPACA) was signed into the U.S. federal law aimed primarily at decreasing the number of uninsured Americans and reducing the overall costs of health care. One of its provisions was to eliminate co-payments, co-insurance, and deductibles for select health care insurance benefits considered to be part of an "essential benefits package" of preventive care (2). Thus, with the advent of the Act, the financial barrier to screening to the majority of the underserved population will diminish (3) (4) and a likely result is that utilization of preventive services will rise. This expected increase in demand for preventive services would put more pressure on the already overstretched preventive services system serving the underserved population and present a challenge to public health officials to design and plan a screening strategy that will be cost- beneficial (i.e. a screening strategy in which the benefits outweigh the costs) with the surge in demand. 8 As the Act aims to reduce the overall costs of health care, cost-benefit of a screening strategy would be especially important as total screening costs would rise in response to the increase in screening rates. The importance of cost-benefit within the health reform framework is underscored by the experience of the state of Massachusetts, where early implementation of the universal health coverage led to burgeoning health care costs as increased demand was not balanced by cost control. Similarly, to ensure the continued viability of this complementary screening strategy, we need to make sure we have a screening strategy that is cost-beneficial in face of increased demand. Cost-benefit of a screening strategy is influenced by how well the affected population is compliant to this strategy. Poor compliance will lead to suboptimal cost-benefit for the population. However, screening strategies often vary in compliance. To achieve maximum cost benefit, especially for screenings that often suffer low compliance, such as Diabetic Retinopathy screening, policymakers need to consider multi-level (e.g. both patient- and provider-level) interventions to improve screening compliance. To meet the need of determining cost-beneficial population-based screening strategies, many different screening strategies need to be considered. Yet, it is costly, impractical, and time- consuming to do clinical trials to test all the different screening strategies. Simulation models provide a more cost- and time-effective way to help determine cost-beneficial screening strategies. In particular, to achieve maximum cost benefit for populations with low compliance, policymakers need to consider screening strategies that include multi-level (e.g. both patient- and provider-level) interventions to improve screening compliance. Although simulation models have been used extensively to address cost-benefit of screening, there are shortcomings to existing simulation models that disallow their use in evaluating such screening strategies. 9 1.2. LIMITATIONS OF CURRENT SIMULATION MODELS Current simulation models are lacking in that: first, they do not have a structure that enables the evaluation of strategies that include policies affecting compliance, even though screening compliance is often low (Eye exam for Diabetic Retinopathy screening, for instance, has a compliance rate of only 50-60% in the underserved population (5)). Thus, strategies that include policies targeting the improvement of compliance may be necessary to achieve maximum cost- benefit; second, current simulation models do not have a structure that can evaluate multi-level strategies (e.g. those targeting patient, providers, and clinics; or students, clinics, and community) even though they are more likely to have sustained or powerful effect than those targeting the individual-level only (6) (7). These limitations prevent the use of simulation models to evaluate screening strategies involving screening behavioral modification at multiple levels. This research will address the first limitation by including compliance as mediating variable in the model, and the second limitation by addressing the hierarchical nature of certain interventions to take effect on health outcome, e.g. the effect of provider-level intervention on patient behavior is manifested through the provider’s compliance with the intervention and through the use of a nesting structure. The following paragraphs will address the importance of addressing the two limitations above. The problem of addressing these two limitations has a multidisciplinary nature, necessitating bringing together different fields (health behavior, ISE, economics) to address it. 10 1.2.1. THE IMPORTANCE OF HAVING A MODEL STRUCTURE THAT ENABLES THE EVALUATION OF STRATEGIES THAT INCLUDE POLICIES AFFECTING COMPLIANCE Compliance or adherence is patients’ realization of clinical advice to use health services appropriately to achieve better health outcomes. In light of the larger role of personal responsibility in preventive services, understanding compliance is important because no matter how effective the intervention is, it would not achieve its maximum benefit in the society if patients do not comply; Treatment effectiveness is compromised by poor compliance, turning this into an important quality of life and health economics issue in population health. In developed countries, only half of chronic disease patients follow treatment recommendations, with even lower figures for preventative therapies (8), making compliance an especially important issue to address in the context of screening. Interventions aimed at improving adherence would yield positive returns through primary prevention of risk factors and secondary prevention of adverse health outcomes (8). Thus, there is a need to understand barriers of compliance, what kind of intervention patients will respond to, and including interventions aimed at improving adherence in screening strategies. Preferences and people’s value systems have always been an important consideration in Industrial and Systems Engineering, such as in the field of decision analysis. Understanding patients’ value system as well as factors and barriers that influence their decisions is a basic requirement as health systems engineers must apply scientific methods in a value-laden setting. This is especially important for screening, in which patient compliance plays a large role. Screening for asymptomatic diseases, such as DR (Diabetic Retinopathy) screening and various forms of cancer, has a higher patient burden than other aspects of medical care for symptomatic diseases, since patients need to seek care before symptoms develop. In the case of DR for instance, previous studies exploring the 11 reasons why patients never had a dilated eye examination found that many diabetic patients believed they had no eye problem or had been told that such examinations were not necessary (9) (10). Without symptoms as a driving factor to seek care, patients have less urgency and a higher decision threshold for seeking preventive care compared to seeking care for symptomatic diseases. In this regard, barriers to screening become very salient factors to take into account as high barriers can make the already high screening decision threshold insurmountable. Thus, it is important to design the screening programs to reduce these barriers and lower this decision threshold for preventive care, so as to maximize the reach of the screening program in the population. Since compliance is an important modifier of health system effectiveness, to accurately measure population health outcomes, compliance rates need to be used to inform planning and project evaluation, along with resource utilization indicators and efficacy of interventions. In fact, significant financial benefits and increased effectiveness attributed to low-cost interventions for improving adherence have been consistently recorded in studies. “Increasing the effectiveness of adherence interventions may have a far greater impact on the health of the population than any improvement in specific medical treatments (11).” A system that addresses determinants of compliance is necessary to realize the potential of medical advances; in turn, a simulation model that is able to evaluate such systems in the context of population health is necessary. To do so, a simulation model needs to address compliance not as an input factor as is typically done in current simulation models, but instead as a mediating factor, with determinants of compliance as the input factors affecting compliance, and impact on population health as outcome. Having a model that can evaluate a preventative health care system that addresses determinants of compliance will allow policymakers to identify strategies that have a more comprehensive reach in the population and quantify their impact on population health. This means that policymakers can 12 plan better preventative strategies involving behavioral interventions to support screening services that are likely to have a greater impact than those that focus on just the screening services alone. In addition, such models are required for better policy assessment as even policies not designed to influence compliance can impact compliance. For instance, policymakers might contemplate recommending a more invasive yet more accurate screening procedure, but this is likely to reduce compliance levels, and not taking into account the effect of such interventions on compliance will overestimate their impact. Thus, by considering compliance as a mediating factor and taking into account the effect of considered interventions on compliance, this model also allows for more accurate impact evaluation of screening interventions in general. In summary, screening compliance is a significant factor that affects outcome, and the outcome of an individual with a preventable disease depends on him/her getting appropriate screening and treatment. Yet, screening compliance in the population is often low. Current simulation models used to evaluate the cost-benefit of population-based screening strategies do not consider policies that address compliance issues and treat compliance as a fixed input variable so they could not evaluate the cost-benefit of screening strategies targeting compliance. This means that policymakers might miss out on or inadequately quantify screening strategies that include policies targeting compliance. Additionally, even interventions which are not specifically designed to improve adherence often affect adherence, and evaluating such interventions without considering their effect on compliance will under- or over-estimate their impacts, and thus there is a need for simulation models with a structure that enables the evaluations of strategies that include policies that affect adherence 13 1.2.2. THE IMPORTANCE OF HAVING A SIMULATION MODEL STRUCTURE THAT CAN EVALUATE MULTI-LEVEL STRATEGIES In terms of system changes to improve compliance, the most promising screening strategies are those that combine several components, including patient education, social support, telephone follow-up, and others, with multi-component strategies showing effective in alleviating poor compliance. In addition, an expert panel of compliance recognizes the importance of a multilevel approach targeting patient, provider, and organization in the effort to improve compliance (7). Thus, a good strategy to improve compliance should ideally be a multi-level, multi-component strategy that addresses these multiple levels of contextual factors influencing compliance. These multi-component compliance strategies would have different impacts on compliance and cost of the program depending on the combination of components used in the strategy. The impact on compliance depends not only on the individual components in the compliance strategy, but also on the potential synergistic or antagonistic effects from certain combinations of those components. The cost of the program is affected not only by the costs of the implementation of the components themselves, but also by the changes in screening compliance which in effect would affect effectiveness and thus program costs. Having certain combinations of components to increase compliance may increase screening program costs markedly, but the rise in compliance may raise the effectiveness to a level that lowers the cost of blindness in the society. The cost trade-off here is also dependent on other components of the screening strategy which are not specifically designed to raise compliance, such as the screening modality and the frequency of screening, as their cost- effectiveness varies according to utilization, which in turn depends on screening compliance levels. Aside from considering multi-component strategies, these multiple components also should be at multiple levels (e.g. targeting patients, providers, and clinics; or students, schools, and community) 14 as multi-level strategies are likely to be most effective in changing behavior, such as compliance with screening. According to socio-ecological model of health behavior, there are multiple levels of influence on patient behavior, and health behavior change programs that target these multiple levels, for instance by having the individual-level health behavior change programs reinforced by providers and clinic policies, are more likely to have powerful or sustained population-wide effect (6). Following the socio-ecological model of health behavior, in the context of screening strategies it is important to consider multi-level strategies because health care system changes can affect compliance, as the behavior of healthcare professionals and the delivery of medical care contribute to patient compliance with any health behavior (8). Policies of healthcare organizations, including hospitals, HMOs (healthcare maintenance organizations), and physicians’ offices, can influence the extent to which preventive services are provided. Meanwhile, healthcare providers, including physicians, nurses, nutritionists, health educators, and psychologists involved in diabetic patient care play a role in enhancing compliance by interpreting recommendations, educating and motivating patients, monitoring patients’ screening behavior, and providing feedback. However, there is the challenge of evaluating such multi-level strategies. One particular challenge is in developing and applying appropriate methods and research designs to deconstruct which components of the multi-level health care delivery result in the greatest improvement in care (12). A simulation model with a structure to evaluate multi-level strategies would be helpful in quantifying the effects of such strategies would serve as a very useful tool for policy planning and evaluation. 15 1.3. PURPOSE OF STUDY In summary, for the purpose of determining cost-beneficial screening strategies, there needs to be a simulation model that addresses limitations of current research and models the impact of behavioral modification on screening compliance at various levels (patient, provider, clinic) through certain interventions. This will be addressed by developing a generic conceptual model structure that can address this problem for screening strategies in general and then illustrating the conceptual model with a Diabetic Retinopathy screening simulation model. By doing so, this research can have broader impacts in the area of healthcare systems engineering and Diabetic Retinopathy. For healthcare systems engineering, the research would propose a generic hierarchical simulation model that can synthesize information about screening strategies that include components that intend to improve screening compliance at the patient and provider level. The model is used to predict expected costs and benefits of long-term screening strategy use within a given population. Such analysis provides valuable information for the decision maker when a screening strategy involving components to improve compliance is being considered. These analyses make use of the best currently available information to guide initial policy design decisions or to aid preliminary policy evaluation, thereby bridging the gap until further information becomes available. When further information becomes available, the model parameters can be updated to reflect the new information and the analyses repeated to get better estimates of the long term impact of the screening strategy. 16 For Diabetic Retinopathy, the model that is constructed to illustrate the generic model can itself be used for several applications such as to evaluate DR screening strategies intended to improve DR screening compliance or to determine multi-level policies that can support the implementation of new DR screening strategies, such as tele-screening / remote screening. An application of the model in the immediate future is to guide policy in the implementation of DR tele-screening in LA County. One potential application is in determining what kinds of multi-level interventions (e.g. appointment reminder at the provider-level combined with changes in insurance coverage at the patient-level) would be most effective in improving compliance and would work best in concert with the DR tele-screening to ensure successful implementation and maximum cost benefit. Another application can exploit the capability of the model to incorporate economy of scale in evaluation of strategies to determine the clinic size threshold at which tele-screening will be cost- beneficial, given the high initial set-up cost of camera for tele-screening and thus better economy of scale for larger clinics. In essence, addressing the impact of behavioral modification on screening at multiple levels will lead to improved policy evaluation and design, and enable policymakers to make a better business case for screening strategies that involve improving compliance. 1.4. STRUCTURE OF DISSERTATION This dissertation begins with introducing the motivation for a simulation model that has a structure to evaluate the impact of behavioral modification on screening at multiple levels, the limitations of current simulation models that render them unable to do so, and purpose of the study in chapter 1. Chapter 2 reviews the literature on the main themes of this dissertation and their related fields to 17 give an overview of current research, limitations, and theories to support the development of the models in the dissertation. Chapter 2 will review literature on compliance and multi-level strategies, and then DR and DR screening, and finally simulation models available to address DR screening. Chapter 3 then delineates the research question and approach in broad terms. Chapters 4 and 5 delve into the approach in more detail. Chapter 4 proposes a general conceptual modeling framework for screening strategies that can address the impact of behavioral modification on screening at multiple levels, explains the different pathways / mechanisms of influence that different screening strategy components can have on the outcome, and proposes a generic modeling schema to elucidate the construction of simulation models based on the conceptual modeling framework. Chapter 5 illustrates the general conceptual model by implementing it in a simulation model for Diabetic Retinopathy screening, the model profile of which is given in chapter 5. Chapter 6 presents preliminary verification and validation for this research, which includes verification and validation from the implementation of a basic Diabetic Retinopathy screening model that will be built upon to accommodate the structure necessary to evaluate the impact of behavioral modification on screening at multiple levels, verification of model outputs for the three models when they are fed with the same input, as well as validation for some model parameters used. Chapter 7 presents results of the illustrative example of DR screening for models 1 through 3, presents the sensitivity analysis results, compares the results from the three models, and performs policy analysis based on the model results. Finally, chapters 8 and 9 presents the discussion and conclusion from this research. 18 Supplementary materials, such as the methods and results of the analysis done to populate model parameters such as screening compliance, full results of the sensitivity analysis, and sample codes are given in the appendix. 19 CHAPTER 2. LITERATURE REVIEW 2.1. COMPLIANCE Compliance has been defined in the literature as “the consistency and accuracy with which someone follows the regimen prescribed by a physician or other health professional (13)”, or “a measure of the extent to which patients follow a prescribed treatment plan – e.g. take drugs, undergo a medical or surgical procedure, exercise or quit smoking (14)”. Patient compliance has also been equated with the terms adherence and concordance, with slight nuances and differences accorded to each term; the term concordance for instance has the additional layer of meaning involving “the process by which a patient and clinicians make decisions together about treatment (15)”. For the purpose of this dissertation, compliance is defined more generally as “how well a patient’s behavior follow medical advice (16)”, specifically with screening recommendations. For Diabetic Retinopathy, this means following the eye examination schedule for patients with Diabetes Mellitus as recommended by the American Academy of Ophthalmology; for Type II diabetics, this means getting yearly eye examinations since time of diagnosis (17). In the context of organizations, it is defined as “The adherence of a particular organization to statutes or mandates from regulatory agencies – governing agencies or bodies – or to an official mandate or obligatory standard (14)”. For this dissertation, this definition is expanded to the context of providers or other parties administering a health intervention (e.g. schools, local organizations) to cover their adherence to an official mandate or obligatory standard, particularly relating to the health intervention. For instance, a component of screening strategy to improve compliance may be public reporting of coordination of screening services at the PCP (Primary Care Physician) or mandating the availability of patient navigators for screening in health clinics. PCPs 20 and clinics may not comply with such mandates or standards, and a good planning or evaluation of the screening strategy should take into account such factors along with patient compliance. The importance of considering compliance has been covered in the introduction chapter above. In essence, poor adherence is a significant worldwide problem, and its ramifications will grow as the rate of chronic diseases grows, with the poor being disproportionately affected. The consequences include poor health outcomes, compromised patients’ safety, and increased health care costs. Interventions to improve adherence could have return of investment of significant magnitude through prevention of risk factors and adverse health effects (8). Some theoretical foundations for predicting and improving screening compliance include the health belief model, theory of reasoned action, transtheoretical model, and prospect theory, although no one theory is sufficient (18). Heuristic frameworks integrating some of the constructs from the different theoretical models are also present in the literature, an example of which is Andersen’s framework of access to medical care (19). Studies on factors influencing compliance usually take the form of surveys to determine factors influencing compliance, which are often analyzed using regression. A major hypothesis attributes the low compliance rates largely to access factors, such as health insurance coverage and having a usual source of care (20) (21) (22) (23) (24). Access can be defined as “those dimensions that describe the entry of a population group into the health care delivery system” (25). Lack of access encompasses lack of insurance coverage, scarcity of providers, linguistic barriers, and health literacy, among others (5) (26) (27) (28). Various other factors influence screening compliance rates, such as demographics and socio- economic factors, disease severity, presence of co-morbidities acculturation, and self-efficacy. How these factors influence compliance rates form the basis of the analysis framework to characterize 21 this relationship that is done to parameterize the simulation model built for the dissertation. Section 1 in the appendix delves into these in more details. Interventions to improve compliance try to remove some of the barriers to screening, such as by having mobile screening outposts and moving specialist screening to the PCP to improve access. Other interventions that can work include appointment reminders to remind patients of their screening appointment (29), contracting with providers (30), providing patient navigators especially for less acculturated patients, and so on. Such behavioral interventions to improve screening compliance will be considered as part of the screening strategies evaluated with the simulation model proposed here, alongside the more traditionally studied medical interventions. 2.2. MULTI-LEVEL SCREENING STRATEGIES As has been introduced in the introduction above, according to socio-ecological model of health behavior, there are multiple levels of influence on patient behavior, and health behavior change programs that target these multiple levels, for instance by having the individual-level health behavior change programs reinforced by providers and clinic policies, are more likely to have powerful or sustained population-wide effect (6). Ecological or social-ecological models are characterized by multiple levels of influence on patient behavior and emphasize environmental and policy influences (31) (32). Although such models have been accepted in smoking cessation and tobacco control (33) and broad public health areas (34), health behavior change programs tend to target the individual patient level only (35). This is true also for efforts to improve compliance with screening rates. 22 This multi-level intervention challenge and the need for more studies on it has been highlighted in several papers in the literature; a recent publication of the Journal of the National Cancer Institute (JNCI) was a special issue on multi-level interventions in the cancer care continuum, from prevention to long-term survival and end-of-life care (36); the American Heart Association (AHA) calls the need for multi-level intervention to improve compliance “the multi-level compliance challenge (7)”. For screening in particular, it has been highlighted that maximum screening compliance can be achieved when strategies are tailored to the steps and interfaces in the screening process “that are most critical for their organizations, the providers who work within them, and the patients they serve (37)”, shining a light on the need for multi-level screening strategies that target patients, providers, and health-care organizations. As the structure and practice of health care delivery changes, alternative models of multilevel interventions and quantification of their effects are necessary (38). In particular, Murray et al drew attention to the need to develop and apply “appropriate methods and research designs that can be used to deconstruct which components of multilayered and multifaceted health-care delivery result in the greatest improvements in care (39)”. A simulation model to evaluate multi-level screening strategies as proposed in this dissertation is a step toward evaluating alternative models of multilevel interventions under consideration, quantifying of their effects, and estimating which components that result in the greatest improvement in care to inform and aid policymakers in designing a cost-beneficial screening strategy with an optimal reach in the population. One aspect of the simulation model structure proposed in this dissertation is borrowed from multilevel models in statistics, also called hierarchical linear model, nested model, mixed model, or random effects model. Multilevel models are appropriate for analyzing data where the data is 23 organized at more than one level in a nested data (40), and the units of analysis are typically individuals (at a lower level) nested within higher level aggregate units. This organization of data at multiple levels in a nested structure is also used in the simulation model proposed in this dissertation, such that individual patients are nested within the providers serving the patients within the simulation model. This way, the effect of provider compliance of the screening strategy components on the patients he served can be simulated, and the overall effect of the screening strategy estimated. 2.3. DR & DR SCREENING 2.3.1. DIABETES AND DR Diabetes is one of the most significant chronic illnesses affecting the US population in terms of prevalence, related health care expenditures, and associated morbidity and mortality (41). According to the 2005-2008 National Health and Nutrition Examination Survey (NHANES), 11.3% of the population aged 20 or older has diagnosed diabetes (42)Among the general US population, approximately 8%, or about 23.6 million people, has diabetes (43). Diabetes increases the risk of developing a number of chronic complications. In particular, diabetic retinopathy (DR), a chronic complication of diabetes that results in retinal damage, is the leading cause of blindness among adults of 20-74 years of age, a subgroup that includes a large swath of working age Americans (44). There is a high prevalence of diabetic retinopathy among patients with diabetes: 20 years after diabetes diagnosis, >90% of patients with type I diabetes and >60% of patients with type 2 diabetes will have some degree of retinopathy. Among American adults 40 24 years and older with DM, the estimated crude prevalence of DR is about 40%, and the estimated crude prevalence for vision-threatening DR is about 8% (45). In the US general population, the estimated prevalence rates for DR and vision-threatening DR are 3.4% and 0.75% respectively (45). For type I diabetes patients, DR onset occurs mostly during the subsequent 2 decades after first diagnosis of diabetes or before puberty, while for type 2 diabetes patients, up to 21% is found to have DR at the time of diabetes diagnosis, and most would then develop DR over subsequent decades. An important factor contributing to annual DR screening non-compliance is the asymptomatic characteristic of severe retinopathy. Previous studies exploring the reasons why patients never had a dilated eye examination found that many diabetic patients believed they had no eye problem or had been told that such examinations were not necessary (21) (20). These data also highlight the need to focus on complications associated with diabetes and on DR screening programs that also include patient education and better physician-patient communication. Not all disease screenings are beneficial, and screening for a disease must meet certain criteria to be medically and financially acceptable. A case can be made for DR screening as DR meets the Wilson-Jungner criteria for appraising the validity of a screening program: it has important adverse effects, there is a detectable early stage and the method of screening is acceptable to the population, the treatment is available and effective – especially at an earlier stage, and there is an agreed policy on whom to treat. Indeed, DR screening can be very effective in identifying patients at high risk for DR for treatment, and treatment for DR is highly effective in halting the progress of the disease and preventing blindness (46) (5). Early detection of DR can be followed with treatments that have been proven to reduce the risk severe vision loss by >90% (41). Thus, DR screening could effectively reduce the incidence of blindness in the community. 25 However, compliance with eye screening guidelines – which is outlined by the NEI (National Eye Institute) to include a comprehensive dilated eye exam at least once a year for people with diabetes - is very low, and this is especially so for the underserved community (26) (46) (5). The DR screening guidelines specify that diabetes patients should have yearly eye examinations subsequent to their initial DR screening, or more frequently if retinopathy is progressing (20). On average, less than half of diabetic patients in the US do not meet the ADA (American Diabetic Association) guidelines for eye screening, and 60% of patients requiring vision-saving surgery do not receive treatment (47). Some barriers to eye screening compliance include socioeconomic factors, insufficient referrals, and lack of access, which encompasses lack of insurance coverage, scarcity of providers, linguistic barriers, and health literacy, among others (21) (41) (48). Since DR is the most prevalent yet preventable disease that can lead to blindness, and the outcome of an individual having DR is highly dependent on him getting appropriate screening and treatment, there is an imperative need to implement appropriate programs to detect and manage DR which would also address these barriers while remaining cost-effective and accessible. 2.3.2. DR SCREENING STRATEGIES There are a number of components making up a DR screening strategy; these components can be broadly categorized into biology / clinical /medical -based and behavioral-based components. Medical-based components include the choice of screening modality, frequency of screening, whether screening is offered at the primary care or the specialist (ophthalmologist) level, disease management to stave off progression of disease, and so on. Behavioral-based components include components to change screening behavior (i.e. improve compliance), such as education on the 26 importance of DR screening, DR awareness programs at school and work, change in insurance coverage and co-pay for DR screening (such as that mandated by the Affordable Care Act), changes in the physical environment to enable easier DR screening, and so on. Most of the studies in the literature focus on medical-based components. There have been studies on improving screening rates for various cancers, but not DR; also, studies typically only focus on either medical-based or behavioral-based screening strategies, not on the synergy between combined screening strategies. There are several components to be considered in the clinical-based DR screening strategy, such as frequency of screening, screening modality, the use of dilation for screening, the setting of the screening, and appointment reminders for DR screening. Some of these are dependent on other components; for instance, the screening modality ophthalmoscopy is usually performed with dilation, and is performed in a specialist’s office as it is typically performed by an ophthalmologist. Screening modalities for diabetic retinopathy include ophthalmoscopy (direct and indirect), stereoscopic color film fundus photography, mydriatic (with pupil dilation) or nonmydriatic (without pupil dilation) digital color photography, and monochromatic photography (17). The gold standard for DR screening is a 30-degree stereoscopic photography of seven standard fields on color film, however, this method has a long turnaround time, is labor-intensive, and requires expensive equipment and trained retinal photographers and readers, and thus not ideal for widespread implementation (49). The traditional modality for DR screening is indirect ophthalmoscopy, where ophthalmologists dilate the pupil and examine the retina for DR (17). Where access to care is insufficient, however, this modality might need to be supplemented, especially as supply of eye care providers is outpaced by the increasing rate of patients with diabetes, and some communities have poor access to ophthalmic care. 27 A promising modality for supplementing traditional ophthalmoscopy in these settings is DRTS (Diabetic Retinopathy Tele Screening), which employs digital retinal imaging tools that can be operated by trained non-specialists, and forwarding of the resulting images to a remote reading center to be graded by an ophthalmologist. Two-field mydriatic and non-mydriatic digital photography seem to perform favorably compared to ophthalmoscopy (50). The digital images may be interpreted on-site by trained readers or at a remote reading center for interpretation and grading (“store and forward”). As DRTS allows screening to be done by non-specialists, removing limits imposed by the availability of eye specialists and improving access, it is thus a promising screening solution for use in the PCP. DRTS is a particularly promising method to be used in the primary care settings (including family practice, internal medicine, and endocrinology offices), as it allows screening to be done by non- specialists, thereby helping to remove the structural barriers of care which limits screening based on the availability of eye specialists, and improving access. DRTS however has expensive initial outlay, and thus is especially useful when compliance and consequently utilization is higher. Retinal imaging can be mydriatic (performed with dilation) or nonmydriatic (performed without dilating the pupil). From a patient’s perspective, dilation may be uncomfortable and time- consuming, as it may preclude the patient from driving and doing other activities until the effects of dilation has worn off. From a provider’s perspective, nonmydriatic cameras may be simpler to operate as it does not require a trained retinal photographer. However, the costs between mydriatic and nonmydriatic cameras differ, as are the sensitivity and specificity of the modalities and thus their effectiveness. Current ADA guidelines recommend an annual screening frequency for DR screening (49). However, the use of new modalities such as DRTS may require re-evaluation of screening frequency, which was developed for ophthalmoscopy. Also, depending on the composition of the 28 population at risk for DR, other screening frequencies may be more cost-effective even for ophthalmoscopy and is worth examining. Other possible clinical components include appointment reminders, more referrals from the primary care setting for eye screening, and more DR screening education at the clinics. The choice of one component over the other, or combinations of some components over other combinations, has several important implications: cost, benefit, and compliance (which ultimately also affect the overall cost-benefit). These are the implications that need to be considered when choosing these different components for the DR screening program. In addition, some combinations of components may yield a more robust program design than other combinations, in that it will still be cost-effective over a range of likely compliance rates. Current DR screening solutions are suboptimal, but are especially so given the changing landscape due to the health care reform – consequently, there is a need better screening strategy design for better DR screening quality. A better evaluation of cost-beneficial DR screening strategies requires simulation model that addresses current limitations, such as one that is proposed by this dissertation. 2.4. DR SCREENING MODELING Modeling for DR screening usually consists of modeling the natural disease (DR) progression and the screening strategy, and then combining the two models. Modeling of natural disease progression of DR involves characterizing the onset and progression of DR in a model. Diabetic retinopathy is a progressive disease, which manifests first as nonproliferative DR, in which the patient is asymptomatic, although it is characterized by retinal 29 microaneurysms, hemorrhages, and exudates. It then progresses to proliferative retinopathy, characterized by growth of abnormal retinal blood vessels and fibrous tissue, which may be accompanied by vitreous hemorrhages or macular traction, which may lead to severe visual loss (49). Modeling of DR is fairly straightforward given its predictable and orderly progression, from mild nonproliferative abnormalities to moderate and severe nonproliferative diabetic retinopathy (NPDR) to proliferative diabetic retinopathy (PDR) (49). Several epidemiological studies describe the onset and progression of DR, of which the Wisconsin Epidemiologic Study of Diabetic Retinopathy (WESDR) is considered a representative one. WESDR follows a cohort of 1210 Type I diabetes patients and 1780 Type II diabetes patients for 4 years with repeated fundus photograph examinations. The DR natural progression model in this dissertation is based on the model in the seminal literature of DR modeling by Dasbach (46), which is based on the results of WESDR. The use of modeling to evaluate the societal effect of DR screening strategies at the patient (individual) level includes the use of Markov model to evaluate different screening strategies for DR is illustrated in Vijan’s and Dasbach’s paper (51) (46), among others such as DES (52) and systems modeling (53). The use of provider-level simulations can also be found in the literature. For instance, Whited uses decision trees to evaluate the effect of annual screening on the yearly cost incurred by three different federal healthcare agencies (54). Meanwhile, M. Porta et al uses statistical analysis on data gathered by direct verification of the procedures and retrospective surveys to compare the cost- benefit of different DR screening strategies in three different clinical settings (55). However, in these papers, the models are all used to evaluate medical / clinical-based interventions / screening strategies. The models treat compliance as a fixed input factor and therefore cannot evaluate behavioral-based screening strategy components to improve patient compliance, nor can 30 they evaluate screening strategies that target multiple levels, such as both patient- and provider- level. Also, many of the papers typically focus on one specific component of the DR screening strategy instead of a multi-component one; Whited’s paper focuses on the cost-effectiveness of different screening modalities, while Vijan’s paper focuses on the different screening intervals. Dasbach’s paper does examine the combination of different modalities and screening intervals, but he does not consider provider-level components of the screening strategy either, such as appointment reminders. A number of studies have examined the feasibility and cost-effectiveness of various DRTS solutions for DR screening, although most of them focus on using DRTS in remote areas (56) (57), in places where ophthalmologists are generally not available, such as prisons (58), or in other specific populations, such as the Veterans population (50) (59), and other federal healthcare agencies (54). Meanwhile, the evaluation of DRTS in the urban setting, particularly one that has a large percentage of underserved population, has not been explored much, and this is a body of study that the model demonstrated here can have potential applications for. 31 CHAPTER 3. RESEARCH QUESTIONS AND APPROACH 3.1. RESEARCH QUESTIONS For the purpose of determining cost-beneficial screening strategies, there needs to be a simulation model that addresses limitations of current research and models the impact of behavioral modification on screening compliance at various levels (patient, provider, clinic) through certain interventions. The research questions considered in this dissertation are: 1. How to represent the following in a simulation modeling framework: • Behavioral modification involving compliance of screening • Multi-level nature of behavioral modification involving compliance of screening 2. Using DR screening as an example, how to build a simulation model to realize the above framework to evaluate strategies that include components that affect compliance at multiple levels? 3. Comparing the simulation model above to those without considering compliance as a mediating variable and the hierarchical structure of interventions, what are the advantages in terms of policy impact assessment? 32 3.2. APPROACH a. Approach for research question 1: To develop a simulation modeling framework that incorporates behavioral modification involving compliance of screening at multiple levels, 3 conceptual models will be developed. First, model 1 that represents the general structure of existing simulation models for screening strategies will be developed by abstracting the structures of existing simulation models for screening strategies described in the literature. Second, model 2 will be developed to incorporate evaluation of behavioral modification of screening by modifying model 1 to change the role of compliance from input to mediating variable. Third, model 3 will be developed to incorporate the multi-level nature of behavioral modification involving compliance of screening by incorporating the hierarchical nature of interventions into model 2. b. Approach for research question 2: The 3 conceptual models will then be implemented in the context of Diabetic Retinopathy screening to illustrate the models. This implementation will take the form of a microsimulation model for DR screening strategy that can evaluate strategies that include components to improve adherence at several levels. This constructed DR screening simulation model will then be used to evaluate DR screening strategies to evaluate adherence at multiple levels to demonstrate its value. c. Approach for research question 3: The three constructed DR screening simulation models will then be used to evaluate several screening strategies to test the 3 models and evaluate their differences in the following dimensions: 33 • The extent to which each model can evaluate different mechanisms of influence of an intervention • The estimation of policy impact assessment using each model • The resultant most cost-beneficial screening strategy determined by each model d. Summarizing the approaches above, the study design is depicted in the diagram below: 1. Develop model 1 (existing model), a conceptual model depicting the general structure of existing simulation models. 2. Develop model 2 (existing model + compliance as mediating variable) by modifying model 1 to change the role of compliance from input to mediating variable. 3. Develop model 3 (existing model + compliance as mediating variable + hierarchical structure) by incorporating the hierarchical nature of interventions in model 2. 4. Construct three DR screening simulation models as demonstration cases to illustrate conceptual models 1, 2, and 3. 5. Evaluate DR screening strategies involving components that affect compliance at multiple levels using the constructed DR screening simulation model to demonstrate the value of the conceptual model 3. 6. Model comparison: Use the 3 DR screening simulations models for policy impact assessment and compare the results. 34 CHAPTER 4. GENERIC MODELING FRAMEWORK This chapter begins with the definition of terms used in the generic modeling framework: screening strategy and cost-benefit. Then, this chapter proposes a general conceptual modeling framework for screening strategies that can address the impact of behavioral modification on screening at multiple levels, explains the different pathways / mechanisms of influence that different screening strategy components can have on the outcome, and proposes a generic modeling schema to elucidate the construction of simulation models based on the conceptual modeling framework. 4.1. SCREENING STRATEGY AND SCREENING STRATEGY COMPONENT DEFINITION Screening strategy components are components of the screening strategy that can be changed to change the outcome of interest (in this case, cost benefit). An example of a screening strategy component is mandating the usage of a different screening modality, or recommending a different screening frequency or starting age for regular screening for a certain disease. Component alternatives are forms of the screening strategy components which are under consideration for implementation. Status quo is usually considered as an alternative. Screening strategy A Screening strategy component 1 Alternative 1 Screening strategy component 2 Alternative 2 Screening strategy component 3 Alternative 2 35 A screening strategy is composed of a combination of multiple screening strategy components. As an example, in the case of DR, a possible DR screening strategy consists of recommending biannual eye screening using mydriatic camera, another possible DR screening strategy consists of recommending annual eye screening using opthalmoscope. The following table gives examples of possible DR screening strategy components and alternatives for each component. Due to the limitations of current simulation models, only the first two of the following screening strategy components can be evaluated, as the last four components in the table affect the resulting cost-benefit through patient compliance and/or provider implementation of the policy. This dissertation will propose and build a simulation model that can evaluate all these components. DR screening strategy A Screening modality Mydriatic camera Screening frequency Biannual DR screening strategy B Screening modality Ophthalmosc ope Screening frequency Annual 36 Screening strategy components Level Alternatives Status quo Alternative 1 Alternative 2 Screening modality Patient Current (Opthalmoscope) Non-mydriatic camera Mydriatic camera Screening frequency Patient Current (Annual) Biannual Biennial Insurance Patient Current ACA Premium support (voucher for eye exam) Patient education Patient Current Targeted mailing Automated DR education calls Provision of patient navigator Provider Current (No patient navigator) Provision of patient navigator for large clinics only Provision of patient navigator for all clinics Appointment reminder Provider Current Automatic call reminder Nurse- administered call appointment reminder 37 4.2. DEFINITION OF COST-BENEFIT The outcomes of interest in this modeling framework are: • costs (i.e. costs of screening and treatment) • benefits (monetary values of desirable consequences of economic policies and decisions; in this case, monetary value of better health outcomes in terms of reduced morbidity and mortality rate in the society). The modeling framework characterizes the relationship between the screening program and societal cost and benefit, with a highlight on the role of compliance in the health care delivery process. Cost and benefits are viewed from the societal perspective in this modeling framework. Together, costs and benefits reflect the resulting changes in individual and social welfare as a consequence of implementing alternative programs. Both direct (i.e. values of desirable health and non-health outcomes directly related to the implementation of proposed interventions) and indirect (i.e. costs and savings resulting from the interventions that are not directly related to them, for instance productivity loss) costs and benefits are included. Intangible benefits (e.g. reductions in health risk, pain, and suffering) which cannot be estimated from market data are excluded. The benefit measure captures how the morbidity and mortality outcome is altered by the screening program; for DR screening, this is measured by the savings associated with altered morbidity outcome: sight years saved or years of averted blindness, i.e. the total number of sight-years beyond what is expected under the natural disease progression (without screening) that is achieved by the cohort; for cancer screening, this is measured by the savings associated with altered morbidity and mortality (i.e. the development of cancer and deaths due to cancer). The benefit measure can also 38 incorporate not just the primary morbidity but also secondary morbidities (co-morbidities); for instance, for DR, although the primary morbidity associated with it is blindness, secondary morbidity associated with DR can include depression and injury associated with vision loss (60). The cost outputs of the model include costs of provision (i.e. costs of screening and treatment) and cost associated with health outcomes (e.g. blindness in the case of DR screening). The cost of screening includes costs associated with screening delivery (space, equipment, personnel, etc) and patient costs (transportation costs, transport time, opportunity costs lost from screening). The cost associated with blindness includes medical and rehabilitation costs as well as productivity loss. The decision of which screening program to choose needs to consider the trade-offs between these two outcomes. To consider this trade-off, cost benefit analysis will be performed to find the net benefit from the screening program to the society using the two cost outcomes from the model. The net benefit will calculate the savings to the society if a screening program is in place compared to if there is no screening program in place. The savings will be in terms of the reduced societal costs incurred by the morbidity and mortality associated with the disease being screened less the cost of provision of the screening program (61). All costs and benefits are adjusted to net present value. Thus, the cost-benefit analysis will be as follows (62): = ( − ) (1 + ) Where: r = discount rate t = year n = analytics horizon (in years) 39 We will then calculate this net benefit for different screening programs, which are essentially various combinations of the different options for the screening programs components; we then want to choose a screening program that will maximize this net benefit to the society. If the net benefits for all the programs considered are negative, we will decide to not have any screening program in place as the benefits of the screening program do not exceed the costs of the program. Benefits from multiple perspectives Perspective Benefit consists of savings from the following averted costs: Medical system Direct medical cost + Indirect medical cost Medicare / Medicaid Direct medical cost + Indirect medical cost + Direct non-medical cost Societal Direct medical cost + Indirect medical cost + Direct non-medical cost + Productivity loss Direct non-medical cost here includes nursing home costs, guide dogs, and so on. Which cost components are included depends on which payer’s perspective is taken (e.g. guide dogs are not covered by Medicare / Medicaid but may be covered by the Veterans Administration). 40 4.3. CONCEPTUAL MODEL FRAMEWORK The diagrams below depict the main causal pathways in the disease screening models for the 3 different models considered. Model 1 depicts the general structure of existing simulation models; model 2 modifies model 1 by moving compliance from input to mediating variable; model 3 builds on model 2 by incorporating the hierarchical nature of interventions into the model. The arrows show how one variable influences another variable in the model. The figure gives a high-level conceptual view on which the detailed model is built on; in other words, it represents a broad summary of the model structure but is not intended to depict all factors and variables responsible for the outcomes associated with screening comprehensively. The model contains other interacting elements not portrayed in the diagram, which would be elaborated on in the following chapters. Variables associated with treatment, including treatment compliance, treatment response, and treatment costs, for instance, are considered in the model implementation, although not included in the diagrams below, since the focus of the conceptual model is on the pathways between screening strategy components and outcome. The conceptual model framework shown here is generalizable to most screening services. A cancer screening model that seeks to elucidate the relationship between components of the screening program and costs would follow the same dynamics, with compliance with cancer screening also playing an important role. For some screening services, the morbidity and mortality may include not just harms caused by developing the disease, but also harms caused by screening itself (for instance, false positives in breast cancer screening may necessitate more invasive follow-up screening procedures such as painful biopsies that (63) may introduce the possibility of infection and scarring (64), and thus costs related to health outcomes include more than just benefits related to the disease that is to be prevented, but also costs related to negative health outcomes due to the 41 screening procedure. In contrast, for some diseases where the adverse effects of screening are negligible, screening does not contribute to more morbidity and mortality, so the benefits can be assumed to comprise of savings from diseases averted by screening; for instance, the adverse effects of DR screening are negligible, as the verification of false positives typically involves another round of eye screening, which is non-invasive, and thus in the model, the benefits can be assumed to comprise mainly savings due to averted blindness. To a more limited extent, the conceptual model framework also is generalizable to other preventive services, especially those that require patients and providers to realize clinical advice or guidelines for better health outcomes for patients. A prevention program that involves having patients at high risk of cardiovascular disease take aspirin as preventive measure, for instance, can use a similar framework, where the compliance considered would involve medication compliance instead of screening compliance. Another example where a similar framework can be used includes an obesity prevention program that encourages students to engage in more physical activity. Below, each model would be introduced, starting with an overview of each model and a diagram showing all the causal pathways, followed with diagrams showing different mechanisms of influence of various screening strategy components on outcome for each model. Different screening strategy components would follow different mechanisms of influence based on their characteristics, what they are designed to do, and who they target. As model 1 is a subset of model 2 and model 2 is a subset of model 3, the mechanisms of influence of model 1 are also subset of mechanisms of influence of model 2, and the mechanisms of influence of model 2 are also subset of mechanisms of influence of model 3. 42 4.3.1. MODEL 1 (GENERAL STRUCTURE OF EXISTING SIMULATION MODELS) 4.3.1.1. Model 1 overview Model 1 is an abstraction of the structure of simulation models for evaluation of screening strategies as described in the literature. Model 1 reflects the structure of existing simulation models and is illustrated in the diagram above. In this model, the course of the disease progression is affected independently by patient characteristics, patient compliance, and components of the screening program. The probability of moving from one state to another in the disease progression is affected by patient characteristics such as age and race /ethnicity, as well as disease severity and presence of co-morbidities. The choice of screening strategy and the rate of patient compliance affect screening costs. The course of disease progression then affects follow-up and treatment costs as well as health outcomes and thus benefits. 43 Whether patient comply with screening or not, they will be progressed through the disease progression sub-model; however, their compliance with screening will affect the disease progression. Those who comply with screening are screened and undergo treatment if they are diagnosed with the disease and comply with treatment, thus altering their natural disease progression and reducing their risk of morbidity and mortality. Meanwhile, those who do not comply with screening are not screened and will follow their natural disease progression. Screening strategies that can be evaluated using model 1 do not include those that target compliance since the evaluation of such screening strategies require the model to include a mechanism of influence involving patient compliance as a mediating variable. This will be addressed in model 2 and 3. 4.3.1.2. Model 1 mechanism of influence 44 There is a single mechanism of influence for model 1. In current simulation models as exemplified by model 1, the mechanism of influence from screening strategy to outcome is mediated by a single process: disease progression. Screening modality and screening frequency are examples of components of the screening program that can affect the course of disease progression by altering how much screening potentially reduces the risk of morbidity and mortality from the disease. Different screening modalities have different sensitivities and specificity, and thus different rates of true or false positives and negatives. This changes the probability of a person being diagnosed with the disease when he does have the disease, making him more or less likely to reduce his risk of morbidity and mortality by undergoing treatment after disease detection. Similarly, a higher screening frequency means that patients are screened at higher rates and are more likely to have their disease detected early, thus altering their natural disease progression and their risk of morbidity and mortality. Oftentimes the underlying assumption that many screening strategy components follow this mechanism of influence is not valid, however. In reality, screening strategy components follow this mechanism of influence only if they do not affect patient compliance. For instance, changing a screening modality to a different one that has different sensitivities and specificities but similar adverse effects and similar screening methods from the patient’s point of view will change disease progression but likely not patient compliance. Yet, many screening components that are not designed to influence patient compliance, such as screening modality and frequency, can affect patient compliance. For instance, too high of a screening frequency may overwhelm patients and lower screening compliance; using a nonmydriatic camera that requires no eye dilation as DR screening modality instead of mydriatic camera that requires dilation may improve screening compliance; using a more invasive screening modality may reduce screening compliance. 45 Consequently, using model 1 to estimate the effects of such screening strategy components under- or over-estimate their effects on cost-benefit as it fails to take into account the influence of such components on compliance, which will also influence the cost-benefit. Such screening strategies follow the first mechanism of influence in model 2, and should be evaluated with model 2 or 3. 4.3.2. MODEL 2 4.3.2.1. Model 2 overview Model 1 does not have a model structure that enables the evaluation of strategies that include policies affecting compliance. Model 2 addresses this limitation by modifying model 1 to change patient screening compliance from an input to a mediating variable. In this model, patient characteristics and components of screening program are factors affecting compliance. Demographics like age and ethnicity, for instance, have been shown in several studies as factors affecting screening compliance (65). Other patient characteristics, such as co-morbidities, 46 disease severity, and self-care, also can affect screening compliance. Finally, the choice of component of the screening program can affect compliance. Model 2 shares most of the pathways of model 1, except that now the course of the disease progression can be affected by patient characteristics and components of the screening program independently, or they can influence disease progression through screening compliance. This means that screening strategies can alter the course of a patient’s disease progression by altering his/her screening compliance, which will eventually have an effect on the outcomes. In effect, model 2 is now able to evaluate strategies that include policies affecting compliance as well as more accurately reflect reality, in that the model now takes into account the effect of screening strategy components, even those not designed to alter compliance, on patient screening compliance. 4.3.2.2. Mechanisms of influence in model 2 In model 2, there are two additional mechanisms of influence from screening strategy to outcome other than the one covered in model 1. 47 Mechanism of influence 1: Screening program components that follow this mechanism of influence are those that affect disease progression while also, intentionally or not, alter screening compliance. Components of such screening program affect the costs of screening through two pathways: directly and indirectly, through its effect on screening compliance. The choice of screening modality, for instance, would incur both fixed and variable costs associated with the delivery of screening. In the case of DR, if a mydriatic camera is chosen as the screening modality, fixed costs such as the purchase cost of the camera, rent of space, and so on, would be incurred regardless of the rate of the screening compliance. Most of the variable costs, however, such as reading center costs and verification of positives, are incurred only when patients are screened as positive, and thus are affected by screening compliance. Similarly, patient costs are only incurred when patients are being screened, and thus are also affected by screening compliance. 48 Such components of the screening program can also affect the benefits of screening through two pathways as well, directly and indirectly through its effect on screening compliance. For instance, as has been explained in model 1, screening modality affects health outcomes directly through its effect on disease progression due to the different sensitivities and specificities. Yet, it also affects health outcomes indirectly through screening compliance: the choice of a screening modality that does not require dilation may increase compliance as patients do not need to undergo uncomfortable dilation and would not need to refrain from working or driving for several hours after the screening exam until the dilation wears off. The choice of a screening modality that allows DR screening to be done by a non-specialist, such as a retinal camera, may also improve compliance if such modality is located in a PCP (primary care physician) office that is more accessible than a specialist’s office (66) (67). Subsequently, the components of the screening strategy have an effect on the benefits through its effect on disease progression and health outcomes, or through its effect on screening compliance, disease progression, and health outcomes. 49 Mechanism of influence 2: In this mechanism of influence, screening strategy components affect disease progression indirectly through patient screening compliance. Screening program components that follow this mechanism of influence are those chosen specifically to improve patient compliance with screening at the patient level, such as changing insurance coverage to waive out-of-pocket costs for screening (such as in the case of the Affordable Care Act), or patient education on the benefits and importance of screening through targeted mailing. 50 4.3.3. MODEL 3 4.3.3.1. Model 3 overview This model reflects the involvement of provider compliance provider-level screening strategy components designed to alter patients’ compliance in altering patients’ disease progression and health outcomes. A screening program may have important provider-oriented components. Such provider-oriented components may include mandatory patient education on benefits and importance of screening by providers, or contracting between providers and patients on screening, use of screening appointment reminders, or mandatory provision of patient navigators to help patients navigate the health care system. With all these screening strategy components, provider compliance in instituting such components also plays a role in the success of the screening strategy in improving patient outcomes. Lack of 51 provider compliance results in that particular provider’s patients not benefiting from the provider- level components. A more general example of provider-level components goes beyond screening services. They can include a program to ensure that patients receive their recommended screening services, counseling, and immunization during their check-ups by including a flowchart listing recommended activities for physicians to follow. Here too, lack of adherence to using the chart flowchart may result in unimproved screening rates among patients; consequently, the program will not have an effect on morbidity and mortality and thus costs associated with health outcomes. Model 3 can be generalized to more than 2 levels by adding other aggregate levels, e.g. clinics, as additional mediating variables. Thus, we can have clinic-level screening strategy components influencing patient compliance through clinic compliance with policy, or clinic-level components that may affect patient compliance through clinic compliance with policy and provider compliance with policy. An example of the former is the coordination of screening services for patients by the clinics. An example of the latter is clinic workshops to support provider-level screening strategy component of patient education on the benefits of screening compliance by providers; such workshops can educate providers on methods to educate patients on screening and help improve provider compliance with the provider-level screening strategy component. 52 4.3.3.2. Mechanism of influence in model 3 In addition to the mechanisms of influence of models 1 and 2 introduced above, model 3 has one additional mechanism of influence involving provider compliance with policy. Provider-level screening strategy components influence provider compliance with policy, which in turn influence patient screening compliance, disease progression, and thus costs and benefits. 4.4. MODELING SCHEMA 4.4.1. HIGH-LEVEL MODELING SCHEMA FOR MODELS 1, 2, 3 The following high-level modeling schema for models 1, 2, and 3 depict how to implement the modeling frameworks shown above in a simulation model. The modeling schema consists of several modules, with model 1 being a subset of model 2, and model 2 being a subset of model 3. Model 3 53 can therefore be built sequentially, starting with model 1, then adding the patient-level compliance module to form model 2, and finally adding the provider-level compliance module to form model 3. Figure 4.4.1.1. High level modeling schema for models 1-3 There are four modules (components) in model 3: compliance module (with provider-level sub- module and patient-level sub-module), disease screening module, cost module, benefit module. 54 4.4.2. DETAILED GENERIC MODELING SCHEMA A detailed generic modeling schema of model 3 is given below. This schema shows the inputs and outputs to the modules and sub-modules as well as the interaction between the different modules and sub-modules. Figure 4.4.2.1. Detailed modeling schema for model 3 As the implementation of the modeling schema takes the form of a microsimulation, this modeling schema shows the generation of cost and benefit outcomes for an individual patient for one simulation period. These outcomes are tracked so the outcomes for each patient at the end of each simulation period are known. At the end of the simulation, the cost and benefit outcomes for all patients for the entire simulation period are added up and cost-benefit analysis performed to obtain societal cost-benefit for a particular screening strategy. 55 As before, screening strategy and patient characteristics are inputs to the model, while cost and benefit outcomes are outputs. The choice of the screening strategy is an input to the compliance module and the disease screening module. Provider-level screening strategy components are inputted to the provider-level sub-module to generate compliance of a specific patient’s provider, which along with patient-level screening strategy components and patient characteristics are inputted to patient-level sub-module to generate the probability of that particular patient’s compliance. In the disease screening module, this probability of patient’s compliance is used as a threshold along with a random number generator to obtain whether this particular patient comply with screening for that particular simulation period. If he/she complies with screening, he will get a disease diagnosis, whose accuracy depends on the screening sensitivity and specificity used. The number of times he gets the disease diagnosis in that particular simulation period depends on the screening frequency used. If the diagnosis is positive and patient complies with screening, this alters the natural disease progression. If patient does not comply with screening, he/she immediately is progressed through the natural disease progression. The final disease state of that patient for this particular simulation period is then fed to the benefit module to determine the benefit outcomes. At each stage of the disease progression module, utilization of services are tracked and fed into the cost module, along with the choice of screening strategy used, to determine the screening costs for this patient for this simulation period. Follow-up visits generally consist of additional screening required to confirm positive findings and incur additional cost. The form of natural disease progression sub-module differs based on the disease; included states or variables within the natural disease progression sub-module depend on characteristics of disease and treatment effects. For instance, a breast cancer or Diabetic Retinopathy disease progression sub-module can take the form of a Markov or Discrete Event Simulation model with the stages of 56 breast cancer or Diabetic Retinopathy progression; a diabetes disease progression sub-module can take the form of differential equations of blood sugar level. In addition, adverse effects can be handled by adding a state in the natural disease progression sub-module which will reduce the benefits of screening for affected patients; for instance, treatment for false positive breast cancer patients are unnecessary and may results in infections, reducing the benefits of screening. To enable the hierarchical structure of the patient- and provider-level sub-modules, patients need to be nested within providers who provide services to these patients, as is illustrated in the diagram below. Thus, compliance of provider with policy only affects the compliance of patients under that provider. Figure 4.4.2.2. Nested structure illustration 4.5. FORM OF COMPLIANCE MODEL In contrast to the disease progression model where the model forms are fairly well-established (e.g. using Markov model to model the progression of diabetes or DR, or using differential equations to model obesity, or using system dynamics model to model the interaction between various dietary Patient level Provider level Organization level Clinic 1 Provider 1 Patient 1 Patient 2 Provider 2 Patient 1 Patient 2 Patient 3 57 elements with blood pressure), the forms of behavioral models for use in simulation models (both in terms of an individual / patient behavior as well as organizational behavior impacting clinic-level implementation of a certain policy) are not as well-established. This section will propose some general model forms for the compliance model. In general, there are three types of models that can be used for the compliance model: 1. Empirical model (i.e. models based on empirical observations, using methods such as regression analysis, artificial neural network, and support vector network) 2. Experimental model (i.e. models formed from the results of choice experiments, such as stated preference discrete choice experiment, that attempts to model the decision process of an individual in a stated context). 3. Theoretical model (i.e. models where there is already an established theory of behavior, e.g. the price elasticity model) If there is already data on how screening strategies of interest impact actual implementation of the strategy in clinics and patient compliance with screening, empirical methods such as regression analysis can be used to come up with the implementation and compliance models (e.g. by using hierarchical regression / nested models to form a screening compliance model where patients are nested within clinics). Sometimes the screening strategies of interest have not yet been implemented (or the new screening strategy of interest involves screening strategy components that have not been implemented before), such that empirical data based on existing implementation cannot be obtained. For instance, current guidelines specify for an annual eye screening frequency for diabetes patients. Policymakers might consider whether a different screening frequency might affect screening compliance, but might not have empirical data on it. One possible way to model compliance in this case is by performing choice experiments to estimate non-market valuation of 58 the different screening frequencies to the patients. Choice experiments are usually used to determine willingness to pay for a certain product / service, although in general they can be used to elicit other forms of preference / values (e.g. uptake for eye screening in our example) (68) (69). The approach relies on answers to carefully worded survey questions of hypothetical scenarios / products (e.g. different screening frequencies). The answers indicating preference of patients are then scaled following an appropriate model of preference, usually in the multinomial logit or probit form. Established models such as behavioral economics models or various behavior change models (e.g. health belief model (70), theory of planned behavior (70), and trans-theoretical model of behavior change (71)) can also be used as a starting point for the patient / provider compliance model. On the clinic implementation side, models from the nascent field of implementation science (e.g. the diffusion of innovation or the Promoting Action on Research Implementation in Health Services (PARIHS) framework) can be used (68) (69). Empirical methods can also be used alongside the theoretical models to populate the parameters of the theoretical model form for a particular context. Alternatively, expert opinion elicitation can be used when data is very scarce to determine the value of the parameters in the model. 4.6. MODEL COMPARISON The three constructed DR screening simulation models will be used to evaluate several screening strategies to test the 3 models and evaluate their differences in the following dimensions: • The extent to which each model can evaluate different mechanisms of influence of an intervention 59 • The estimation of policy impact assessment using each model • The resultant most cost-beneficial screening strategy determined by each model 4.6.1. THE EXTENT TO WHICH EACH MODEL CAN EVALUATE DIFFERENT MECHANISMS OF INFLUENCE OF AN INTERVENTION Models 1, 2, 3 can evaluate different mechanisms of influence of an intervention. The structure of model 1 does not allow for evaluation of multi-level screening strategy components addressing behavioral modification involving screening compliance, and the structure of model 2 does not allow for evaluation of provider-level screening strategy components addressing behavioral modification involving screening compliance. To demonstrate this, model 2 will be used to evaluate screening strategies that include components that cannot be evaluated by model 1, and model 3 will be used to evaluate screening strategies that include components that cannot be evaluated by model 1 and 2. 4.6.2. THE ESTIMATION OF POLICY IMPACT ASSESSMENT USING EACH MODEL As has been explained above, the use of model 1 to evaluate screening strategies that have components which actually has a mechanism of influence outside the purview of model 1 can result in under- or over-estimation of policy impact. To demonstrate this, both model 1 and 2 will be used to evaluate the same screening strategy including screening modality as a component. In model 1, its effect on compliance will not be taken into account, while it will be in model 2. The differential 60 impact on cost-benefits of evaluation of this same screening strategy using models 1 and 2 will be shown. Similarly, using model 2 to evaluate screening strategies that have components which should be evaluated using model 3 will result in under- or over-estimation of policy impact. Model 2 does not take into account provider compliance, in effect assuming that all providers will comply with provider-level screening strategy components and all patients will benefit from these components. Models 2 and 3 will be used to evaluate the same screening strategy including provider-level components and the differential impacts on cost-benefits using the two models shown. 4.6.3. THE RESULTANT MOST COST-BENEFICIAL SCREENING STRATEGY DETERMINED BY EACH MODEL Models 1, 2, and 3 will be used to determine the most cost-beneficial screening strategy among all the screening strategies that can be evaluated using each model. The resultant most cost-beneficial screening strategy determined by model 1 will be compared to that determined by models 2 and 3. 61 CHAPTER 5. APPLICATION OF GENERIC MODEL TO DR SCREENING EXAMPLE: MODEL PROFILE FOR DR SCREENING The application of the generic model above will be illustrated within the context of DR screening. The following model profile delineates the construction of such a model, including the model parameters, structure, and constructions of the modules and sub-modules. Only the model profile for model 3 will be shown, since models 1 and 2 are subsets of model 3. The appropriate compliance modules or sub-modules can be taken out of model 3 to form models 1 and 2 respectively. The model profile is intended to provide insight into ongoing research. Model parameters, structure, and results contained here should be considered representative but preliminary in nature. The model profile begins with stating the model purpose, the model overview, and assumption overview. It then gives an overview of the model components, parameters, and outputs. 5.1. MODEL PURPOSE This model intends to estimate and compare the impact of various DR screening strategies, including medical intervention components as well as behavioral intervention components (i.e. strategies aiming to improve screening compliance at the patient and provider level) on a population in terms of costs and benefits. The model can be used to compare the impact of 62 implementing a DR screening strategy to the status quo, or to aid comparative effectiveness evaluation of different DR screening strategies. The model can also be used as a design tool to aid policy design for DR screening strategies. 5.2. MODEL OVERVIEW The system that is modeled This simulation model models a population with Type II diabetes consisting of individual patient life histories in which Diabetic Retinopathy may develop and the effect of various screening strategies on compliance, time of diagnosis, blindness, and overall societal cost-benefit. The screening strategies here include medical intervention components as well as behavioral intervention components. With regards to the behavioral intervention components, the model also models the compliance of providers with screening strategy components and its effect on patient compliance and thus the outcomes of the screening strategies. The patient life histories consist of: • Baseline screening compliance based on patient characteristics (demographics, diabetes severity, self-care, self-efficacy). • Development of DR as influenced by age and diabetes severity (using insulin use as a proxy). • Timing of diagnosis of high-risk (proliferative) DR. • Net blindness from DR. 63 • Influence of various screening strategies on screening compliance, time of diagnosis, blindness, and societal costs and benefits. A patient’s individual life history starts with these patient characteristics: starting age (age at which patient enters the simulation), demographics, disease severity, self-care, and self-efficacy (for this particular model, the distribution of these characteristics is taken from CHIS 2009 survey to represent California population). Baseline patient screening compliance is then predicted based on these patient characteristics. Age of death is determined from a life table, adjusted by diabetes risk factors. The corresponding age-specific risk factors are applied to modify the transition rates in the multistage DR progression model to DR and blindness. These transition rates are modified by DR screening at rates determined by the screening strategy used. General modeling methodology This simulation model is a microsimulation of individual patient life histories to constitute a relevant cohort population. Microsimulation is a modeling technique that operates at the level of individual units such as persons and within the model each unit is represented by a record containing a unique identifier and a set of associated attributes such as age and sex (68). A set of rules (transition probabilities) are then applied to these units leading to simulated changes in state and behavior. The result is an estimate of the outcomes of applying these rules over many time steps, including both total overall aggregate change and the way this change is distributed in the population that is being modeled. In the DR model, there are two levels of units – providers and patients, in that patients are nested within providers in the simulation to enable the modeling of the effect of provider compliance of a certain policy on the compliance of patient under that specific provider. The microsimulation will be constructed in Matlab 2009. Primary unit of analysis 64 This estimates made with this model will be primarily on population level. By simulating large numbers of individuals, modeled statistics are generated for the population that those individuals represent. Major components There are four modules in this model: compliance (with provider-level and patient-level sub- modules), disease progression module, cost module, and benefit module. These modules are described in more detail in component overview. Inputs and outputs Inputs into the models are the screening strategies considered and individual patient characteristics constituting the population considered. For more detail, see the input parameters chapter in the parameter overview. The main output of the model is societal cost-benefit for various screening strategies after the length of the period as simulated in the model. Secondary output includes blindness rates in the population with various screening strategies (i.e. health outcomes). 5.3. ASSUMPTION OVERVIEW For the purpose of this dissertation, we use the simplifying assumption that each clinic has one provider, such that provider-level and clinic-level components are equivalent (i.e. clinic and provider levels are condensed into just one level, which is referred to as ‘provider-level’ here). In addition, we make a simplification in terms of capacity issue in that one camera and one patient navigator is assumed to be enough to serve all patients in the clinics. This is currently a reasonable 65 assumption as patients only go for screening once a year or less frequently, the number of diabetic patients served per clinic is not too large, and eye screening takes only a few minutes to perform for the medical providers. However, this assumption needs to be revisited if the number of diabetic patients served per clinic grow or it is determined that patients need to screen more frequently. DR is assumed to be strictly progressive, so no patient regresses from high-risk (proliferative) DR to low-risk (non-proliferative) DR. We assume here that screening frequency and modality choice do not have an effect on screening compliance to simplify the compliance model. However, there is evidence that they likely affect compliance, especially if changing the modality to non-mydriatic camera also involves changing the screening location to more accessible primary care clinics. We assume that the Los Angeles population with Type II diabetes using Los Angeles County Health Services (LACHS) clinics are similar to the California population, since there is better data for the variables of interest at the California level than at the county level. However, this might have an effect of under-representing the underserved population who tend to use LACHS safety-net clinics. Also, we assume that the effect of having health insurance on compliance with eye screening in the context evaluated here (i.e. Los Angeles population using Los Angeles County Health Services clinics) is similar to that for the California population, i.e. that having health insurance will increase screening rates. However, as LACHS are safety-net clinics, this might not have the same effect within the LACHS network since patients with health insurance might actually be less likely to go for screening within the LACHS clinics since they can go to non-safety-net clinics outside LACHS. This effect would vary based on the type of insurance (e.g. Medicaid vs private insurance). We assume that patient cost are negligible and thus are not considered in the cost module, as patient cost due to waiting time and time in clinic are relatively small according to the VA study where cost parameter estimates are derived. 66 We simplify the benefit model for averted indirect medical costs as being linearly dependent on the number of years it takes to adjust to blindness, even though patients are likely to adjust to their blindness more and more as time passes and incur less and less medical costs due to secondary complications of blindness. Also, the number of years it takes to adjust to blindness is assumed to be the same for everyone even though it probably vary with patient characteristics. To refine the benefit model in the future, the averted indirect medical costs should reflect the diminishing costs with the passing years of blindness, and the period of adjustment to blindness should have a probability distribution with patient characteristics as predictors. For the screening frequency component of the screening strategy, we assume that no patients screen in the non-screening years (i.e. if the screening frequency is biennial / every 2 years, patients screen in the first year, and then no patient go for eye screening in the second year, and so on, instead of distributing some patients to screening in one year and others in another year). This has ramifications in the discounting of the costs and benefits and therefore the resulting NPV, although for the screening frequency alternative considered (every 2 years vs every 1 year), this effect is minor. This would also have an effect on capacity utilization, although not for this particular example as we assume that clinics do not hit the capacity ceiling even for the highest screening frequency. This assumption will need to be revisited for lower screening frequencies (e.g. if we are evaluating colonoscopy screening strategies, in which the current guidelines specify a screening frequency of 10 years) as this would affect the discounting and capacity utilization significantly. Study data on the impact of implementation of patient navigators to increase eye screening is not available, thus the odds ratio estimates reported from the above mammography screening study is used as a proxy for the odds ratio of having patient navigator for increasing eye screening rates in the model. 67 Some screening strategy components can interact with each other in a crowding out or substitution effect fashion. For example, patient education at the provider level and patient education at the patient level, if offered together, will likely have crowding out effect where the effects of one is diminished by the other. For this dissertation, it is assumed that there is no crowding out or substitution effect between the screening strategy components. However, if crowding out or substitution effect is known (and the effect can be estimated), the combination of affected components should be considered in a single screening strategy component and the combined effects considered to take into account the crowding out effect. 5.4. COMPONENT OVERVIEW 68 The relationships between the different modules are shown above. This diagram has been explained in chapter 4 above. The individual modules are explained in more details below. 5.4.1. COST MODEL Fixed program costs include cost of the equipment (e.g. fundus camera or ophthalmoscope) and space. Variable program costs include personnel, verification of positive diagnosis, maintenance, training, overhead, and Reading Center fees. Patient opportunity costs (time spent traveling to and from the clinic, recovering from dilation, being screened and verified) are not included in the model. The calculation of total cost is done according to the following series of equations: Total cost = screening program cost + follow up visit cost + treatment cost Screening program cost = fixed program cost + variable program cost Fixed program cost = cost of space + cost of diagnosis equipment (e.g. camera) + cost of interpretation equipment (e.g. reading center equipment) + cost of other equipment Variable program cost = utilization of screening program x (unit personnel cost + unit reading center charge + unit verification of positives cost + unit training cost) Follow up visit cost = utilization of follow up visit x unit follow up visit cost Treatment cost = utilization of treatment x unit treatment cost 69 5.4.2. BENEFIT MODEL The benefits include savings from averted direct and indirect medical and non-medical costs as well as savings from averted productivity losses. Social costs of blindness, which are harder to quantify, are excluded. Social costs of blindness include the interference with various activities of daily living, limited participation in a host of leisure activities, loss of status and self-esteem, erosion of quality of life of both the affected individuals and their families (69). The costs for programs for the blind (including programs similar to the Department of Education’s Independent Living Services for Older, Blind Individuals, the American Printing House for the Blind, and the Library of Congress’ National Library Service for the Blind and Physically Handicapped) are excluded since they are small compared to the other cost components considered. Direct medical costs include both eye-related medical costs, such as eyeglasses and ophthalmologist visits. Indirect medical costs include non eye-related costs, which are medical costs due to secondary outcomes of blindness, including depression and injury. Direct nonmedical costs include nursing home care and long term care include visits from nurses and counselors and home rehabilitation and therapy as well as other types of home care. Productivity losses are defined as the incremental costs of lower labor force participation and lower wages for visually impaired and blind individuals compared with those in the same age group with normal vision. In addition to the costs shown in the diagram above, based on the perspective of the DR screening program designer, other costs may be considered as well (although they are not included in this model). For instance, the state or federal government may also consider tax losses from additional 70 deductions due to blindness, supplemental security income, social security disability insurance, and food stamps as part of the cost of blindness. 5.4.3. DR SCREENING MODULE The natural disease progression model as described below is based on the model structure from a seminal paper by Dasbach. This natural disease progression model simulates the progression of retinopathy in a population cohort with different population subgroup risk profiles for DR progression in a Markov process. The different population subgroups have different risk profiles for DR disease progression and are based on diabetes severity with treatment (with/without insulin) as proxy. Four states are considered in the disease model: low-risk DR (no retinopathy / non sight- threatening / non-proliferative retinopathy), high-risk DR (sight-threatening / proliferative retinopathy), blind (visual acuity worse than 20/200 in both eyes), and dead. 71 Simulation begins with patients in the low-risk or high-risk states. Wisconsin Epidemiologic Study of Diabetic Retinopathy (WESDR) data is used to construct the three-state (low risk DR, high risk DR, and blind) transition probability matrix, then mortality is incorporated into the matrix using life tables data by adjusting age-specific general population mortality probabilities by the relative risks observed in the WESDR. Patients experience increased mortality risk appropriate to their age as they age in the simulation; also, as individuals in the simulation experience a further increased mortality risk as they progress to higher-risk states of retinopathy. Simplification was made in that mortality was considered separately from retinopathy progression, although it partly reflects the progression of the underlying diabetes. This model form allows for the estimation of the probability that a person will be in one of the four health states at the end of a cycle time of the model given a person’s current health state and age. A cycle time of 1 year was used, applied over the remaining lifetime of the cohort. As this simulation follows a cohort, it does not include incident cases of diabetes over the simulation period. Screening can alter the natural disease progression by reducing the risk of blindness for patients whose DR are caught early and then undergo treatment. To capture that in the model, one additional state (treated) is added to the 4 states of the natural disease progression model. Screening can move patients from high-risk to treated, halving the relative risk of blindness. The transition probability from high-risk to treated is determined by the sensitivity and specificity of the screening modality. 72 73 Parameters affecting the probability of moving from one state to another (compliance with screening and treatment) are shown in the diagram. Patients in the low-risk-DR state with true negative results (correctly diagnosed as not having high-risk DR) stay in that state. Patients in the low-risk-DR state with false positive results (incorrectly diagnosed as having high-risk DR) go for an extended visit (incurring an extended visit cost) and undergo a gold standard exam, eventually being diagnosed correctly as low-risk DR and thus staying in that state. High-risk DR patients with 74 false negative screening results would stay in that state due to incorrect diagnosis, while those with true positive results undergo extended follow-up visit for gold standard exam, confirming the initial diagnosis. Then, if they comply with treatment, they will move to the ‘treated’ state; otherwise, they will stay in the high-risk DR state. The gold standard exam is considered to be 100% sensitive and specific for the purposes of these analyses. In this screening sub-model, treatment includes the ongoing process of repeated follow-up and panretinal photocoagulation sessions, which reduces subsequent risk of blindness. 5.4.4. COMPLIANCE MODULE The compliance module is based on an empirical model. Probability of compliance is described as a series of equations. First, the baseline probability of compliance of each patient is calculated; then, the probability of compliance of each patient after patient-level and provider-level screening strategy components are applied is calculated with functions involving the baseline compliance probability. To populate the coefficients in this compliance module, we need to know the effect of patient characteristics and patient-level screening strategy on compliance. To do so, logistic regression analysis is done on data from CHIS 2009 with the following variables: Dependence variable: Compliance Independent variables: • Demographics (age, race, sex, living within 200% poverty level, married, education level, employment status) 75 • Comorbidity / disease severity (high blood pressure, heart disease, on insulin) • Access (having insurance in the past 12 months) • Quality of Care (doctor checking HbA1C level, doctor checking feet, doctor helping to coordinate services) • Self-care (smoking, frequency of physical exercise, freq. of eating fast food, freq. of eating fruits & vegetables, freq. of checking HbA1C level) • Self-efficacy (confidence to control and manage diabetes) The results of the regression analysis are used to populate the coefficients of the baseline compliance equation reflecting the relationship between compliance and patient characteristics and the coefficients for the patient-level compliance module equation reflecting the relationship between compliance and patient-level screening strategy component for case 2 below. The method and results for this regression analysis are described in detail in the appendix. 5.4.4.1. Inputs and outputs of this module The inputs to this module are patient-level and provider-level screening strategy components and patient characteristics. The output of this module is probability of patient compliance given the baseline characteristics and screening strategy used. 5.4.4.2. Patient-level compliance module and provider-level implementation modules 76 We will describe empirical data-based patient- and provider-level module forms for different cases here based on the empirical data available. a.i. Case 1 (We have empirical data that allows us to model the relationships between demographics, all patient-level and provider-level screening strategy components, and provider- level implementation and patient-level compliance) Use hierarchical logical regression (70): Level 1: For a patient with covariate patient characteristics vector x, corresponding to the kth cluster (e.g. provider or clinic), the probability of observing Y = 1 (i.e. compliance) is ( = 1| ; ) = exp (η%&, u ) *) 1 + exp (η%&, u ) *) Level 2: For that individual, the second-level equation is η(x, u k) = βx + u k, where u k ∼ N(0, σ 2 ). a.ii. Case 2 (We have more limited empirical data: we have information on the relationships between patient characteristics, patient-level screening strategy components, and patient compliance, as well as provider-level implementation and patient compliance, but the data is not nested, so hierarchical regression cannot be performed) This case is the one used for the implementation of the DR model described in this dissertation. First, the baseline patient compliance is calculated based on patient characteristics using the following logistic regression equation: 77 Baseline patient compliance equation: ,- %. /0 * = 1 + 1 2 2/0 + 1 3 3/0 + ⋯ + 1 5 5/0 Where: . /0 = 6, . 6 , 7 89.,68 .6 : .;: < 1 5 = 88 ;66 , 5 5/0 = :.: ;66 , 9 (. -. 6-, >ℎ ℎ ,, ℎ6;- 68, 8 ) .6 : .;: < Patient-level equation: ,- %. /0 ∗ * = 1 + 1 2 2/0 + 1 3 3/0 + ⋯ + 1 5 5/0 ∗ Where: 5/0 ∗ = A 1, 8- 6 -7 89. 9 6- .6 89.,68 5/0 , 8- 6 -7 89. 9 : 6- .6 89.,68 . /0 ∗ = 9:: . 6 , 7 89.,68 .6 : .;: < 6 .6 − ,;, 8- 6 -7 89. 6 6..,: Provider-level equations: ,- %. /0 ∗∗ * = 1 + 1 2 2/0 + 1 3 3/0 + ⋯ + 1 5 5/0 ∗ + 1 /0 Or ,- %. /0 ∗∗ * = ,- %. /0 ∗ * + 1 /0 Where 78 /0 = B 1, 8,8 < ℎ6 .6 ,- 9 ℎ C ℎℎ,: DE 9.,9 ℎ .;: − ,;, 8- 6 -7 89. 0, ℎ> a.iii. Case 3 (We have prior limited information only on the relationships between screening strategy components and compliance, e.g. all we know is that using a nonmydriatic camera will increase compliance by 10%) Patient-level equation: . /0 ∗ = G . /0 H(I 0J (1 + K J )) 2 , . /0 H(I 0J (1 + K J )) 2 ≤ 1 1, ℎ> I 0J = M 1, .,87 6- .6 89.,68 N NOP QR ,/S TUJ/VW XUYZ U [\]Y T[/Y VU5TJ/[VY K J = ,6 ; 8ℎ6- 89.,68 .8 : 8 - 6 -7 89. ^, 6, 6 ; , Provider-level equation: . /0 ∗∗ = G . /0 ∗ H(1 0J (1 + K J )) 2 , . /0 ∗ H(1 0J (1 + K J )) 2 ≤ 1 1, ℎ> 1 0J = M 1, .;: < _ ℎ6; 8ℎ6- (6 0J = 1) N NOP QR ,/S T\U`/XY\ 0 a Z bYc[`/U\ XUYZ U Vc[]Y ([ dQR ) K J = ,6 ; 8ℎ6- 89.,68 .8 : 8 - 6 -7 89. ^, 6, 6 ; , 79 . /0 ∗∗ = 9:: . 6 , 7 89.,68 .6 : .;: < 6 .6 − ,;, 6: .;: − ,;, 8- 6 -7 89. 6 6..,: 6 0J = 89.,68 .;: < 6 -7 89. ^ , 6, 6 ; , The probability distribution of a jkl depends on the screening strategies applied and may also depend on other factors such as provider characteristics, characteristics of the patient being treated, peer influence, and so on. In this model, a simplification of the distribution is made in that the provider compliance only depends on the screening strategy component alternative applied and follows a uniform distribution: 6 0J = A 1, e J ≤ ≤ 1 0, ℎ> The probability distribution for a jkl can be updated in the future when there is more information on determinants of provider behavior under certain screening strategies. e J = 89.,68 ℎℎ,: 6 -7 89. ^, 6, 6 ; , 5.5. PARAMETER OVERVIEW 5.5.1. INPUT PARAMETERS (POPULATION COHORT, SCREENING STRATEGY) Population cohort and nesting within clinics The patient population cohort simulated here represents the Los Angeles population with Type II diabetes using the representing Los Angeles County Department of Health Services (LACDHS) 80 clinics, 42265 patients in total. The patient population cohort used in this simulation model is representative of California population with Type II diabetes (it is assumed that the Los Angeles diabetic population is similar to the California diabetic population) and is based on the 2009 California Health Interview Survey (CHIS), a representative telephone survey of 47614 adults, 12 324 children and adolescents from over 49000 households across California. The patients are nested within providers / clinics, with the clinics representing 21 LACDHS clinics. As stated above, a simplifying assumption is made here that each clinic has one provider handling DR and as such that provider-level components are equivalent to clinic-level components. The number of patients distributed in each clinic is based on empanelment reports for the LACDHS clinics that list the total number of diabetics per clinic. The distribution of patients among the clinics is shown in the table below. 81 Clinic No. of diabetic patients % of panel Antelope Valley 159 8 Bellflower 905 17 Dollarhide 413 12 Glendale 392 25 Hudson 3841 21 High Desert 991 10 Humphrey 4587 23 Harbor-UCLA 4078 21 LAC-USC 2791 13 Long Beach 3378 26 Lake Los Angeles 133 13 La Puente 485 16 Littlerock 99 13 Martin Luther King 1919 18 MidValley 3486 18 OV-UCLA 2966 25 El Monte 4501 23 Roybal 3110 26 San Fernando 1323 21 South Valley 1078 9 Wilmington 1630 31 82 Screening strategy Screening strategy components Level Alternatives Status quo Alternative 1 Screening modality Patient Current (Opthalmoscope) Non-mydriatic camera Screening frequency Patient Current (Annual) Biennial (every 2 years) Insurance Patient Current Mandatory health insurance Provision of patient navigator Provider Current (No patient navigator) Provision of patient navigator for clinics with size above a certain threshold The medical intervention components considered here are screening modality and screening frequency. Two different screening frequencies (annual and biennial) are considered. For annual screening, patients are screened once during the cycle time, and for biennial screening, patients are screened once every two years (i.e. once every other cycle time). Other components considered in the screening program are behavioral components intended to raise patient compliance with DR screening, each with different costs associated with them as well as different effects on compliance. These components include patient-level and provider-related 83 interventions. Patient-level components include changes in insurance coverage to include eye screening. Provider-level components include provision of patient navigator for clinics that are above a certain threshold size-wise, the threshold of which can be specified as input. As in reality implementation of such provider-related intervention would not be carried out 100% of the time by the providers involved, in this model these provider-related interventions are also subject to provider compliance with these interventions. Thus, the user can also specify the percentage of clinics meeting the size threshold that are likely to implement the policy (i.e. probability of implementation for clinics meeting size threshold). If no provider-level implementation input parameters are specified by the user, default values will be used. The default for clinic size threshold is 2000 patients, and the default for probability of implementation for clinics meeting size threshold is 0.8. 5.5.2. LENGTH OF SIMULATION PERIOD AND NUMBER OF SIMULATION ITERATIONS 5.5.2.1. Length of simulation period Although the length of the simulation period can be user-adjusted to reflect the duration in which the screening strategy is instituted, the default length of simulation period is 30 years. 5.5.2.2. Number of simulation iterations The entire simulation is run numerous times as a Monte Carlo simulation. To find out the number of simulation iterations required, the simulation is run with different numbers of iterations, and each 84 time, the maximum SEM/mean (71) for the main primary outcomes (sight years and screening & treatment cost) obtained for that particular number of iterations is calculated. The SEM/mean vs number of iterations is then plotted to find a point at which result variance is at an acceptable confidence level. The graph plateaued around SEM/mean of 0.03 (corresponding to confidence level of 94%) at 30 runs. Given that the simulation seems to have reached near optimal precision in the results at that point, and that the confidence level of 94% is an acceptable confidence level, the number of iterations used for the simulation is 30. 85 5.5.3. DR PROGRESSION PARAMETERS DR natural history parameter Disease stage transition probabilities table that lists the transition probabilities from one disease stage to another is given below. Disease stage transition probabilities (annual) Type II diabetes (insulin- taking) Type II diabetes (not taking insulin) From low risk to low risk 0.973 0.990 From low risk to high risk 0.021 0.006 From low risk to blind 0.005 0.004 From high risk to low risk 0.000 0.000 From high risk to high risk 0.932 0.841 From high risk to blind 0.068 0.159 From treated to remain sighted 0.968 0.926 From treated to blind 0.032 0.074 Data source: Dasbach, et al. Cost-effectiveness of Strategies for detecting Diabetic Retinopathy. Medical Care, Vol 29, No 1, Jan 1991, pp20-39 Age-specific, general population mortality rates are adjusted for the relative risks of 2.6 for patients in the low retinopathy risk state and 7.2 for patients in the high retinopathy risk state. 86 DR screening history parameter The probability of obtaining true positive, false positive, true negative, and false negative diagnosis are calculated based on the sensitivity and specificity of the modality used. Modality sensitivity and specificity table Method Ophthalmolscopy Non-mydriatic camera Sensitivity 0.794 0.929 Specificity 0.989 0.85 Data source: Dasbach, et al. Cost-effectiveness of Strategies for detecting Diabetic Retinopathy. Medical Care, Vol 29, No 1, Jan 1991, pp20-39 In this screening sub-model, treatment includes the ongoing process of repeated follow-up and panretinal photocoagulation sessions, which reduces subsequent risk of blindness by about 50%. For patients not responding to photocoagulation (assumed to be 15%), patients will undergo vitrectomy, out of which 44% will have their vision loss averted. 87 5.5.4. COST-BENEFIT PARAMETERS All dollar values in the cost and benefit parameters are adjusted to 2012 dollars using the CPI index. CPI index (from specified year to 2012) Year CPI index 1990 1.76 1992 1.64 2003 1.25 2004 1.22 Data source: Bureau of Labor Statistics In the simulation, all future dollars are discounted to present value at the standard rate of 5%. 5.5.4.1. Cost parameters 5.5.4.1.1 Screening cost parameters Screening program costs table is given below. Utilization of equipment is assumed to be 100%. Additional overhead is computed at 29.6% of total costs. Other costs include camera maintenance. Clinic space was priced at market rental rate of $.78 per square foot per month. Space was utilized for camera storage, camera exams, and ophthalmologic exams. The non-mydriatic camera costs $12000. All these costs are amortized and converted to 2012 dollars. A simplifying assumption here is made such that the fixed cost/clinic is the same irrespective of clinic size. This is a reasonable assumption given that only clinic space needed for DR screening is 88 considered here, and even the larger clinics tend to only have space for one ophthalmologist office or one retinal screening camera in the clinic. Opthalmoscopy Nonmydriatic camera Fixed costs (1992 $/clinic) Space 119.33 627.33 Camera 0.00 2214.67 Other equipment 100.67 291.33 Overhead 1235.33 2090.67 Variable costs (1992 $/patient) Personnel 11.09 5.32 Reading center charge 0.00 3.42 Other 0.27 2.22 Verification of positives 2.45 6.78 Training 0.00 0.77 Data sources: Lairson DR, Pugh JA, Kapadia AS, Lorimor RJ, Jacobson J, Velez R. Cost-effectiveness of Alternative Methods for Diabetic Retinopathy Screening, Diabetes Care, Vol 15, No 10, October 1992. 89 Breakdown of personnel costs are shown in the table below. Ophthalmoscopy Non-mydriatic camera Minutes Minutes Person Cost/min Mean Std Dev Person Cost/min Mean Std Dev Preliminary Intake / history Clerk 0.1655 4.41 2.1 Clerk 0.1655 4.41 2.1 Physical exam PA/NP 0.3914 6.99 2.68 PA/NP 0.3914 6.99 2.68 Dilation PA/NP 0.3914 2 Screening Opthalmologist 1.9 6.52 5.07 PA/NP 0.3914 8.09 4.17 Interpretation N/A - - - PCP 0.6412 1.23 0.92 Data sources: Lairson DR, Pugh JA, Kapadia AS, Lorimor RJ, Jacobson J, Velez R., Cost-effectiveness of Alternative Methods for Diabetic Retinopathy Screening, Diabetes Care, Vol 15, No 10, October 1992. 5.5.4.1.2 Screening cost parameters Data sources for follow-up and treatment costs below are from Dasbach (46). Follow-up (verification of positive diagnosis) costs include an extended ophthalmologic visit cost for a gold standard test. Treatment costs include panretinal photocoagulation treatment visit cost 90 and vitrectomy cost for patients who do not respond to panretinal photocoagulation treatment, as well as the cost of post-treatment visit. Follow up and treatment intervention costs (Dasbach) Adj. cost estimates (2012 $) Original cost estimates (1990 $) Extended ophthalmologic clinic visit cost 128.76 73.16 Panretinal photocoagulation treatment visit cost 312.28 177.43 Vitrectomy cost 6886.88 3913 Post-treatment clinic visit cost 121.88 69.25 Average treatment cost 890.77 506.122 Data source: Dasbach, et al. Cost-effectiveness of Strategies for detecting Diabetic Retinopathy. Medical Care, Vol 29, No 1, Jan 1991, pp20-39 The following table gives the response rates for the treatment options to determine the appropriate amount of costs incurred by patients undergoing treatment. Positive response rate for photocoagulation 85% Percent of patients not responding to photocoagulation going on to have vitrectomy 100% Positive response rate for vitrectomy 44% Data source: Dasbach, et al. Cost-effectiveness of Strategies for detecting Diabetic Retinopathy. Medical Care, Vol 29, No 1, Jan 1991, pp20-39 91 5.5.4.1.3 Patient navigator cost parameters The cost incurred to hire a patient navigator is a fixed cost incurred annually by the clinics and is estimated to be $28860, the median annual income for medical assistants, as obtained from the Bureau of Labor Statistics. 5.5.4.2. Benefit parameters The benefit associated with eye screening will be the economic burden of blindness that is averted through averted blindness due to eye screening. Direct medical cost includes medical costs associated directly with blindness, i.e. eye-related care, such as ophthalmologic visits and eyeglasses. All patients who become blind will incur this cost. Indirect medical cost includes medical costs associated with injuries, depression, and other secondary comorbidities related to the blindness. In the simulation, it is only incurred for the first 3 years after patient becomes blind, as it is assumed that patients need 3 years to adjust to blindness and in those 3 years they would incur such non-eye related care medical costs. Direct nonmedical cost includes nursing home use that would otherwise not be required without the occurrence of the disability (blindness). Only patients who are blind and above the age of 65 are considered. It is estimated that about 35.7% of such patients will require the nursing home due to their blindness and costs will only be incurred for this portion of patients. Productivity losses include losses due to reduced wages and lost wages and are only considered for blind patients under age 65. It is estimated that about 55% of blind patients under the age of 65 would have reduced wages due to having to switch to a lower pay job or reducing work hours because of their blindness, and that 30% of blind patients under the age of 65 would not be able to work due to their blindness and would have lost wages. 92 The following summary table gives the economic burden of blindness which is used in calculating the benefits from averted blindness. The original estimates from the literature are given (72) (73) (74), as well as the estimates that have been adjusted to 2012 dollars for use in the model. Economic burden of blindness Adjusted estimate (2012 $/patient) Original estimate (2004 $/patient) † Original estimate (2003 $/patient) ‡ Portion of blind patients with burden † Direct medical cost 564.81 462.96 - 1 Indirect medical cost 5420.46 - 4443 1 Direct nonmedical cost 31472.34 25797* - 0.357 Productivity losses – unemployment (lost wages) 40497.90 33195 - 0.55 Productivity losses – reduced wages 14787.62 12121.00 - 0.3 Data sources: † Rein, et al. The Economic Burden of Major Adult Visual Disorders in the United States. Arch Opthalmology, Vol 124, Dec 2006. *Cooney, et al. Comparative Assessment of Cost and Care Outcomes Among Georgia’s Community- Based and Facility-Based Long-term Care Programs. Atlanta: Georgia State University, 2004. 93 ‡ Javitt JC, Zhou Z, Willke RJ. Association between vision loss and higher medical care costs in Medicare beneficiaries: costs are greater for those with progressive vision loss. Ophthalmology. 2007 Feb;114(2):238-45. 5.5.5. COMPLIANCE MODULE PARAMETERS The following compliance module parameters are obtained by regression analysis as described in the appendix. These parameter estimates from analysis of maximum likelihood estimates are used as coefficients for the baseline patient compliance equation. The odds ratio point estimates can be found in the appendix A. Variable Parameter estimate Standard Error Pr > ChiSq Intercept -0.382 0.2966 0.1978 Age 1: 18-49 -0.4344 0.0888 <.001 2: 50-64 0.00947 0.063 0.8806 3: 65 and above Ref Less than 200% poverty level -0.2342 0.0982 0.0171 On insulin -0.3084 0.1085 0.0045 94 Having insurance in the past 12 months 0.672 0.1507 <.0001 Doctor checking HbA1C level 0.4277 0.1008 <.0001 Doctor checking feet 0.5181 0.0885 <.0001 Frequency of checking own HbA1C level 1:Never -0.2332 0.0751 0.0019 2:Less than 30 times / month 0.0475 0.0652 0.4664 3:More than 30 times / month Ref Frequency of fast food consumption 1:None Ref 2:Once a week 0.0186 0.0611 0.7604 3:Twice a week or more -0.123 0.0619 0.0468 Confidence to control & manage diabetes 1:Very confident 0.1917 0.0661 0.0037 2:Somewhat confident -0.0351 0.0673 0.6013 3:Not too confident / Not at all confident Ref 95 5.5.6. IMPLEMENTATION MODULE PARAMETERS The following implementation module parameters (odds ratio point estimates) are used to construct the coefficients for the provider implementation module equation. Variable Odds ratio point estimate 95% confidence limit (lower bound) 95% confidence limit (upper bound) Having patient navigator 2.5 1.9 3.2 Data source: Phillips CE, Rothstein JD, Beaver K, Sherman BJ, Freund KM, Battaglia TA. Patient navigation to increase mammography screening among inner city women. J Gen Intern Med. 2011 Feb;26(2):123-9. doi: 10.1007/s11606-010-1527-2. Epub 2010 Oct 8. 5.5.7. LIMITATIONS There are limitations in the knowledge that inform the model, which mainly regards uncertainty in parameter estimates, as well as the compliance module, especially the provider-level implementation, which has not been well-established in the literature. Lauren P Daskivich, MD, was consulted as a subject expert to confirm that the parameter estimates used in the model are reasonable. Dr. Lauren Daskivich is the Director of Ophthalmology and Eye Health Program for the Los Angeles County Department of Health Services and attending physician at the department of ophthalmology at the VA Greater Los Angeles Healthcare System. 96 The model is constructed to be flexible enough to accommodate different levels of data availability. As models 1 and 2 are subsets of model 3, if limited data is available, model 3 can be specified with data at the level of models 1 and 2, and would produce the same results as models 1 or 2. Conversely, if more data becomes available, the methods proposed here can still be utilized to incorporate these updated data into the program. Although subject to lack of available trial data for accurate estimation of some parameters, this model is intended as an illustration of the methods proposed here and as a stepping stone towards designing a comprehensive DR screening strategy encompassing behavioral and medical intervention components that is also cost-beneficial. The model has a limited amount of detail, particularly in the regards of the effect of biology and the underlying diabetes on the disease, since the focus of the model is mainly to illustrate a simulation model to model screening strategy including medical and behavioral interventions, not on the biology of the disease. However, if this is added on to the model, the model would then be able to evaluate the synergy of screening strategies combined with preventive interventions targeting the biology of the disease that can reduce DR rate. For instance, it can evaluate the cost-benefit of interventions where the screening strategy is implemented alongside interventions to maintain blood glucose, cholesterol, and blood pressure levels (such as patient education, automated call reminders to check blood glucose levels, social support network, and so on), which reduces the risk for microvascular complications of diabetes, including DR (75) (76). Other limitations are due to simplifications made which can be referred to in the assumption overview. There is a need for more research and data on the effects and interactions of screening strategy components and their outcomes, since the true value of the methods in this program design is 97 especially apparent when we can show the synergies or antagonistic effects between the various in the strategies making up the DR screening program. 5.6. OUTPUT OVERVIEW The main output consists of the societal cost-benefit of various screening strategies and the secondary output consists of simulated events (e.g. the rate of provider and patient compliance in a given simulation period, the number of cases diagnosed in the cohort, the number of cases missed by screening in the cohort, etc). The output is tracked by year in the simulation period and stored in Matlab data files for further calculations with the simulated results. The main output and selected secondary outputs are also exported automatically to Microsoft Excel for output display and further processing. Output listing Primary outputs: Societal cost-benefits of various screening strategies under consideration Main secondary output: Averted blindness rate under various screening strategies Other secondary outputs: • Rate of provider compliance with provider-level screening strategy components in a given year • Rate of patient screening compliance in a given year • Number of positive and negative (and number of false or true) test results per disease state and per year • Incidence per disease state (DR and blindness) and per year 98 • Screening costs in a given year • Economic burden from blindness in a given year • Cost-benefit of screening in a given year All outputs are produced for the screening strategies under consideration as well as the case of no screening at all so that cost-benefit analysis of each screening strategy can be performed. 99 CHAPTER 6. MODEL VERIFICATION & VALIDATION 6.1. PRELIMINARY MODEL VERIFICATION Before building the full-fledged model, to ensure that the validity of the disease progression module, a simplified version of model 1 in the context of DR has been built, verified, and validated. This simplified version of model 1 is a macrosimulation with a simplified cost and benefit modules. This will be converted into a microsimulation and the cost and benefit modules will be updated later on to form the full-fledged model 1. This simplified model 1 is used to evaluate the cost-benefit of screening strategies with the following components: Screening strategy components Level Alternatives Status quo Alternative 1 Alternative 2 Screening modality Patient Current (Opthalmoscope) Non-mydriatic camera Mydriatic camera Screening frequency Patient Current (Annual) Biannual - The model is thus used to evaluate the cost-benefit of six different screening strategies on the cohort population. This is for a cohort with advanced starting age (63-64 years), compliance of 65% and a simulation period (and therefore screening program period) of 60 years. 100 The preliminary model verification and validation from this simplified version of model 1 are presented below. To verify the model, the model outputs will be examined to ensure that they are reasonable over a range of input parameters (77) (78). Here, various model outputs from the simplified DR screening simulation model 1 are plotted and checked if their interpretation matches the expected model behavior. This will be done for a range of inputs of all the 6 possible screening strategy and for 2 different population subgroups: insulin- taking and non-insulin-taking. The following two graphs (fig 6.1.1 and 6.1.2) show the percentage of the cohort in each state throughout the simulation period for insulin-taking diabetic patients. The first graph shows the results for annual ophthalmoscope screening, while the second graph shows the results for no screening (natural disease progression). As expected, as shown in fig 6.1.1, screening moves some people from the high-risk to the treated group (thus the decreasing nature of the ‘high-risk’ curve in the screening graph compared to the ‘high-risk’ curve in the no screening graph, which peaks after a while), which results in fewer people becoming blind (as illustrated by the lower peak of the ‘blind’ curve in the screening graph vs the no screening graph). Before screening takes place (i.e. year 0), it is assumed that there is nobody in the treated group, but once screening starts and people are identified for treatment, as expected, this number jumps from 0 to some number (and accordingly the number of people in the ‘high risk’ state drops sharply). After the first year, the number of people being treated changes with a smoother gradient. This can be seen in the shape of the curve of the ‘treated’ state in the state probability graphs for the different screening methods, where the gradient from 0 th to 1 st year is quite steep, but the curve smoothens out after that. 101 In contrast, when there is no screening, as in fig 6.1.2, the number of people in the treated group remains 0 over the duration of the simulation since there is no screening, diagnosis, and therefore treatment. In both graphs, as reasonably expected, the percentage of people in the low-risk group decreases as simulation progresses as patients move to other states, while the percentage of blind people increases from 0 after year 0. The percentage of people in the dead state increases monotonically as expected, and by the end of the simulation period (60 years), almost the entire cohort is in the dead state. Fig 6.1.1. Percentage of insulin-taking cohort in each state for annual ophthalmoscope screening 102 For the non-insulin-taking cohort, the model outputs as given in fig 6.1.3 and 6.1.4 below are also reasonable, with similar interpretations as for the insulin-taking cohort above. Fig 6.1.2. Percentage of insulin-taking cohort in each state for no screening Fig 6.1.3. Percentage of non-insulin-taking cohort in each state for opthalmoscope screening 103 The following two graphs (fig 6.1.5. and 6.1.6) show the number of sight years saved as the simulation progresses for different screening strategies and for different population subgroups respectively. For the same population group and frequency of screening, sight years gained is the highest for mydriatic camera screening, followed by nonmydriatic camera screening and opthalmoscope screening, which is expected since sensitivity and specificity of the screening modality is highest for mydriatic camera, followed by nonmydriatic camera, and opthalmoscope. For the same screening method and for the same population group, annual screening yields more sight years gained than biennial screening. Since patients are screened twice as often in annual screening compared to biennial screening, it is reasonable to expect that more people would be correctly diagnosed and treated, reducing the number of people going blind each year. As expected, as shown in figure 6.1.4, total sight years are higher for the non-insulin taking group than the insulin-taking group. This is due to the much lower prevalence of people in high-risk states Fig 6.1.4. Percentage of non-insulin-taking cohort in each state for no screening 104 who are likely to become blind in the non-insulin taking group compared to those in the insulin- taking group. Accordingly, screening would benefit the insulin-taking group more (as reflected by the higher number of sight years gained), as it has a higher prevalence of high-risk states, since many of the high-risk patients would be diagnosed with screening, thus reducing the probability of going blind. 0 10 20 30 40 50 60 0 5 10 15 20 25 30 35 year sight yrs saved Sight yrs saved Opt Nmc Myd Opt b Nmc b Myd b Fig 6.1.5. Number of sight years saved for different screening strategies 105 6.2. MODELS 2 AND 3 VERIFICATION As model 1 is a subset of model 2, and model 2 is a subset of model 3, and the models were designed to be flexible about the inputs they can take, we expect that models 2 and 3 would yield the same results as model 1 if given the same “lowest common denominator” inputs. In other words, if we only have information that is sufficient to specify model 1 but not the higher-level models, such as knowing only the average compliance rate without knowing the distribution of the compliance in the population, inputting this information into all the models should yield the same results. To verify that the models indeed behave as such, models 1, 2, 3 were used to evaluate the cost- benefit of screening strategies with 2 components that can be evaluated by all models: screening frequency and screening modality. The three models were fed the same inputs, with all patients Fig 6.1.6. Number of sight years saved for different population subgroups 106 having the same screening compliance probability, and all costs attributed to variable costs (i.e. zero fixed costs). The models indeed yield the same results as seen below. Mod ality Scr_ freq Mo del Averted blindness Costs of screening Direct Medical Indirect Medical Direct Non- medical Productivi ty - lost wages Productivi ty - reduced wages 1 1 1 3560.466 19107975 2010973 7597219 12032770 55451137 11044214 2 1 1 4280.094 25600087 2417424 8973774 13763876 68048233 13553180 1 2 1 2727.853 11468843 1540708 5766172 8669427 43573230 8678489 2 2 1 3344.789 15339624 1889158 7179781 10755438 53179407 10591753 1 1 2 3560.466 19107975 2010973 7597219 12032770 55451137 11044214 2 1 2 4280.094 25600087 2417424 8973774 13763876 68048233 13553180 1 2 2 2727.853 11468843 1540708 5766172 8669427 43573230 8678489 2 2 2 3344.789 15339624 1889158 7179781 10755438 53179407 10591753 1 1 3 3560.466 12730198 2010973 7597219 12032770 55451137 11044214 2 1 3 4280.094 18000720 2417424 8973774 13763876 68048233 13553180 1 2 3 2727.853 8041138 1540708 5766172 8669427 43573230 8678489 2 2 3 3344.789 11251813 1889158 7179781 10755438 53179407 10591753 107 6.3. PRELIMINARY MODEL VALIDATION To validate the model, model predictions are compared to past performance of the actual system. In this case, validating the model by comparing the overall cost-benefit for each scenario is difficult due to variability in included costs and the way costs are measured in different studies, as well as the variability in the implementation of the screening strategies used. Instead, model validation is done here for the individual modules. 6.3.1. DISEASE PROGRESSION MODULE VALIDATION For the disease progression module, model validation is done for the health outcomes by comparing the incidence of blindness as predicted in the model to figures of epidemiologic incidences of blindness from the literature. According to a study by Moss, et al. the 4-year incidence of blindness and vision loss in a population with diabetes mellitus is 3.2% and 2.7% among Type II diabetes patients taking insulin and not taking insulin respectively (79). This model predicts a 4-year incidence of blindness of 3.29% and 2.08% among Type II diabetes patients taking insulin and not taking insulin respectively with no screening, which is comparable. A Swedish study reports a 3-year incidence rate of blindness and visual impairment of about 1.6% in a population with diabetes mellitus with an ophthalmological control and screening program in place (80). This model predicts a 3-year incidence rate of blindness and visual impairment of 1.5% 108 and 2.3% respectively for Type II diabetes patients taking insulin and not taking insulin respectively with an annual ophthalmoscope screening in place, which again is comparable. 6.3.2. COMPLIANCE MODULE VALIDATION The modeled average screening compliance rate for eye screening with the status quo screening strategy in California is 68%. In comparison, the literature reports a range of compliance with eye screening guidelines. Baseline findings from the Diabetic Retinopathy Awareness program in New York report 65% of adherence with eye screening guidelines (81), which is comparable to the 68% modeled compliance rate. On average, previous research studies show 60% of diabetic patients in the US following eye screening guidelines (82) (48) (23) (23) (83), which again is comparable. That said, for the underserved population, such as the US urban safety net setting, several studies found annual eye examination rates to be lower than 35% (5) (84) (83). The compliance module thus would need to be adjusted accordingly to fit the population to be studied if the population under study were likely to be significantly different from the general population in terms of compliance, such as for inner-city diabetic patients. 6.3.3. COST MODULE VALIDATION The parameters used for screening costs associated with different modalities of screening are taken from a VA study (ref). For models 1 and 2, the fixed costs were amortized and divided by the average number of patients in a clinic and then added to per patient variable cost to obtain a per 109 patient overall cost. This per patient overall cost, in 2012 dollars, is $33.20 for opthalmoscope screening visit cost, and $49.26 for nonmydriatic camera screening visit cost. This is comparable to the screening visit costs from the WESDR study reported in Dasbach’s paper, which are $34.23 and $53.06 (in 2012 dollars) for nonmydriatic camera screening visit cost. 110 CHAPTER 7. RESULTS 7.1. RESULTS OF MODELS 1-3 Screening strategy components Level Alternatives 1 2 Screening modality Patient Current (Opthalmoscope) Non-mydriatic camera Screening frequency Patient Current (Annual) Biennial (every 2 years) Insurance Patient Current Mandatory health insurance Provision of patient navigator Provider Current (No patient navigator) Provision of patient navigator for clinics with size above a certain threshold The table above lists the screening component alternative referred to by alternatives 1 and 2 in the results tables below. For instance, screening modality = 1 denotes the choice to use of ophthalmoscope screening modality in the screening strategy scenario. 111 7.1.1 RESULTS OF MODEL 1 Table 7.1.1.1 shows the raw outcomes: the costs of screening and treatment, and the number of years of averted blindness (i.e. years of sight years saved) in the cohort after 30 years, and the breakdown of the years of sight years saved into sight years saved when under 65 years of age and when over 65 years of age. The breakdown of the sight years saved into the two age categories facilitate the calculation of the benefit, as nursing home costs and productivity loss are age dependent. Scenario 3 (biennial opthalmoscope screening) results in the lowest amount of averted blindness, as it uses a less sensitive modality and a lower screening frequency. In contrast, the scenario with the most sensitive modality and the highest screening frequency (scenario 2, or annual nonmydriatic camera screening) results in the highest amount of averted blindness. However, the more sensitive modality (nonmydriatic camera, or modality 2) is also the more expensive modality, and screening every year instead of every 2 years incur a higher cost of screening and therefore treatment; accordingly, the costs of screening and treatment is the lowest for scenario 3 and highest for scenario 2. Scenario no Scenario Averted blindness Costs of screening & treatment Modality Scr_freq Total Under 65 Over 65 1 1 1 3714 2549 1164 $14,275,600 2 2 1 4341 2989 1353 $22,374,990 3 1 2 2784 1945 839 $8,895,480 4 2 2 3443 2399 1044 $13,577,702 Table 7.1.1.1 Model 1 results: averted blindness and costs of screening and treatment 112 The following table shows the benefits breakdown. The direct medical and indirect medical benefits include averted medical costs consisting of eye- and non-eye-related medical costs related to blindness, and are therefore increasing in magnitude with increasing years of sight saved. The direct nonmedical costs include averted nursing home costs for blind patients over 65, and therefore are highest for scenarios that result in the highest number of sight years saved when 65 years or older, which is scenario 2 in this case. On the other hand, the benefits of averted lost and reduced wages from productivity loss are increasing in magnitude with increasing years of sight saved below 65 years of age. Scen ario no Scenario Benefits Modal ity Scr_fre q Direct Medical Indirect Medical Direct Nonmedical Productivity - lost wages Productivity - reduced wages 1 1 1 $2,097,491 $8,083,495 $13,083,129 $56,780,816 $11,309,046 2 2 1 $2,451,955 $9,411,815 $15,196,748 $66,569,455 $13,258,651 3 1 2 $1,572,510 $6,001,610 $9,425,924 $43,327,688 $8,629,584 4 2 2 $1,944,475 $7,445,281 $11,729,565 $53,429,785 $10,641,621 Table 7.1.1.2. Model 1 results: breakdown of benefits Table 7.1.1.3 summarizes the results above in terms of net benefit. As the driver of the net benefit from the societal perspective is the benefit from averted productivity loss, which is high in magnitude compared to benefits from other sources and compared to the costs of screening and 113 treatment, the net benefit from the societal perspective is the highest for the scenario that results in the highest sight years saved under 65 years of age, which is scenario 2. However, scenario 2 also is the most expensive out of the 4 alternatives in terms of costs of screening and treatment. As much of the costs of screening and treatment are borne by the medical system while the benefits from the perspective of the medical system (consisting of the averted direct and indirect medical costs) are comparatively low, scenario 2 also results in the lowest net benefit from the medical system perspective. On the other hand, the scenario with the lowest outlay in terms of costs of screening and treatment, scenario 3, results in the highest net benefit from the medical system perspective. From the medical system perspective, none of the scenarios result in a positive net benefit. Unlike the societal perspective where the net benefit is driven by averted productivity loss, and the medical system perspective where the net benefit is driven by costs of screening and treatment, the Medicare/Medicaid perspective does not have a single driver of net benefit, as no single component of the cost or benefit component in this perspective has a magnitude that dwarves other components. From the Medicare / Medicaid perspective, the status quo (scenario 1, i.e. annual ophthalmoscope screening) results in the highest net benefit, and scenario 2 (annual nonmydriatic camera screening) results in the lowest net benefit. 114 Scenario no Scenario Net benefit Modality Scr_freq Medical system perspective Medicare / Medicaid perspective Societal perspective 1 1 1 -$4,094,614 $8,988,515 $77,078,377 2 2 1 -$10,511,220 $4,685,528 $84,513,634 3 1 2 -$1,321,359 $8,104,565 $60,061,836 4 2 2 -$4,187,946 $7,541,619 $71,613,024 Table 7.1.1.3. Model 1 results: net benefit 7.1.2. RESULTS OF MODEL 2 As shown in table 7.1.2.1, for the first 4 scenarios that can also be evaluated by model 1, the results show similar trends as the results of model 1, with scenario 3 resulting in the highest averted blindness and costs of screening and treatment, while scenario 3 results in the lowest averted blindness and also lowest costs of screening and treatment. Model 2 allows the evaluation of four additional scenarios compared to model 1. Scenarios 5-8 evaluate scenarios 1-4 with one additional component of health insurance; scenarios 1-4 consider status quo insurance, while scenarios 5-8 consider mandatory health insurance as a patient-level intervention to increase screening compliance. As mandatory health insurance will increase screening rates across the board, scenarios 5-8 each have higher averted blindness and also higher costs of screening and treatment than each of their counterparts in scenarios 1-4 where there is no health insurance intervention. Thus, out of the 8 scenario alternatives, the alternative with the 115 highest averted blindness as well as the highest of cost of screening and treatment is scenario 6 (annual nonmydriatic camera screening with mandatory health insurance), the counterpart of scenario 2 with an additional intervention in the form of mandatory health insurance. Scenario no Scenario Averted blindness Costs of screening & treatment Modality Scr_freq Insurance Total Under 65 Over 65 1 1 1 1 3301 2172 1129 $13,423,416 2 2 1 1 3912 2583 1329 $21,020,459 3 1 2 1 2393 1579 814 $8,289,272 4 2 2 1 2991 1966 1025 $12,678,415 5 1 1 2 3440 2280 1160 $13,869,158 6 2 1 2 4057 2698 1359 $21,741,748 7 1 2 2 2520 1679 840 $8,583,082 8 2 2 2 3132 2079 1053 $13,120,004 Table 7.1.2.1 Model 2 results: averted blindness and costs of screening and treatment Accordingly, as seen in table 7.1.2.2, the magnitude of the benefits reflect the magnitude of the averted blindness in the appropriate age categories. Thus, scenario 6 results in the highest benefit components across the board. 116 Sce nar io no Scenario Benefits Mo dali ty Scr _fr eq Ins ura nce Direct Medical Indirect Medical Direct Nonmedical Productivity - lost wages Productivity - reduced wages Total 1 1 1 1 $1,864,587 $7,528,602 $12,682,893 $48,389,377 $9,637,722 $80,103,181 2 2 1 1 $2,209,703 $8,844,092 $14,936,073 $57,532,722 $11,458,803 $94,981,394 3 1 2 1 $1,351,665 $5,469,403 $9,142,949 $35,179,365 $7,006,681 $58,150,063 4 2 2 1 $1,689,122 $6,831,931 $11,517,671 $43,779,680 $8,719,607 $72,538,011 5 1 1 2 $1,943,132 $7,765,593 $13,033,827 $50,791,202 $10,116,093 $83,649,846 6 2 1 2 $2,291,237 $9,092,364 $15,268,983 $60,088,146 $11,967,768 $98,708,499 7 1 2 2 $1,423,223 $5,668,494 $9,443,211 $37,406,084 $7,450,177 $61,391,189 8 2 2 2 $1,769,043 $7,066,364 $11,832,717 $46,306,920 $9,222,958 $76,198,003 Table 7.1.2.2. Model 2 results: breakdown of benefits Consequently, scenario 6 also results in the highest net benefit from the societal perspective, and conversely the lowest net benefit from the medical system perspective. Scenario 3 remains the scenario with the highest net benefit as it remains the one with the lowest cost of screening and treatment. From the Medicare / Medicaid perspective, scenario 5 is the one with the highest net benefit, and scenario 6 is the one with the lowest net benefit. 117 Scenario no Scenario Net-benefit Modality Scr_freq Insurance Medical system perspective Medicare / Medicaid perspective Societal perspective 1 1 1 1 -$4,030,227 $8,652,666 $66,679,765 2 2 1 1 -$9,966,664 $4,969,408 $73,960,934 3 1 2 1 -$1,468,204 $7,674,745 $49,860,791 4 2 2 1 -$4,157,362 $7,360,309 $59,859,596 5 1 1 2 -$4,160,433 $8,873,393 $69,780,689 6 2 1 2 -$10,358,147 $4,910,836 $76,966,750 7 1 2 2 -$1,491,365 $7,951,846 $52,808,107 8 2 2 2 -$4,284,597 $7,548,120 $63,077,999 Table 7.1.2.3. Model 2 results: net-benefit 7.1.3. RESULTS OF MODEL 3 As seen in table 7.1.3.1, model 3 can evaluate 8 additional scenarios compared to model 2, as we add one more screening strategy component, a provider-level intervention to improve screening compliance, the hiring of patient navigator. As hiring a patient navigator (i.e. navigator setting = 2) increases screening compliance rates, this also increases the averted blindness across the board, as well as the costs of screening and treatment. Thus, scenario 14, which is the counterpart of scenario 6 in model 2, but with the additional hiring of patient navigator as part of the screening strategy, is now the scenario with the highest averted blindness and also highest costs of screening and 118 treatment. Accordingly, as shown in table 7.1.3.2, scenario 14 also has the highest benefits across the board. Scenario no Scenario Averted blindness Costs of screening & treatment Modal ity Scr_fr eq Insura nce Navigator setting Total Under 65 Over 65 1 1 1 1 1 3301 2172 1129 $13,283,721 2 2 1 1 1 3912 2583 1329 $20,553,707 3 1 2 1 1 2393 1579 814 $8,584,401 4 2 2 1 1 2991 1966 1025 $13,698,719 5 1 1 2 1 3440 2280 1160 $13,693,230 6 2 1 2 1 4057 2698 1359 $21,151,104 7 1 2 2 1 2520 1679 840 $8,859,053 8 2 2 2 1 3132 2079 1053 $14,074,791 9 1 1 1 2 3645 2436 1209 $14,633,201 10 2 1 1 2 4246 2845 1401 $22,535,558 11 1 2 1 2 2749 1848 901 $9,516,993 12 2 2 1 2 3369 2258 1111 $14,988,109 13 1 1 2 2 3770 2540 1230 $14,979,307 14 2 1 2 2 4355 2936 1419 $23,039,700 15 1 2 2 2 2842 1924 918 $9,751,752 16 2 2 2 2 3478 2346 1132 $15,316,605 Table 7.1.3.1 Model 3 results: averted blindness and costs of screening and treatment 119 Sce nari o no Scenario Benefits Mo dali ty Sc r_ fr eq Ins ura nce Na vig ato r Direct Medical Indirect Medical Direct Nonmedical Productivity - lost wages Productivity - reduced wages 1 1 1 1 1 $1,864,587 $7,528,602 $12,682,893 $48,389,377 $9,637,722 2 2 1 1 1 $2,209,703 $8,844,092 $14,936,073 $57,532,722 $11,458,803 3 1 2 1 1 $1,351,665 $5,469,403 $9,142,949 $35,179,365 $7,006,681 4 2 2 1 1 $1,689,122 $6,831,931 $11,517,671 $43,779,680 $8,719,607 5 1 1 2 1 $1,943,132 $7,765,593 $13,033,827 $50,791,202 $10,116,093 6 2 1 2 1 $2,291,237 $9,092,364 $15,268,983 $60,088,146 $11,967,768 7 1 2 2 1 $1,423,223 $5,668,494 $9,443,211 $37,406,084 $7,450,177 8 2 2 2 1 $1,769,043 $7,066,364 $11,832,717 $46,306,920 $9,222,958 9 1 1 1 2 $2,058,655 $8,176,734 $13,586,147 $54,252,058 $10,805,393 10 2 1 1 2 $2,398,314 $9,463,354 $15,742,223 $63,372,676 $12,621,948 11 1 2 1 2 $1,552,924 $6,157,367 $10,125,483 $41,168,436 $8,199,525 12 2 2 1 2 $1,902,704 $7,558,686 $12,483,761 $50,287,347 $10,015,741 13 1 1 2 2 $2,129,538 $8,366,197 $13,825,122 $56,573,683 $11,267,791 14 2 1 2 2 $2,459,742 $9,633,669 $15,945,620 $65,391,938 $13,024,125 15 1 2 2 2 $1,605,044 $6,310,342 $10,313,408 $42,851,327 $8,534,707 16 2 2 2 2 $1,964,210 $7,741,038 $12,718,068 $52,248,400 $10,406,324 Table 7.1.3.2 Model 3 results: benefits breakdown 120 Scenario no Scenario Cost-benefit Modality Scr_freq Insurance Navigator setting Medical system perspective Medicare / Medicaid perspective Societal perspective 1 1 1 1 1 -$3,890,532 $8,792,361 $66,819,460 2 2 1 1 1 -$9,499,912 $5,436,161 $74,427,686 3 1 2 1 1 -$1,763,332 $7,379,617 $49,565,662 4 2 2 1 1 -$5,177,666 $6,340,005 $58,839,292 5 1 1 2 1 -$3,984,506 $9,049,321 $69,956,616 6 2 1 2 1 -$9,767,503 $5,501,480 $77,557,394 7 1 2 2 1 -$1,767,336 $7,675,875 $52,532,136 8 2 2 2 1 -$5,239,384 $6,593,333 $62,123,212 9 1 1 1 2 -$4,397,812 $9,188,336 $74,245,786 10 2 1 1 2 -$10,673,890 $5,068,333 $81,062,958 11 1 2 1 2 -$1,806,702 $8,318,781 $57,686,743 12 2 2 1 2 -$5,526,719 $6,957,042 $67,260,130 13 1 1 2 2 -$4,483,572 $9,341,550 $77,183,025 14 2 1 2 2 -$10,946,290 $4,999,330 $83,415,394 15 1 2 2 2 -$1,836,366 $8,477,042 $59,863,076 16 2 2 2 2 -$5,611,358 $7,106,710 $69,761,434 Table 7.1.3.3 Model 3 results: net-benefits 121 Consequently, as seen in table 7.1.3.3, scenario 14 has the highest net benefit from the societal perspective and the lowest net benefit from the medical system perspective, while scenario 3 remains the scenario with the lowest net benefit from the societal perspective but the highest net benefit from the medical system perspective. From the Medicare/Medicaid perspective, scenario 14 has the lowest net benefit, while scenario 13 has the highest net benefit. The figure below illustrates the net benefits for the 16 scenarios for the three perspectives in graphical format. 7.2. SENSITIVITY ANALYSIS Sensitivity analysis is performed to examine how uncertainties in some model parameters or inputs affect the outputs by running model 3 with different values of the model parameters that are -20 0 20 40 60 80 100 1 2 4 5 6 7 8 9 10 11 12 13 14 15 16 Millions Screening strategy scenario Model 3: Net-benefit from different perspectives Societal perspective Medicare / Medicaid perspective Medical system perspective 122 subject to uncertainty. The analysis done here is an OFAT (one factor at a time) sensitivity analysis (i.e. only one parameter / factor is varied at a time while keeping others at their baseline values) to see what effect changes in each variable / parameter have on the output (85). The list of parameters varied in the sensitivity analysis and the lower and upper bounds explored for each parameter are given in the table below. This section highlights the important results in the sensitivity analysis. The full results of the sensitivity analysis is given in the appendix. Parameter to vary in sensitivity analysis Lower bound Upper bound Years to adjust to blindness 1 year 5 years Patient compliance model parameters 95% confidence level lower bound estimates for all regression coefficients 95% confidence level upper bound estimates for all regression coefficients Patient navigator model parameter 95% confidence level lower bound for the odds ratio estimate 95% confidence level upper bound for the odds ratio estimate Percentage of eligible clinics that implement the screening strategy component 60% 100% Clinic size threshold for patient navigator implementation 1000 3000 Discount factor 2% 8% Table 7.2.1. Parameters varied in the sensitivity analysis 123 One of the model parameters that is associated with the most uncertainty due to the lack of reliable data is the number of years for a patient who recently became blind to adjust to his/her blindness (denoted as YrAdjBlind in the table). This parameter is used in the calculation to estimate the indirect medical costs; the longer it takes for a patient to adjust to being blind, the more he/she incurs in indirect medical costs due to secondary complications of blindness such as higher rates of injuries and depression. The lower and upper bounds of number of years to adjust to blindness tested below are 1 and 5. As we can see in table 7.2.2, the net benefit for the medical system and Medicare/Medicaid perspectives are very sensitive to changes in this parameter. For the medical system perspective, using the baseline value for this parameter results in no scenario worth considering as the net benefits are all negative; however, if it takes 5 years to adjust to blindness, scenarios where the cheaper screening modality (ophthalmoscope) is used become cost-beneficial. For the Medicare/Medicaid perspective, if it takes only one year to adjust to blindness, the net benefit of some scenarios end up being negative and not worth implementing. 124 Scen ario no Relevant parameters Net-benefit (medical system perspective) / million $ Net-benefit (Medicare / Medicaid perspective) / million $ Net-benefit (societal perspective) / million $ YrAdjBli nd=1 YrAdjBli nd=3 YrAdjBli nd=5 YrAdjBli nd=1 YrAdjBli nd=3 YrAdjBli nd=5 YrAdjBli nd=1 YrAdjBli nd=3 YrAdjBli nd=5 1 -8.785 -3.891 0.535 3.898 8.792 13.218 61.925 66.819 71.245 2 -15.250 -9.500 -4.299 -0.314 5.436 10.637 68.678 74.428 79.629 3 -5.319 -1.763 1.448 3.824 7.380 10.591 46.010 49.566 52.777 4 -9.619 -5.178 -1.164 1.899 6.340 10.354 54.398 58.839 62.853 5 -9.034 -3.985 0.581 4.000 9.049 13.615 64.907 69.957 74.522 6 -15.680 -9.768 -4.421 -0.411 5.501 10.848 71.645 77.557 82.904 7 -5.452 -1.767 1.560 3.991 7.676 11.004 48.847 52.532 55.860 8 -9.833 -5.239 -1.088 1.999 6.593 10.744 57.529 62.123 66.274 9 -9.713 -4.398 0.411 3.873 9.188 13.997 68.930 74.246 79.055 10 -16.827 -10.674 -5.107 -1.084 5.068 10.636 74.910 81.063 86.630 11 -5.809 -1.807 1.809 4.317 8.319 11.935 53.685 57.687 61.303 12 -10.440 -5.527 -1.085 2.044 6.957 11.399 62.347 67.260 71.702 13 -9.923 -4.484 0.436 3.902 9.342 14.261 71.744 77.183 82.103 14 -17.210 -10.946 -5.280 -1.264 4.999 10.666 77.152 83.415 89.082 15 -5.938 -1.836 1.868 4.376 8.477 12.182 55.762 59.863 63.568 16 -10.643 -5.611 -1.064 2.075 7.107 11.654 64.729 69.761 74.309 Table 7.2.2. Sensitivity analysis results from varying the parameter “years to adjust to blindness” 125 The simulation is also sensitive to the compliance parameter. The following sensitivity analysis uses 95% lower bound estimates for all the coefficient estimates of the regression model that make up the patient screening compliance model, and then does the same using the 95% upper bound estimates for all the regression coefficient estimates. While the general trends do not change (i.e. the net benefit for the medical system perspective is still all negative, and the scenarios with the highest and lowest net benefits remain the same for all perspectives), the magnitudes of the estimates differ from the baseline estimates by as much as 50%. In the future, uncertainty in the compliance model should be incorporated by “perturbing” the coefficients instead of using point estimates. Bootstrapping could also then be used to establish confidence intervals of the outcome estimates that reflect the level of uncertainty in the estimation of the model parameters. 126 Scenario no Relevant outcomes Net benefits (medical system perspective) / million $ Net benefits (Medicare / Medicaid perspective) / million $ Net benefits (societal perspective) / million $ Lower bound Default Upper bound Lower bound Default Upper bound Lower bound Default Upper bound 1 -2.723 -3.891 -4.999 6.288 8.792 9.532 37.464 66.819 82.495 2 -6.658 -9.500 -12.047 4.119 5.436 4.556 43.775 74.428 88.285 3 -1.696 -1.763 -1.912 4.019 7.380 9.137 23.579 49.566 67.113 4 -4.471 -5.178 -5.938 2.926 6.340 7.648 28.691 58.839 77.884 5 -2.715 -3.985 -5.138 6.774 9.049 9.581 40.632 69.957 83.563 6 -6.764 -9.768 -12.311 4.517 5.501 4.442 47.235 77.557 89.338 7 -1.669 -1.767 -1.940 4.398 7.676 9.327 25.914 52.532 68.408 8 -4.501 -5.239 -6.005 3.266 6.593 7.836 31.241 62.123 79.578 9 -3.233 -4.398 -5.331 7.424 9.188 9.507 48.690 74.246 84.237 10 -7.881 -10.674 -12.671 4.848 5.068 4.218 55.321 81.063 90.199 11 -1.737 -1.807 -2.005 5.469 8.319 9.422 33.549 57.687 69.492 12 -4.785 -5.527 -6.177 4.451 6.957 7.794 40.348 67.260 80.982 13 -3.340 -4.484 -5.395 7.618 9.342 9.663 50.936 77.183 85.642 14 -8.113 -10.946 -12.837 4.884 4.999 4.253 58.168 83.415 91.309 15 -1.717 -1.836 -1.984 5.769 8.477 9.672 35.866 59.863 71.321 16 -4.822 -5.611 -6.186 4.659 7.107 8.003 42.971 69.761 82.828 Table 7.2.3. Sensitivity analysis results from using lower bound estimates and upper bound estimates of the compliance model 127 Varying the discount factor used in the financial discounting of the cost and benefit estimates from 5% to 2% and 8% has roughly the same magnitude of effects as using the lower and upper bound estimates of the compliance model above. The general trends as to which scenarios are deemed least and most beneficial also do not change when the discount factors are changed. Sensitivity analysis for the implementation module parameters is done for three different parameters. In each case, the effect on net benefit is relatively minor. The results of the sensitivity analysis for these parameters can be found in the appendix. 7.3. MODEL COMPARISON 7.3.1. INTRINSIC DIFFERENCES BETWEEN THE THREE MODELS AND EXPECTED EFFECTS ON OUTCOMES The three models are built with increasingly complex model specifications to illustrate the benefit of specifying a more realistic compliance model (in model 2) and a multi-level structure with a more realistic cost structure (in model 3). The benefits are twofold: it allows for the evaluation of a larger range of scenarios, specifically those involving aimed at modifying behavior to increase utilization of preventive services, as seen in section2 7.1-7.3 above, and it also improves precision of outcome estimates, as will be shown in this chapter. 128 Table 7.3.1.1 summarizes the differences in model specifications between the three models, and table 7.3.1.2 summarizes the expected effects of these differences on functionality and precision of outcome estimates. Model specification Model 1 Model 2 Model 3 Compliance Average compliance, every patient has the same probability of compliance Each patient has his/her own baseline probability of compliance, based on patient characteristics Baseline compliance probability may be modified by patient-level screening component strategies to improve compliance Each patient has his/her own baseline probability of compliance, based on patient characteristics Baseline compliance probability may be modified by patient-level screening component strategies to improve compliance & provider-level screening component strategies Cost Costs are all variable Costs are all variable Costs have fixed & variable components Table 7.3.1.1. Differences in model specifications between models 1, 2, and 3 129 Effect on Model 1 Model 2 Model 3 Functionality Cannot evaluate screening strategies that intends to improve patient compliance at the patient- and provider-level Cannot evaluate screening strategies that intends to improve patient compliance at the provider-level Can evaluate screening strategies that intends to improve patient compliance at the provider-level Precision of estimates: Averted blindness Overestimates averted blindness as it ignores correlation of compliance with other parameters Provides more precise estimates of averted blindness Precision of estimates: Screening cost Underestimates true screening cost for biennial screening as fixed cost is still incurred during years of non-screening Ignores economy of scale in estimation of screening cost Provides more precise estimates of screening cost Estimates of screening cost takes into account economy of scale Table 7.3.1.2. Expected effects of model specifications differences on outcomes 130 7.3.2. EFFECTS OF MODEL DIFFERENCES IN PRECISION OF OUTCOME ESTIMATES This section will list the percentage differences in the outcomes of the different models, illustrating how the different model specifications, in terms of compliance model and cost structure in particular, affect the precision of estimates of the model outcomes. The comparison will only be performed for scenarios that can be evaluated by both models being compared. 7.3.2.1. Model comparison: Model 1 vs model 2 The figures below illustrates aspects of differences in the compliance model specification between model 1 and model 2. Probability of screening compliance is correlated with some factors that influence outcomes, such as age and insulin use. Figure 7.3.2.1.1 plots the average probability of screening compliance of the population at each age group at the beginning of the simulation. For model 2, we can see that there is a positive correlation between age and probability of screening compliance, with older patients more likely to comply with screening. For model 1, everyone in the simulation has the same probability of screening compliance regardless of age (i.e. the probability of screening compliance is not age dependent. This non-age-dependent probability of screening compliance in model 1 is specified to be the overall mean of probability of screening compliance in model 2. Figure 7.3.2.1.2 shows the average probability of screening compliance of the population for those on insulin and those not on insulin at the beginning of the simulation. For model 2, we can see that there is a negative correlation between insulin use and probability of screening compliance, while 131 again, for model 1, everyone in the simulation has the same probability of screening compliance regardless of insulin use. Figure 7.3.2.1.1. Graphs of probability of screening compliance vs age for models 1 and 2 Figure 7.3.2.1.2. Graphs of probability of screening compliance vs insulin use for models 1 and 2 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 18 21 24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62 64 66 68 70 72 74 76 78 80 82 84 Average probability of screening compliance Age Probability of screening compliance vs age Model 2 Compliance Model 1 Compliance 0.64 0.66 0.68 0.7 0.72 0.74 0.76 0.78 0.8 No Yes Average probability of screening compliance On insulin Probability of screening compliance vs being on insulin for models 1 and 2 Model 2 Model 1 132 Table 7.3.2.1 shows the percentage differences in model outcomes between models 1 and 2. In all scenarios, model 1 overestimates the averted blindness by 10-14%. This overestimation is magnified when using a screening strategy scenario that uses less sensitive modality or lower screening frequency. There are two major reasons related to the difference in the compliance model structure that leads to the overestimation. First, since in model 1 the probability of screening compliance is the same regardless of patient’s age, as we can see in figure 7.3.2.1, model 1 ends up overestimating the probabilities of screening compliance of the younger patients and underestimating those of the older patients. As a result, younger patients are more likely to undergo screening and thus have their blindness averted in model 1 than in model 2, while older patients are more likely to undergo screening and have their blindness averted in model 2 than model 1. However, since younger patients would live longer and have more years of averted blindness than the older patients, the number of sight years saved in model 1 ends up higher than it should be due to the inaccurate specification of compliance model. The second major reason leading to the overestimation is the correlation of screening compliance with diabetes severity (with insulin use as an indicator). Patients on insulin are less likely to go for screening than those not on insulin (perhaps due to lower functional status), but they are also at a higher risk of developing DR and blindness. As seen in figure 7.3.2.2, in model 1, as all patients have the same probability of screening compliance regardless of disease severity, the model ends up inflating the screening compliance probability of patients on insulin and deflating the screening compliance probability of patients not on insulin. Again, this has the result of overestimating the resulting averted blindness since patients on insulin who are more likely to have DR are more likely to get screened and thus have their DR treated in model 1 than in model 2. This is also the reason 133 the costs of screening are overestimated in model 1 by about 6-7%: in model 1, patients on insulin are more likely to go for screening, but such patients are also more likely to have DR and thus to require treatment, thus incurring higher costs of treatment overall than in model 2. As the precision of estimates that is most affected by the model specification differences in this case is averted blindness, in terms of net benefit, this also results in the highest impact from the societal perspective, with differences between estimates of model 1 and model 2 ranging from 12–17%. Scenario no Scenario Averted blindness (% difference) Costs of screening & treatment (% difference) Net benefit (% difference) Modality Scr_freq Medical system perspective Medicare / Medicaid perspective Societal perspective 1 1 1 -11.10 -5.97 -1.57 -3.74 -13.49 2 2 1 -9.88 -6.05 -5.18 6.06 -12.49 3 1 2 -14.04 -6.81 11.11 -5.30 -16.98 4 2 2 -13.13 -6.62 -0.73 -2.40 -16.41 Table 7.3.2.1. Percentage difference in outcomes of model 1 vs model 2 In model 2 above, the compliance model considered is dynamic, in that patient’s probability of compliance increases (i.e. does not remain static) as they age, reflecting patient’s tendency to adhere to screening guidelines more as they age (this has been shown in different screening cases beyond DR screening, both in cross-sectional (93) and longitudinal studies (94)). The dynamicity of the compliance model also has an effect on the compliance and thus outcomes. Figure 7.3.2.1.3 134 below shows the effect this has on the number of patients undergoing screening over the years in the simulation for one repetition of the simulation. Both cases start with the same number of people undergoing screening which declines with passing simulation years as patients die in the simulation. However, in the static compliance model, patients retain the compliance probability that they start with in the simulation throughout the simulation. As the older patients who start out with a higher probability of screening compliance die in the simulation, the simulation is left with a pool of patients that are less likely to undergo screening than the one it starts with. In contrast, with the dynamic compliance model, as the older patients die in the simulation, the remaining patients left in the simulation has an increased probability of compliance as they age. Consequently, the case using the static compliance model declines at a lower rate. Figure 7.3.2.1.3. Number of patients undergoing screening over the years of the simulation with static vs dynamic compliance model for one repetition of the simulation This difference in the compliance rate of screening using the two models consequently impact the outcomes. As fewer patients comply with screening in the case of the static model, the averted years 135 of blindness is lower too (as shown in the figure 7.3.2.1.4 below), affecting the costs and benefits. Thus, the dynamicity of the compliance model is another important consideration in the model specification. Figure 7.3.2.1.4. Sight years saved over the years of the simulation with static vs dynamic compliance model for one repetition of the simulation 7.3.2.2. Model comparison: Model 2 vs model 3 Both models 2 and 3 have the same compliance model specifications, thus there is no difference in the estimates of the averted blindness. However, the simplified cost structure in model 2 compared to model 3 results in differences in the estimates of costs of screening and treatment between 136 model 2 and 3. For scenarios involving annual screening frequency the differences are minor (1%- 2% difference). These differences arise from the consideration of economy of scale in the cost structure, wherein the fixed costs associated with clinic space and modality are incurred per clinic, and thus screening is more economical in clinics that serve more diabetic patients (assuming capacity constraint has not been reached). Whether this results in under- or overestimation of the costs of screening and treatment and the magnitude of the difference in cost estimates depend on the distribution of patients among the clinics in the population considered. However, for scenarios involving biennial screening frequency, the difference in the costs of screening and treatment estimate is as large as 8%. Model 2 underestimates the costs of screening and treatment in the case of scenarios involving screening frequency of every 2 years because it considers that all costs are variable, and that screening costs are only incurred in years of screening, even though fixed costs (such as clinic space rental) are still incurred in years of non- screening. The underestimation is especially high for scenarios 4 and 8 that involve a more expensive modality (nonmydriatic camera). In model 3, amortization expense of the camera is recorded each year regardless of whether patients come in for screening, but in model 2 this is only done in screening years. As the net benefit from the medical system perspective is driven by the costs of screening and treatment, the differences in the precision of estimates of costs of screening and treatment have the largest impact on the net benefit estimates from the medical system perspective, with differences of estimates ranging from 3-25%. The net benefit from the societal perspective is barely affected as its main driver is the number of sight years saved and thus averted productivity loss. 137 Scenari o no Scenario Averted blindness (% differenc e) Costs of screening & treatmen t (% differenc e) Net benefit (% difference) Modalit y Scr_fre q Insuranc e Medical system perspectiv e Medicare / Medicaid perspectiv e Societal perspectiv e 1 1 1 1 0.00 -1.04 -3.47 1.61 0.21 2 2 1 1 0.00 -2.22 -4.68 9.39 0.63 3 1 2 1 0.00 3.56 20.10 -3.85 -0.59 4 2 2 1 0.00 8.05 24.54 -13.86 -1.70 5 1 1 2 0.00 -1.27 -4.23 1.98 0.25 6 2 1 2 0.00 -2.72 -5.70 12.03 0.77 7 1 2 2 0.00 3.22 18.50 -3.47 -0.52 8 2 2 2 0.00 7.28 22.28 -12.65 -1.51 Table 7.3.2.2. Table of percentage differences in outcomes of model 2 and model 3 7.3.2.3. Model comparison: Model 1 vs model 3 As models 1 and 3 are the most differences in model specification, the differences in estimates between model 1 and model 3 are also more manifold than between model 1 and 2 or between model 2 and 3. The differences of net benefit estimates between model 1 and 3 range from 5-33% for the medical system perspective, 2-16% for the Medicare / Medicaid perspective, and 12-18% for the societal perspective. 138 Scenario no Scenario Averted blindness (% difference) Costs of screening (% difference) Net benefit (% difference) Modality Scr_freq Medical system perspective Medicare / Medicaid perspective Societal perspective 1 1 1 -11.10 -6.95 -4.98 -2.18 -13.31 2 2 1 -9.88 -8.14 -9.62 16.02 -11.93 3 1 2 -14.04 -3.50 33.45 -8.94 -17.48 4 2 2 -13.13 0.89 23.63 -15.93 -17.84 Table 7.3.2.3. Table of percentage differences in outcomes of model 1 and model 3 7.4. POLICY ANALYSIS AND CONTEXT-BASED DECISION MAKING 7.4.1. INCREMENTAL ANALYSIS: CHOOSING FROM THE POOL OF SCREENING STRATEGIES As there are multiple alternatives of screening strategies to choose from, incremental analysis is performed to compare the alternatives. First, the scenarios are ordered from lowest to highest cost, as shown below. 139 Scenario no Scenario Cost of screening Modality Screening frequency Insurance Navigator setting 3 1 2 1 1 $8,584,400.71 7 1 2 2 1 $8,859,052.69 11 1 2 1 2 $9,516,993.01 15 1 2 2 2 $9,751,752.32 1 1 1 1 1 $13,283,721.25 5 1 1 2 1 $13,693,230.00 4 2 2 1 1 $13,698,718.76 8 2 2 2 1 $14,074,791.50 9 1 1 1 2 $14,633,200.82 13 1 1 2 2 $14,979,306.87 12 2 2 1 2 $14,988,108.99 16 2 2 2 2 $15,316,604.79 2 2 1 1 1 $20,553,707.39 6 2 1 2 1 $21,151,104.48 10 2 1 1 2 $22,535,557.93 14 2 1 2 2 $23,039,699.91 Incremental analysis is then performed with sight years saved as the effectiveness measure by first applying principles of strong dominance followed by principles of extended dominance (86) (87). A scenario of no screening (denoted by NS in the tables below) is included as an alternative. 140 1 Scenario Cost Effectiveness ICER ($/sight rys saved) Decision NS 0 0 3 $8,584,401 2,393 3587.074 weakly dominated 7 $8,859,053 2,520 2167.832 11 $9,516,993 2,749 2865.126 15 $9,751,752 2,842 2543.978 1 $13,283,721 3,301 7686.130 5 $13,693,230 3,440 2944.717 4 $13,698,719 2,991 strongly dominated 8 $14,074,791 3,132 strongly dominated 9 $14,633,201 3,645 4595.635 13 $14,979,307 3,770 2757.809 12 $14,988,109 3,369 strongly dominated 16 $15,316,605 3,478 strongly dominated 2 $20,553,707 3,912 39274.517 6 $21,151,104 4,057 4138.317 10 $22,535,558 4,246 7302.707 14 $23,039,700 4,355 4635.402 After the first pass, scenarios 4, 8, 12, and 16 are strongly dominated (scenario 5 results in better benefits for less cost than scenarios 4 and 8; scenario 13 results in better benefits for less cost than scenarios 12 and 16). In addition, scenario 3 is weakly dominated (it is less costly than scenario 7, 141 but has a higher ICER, and a more effective scenario with a lower ICER is preferred by the decision maker). In the subsequent passes, we’d iteratively eliminate the weakly dominated scenarios from the analysis starting from the scenario with the least cost. 2 Scenario Cost Effectiveness ICER ($/sight rys saved) Decision NS 0 0 3 $8,584,401 2,393 weakly dominated 7 $8,859,053 2,520 3515.716 weakly dominated 11 $9,516,993 2,749 2865.126 15 $9,751,752 2,842 2543.978 1 $13,283,721 3,301 7686.130 5 $13,693,230 3,440 2944.717 4 $13,698,719 2,991 strongly dominated 8 $14,074,791 3,132 strongly dominated 9 $14,633,201 3,645 4595.635 13 $14,979,307 3,770 2757.809 12 $14,988,109 3,369 strongly dominated 16 $15,316,605 3,478 strongly dominated 2 $20,553,707 3,912 39274.517 6 $21,151,104 4,057 4138.317 10 $22,535,558 4,246 7302.707 14 $23,039,700 4,355 4635.402 142 3 Scenario Cost Effectiveness ICER ($/sight rys saved) Decision NS 0 0 3 $8,584,401 2,393 weakly dominated 7 $8,859,053 2,520 weakly dominated 11 $9,516,993 2,749 3461.379 weakly dominated 15 $9,751,752 2,842 2543.978 1 $13,283,721 3,301 7686.130 5 $13,693,230 3,440 2944.717 4 $13,698,719 2,991 strongly dominated 8 $14,074,791 3,132 strongly dominated 9 $14,633,201 3,645 4595.635 13 $14,979,307 3,770 2757.809 12 $14,988,109 3,369 strongly dominated 16 $15,316,605 3,478 strongly dominated 2 $20,553,707 3,912 39274.517 6 $21,151,104 4,057 4138.317 10 $22,535,558 4,246 7302.707 14 $23,039,700 4,355 4635.402 143 4 Scenario Cost Effectiveness ICER ($/sight rys saved) Decision NS 0 0 3 $8,584,401 2,393 weakly dominated 7 $8,859,053 2,520 weakly dominated 11 $9,516,993 2,749 weakly dominated 15 $9,751,752 2,842 3431.588 1 $13,283,721 3,301 7686.130 weakly dominated 5 $13,693,230 3,440 2944.717 4 $13,698,719 2,991 strongly dominated 8 $14,074,791 3,132 strongly dominated 9 $14,633,201 3,645 4595.635 13 $14,979,307 3,770 2757.809 12 $14,988,109 3,369 strongly dominated 16 $15,316,605 3,478 strongly dominated 2 $20,553,707 3,912 39274.517 6 $21,151,104 4,057 4138.317 10 $22,535,558 4,246 7302.707 14 $23,039,700 4,355 4635.402 144 5 Scenario Cost Effectiveness ICER ($/sight rys saved) Decision NS 0 0 3 $8,584,401 2,393 weakly dominated 7 $8,859,053 2,520 weakly dominated 11 $9,516,993 2,749 weakly dominated 15 $9,751,752 2,842 3431.588 1 $13,283,721 3,301 weakly dominated 5 $13,693,230 3,440 6584.597 weakly dominated 4 $13,698,719 2,991 strongly dominated 8 $14,074,791 3,132 strongly dominated 9 $14,633,201 3,645 4595.635 13 $14,979,307 3,770 2757.809 12 $14,988,109 3,369 strongly dominated 16 $15,316,605 3,478 strongly dominated 2 $20,553,707 3,912 39274.517 6 $21,151,104 4,057 4138.317 10 $22,535,558 4,246 7302.707 14 $23,039,700 4,355 4635.402 145 6 Scenario Cost Effectiveness ICER ($/sight rys saved) Decision NS 0 0 3 $8,584,401 2,393 weakly dominated 7 $8,859,053 2,520 weakly dominated 11 $9,516,993 2,749 weakly dominated 15 $9,751,752 2,842 3431.588 1 $13,283,721 3,301 weakly dominated 5 $13,693,230 3,440 weakly dominated 4 $13,698,719 2,991 strongly dominated 8 $14,074,791 3,132 strongly dominated 9 $14,633,201 3,645 6078.059 weakly dominated 13 $14,979,307 3,770 2757.809 12 $14,988,109 3,369 strongly dominated 16 $15,316,605 3,478 strongly dominated 2 $20,553,707 3,912 39274.517 6 $21,151,104 4,057 4138.317 10 $22,535,558 4,246 7302.707 14 $23,039,700 4,355 4635.402 146 7 Scenario Cost Effectiveness ICER ($/sight rys saved) Decision NS 0 0 3 $8,584,401 2,393 weakly dominated 7 $8,859,053 2,520 weakly dominated 11 $9,516,993 2,749 weakly dominated 15 $9,751,752 2,842 3431.588 1 $13,283,721 3,301 weakly dominated 5 $13,693,230 3,440 weakly dominated 4 $13,698,719 2,991 strongly dominated 8 $14,074,791 3,132 strongly dominated 9 $14,633,201 3,645 weakly dominated 13 $14,979,307 3,770 5629.340 12 $14,988,109 3,369 strongly dominated 16 $15,316,605 3,478 strongly dominated 2 $20,553,707 3,912 39274.517 weakly dominated 6 $21,151,104 4,057 4138.317 10 $22,535,558 4,246 7302.707 14 $23,039,700 4,355 4635.402 147 8 Scenario Cost Effectiveness ICER ($/sight rys saved) Decision NS 0 0 3 $8,584,401 2,393 weakly dominated 7 $8,859,053 2,520 weakly dominated 11 $9,516,993 2,749 weakly dominated 15 $9,751,752 2,842 3431.588 1 $13,283,721 3,301 weakly dominated 5 $13,693,230 3,440 weakly dominated 4 $13,698,719 2,991 strongly dominated 8 $14,074,791 3,132 strongly dominated 9 $14,633,201 3,645 weakly dominated 13 $14,979,307 3,770 5629.340 12 $14,988,109 3,369 strongly dominated 16 $15,316,605 3,478 strongly dominated 2 $20,553,707 3,912 weakly dominated 6 $21,151,104 4,057 21557.719 weakly dominated 10 $22,535,558 4,246 7302.707 14 $23,039,700 4,355 4635.402 148 9 Scenario Cost Effectiveness ICER ($/sight rys saved) Decision NS 0 0 3 $8,584,401 2,393 weakly dominated 7 $8,859,053 2,520 weakly dominated 11 $9,516,993 2,749 weakly dominated 15 $9,751,752 2,842 3431.588 1 $13,283,721 3,301 weakly dominated 5 $13,693,230 3,440 weakly dominated 4 $13,698,719 2,991 strongly dominated 8 $14,074,791 3,132 strongly dominated 9 $14,633,201 3,645 weakly dominated 13 $14,979,307 3,770 5629.340 12 $14,988,109 3,369 strongly dominated 16 $15,316,605 3,478 strongly dominated 2 $20,553,707 3,912 weakly dominated 6 $21,151,104 4,057 weakly dominated 10 $22,535,558 4,246 15878.726 weakly dominated 14 $23,039,700 4,355 4635.402 After applying principles of dominance, scenarios 15, 13, and 14 remain. This is illustrated in the cost-effectiveness graph below. The concave hull / frontier represent programs that have the highest effect for a given level of cost. The scenarios enveloped within the concave hull are either 149 strongly or weakly dominated by the scenarios on the concave hull (87). The scenario of choice is then determined by finding a scenario with tangency between the frontier and the highest line with a slope representing the maximum willingness to pay to avert blindness. Weakly dominated alternatives lying close to the concave hull may also be considered for non-economic reasons (acceptability, availability, etc). 3 7 11 15 1 5 4 8 9 13 12 16 2 6 10 14 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 $0 $5,000,000 $10,000,000 $15,000,000 $20,000,000 $25,000,000 Effectiveness (sight yrs saved) Cost Effectiveness (sight yrs saved) vs Cost of screening and treatment 150 7.4.2. COMPARATIVE ANALYSIS: COMPARISON TO STATUS QUO The 15 alternative scenarios (i.e. scenarios 2-16) are then compared to status quo (scenario 1). To see how each of the scenarios 2-16 fare compared to the status quo, the costs and benefits of each scenario relative to scenario 1 are plotted in a cost-benefit plane, an example of which is plotted in figure 9.2.1 (88). The x-axis represents the difference (new treatment versus comparator, which is the status quo here) in the mean effect or mean benefit (∆ benefit) while the y-axis represents the difference in mean cost (∆ cost) (89) (90). Any programs falling in the northwest quadrant (region II) are less beneficial and more costly compared to the status quo and should never be adopted. Any programs falling in the southeast quadrant (region IV) are more beneficial and less costly and should always be adopted. Oftentimes, programs fall either in the northeast quadrant, offering more benefit at a higher cost, or the southwest quadrant, offering less benefit at a lower cost. The dotted line indicates the line where relative cost = relative benefit (i.e. x=y line). Below the x-y line, y>x, or the relative benefit compared to status quo exceeds the relative cost of the scenario as compared to the status quo. Thus, regions IB and IIIA also indicate regions of non-cost-beneficence, while regions IA and IIIB indicate regions of cost-beneficence. In region IIIB, programs offer less benefit at less cost and in region IA, programs offer more benefit at more cost; in both regions, the cost to obtain a unit of benefit is lower than the status quo. 151 Figure 7.4.2.1. Cost-benefit plane. The dotted line K divides the plane into cost-beneficial (lower right) and non-cost-beneficial (upper right) regions. From Black, W., The CE Plane: A graphic representation of cost-effectiveness, Med Decis Making, 1990, 10:212. The cost-benefit plane for the different perspectives are given below. From the medical systems perspective, only 4 scenarios (3, 7, 11, 15, i.e. all scenarios involving biennial ophthalmoscope screening) fall below the x=y dotted line. From the Medicare / Medicaid perspective, scenarios 5, 9, and 13 (i.e. scenarios involving annual ophthalmoscope screening) may be considered to replace the status quo. From the societal perspective, scenarios 5, 16, 9, 13, 2, 6, 10, and 14 (i.e. all scenarios that involve annual screening, and one scenario involving biennial screening – scenario 16 - which is biennial non-mydriatic camera screening with mandatory health insurance and patient navigator). 152 Figure 7.4.2.2. Cost-benefit plane (medical system perspective) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 -$6.00 -$4.00 -$2.00 $0.00 $2.00 $4.00 $6.00 $8.00 $10.00 $12.00 -$3.00 -$2.00 -$1.00 $0.00 $1.00 $2.00 $3.00 Cost of screening (relative to scenario 1) Millions Benefit (relative to scenario 1) Millions Cost-benefit plane (medical systems perspective) 153 Figure 7.4.2.3. Cost-benefit plane (Medicare / Medicaid system perspective) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 -$8.00 -$6.00 -$4.00 -$2.00 $0.00 $2.00 $4.00 $6.00 $8.00 $10.00 $12.00 -$8.00 -$6.00 -$4.00 -$2.00 $0.00 $2.00 $4.00 $6.00 $8.00 Cost of screening (relative to scenario 1) Millions Benefit (relative to scenario 1) Millions Cost vs benefit relative to scenario 1 (Medicare/Medicaid perspective) 154 Figure 7.4.2.4. Cost-benefit plane (societal perspective) 7.4.3. THE CASE FOR CONTEXT-BASED DECISION MAKING As seen above, the decisions on which screening strategies are worth implementing differ based on the perspective taken. Having the evaluation done for multiple perspectives provide the decision makers at the highest level with the contextual perspective of parties that will implement the screening strategy at the lower levels. As different parties have different interests but cooperation from all is required for successful implementation, having information from different perspectives is useful to determine amount of 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 -$30.00 -$20.00 -$10.00 $0.00 $10.00 $20.00 $30.00 -$30.00 -$20.00 -$10.00 $0.00 $10.00 $20.00 $30.00 Cost of screening (relative to scenario 1) Millions Benefit (relative to scenario 1) Millions Cost vs benefit relative to scenario 1 (societal perspective) 155 subsidies / incentives that may be needed to facilitate successful implementation / adoption of a screening strategy. 156 CHAPTER 8. DISCUSSION This paper has demonstrated the need for better reflection of compliance and cost structure as well as the multi-level nature of the system in a modeling approach that considers the system perspective. From the modeler’s and the health services researcher’s perspective, this new modeling paradigm will allow the evaluation of a wider range of preventive strategies scenarios, especially those involving behavioral interventions, at multiple levels. This would allow the policymaker to consider the inclusion of screening strategy components that have different mechanisms to increase compliance to increase screening uptake and maximize the net benefit from screening. In addition, it will also increase precision of modeling estimates even for interventions that could be previously evaluated with existing model structures. The paper has highlighted the value of multi-level modeling: implementation / compliance at a higher level (e.g. provider implementation of a screening strategy policy) affects compliance at a lower level (e.g. patient compliance with screening) and therefore health and economic outcomes; therefore, interventions whose implementation span multiple levels, or that has effects at lower levels, should be evaluated with models reflecting the multi-level nature of the system. Being able to consider multiple levels simultaneously in the model is especially important when there is a lot of heterogeneity in the clinic and population. The chapter on model comparison has indicated that an incomplete model specification that oversimplifies the compliance and cost structure result in significantly imprecise model estimates in some cases. The degree of imprecision is dependent on the underlying characteristics of compliance and cost structures. This also highlights the importance of considering the heterogeneity and dynamicity of population compliance in the model. In the illustrated DR 157 screening example, specifying the relationship between patient screening compliance with age and insulin use in the model results in fairly significant effects on the model estimates than using a simplified compliance model of using the same average population compliance for every individual in the model. Thus, using a compliance model that reflects the heterogeneity of compliance in the population is important to obtain more precise estimates. However, in the interest of model parsimony and simplicity, in cases where there is no underlying relationship between patient screening compliance with other factors that influence outcomes of interest, the modeler might consider a vastly simplified compliance model (such as the one used in model 1, which uses the population compliance as the prevailing compliance probability for every individual). The modeler would need to determine to what extent compliance heterogeneity in the model is warranted to balance model parsimony with precision of the estimates. For instance, in the DR screening example above, we can consider using a simplified compliance model with only components that have the most significant impact on the averted blind years (age and insulin) with minimal loss of precision in model estimates. Similarly, if the interventions of interest have very little associated fixed costs (i.e. the associated costs are mostly quantity / volume-related), a simplified cost structure considering only the variable costs that will scale with utilization could be used for a more parsimonious model with acceptable explanatory power. To facilitate the construction of a more parsimonious or more complex model when warranted, the abstraction introduced in the general framework for model 3 specifies the essential components while allowing for the specifications of modules of varying complexity. In the instance where the modeler believes there is no underlying relationship between patient screening compliance with other factors influencing outcome for interest for example, the compliance module will be specified to be a very simple model (i.e. patient compliance = a constant). On the other hand, the disease 158 progression module can take on a more complex form when the relationship between the intervention and the disease is more complicated (such as in evaluation of breast cancer screening, where the negative effects of false positive tests and unnecessary interventions need to be modeled within the disease progression module). The modeler would have to make decisions on the appropriate complexity of each module, ideally with help from experts in the various domains. In practice, often the limitation of data availability also forces model simplicity. Recognizing this limitation, model 3 was built to be flexible enough to accept input data / parameter specifications based on differing levels of data availability and provide the best estimates from the best available information. For instance, sometimes available data only allows the specification of model 1 or 2, but not model 3. However, since model 3 is a superset of models 1 and 2, the available data can be used as input into model 3 to obtain the same results as would have been obtained by model 1 or 2. Then, when we do have more information to specify model 3, we can easily migrate from model 1 or 2 to model 3 to obtain more precise outcome estimates. To reduce computational cost (but at the expense of flexibility in adding additional model components), it might be worth developing some analytical framework for some modules in the model while retaining simulation for the rest of the model structure to retain flexibility (as it is generally easier to refine simulation models than analytical models as we want to incorporate better information about the model structure or want to add model components or input/outputs). More well-established modules like disease progression, which is rooted in biology instead of human behavior, may be a likely candidate for using analytical framework. In general, analytical framework can be used in place of simulation where the behavior of the subsystem is fairly linear and closed-form solutions exist to model such subsystems. The general framework for models 1, 2, and 3 are also extendable to evaluate strategies involving other preventive services (i.e. not just the DR screening illustrated in this dissertation), including 159 other types of screening, obesity prevention interventions, and immunization services. The forms of the individual modules, the number of levels in the model, and the strategy components will vary depending on the strategies being evaluated, with some suggestions being made in chapter 4. With some modifications, the general framework can also be extended to other intervention strategies involving behavioral activation / modification (such as smoking cessation and substance abuse interventions). For instance, the smoking cessation case might include the inclusion of a community level instead of a provider level as the implementation of the smoking cessation interventions is largely at the community level (or, in multi-level interventions that also include providers, the community level can be included in addition to the provider level) and the impact of the intervention on the lower patient level depends on the extent of implementation at the community level. From the policymakers’ perspective, different perspectives should be considered to gain a fuller picture of the system and promote better decision making. This can facilitate more successful implementation as policymakers at the highest level now have comprehensive and quantitative information on how the interventions considered would affect different parties who would implement the intervention, and institute steps to facilitate implementation at all levels through subsidies, incentives / disincentives, or other forms of support if necessary. This also points to the need for more collaboration between the modeler and decision makers at various levels, especially at the model requirement gathering stage. Having decision makers from different levels who would have different perspectives on the intervention involved at the requirement gathering stage would ensure that all the relevant costs and benefits for the different parties are explicitly modeled and the model more applicable to the decision makers. This research points to some areas where availability of accurate data is limited, such as the duration to adjust to the morbidity and data required to parameterize the implementation module. 160 Future research and data collection would be required to fulfil the data requirements for future models and to refine the model. In addition, information on how some screening components might interact and affect the outcomes should also be collected (for example, patient education at the provider level and patient education at the patient level, if offered together, will likely have crowding out effect where the effects of one is diminished by the other, and the quantification of the effect is important to build a more accurate model), as this is another area of research where data is lacking. With that information, the combination of affected screening strategy components could be considered in a single screening strategy component and the combined effects considered to take into account the crowding out / substitution effect for more accurate evaluation of the policies. Discrete choice experiment or other standard preference methods might help to fill the gap in this area when real world data on specific combinations of the screening strategy components are not available yet. Other than data collection, more research needs to be done in the area of compliance and implementation. This research has demonstrated the importance of considering the effects that differing levels of compliance and implementation can have on the outcomes, and how differences in model specifications in the compliance and implementation would affect precision of outcome estimates. Future research could also delve into model refinement and into determining the necessary principles underlying each module components and proposed module structure / form. This includes determining how much complexity or parsimony is required in each module, and how many levels are required (e.g. clinic-level, provider-level, community-level, and patient-level). Different principles might be required for different applications (e.g. evaluation of strategies to improve screening for chronic diseases vs smoking cessation). 161 More research is especially necessary to determine the forms of the compliance and implementation modules, including the extent of heterogeneity of compliance among patients and implementation among providers that should be taken into account in the model, the level of dynamicity required (i.e. the change of compliance or implementation over time: patients often become more likely to go for screening as they age, or they might become less likely to go for screening if they have been getting negative diagnosis in their past results; clinics may have a low level of implementation when a policy has just been rolled out but would gradually increase levels of implementation, perhaps following a diffusion of innovation curve), and so on. Future research is also necessary on the effect of the environment and system barriers on behavior and implementation. Future work is also necessary to expand the framework to model the synergies of various strategies on multiple health conditions and how employing the combinations of these strategies affect the outcomes (e.g. blood glucose control can control both the underlying diabetes as well as DR, tele- screening now can be performed in regular clinics so can schedule foot screening at the same time and also improve diabetes, patient navigator will also increase compliance with other diabetes preventive services). Finally, even though this framework is positioned as an evaluation tool, it could also be a powerful tool for intervention design and policy planning. It can be also be used to determine the optimal clinic size threshold for implementing a certain intervention, or the optimal staffing level for an intervention. More importantly, the model can be used to determine which interventions are robust given the uncertainties in compliance and implementation, or robust in the perspectives of all the parties involved in the implementation of the interventions. 162 CHAPTER 9. CONCLUSIONS This research has proposed a generic simulation model structure that can evaluate screening strategies that include components that intend to improve screening compliance at the multiple levels and provide more precise outcome estimates. The ability to evaluate screening strategies that involve behavioral modification components is important to optimize screening reach and benefit, especially for screening that currently suffer from low compliance. The consideration of a multi- level structure in the model is also important as strategies implemented at multiple levels are often more effective than those implemented at a single level, and it is necessary to consider how implementation at a higher level impact implementation or compliance at a lower level, eventually affecting patient’s compliance with the preventive services being considered, and affecting outcome. The use of a more realistic compliance and cost structure in the model framework also leads to better precision of outcome estimates, which enables better decision making at the policy level. This generic simulation model structure has also been illustrated with its implementation for the case of evaluation of Diabetic Retinopathy screening strategies. However, the generic simulation model framework itself is generalizable to other preventive services and interventions involving behavioral activation or modification, and suggestions on how to adapt the individual modules to fit other contexts were given in the paper. In addition, this research has also proposed a flexible and comprehensive model that makes use of the best currently available information to guide initial policy design decisions or to aid preliminary policy evaluation, thereby bridging the gap until further information becomes available. 163 Finally, this research has made a case for context-based decision making, to analyze cost-benefit from multiple perspectives to understand system barriers that need to be addressed to facilitate successful implementation. 164 ACKNOWLEDGMENTS Over the past six years I’ve received much support from many individuals. My advisor, Dr. Shinyi Wu, has been a great mentor, colleague, and friend. I would like to thank my dissertation chair (Dr. Shinyi Wu), dissertation committee members (Dr. Carl Kesselman and Dr. Chih-Ping Chou), and qualifying committee members (Dr. Azad Madni and Dr. Jennifer Unger) for their guidance. I would also like to thank fellow doctoral students, including those who have moved on to other pastures, for their feedback and friendships. I would like to thank other friends both in Los Angeles and elsewhere in the world for being a source of inspiration and strength. Last but most certainly not least, to my husband and my family, thank you for your love, support, and encouragement. 165 BIBLIOGRAPHY 1. The cost of health system change: public disontent in five nations. Donelan, K, et al. 3, s.l. : Health Affairs, 1999, Vol. 18. 2. Summary of the Affordable Care Act. Meltzera, C.C. s.l. : AJNR, 2011 йил, Vol. 32. 3. Shearer, G. Prevention Provisions in the Affordable Care Act. s.l. : American Public Health Association (APHA) Issue Brief, 2010. 4. Prevention in Health Care Reform: The Time Has Come. Engel, J. 3, s.l. : NC Medical Journal, 2010 йил, Vol. 71. 5. Noncompliance with Vision Care Guidelines in Latinos with Type 2 Diabetes Mellitus. Paz, S., et al. 8, s.l. : Journal of Ophthalmology, 2006, Vol. 113. 6. Sallis JF, Owen N. Ecological Models of Health Behavior. [book auth.] et al. In Glanz. Health Behavior and Health Education: Theory, Research, and Practice (3rd ed). . 7. The Multilevel Compliance Challenge: Recommendations for a call of action, a statement to professionals. Miller, N, et al. s.l. : Circulation, 1997, Vol. 95. 8. World Health Organization. Adherence to Long term Therapies: Evidence for Action. Geneva : World Health Organization, 2003. 92-4-154599-2. 9. Barriers to prevention of vision loss caused by diabetic retinopathy. R, Klein. 912-20, s.l. : Archives Ophthalmology, 2004, Vol. 111. 10. Eye care in African Americans with Type I diabetes. The New Jersey 725. M.S, Roy. 914-20, s.l. : Ophthalmology , 2004, Vol. 111. 166 11. Interventions for helping patients to follow prescriptions for medications. RB, Haynes, et al. s.l. : Cochrane Database Syst Rev., 2002, Vol. 2. 12. Designing studies that would address the multilayered nature of health care. Murray D, Pennell M, Rhoda D, Hade E, Paskett E. 90-96, s.l. : J Natl Cancer Inst Monogr, 2010, Vol. 40. 13. Stedman's Medical Dictionary. Copyright . Wilkins, Lippincott Williams &. 2006. 14. McGraw-Hill Concise Dictionary of Modern Medicine. . s.l. : The McGraw-Hill Companies, Inc., 2002. 15. Medicines concordance (involving patients in decisions about prescribed medicines). s.l. : National Institute for Health and Clinical Excellence, 3 March 2008. 16. Gale Encyclopedia of Medicine. . s.l. : The Gale Group, Inc., 2008. 17. American Academy of Ophthalmology Retina/Vitreous Panel, Preferred Practice Patterns Committee. Diabetic retinopathy. . San Francisco (CA) : American Academy of Ophthalmology (AAO), 2008. 18. Theoretical models for predicting and improving compliance with breast cancer screening. Curry, Susan J. and Emmons, Karen M. 302-316., s.l. : Annals of Behavioral Medicine , 1994, Vol. 16(4). 19. A Framework for the Study of Access to Medical Care. Aday, L A and Andersen, R. 208–220, s.l. : Health Serv Res., 1974, Vol. 9(3). PMCID: PMC1071804. 20. Patient Adherence to Guidelines for diabetes eye care: results from the eye disease follow-up study,. Will, J.C., et al. 10, s.l. : American Journal of Public Health, , 1990, Vol. 84. 21. Barriers to prevention of vision loss caused by diabetic retinopathy,. Klein, R. 8, s.l. : Archives of Ophthalmology, 1997, Vol. 115. 167 22. The relationship of a diabetes telehealth eye care program to standard eye care and changes in diabetes health outcomes. Fonda, S.J., et al. 6, s.l. : Telemedicine Journal and e-Health, 2007, Vol. 13. 23. Factors Associated with Having Eye Examinations in Persons with Diabetes,. Moss, S.E., Klein, R. and and Klein, B. 6, s.l. : Archives of Family Medicine, 1995, Vol. 4. 24. Factors that influence the patient uptake of diabetic retinopathy screening,. Dervan, E., et al. 4, s.l. : Factors that influence the patient uptake of diabetic retinopathy screening, 2008, Vol. 177. 25. Access to Medical Care in the US: Who has it, who doesn’t. Aday, LA., GV., Fleming and R., Andersen. Chicago, IL: : Pluribus, University of Chicago, Center for Health Administration Studies, 1984. 26. Bullough, B. and Bullough, VL. “Poverty, Ethnic Identity, and Health Care.” . New York, : Appleton-Century Crofts, , 1972. 27. “A simple language-based acculturation scale for Mexican Americans: Validation and application to health care research”, . Deyo, RA., et al. s.l. : American Journal of Public Health, 1985, Vol. 75. 28. “Health Care usage by Hispanic outpatients as function of primary language” . Hu, DJ. and Covell RM. 4, s.l. : West J. Med., Vol. 144. 29. Promoting Screening Mammography in Inner-City Settings: A Randomized Controlled Trial of Computerized Reminder as a Component of a Program to Facilitate Mammography. R, Burack, et al. 6, s.l. : Medical Care, 1994, Vol. 32. 30. Understanding and Enhancing Patient Compliance with Diabetic Regimens. Irwin M Rosenstock, Ph.D. 6, s.l. : Diabetes Care, 1985, Vol. 8. 31. A Description of the Socio-Ecological Framework Used in The Trial of Activity for Adolescent Girls (TAAG). Elder, J and al., et. 2, s.l. : Health Educ. Res., 2007, Vol. 22. 168 32. Environmental and societal factors affect food choice and physical activity: rationale, influences, and leverage points. SL, Booth, et al. S21-39, s.l. : Nutr Rev, 2001, Vol. 59. 33. Office on Smoking and Health. Best Practices for Comprehensive Tobacco Control. . Atlanta : Centers for Disease Control and Prevention (CDC)., 1999. 34. (USDHHS)., US Department of Health and Human Services. Healthy People 2010. s.l. : US Department of Health and Human Services (USDHHS)., 2000. 35. Assessment of the integration of the ecological approach in health promotion programs. L, Richard, et al. 318-28, s.l. : Am J Health Promot, 1996, Vol. 10. 36. Introduction: Understanding and Influencing Multilevel Factors Across the Cancer Care Continuum. Taplin SH, Anhang Price R, Edwards HM, Foster MK, Breslau ES, Chollette V, Prabhu Das I, Clauser SB, Fennell ML, Zapka J. s.l. : J Natl Cancer Inst Monogr., 2012, Vol. 44. doi: 10.1093/jncimonographs/lgs008.. 37. Organizational Factors and the Cancer Screening Process. R, Anhang Price, et al. s.l. : J Natl Cancer Inst Monogr, 2010, Vol. 40. doi: 10.1093/jncimonographs/lgq008.. 38. Foreword. Ballard-Barbash, Rachel. 1-2, s.l. : J Natl Cancer Inst Monogr , 2010, Vol. 40. doi:10.1093/jncimonographs/lgq005. 39. Designing studies that would address the multilayered nature of health care. D, Murray, et al. 90- 96, s.l. : J Natl Cancer Inst Monogr, 2010, Vol. 40. 40. Fidell, Barbara G. Tabachnick, Linda S. Using multivariate statistics (5th ed.). Boston : Montreal: Pearson/ A & B, 2007. 41. Diabetic Retinopathy Screening Update. Seema Garg, MD, PhD and Richard M. Davis, MD. 4, s.l. : Clinical Diabetes, 2009, Vol. 27. 169 42. National Diabetes Fact Sheet. (CDC), Center of Disease Control and Prevention. s.l. : Center of Disease Control and Prevention (CDC), 2011. 43. Association, American Diabetes. All about diabetes. [Online] http://www.diabetes.org/about- diabetes.jsp. 44. Preventive eye care in people with diabetes is cost-saving to the federal government. Implications for health-care reform,. Javitt, J.C., Aiello, L.P., Chiang, Y., Ferris, F.L. 3rd, Canner, J.K., Greenfield, S. 8, s.l. : Diabetes Care, 1994, Vol. 17. 45. The Prevalence of Diabetic Retinopathy Among Adults in the United States. Kempen, J., et al. 4, s.l. : Arch Opthalmology, 2004, Vol. 122. 46. Dasbach, E. and al., et. Cost-Effectiveness of Strategies for Detecting Diabetic Retinopathy. Medical Care. 1991, Vol. 29, 1. 47. Treatment of diabetic retinopathy. Ferris FL, 3rd, MD, Davis and LM, Aiello. 667-678, s.l. : N Engl J Med , 1999, Vol. 341. 48. Patterns of adherence to diabetes vision care guidelines: baseline findings from the Diabetic Retinopathy Awareness Program. Schoenfeld ER, Greene JM, Wu SY, Leske MC. 3, s.l. : Ophthalmology, 2001, Vol. 108. 49. Diabetic Retinopathy. American Diabetes Association. 1, s.l. : Diabetes Care, 2002, Vol. 25. 50. Cavallerano, A and Conlin, P. Teleretinal Imaging to Screen for Diabetic Retinopathy in Veterans Health Administration. Journal of Diabetes Science and Technology. 2008, Vol. 2, 1. 51. Vijan, S, Hofer, T and Hayward, R. Cost-Utility Analysis of Screening Intervals for Diabetic Retinopathy in Patients with Type II Diabetes Mellitus. Journal Of American Medical Association. 2000, Vol. 283, 7. 170 52. Simulation Modeling for Evaluating Screening Services for Diabetic Retinopathy. Davies, R, et al. 4, s.l. : The Journal of the Operational Research Society, 2000, Vol. 51. 53. A Systems Modelling Approach for the Prevention and Treatment of Diabetic Retinopathy. Harpera, P.R., et al. 1, s.l. : European Journal of Operational Research, 2003, Vol. 150. 54. Whited, J, et al. A modeled economic analysis of a Digital Teleopthalmology System As Used by Three Federal Healthcare Agencies for Detecting Proliferative Diabetic Retinopathy. Telemedicine and e-Health. 2005, Vol. 11, 6. 55. Porta, M and al., et. Comparison of the Cost-effectiveness of Three Approaches To Screening For and Treating Sight-Threatening Diabetic Retinopathy. Diabetes and Metabolism (Paris). 1999, Vol. 25. 56. Chen, LS, Tsai, CY and al., et. Feasibility of Teleopthalmology for screening for eye disease in remote communities. Journal of Telemedicine and Telecare. 2004, Vol. 10, 6. 57. Tennant, M., et al. Teleopthalmology via Stereoscopic Digital Imaging: A Pilot Project. Diabetes Technology and Theraupetics. 2000, Vol. 2, 4. 58. Aoki, N, et al. Cost-effectiveness Analysis of Telemedicine to Evaluate Diabetic Retinopathy in a Prison Population. Diabetes Care. 2004, Vol. 27, 5. 59. Cavallerano, AA, Cavallerano, JD and al., et. A Telemedicine program for Diabetic Retinopathy in a Veterans Affair Medical Center - the Joslin Vision Network Eye Health Care Model. American Journal of Opthalmology. Vol. 139, 4. 60. Eye Disease and the Older Diabetic. Nina Tumosa, PhD. 3, s.l. : Clinics in Geriatric Medicine, 2008, Vol. 24. 61. [Online] http://www.cdc.gov/owcd/eet/CBA/Fixed/4.html. 171 62. [Online] http://www.cdc.gov/owcd/eet/CBA/Fixed/4.html . 63. American Cancer Society Guidelines for Breast Cancer Screening . Smith, R and al, et. 3, s.l. : CA: A Cancer Journal for Clinicians, 2003, Vol. 53. DOI: 10.3322/canjclin.53.3.141. 64. Cancer Screening in the Elderly: A Framework for Individualized Decision Making. Walter, L and Covinsky, K. 21, s.l. : JAMA, 2001, Vol. 285. doi:10.1001/jama.285.21.2750. 65. Understanding Non-compliance to Colorectal Cancer Screening: a Case Control Study, Nested in a Randomized Trial. Rossi, PG, et al. 139, s.l. : BMC Public Health, 2005 йил, Vol. 5. 66. Mass Community Screening for Diabetic Retinopathy Using a Nonmydriatic Camera with Telemedicine. Boucher, M, Nguyen, Q and Angioi, K. 6, s.l. : Canadian Journal of Ophthalmology, 2005, Vol. 40. 67. Examination Compliance and Screening for Diabetic Retinopathy: A 2-year Follow-up Study. Lee, SJ, et al. 3, s.l. : Clin Experiment Ophthalmol., 2000, Vol. 28. 68. Using stated preference discrete choice modelling to evaluate the introduction of varicella vaccination. Hall, J, et al. 5, s.l. : Health Economics, 2002, Vol. 11. 69. Using stated preference discrete choice modelling to inform health care decision-making: A pilot study of breast screening participation. Gerard, K, Shanahan, M and Louviere, J. 9, s.l. : Applied Economics, 2003, Vol. 35. 70. The Health Belief Model: A decade later. Janz, N.K.. and Becker, M. 1, s.l. : Health Education Behavior, 1984, Vol. 11. 71. The theory of planned behavior. Ajzen, I. 2, s.l. : Organizational Behavior and Human Decision Processes, 1991, Vol. 50. 172 72. The transtheoretical model of health behavior change. Prochaska, J.O. and Velicer, W.F. 1, s.l. : Am J Health Promot, 1997, Vol. 12. 73. The PARIHS framework - a framework for guiding the implementation of evidence-based practice. Rycroft-Malone, J. 4, s.l. : J Nurs Car Qual, 2004, Vol. 19. 74. Diffusion of innovations. Rogers, E.M. Glencoe : Free Press, 1962. ISBN 0-612-62843-4. 75. International Microsimulation Association. What is microsimulation? www.microsimulation.org. [Online] [Cited: 7 1, 2013.] http://www.microsimulation.org/IMA/What%20is%20microsimulation.htm. 76. World Health Organization. Blindness and Visual Disability Part IV of VII: Socioeconomic Aspects. s.l. : WHO factsheet Fact Sheet N 145, 1997. 77. Appropriate Assessment of Neighborhood Effects on Individual Health: Integrating Random and Fixed Effects in Multilevel Logistic Regression,. Klaus Larsen and Juan Merlo. 1, s.l. : Am. J. Epidemiol., 2005, Vol. 161. 78. al., Banks et. Discrete-Event System Simulation, 3rd edition. s.l. : Prentice Hall, 2000. 79. The economic burden of major adult visual disorders in the United States. Rein DB, Zhang P, Wirth KE, Lee PP, Hoerger TJ, McCall N, Klein R, Tielsch JM, Vijan S, Saaddine J. 12, s.l. : Arch Ophthalmol., 2006, Vol. 124. 80. Association between vision loss and higher medical care costs in Medicare beneficiaries costs are greater for those with progressive vision loss. Javitt JC, Zhou Z, Willke RJ. 2, s.l. : Opthalmology, 2007, Vol. 114. 173 81. Cooney, J., Landers, G., Rein, D., Bae, J., Curry, R. Comparative Assessment of Cost and Care Outcomes Among Georgia's Community-Based and Facility-Based Long-term Care Programs. 2004 : Georgia State University, Atlanta. 82. Intensive blood glucose control with sulphonylureas or insulin compared with conventional treatment and risk of complications in patients with type 2 diabetes. UKPDS 33. Group, UK Prospective Diabetes Study (UKPDS). s.l. : Lancet, 1998a, Vol. 352. 83. Risk of developing diabetic retinopathy in Diabetes Control and Complications Trial type 1 diabetic patients with good or poor metabolic control. Zhang L, Krzentowski G, Albert A & Lefebvre PJ. s.l. : Diabetes Care, 2001, Vol. 24. 84. Banks, J, et al. Discrete Event Simulation, Fifth edition. Upper Saddle River : Pearson Education, Inc, 2010. ISBN 0136062121. 85. Model Verification and Validation. Carson, J. s.l. : Proceedings of the 2002 Winter Simulation Conference, 2002. 86. The incidence of vision loss in a diabetic population. Moss, SE, Klein, R and Klein, BE. 10, s.l. : Ophthalmology, 1988 йил, Vol. 95. 87. Incidence of blindness and visual impairment in diabetic patients participating in an ophthalmological control and screening programme. Henricssson, M, et al. 6, s.l. : Acta Ophthalmol Scand, 1996 йил, Vol. 74. 88. Patterns of Adherence to Diabetes Vision Care Guidelines,. Schoenfeld, E.R1., Greene, J.M., Wu, S.Y., Leske, M.C. et al.,. 3, s.l. : Ophthalmology, 2001, Vol. 108. 89. Ophthalmic examination among adults with diagnosed diabetes mellitus. Brechner RJ, Cowie CC, Howie LJ, Herman WH, Will JC, Harris MI. 14, s.l. : JAMA, 1993, Vol. 270. 174 90. Teleretinal screening for diabetic retinopathy in six Los Angeles urban safety-net clinics: initial findings. Ogunyemi O, Terrien E, Eccles A, Patty L, George S, Fish A, et al. s.l. : AMIA Annu Symp Proc., 2011. 91. Ophthalmic referral rates for patients with diabetes in primary-care clinics located in disadvantaged urban communities. . Wylie-Rosett J, Basch C, Walker EA, Zybert P, Shamoon H, Engel S, et al. 1, s.l. : Diabetes Complications, 1995, Vol. 9. 92. How to avoid a perfunctory sensitivity analysis. Saltelli, A., Annoni, P. s.l. : Environmental Modeling and Software, 2010, Vol. 25. 93. MS, Kamlet. A framework for cost-utility analysis of government healthcare programs: Office of Disease Prevention and Health Promotion. s.l. : Public Health Service, U.S. Department of Health and Human Services, 1992. 94. Gold, MR., Siegel, JE., Russell, LB., Weinstein, MC. Cost-effectiveness in health and medicine. s.l. : Oxford University Press, 1996. 95. The CE plane: a graphic representation of cost-effectiveness. Black, WC. 212, s.l. : Med Decision MAking, 1990, Vol. 10. DOI: 10.1177/0272989X9001000308. 96. Lipscomb, J, Gotay, CC and Snyder, C. Outcomes Assessment in Cancer: Measures, Methods and Applications. s.l. : Cambridge University Press, 2005. 97. Drummond, M.F. Methods for the Economic Evaluation of Health Care Programmes. s.l. : Oxford University Press, 2005. 98. Revisiting the behavioral model and access to medical care: Does it matter? Andersen, RM.,. s.l. : Journal of Health Society Behavior, 1995, Vol. 36. 175 99. The summary of diabetes self-care activities measure: results from 7 studies and a revised scale. Toobert, DJ., Hampson, SE. and Glasgow, RE. 7, s.l. : Diabetes Care, 2000, Vol. 23. 100. Language spoken and differences in health status, access to care, and receipt of preventive services among US Hispanics. DuBard, C.A. and Gizlice, Z. 11, s.l. : American Journal of Public Health, 2008, Vol. 98. 101. Does patient behavior or access factors have the largest influence on screening in type I diabetes. Dorsey, R., et al. 4, s.l. : Diabetes Care, 2007, Vol. 30. 102. Development of a questionnaire identifying factors responsible for successful self-management of insulin-treated diabetes. Day, J., Bodmer, C. and Dunn, O. s.l. : Diabetes Medicine, , 1996, Vol. 6. 103. CHIS 2007 Sample Design . [Online] http://www.chis.ucla.edu/pdf/sample_desc_2007.pdf. 104. Standard of care in diabetes. American Diabetes Association. 1, s.l. : Diabetes Care, 2007, Vol. 30. 105. Self-reports of health care utilization compared to provider records. Ritter, P.L., et al. s.l. : Journal of Clinical Epidemiology , 2001, Vol. 54. 106. Comparison of self-reported and medical record health care utilization measures. Roberts, R.O., et al. s.l. : Journal of Clinical Epidemiology , 1996, Vol. 49. 107. Eye care in African Americans with Type I diabetes. The New Jersey 725. M.S., Roy. 914-20, s.l. : Archives Ophthalmology, 2004, Vol. 111. 108. The Wisconsin Epidemiologic Study of Diabetic Retinopathy. III. Prevalence and risk of diabetic retinopathy when age at diagnosis is 30 or more years. R, Klein, et al. 527-532, s.l. : Arch Ophthalmol, 1984, Vol. 102. 176 109. The Prevalence and Risk Indicators of Uncorrected Refractive Error and Unmet Refractive Need in Latinos: The Los Angeles Latino Eye Study. R, Varma, et al. s.l. : Investigative Ophthalmology and Visual Science, 2008, Vol. 49. 110. Cost-effectiveness of Alternative Methods for Diabetic Retinopathy Screening. Lairson DR, Pugh JA, Kapadia AS, Lorimor RJ, Jacobson J, Velez R. 10, s.l. : Diabetes Care, 1992, Vol. 15. 177 APPENDIX A: PARAMETERIZING THE COMPLIANCE MODULE - DETERMINANTS OF COMPLIANCE To populate the compliance module parameters, it is necessary to establish the relationship between screening compliance and determinants of compliance, which include patient characteristics and characteristics of screening strategy components (which will influence provider and system characteristics) in the conceptual model framework. To do this, an analysis of determinants of compliance is performed using logistic regression. The details of the analysis are given below. A.1. FRAMEWORK FOR ANALYSIS Traditional access indicators include having health insurance and having a regular source of care. According to Andersen’s framework, some of the access indicators considered here, such as health insurance and regular source of care, can be thought of as potential access, enabling resources for health care utilization. Dimensions of potential access can be conceptualized as determinants of realized access, which are the actual services used and satisfaction with services (91). In other words, access is an important determinant of health utilization, which includes screening. For this analysis, the traditional definition of care is expanded to include not just factors that influence the ease of medical care, but also access to quality medical care. Thus, the access indicators considered here include not just health care system related factors (type of health insurance and type of usual source of care) but also provider-related factors, including physician 178 adherence to evidence-based guidelines, which include provider checking patients’ HbA1C level and foot sores, developing diabetes care plan, helping to coordinate services with other medical services, and ease of communication with patients. Additionally, this analysis would also investigate the effect of access and self-care indicators that have not been thoroughly studied in previous literature on determinants of DR screening compliance. For instance, among access indicators in general, insurance coverage has often been shown as a major predictor of DR screening (5), but as many diabetics in California lack continuous insurance coverage and eye screening is done yearly, we would specifically use the measure of having any insurance coverage in the past 12 months. The self-care indicators used in this study, which are proxies of the Summary of Diabetes Self-care Activities (SDSCA) constructs (92), including smoking, exercise, diet, and blood-glucose control, have also not been studied extensively in the literature. Demographics and socioeconomics likely also impact DR screening compliance, so they are also included as independent predictors. In Andersen’s framework, demographics and socioeconomic factors such as age, education, and poverty level can be thought of as predisposing and as indicators of need, in that they suggest the need for screening, health literacy, and material resources (91). Access as enabling factors addresses these needs and facilitates the ability of individuals to obtain health care by removing financial and structural barriers to care, so the association between demographics and screening compliance should be diluted or eliminated when access factors are taken into account. Co-morbidities and severity of diabetes also are possible predictors of DR screening compliance, as those with common co-morbidities of diabetes, such as high blood pressure and heart disease, and more severe form (e.g. insulin-dependent) form of diabetes may suffer a higher burden of the disease which may be barriers to screening. On the other hand, having these co-morbidities and 179 more severe diabetes also predisposes patients to DR, and may act as motivation to these patients to obtain regular eye screening. The diverse California population also necessitates the importance of considering language fluency as a predictor of utilization of screening services (93). In particular, lack of access is a barrier for all population groups, but the barrier may be significantly larger for those not fluent in English, as the process of selecting private coverage, for instance, can be a challenge for immigrants who came from countries with different health care systems, and this challenge is compounded by language difficulty. Even among patients with high levels of access, setting appointments, completing forms, and communication with health providers would also present more of a challenge for those without than those with the language proficiency, and may eventually affect screening rates. Thus, we hypothesize that the impact of access on DR screening compliance differs among people with different language fluency, or in other words, that language fluency moderates the effect of access on DR screening compliance. Diabetes is unique in that much of disease management rests with the patient. Patients with better self-care and self-efficacy probably would better utilize preventive services, and a recent study found that patients with better glycemic monitoring and control are more likely to receive dilated eye exam, foot exam, and urine protein screen (94). Patients who check their blood glucose levels more regularly had higher levels of self-efficacy, emotional adjustment to disease, and practical self- management skills, and these characteristics may also predispose such patients to higher utilization of medical care, including screening (95). Thus, we would also like to examine how patient behavior, self-care activities and self-efficacy in particular, influence adherence with DR screening. Lack of access may present a bigger barrier for patients with worse self-care, as these patients typically have lower levels of self-efficacy and practical self-management skills (95), and so, being less proactive in their diabetes management, they may be less likely to be proactive in seeking care. 180 For these people, setting appointments and fulfilling them may present a challenge, especially when access barriers are present; as a result, their use of preventive services may be lower than those with better self-care. Thus, we also hypothesize that self-care moderates the relationship between access and compliance with DR screening. The analysis would assess how screening for diabetic retinopathy is dependent on a combination of the factors outlined above: demographics, co-morbidities and disease severity, socio-economic factors, access, language fluency, as well as self-care and self-efficacy. First, these factors are considered as independent predictors, and then the moderation pathways involving language fluency and self-care which affect the relationship between access and compliance with DR screening are studied. Thus, the analysis examines the effect of patient (in terms of demographics, co-morbidities and disease severity, self-care and self-efficacy, and language fluency), provider, and system (in terms of access to quality care) on adherence with DR screening guidelines. A.2. METHODS A.2.1. DATA SOURCE Data for this study was taken from the 2009 California Health Interview Survey (CHIS), a representative telephone survey of 47614 adults, 12 324 children and adolescents from over 49000 households across California. CHIS is the largest state health survey in the nation conducted biennially since 2001 utilizing a random-digit-dial (RDD) design and since the fourth CHIS collection in 2007 (96), a statewide cell-phone survey only sample was introduced to robustly samples that may be hard to reached by a RDD only design. 181 During the telephone interview, participants were asked specifically if a doctor had ever told them that they had diabetes or (for women only) diabetes during times other than pregnancy. Gestational only and borderline cases were not coded as diabetes. We focused on all Type II diabetes patients for this analysis. A.2.2. DEPENDENT VARIABLES The American Diabetes Association (ADA) (97) guidelines for vision care recommend that persons with Type I diabetes have a dilated eye examination within 3 years of diagnosis while persons with Type II diabetes have the eye examination at time of diagnosis. Subsequent eye examinations for both types of diabetes should be repeated annually thereafter. All participants who reported being told by a doctor of having diabetes were also asked when they had received their last dilated eye examination in which bright lights were used to look into the back of their eyes (within the past month, within the past year, 1-2 years ago, >2 years ago, never). For the purpose of this study, non- compliance was defined as either never had a dilated eye examination or had their last dilated eye exam more than 12 months ago at the time of the interview. Compliance was defined as having had a dilated eye examination in the past 12 months at the time of the interview. A.2.3. INDEPENDENT VARIABLES Access (to quality medical care) measures included health care system factors such as having insurance coverage in the past 12 months and type of usual source of care (none, private doctor/ 182 HMO, government clinic/hospital); provider related factors including whether physicians checked for Hemoglobin A1C (HbA1C) at least once in the past year, whether physicians checked for sore feet at least once in the past year, whether physicians created a personalized diabetes management care plan with the participant, whether doctor’s office helped coordinated services (yes, no, or no usual source of care); as well as difficulty in understanding the doctor. We measured language fluency using a binary variable that indicate whether individuals were fluent in English. Adapting from the SDSCA scale (92) that measures the level of personal involvement in diabetes care, we used non-doctor related diabetes care and other lifestyle variables included in the 2009 CHIS survey to create proxy diabetes self care factors such as frequency of checking blood glucose level, whether the subject was a current smoker, average number of servings of fruits and vegetables per day, average number of times the individual ate fast food per week (as a proxy for high cholesterol food) and physical activity (sedentary, some physical activity, regular physical activity). In addition, the construct of self-efficacy is measured using the variable confidence to control and manage diabetes. Demographics and socio-economic variables included age (categorical); race; gender; education (having a high school diploma vs. not); employment status (employed, unemployed, not in labor force – reflecting resources as well as available leave time to receive diabetes retinopathy screening); living below 200% federal poverty level (vs higher); and living alone versus others (to reflect the level of social support and the presence of additional resources). 183 A.2.4. ANALYSIS The independent relationship between demographics, co-morbidities and disease severity, socio- economic factors, access, language fluency, as well as self-care and self-efficacy on DR screening compliance is examined using logistic regression. Reference categories were redefined and collapsed when there are small cell sizes resulting in unreliable coefficients. To examine the hypothesis that language fluency and self-care moderate access, we included access and language fluency as well as self-care and access interaction terms, and tested if they were significant. All analyses were conducted in SAS 9.2. A.3. RESULTS Among demographics and socioeconomic indicators, only age (being older) and not being within 200% of poverty level positively predict compliance with DR screening. Among co-morbidities and disease severity indicators, only insulin use (which is an indicator of diabetes severity) negatively predict compliance with DR screening. Among access factors, having insurance in the past 12 months increases the likelihood of compliance with DR screening, which is consistent with previous studies (5). Having insurance in the past 12 months is among the strongest predictors for DR screening in this analysis. Aspects of physician adherence to evidence based guidelines, the doctor checking for blood glucose level and sore feet, increase the likelihood of compliance with DR screening. 184 English fluency does not influence the likelihood of compliance. Among self-care indicators, never checking blood glucose level and fast-food intake of twice or more a week negatively predict compliance with DR screening, while regular physical activity positively predict compliance with DR screening. The self-efficacy indicator, having more confidence to control and manage diabetes, also positively predicts compliance with DR screening. Interaction terms were found to be insignificant, indicating that access factors were not moderated by language fluency or self-care. The results are presented in the table below. Only variables that are significantly related to compliance with DR screening are presented. Variable Odds ratio point estimate 95% Wald Confidence Limit Age 1: 18-49 0.423 0.312 0.575 2: 50-64 0.660 0.530 0.822 3: 65 and above Ref Less than 200% poverty level 0.791 0.653 0.959 On insulin 0.735 0.594 0.909 Having insurance in the past 12 months 1.958 1.457 2.631 185 Doctor checking HbA1C level 1.534 1.259 1.869 Doctor checking feet 1.679 1.412 1.997 Frequency of checking own HbA1C level 1:Never 0.658 0.520 0.833 2:Less than 30 times / month 0.871 0.714 1.063 3:More than 30 times / month Ref Frequency of fast food consumption 1:None Ref 2:Once a week 0.918 0.754 1.117 3:Twice a week or more 0.797 0.653 0.972 Confidence to control & manage diabetes 1:Very confident 1.417 1.045 1.921 2:Somewhat confident 1.129 1.129 1.535 3:Not too confident / Not at all confident Ref 186 A.4. DISCUSSION As the 2009 CHIS survey is a single time-point cross-sectional study design involving self-reported responses, it is difficult to determine the accuracy of questionnaire responses. No clinical confirmation of diabetes would exclude those study participants who were unaware of their diabetes. This population would increase the number of diabetic persons who had not received adequate vision care. However, because they do not know that they have diabetes, it would be inappropriate to consider compliance with vision care guidelines in this group. Accuracy of other self-report measures is also questionable especially relating to self-care, access and our compliance outcome where participants might have forgotten or might not have fully comprehended the questions relating to detailed dietary habits or procedures conducted at a doctor’s visit in the past year. However, previous studies have noted that self-reported data on medical conditions and utilization of medical services can provide a reasonably accurate assessment (98) (99). Aspects of physician adherence to evidence based guidelines, the doctor checking for blood glucose level and sore feet, increase the likelihood of compliance with DR screening, and these are among the strongest predictors for DR screening. Physicians who adhere to evidence-based guidelines may be more likely to offer better care in general, and they may be more likely to educate patients on the advantages of screening and recommend eye screening for these patients, which explains the higher likelihood of compliance with DR screening of their patients. This indicates the need to consider expanding the traditional definition of access to include access to quality care, in particular the provider adherence to evidence based guidelines, as they have a significant impact on screening utilization, and possibly use of other preventive services. This also lends support to the importance of multi-level screening strategies that include physician-level interventions and to the 187 importance of considering provider compliance with guidelines and its impact on patient’s compliance. Future work can also examine the relative impact of access, acculturation, and self-care on DR screening compliance across the different socioeconomic subgroups, as access indicators may have a differential impact on different subgroups, and if so, these differences should be taken into account and integrated into improvements in delivery systems and interventions to improve DR screening rates; the impact of these targeted screening strategies can then be estimated using the model proposed. 188 APPENDIX B: SENSITIVITY ANALYSIS RESULTS Sensitivity analysis results for a number of parameters are given below. Parameter to vary in sensitivity analysis Lower bound Upper bound Years to adjust to blindness 1 year 5 years Patient compliance 10% decrease in patient compliance 10% increase in patient compliance Patient compliance model parameters 95% confidence level lower bound estimates for all regression coefficients 95% confidence level upper bound estimates for all regression coefficients Patient navigator model parameter 95% confidence level lower bound for the odds ratio estimate 95% confidence level upper bound for the odds ratio estimate Percentage of eligible clinics that implement the screening strategy component 60% 100% Clinic size threshold for patient navigator implementation 1000 3000 Discount factor 2% 8% 189 B.1. TIME TO ADJUST TO BLINDNESS Scena rio no Scenario Relevant parameters Averted indirect medical costs / % difference Net benefits (medical system perspective) / % difference Net benefits (Medicare / Medicaid perspective) / % difference Net benefits (societal perspective) / % difference Modal ity Scr_fr eq Insura nce Naviga tor setting YrAdjBlin d=1 YrAdjBlin d=5 YrAdjBlin d=1 YrAdjBlin d=5 YrAdjBlin d=1 YrAdjBlin d=5 YrAdjBlin d=1 YrAdjBlin d=5 1 1 1 1 1 -65.01 58.79 125.80 -113.76 -55.66 50.34 -7.32 6.62 2 2 1 1 1 -65.02 58.81 60.53 -54.75 -105.77 95.67 -7.73 6.99 3 1 2 1 1 -65.01 58.72 201.63 -182.12 -48.18 43.52 -7.17 6.48 4 2 2 1 1 -65.01 58.76 85.78 -77.53 -70.05 63.31 -7.55 6.82 5 1 1 2 1 -65.02 58.79 126.73 -114.58 -55.80 50.45 -7.22 6.53 6 2 1 2 1 -65.02 58.80 60.53 -54.74 -107.46 97.18 -7.62 6.89 7 1 2 2 1 -65.01 58.71 208.51 -188.30 -48.01 43.35 -7.01 6.33 8 2 2 2 1 -65.01 58.74 87.68 -79.23 -69.68 62.96 -7.39 6.68 9 1 1 1 2 -65.01 58.81 120.87 -109.34 -57.85 52.34 -7.16 6.48 10 2 1 1 2 -65.02 58.83 57.64 -52.16 -121.40 109.84 -7.59 6.87 11 1 2 1 2 -64.99 58.73 221.50 -200.15 -48.11 43.47 -6.94 6.27 12 2 2 1 2 -65.00 58.76 88.90 -80.36 -70.62 63.84 -7.31 6.60 13 1 1 2 2 -65.01 58.80 121.32 -109.72 -58.23 52.66 -7.05 6.37 14 2 1 2 2 -65.02 58.82 57.22 -51.76 -125.29 113.34 -7.51 6.79 15 1 2 2 2 -64.99 58.71 223.34 -201.74 -48.38 43.70 -6.85 6.19 16 2 2 2 2 -65.00 58.75 89.67 -81.04 -70.81 63.99 -7.21 6.52 Percentage difference in outcomes obtained by using default value vs lower / upper bound in the sensitivity analysis parameter 190 Scena rio no Relevant outcomes Averted indirect medical costs / million $ Net benefits (medical system perspective) / million $ Net benefits (Medicare / Medicaid perspective) / million $ Net benefits (societal perspective) / million $ YrAdjBli nd=1 YrAdjBli nd=3 YrAdjBli nd=5 YrAdjBli nd=1 YrAdjBli nd=3 YrAdjBli nd=5 YrAdjBli nd=1 YrAdjBli nd=3 YrAdjBli nd=5 YrAdjBli nd=1 YrAdjBli nd=3 YrAdjBli nd=5 1 2.634 7.529 11.954 -8.785 -3.891 0.535 3.898 8.792 13.218 61.925 66.819 71.245 2 3.094 8.844 14.045 -15.250 -9.500 -4.299 -0.314 5.436 10.637 68.678 74.428 79.629 3 1.914 5.469 8.681 -5.319 -1.763 1.448 3.824 7.380 10.591 46.010 49.566 52.777 4 2.391 6.832 10.846 -9.619 -5.178 -1.164 1.899 6.340 10.354 54.398 58.839 62.853 5 2.716 7.766 12.331 -9.034 -3.985 0.581 4.000 9.049 13.615 64.907 69.957 74.522 6 3.180 9.092 14.439 -15.680 -9.768 -4.421 -0.411 5.501 10.848 71.645 77.557 82.904 7 1.983 5.668 8.996 -5.452 -1.767 1.560 3.991 7.676 11.004 48.847 52.532 55.860 8 2.472 7.066 11.217 -9.833 -5.239 -1.088 1.999 6.593 10.744 57.529 62.123 66.274 9 2.861 8.177 12.986 -9.713 -4.398 0.411 3.873 9.188 13.997 68.930 74.246 79.055 10 3.311 9.463 15.031 -16.827 -10.674 -5.107 -1.084 5.068 10.636 74.910 81.063 86.630 11 2.155 6.157 9.773 -5.809 -1.807 1.809 4.317 8.319 11.935 53.685 57.687 61.303 12 2.645 7.559 12.000 -10.440 -5.527 -1.085 2.044 6.957 11.399 62.347 67.260 71.702 13 2.927 8.366 13.286 -9.923 -4.484 0.436 3.902 9.342 14.261 71.744 77.183 82.103 14 3.370 9.634 15.300 -17.210 -10.946 -5.280 -1.264 4.999 10.666 77.152 83.415 89.082 15 2.209 6.310 10.015 -5.938 -1.836 1.868 4.376 8.477 12.182 55.762 59.863 63.568 16 2.709 7.741 12.289 -10.643 -5.611 -1.064 2.075 7.107 11.654 64.729 69.761 74.309 Summary results of the effect of varying the uncertain parameter on the most relevant affected outcomes for the particular parameter 191 B.2. PATIENT COMPLIANCE MODEL Scenari o no Scenario Relevant outcomes Averted blindness / % difference Cost of screening & treatment / % difference Net benefits (medical system perspective) / % difference Net benefits (Medicare / Medicaid perspective) / % difference Net benefits (societal perspective) / % difference Modalit y Scr_fre q Insuranc e Navigato r setting Lower boun d Upper boun d Lower boun d Upper boun d Lower boun d Upper boun d Lower boun d Upper boun d Lower boun d Upper boun d 1 1 1 1 1 -5.70 6.13 -7.51 7.40 -12.61 11.75 -3.34 3.44 -5.18 5.83 2 2 1 1 1 -4.90 4.83 -7.13 7.05 -9.68 9.75 -0.19 -3.86 -3.92 4.21 3 1 2 1 1 -8.72 8.12 -7.98 7.77 -6.07 9.16 -9.33 6.89 -8.82 8.28 4 2 2 1 1 -7.37 7.35 -6.94 6.86 -6.09 6.73 -9.89 6.72 -7.26 7.57 5 1 1 2 1 -6.03 5.47 -7.50 7.38 -12.22 12.73 -3.37 2.17 -5.66 5.08 6 2 1 2 1 -5.02 4.46 -7.17 7.05 -9.78 10.00 1.35 -5.25 -4.22 3.75 7 1 2 2 1 -8.89 7.75 -7.90 7.73 -7.58 9.17 -9.04 6.83 -8.97 7.82 8 2 2 2 1 -7.56 7.01 -6.93 6.88 -6.67 7.12 -9.28 6.67 -7.49 7.03 9 1 1 1 2 -4.04 5.78 -5.94 8.76 -8.94 16.29 -2.52 0.81 -3.61 5.14 10 2 1 1 2 -3.29 5.39 -5.73 8.55 -7.82 12.38 5.63 -9.68 -2.64 4.50 11 1 2 1 2 -6.50 8.48 -6.25 9.20 -2.92 14.94 -8.91 6.38 -6.29 8.41 12 2 2 1 2 -5.50 8.45 -5.57 8.29 -4.71 9.66 -6.95 6.59 -5.45 8.49 13 1 1 2 2 -4.20 5.02 -6.01 8.61 -9.74 16.65 -1.99 0.52 -3.80 4.16 14 2 1 2 2 -3.14 4.78 -5.78 8.46 -8.31 12.53 8.06 -11.08 -2.45 3.67 15 1 2 2 2 -5.76 8.44 -6.29 9.02 -6.65 12.91 -6.99 7.86 -5.46 8.23 16 2 2 2 2 -5.15 8.04 -5.63 8.15 -5.56 9.61 -5.47 7.50 -5.03 7.81 Percentage difference in outcomes obtained by using default value vs lower / upper bound in the sensitivity analysis parameter 192 Sce nar io no Relevant outcomes Averted blindness Costs of screening & treatment / million $ Net benefits (medical system perspective) / million $ Net benefits (Medicare / Medicaid perspective) / million $ Net benefits (societal perspective) / million $ Lowe r boun d Defa ult Uppe r boun d Lower bound Default Upper bound Lower bound Default Upper bound Lower bound Defa ult Uppe r boun d Lower bound Default Upper bound 1 3113 3301 3504 12.286 13.284 14.266 -3.400 -3.891 -4.348 8.499 8.792 9.095 63.360 66.819 70.718 2 3721 3912 4101 19.088 20.554 22.003 -8.580 -9.500 -10.426 5.426 5.436 5.226 71.508 74.428 77.564 3 2184 2393 2587 7.899 8.584 9.251 -1.656 -1.763 -1.925 6.691 7.380 7.888 45.192 49.566 53.672 4 2770 2991 3210 12.748 13.699 14.639 -4.862 -5.178 -5.526 5.713 6.340 6.766 54.565 58.839 63.296 5 3233 3440 3629 12.666 13.693 14.703 -3.498 -3.985 -4.492 8.744 9.049 9.245 65.995 69.957 73.507 6 3853 4057 4238 19.635 21.151 22.641 -8.813 -9.768 -10.744 5.576 5.501 5.213 74.284 77.557 80.469 7 2296 2520 2715 8.159 8.859 9.544 -1.633 -1.767 -1.929 6.982 7.676 8.200 47.820 52.532 56.638 8 2895 3132 3352 13.100 14.075 15.043 -4.890 -5.239 -5.612 5.981 6.593 7.033 57.470 62.123 66.491 9 3498 3645 3855 13.764 14.633 15.915 -4.005 -4.398 -5.114 8.957 9.188 9.263 71.564 74.246 78.064 10 4107 4246 4475 21.245 22.536 24.463 -9.839 -10.674 -11.996 5.353 5.068 4.578 78.923 81.063 84.708 11 2571 2749 2983 8.922 9.517 10.392 -1.754 -1.807 -2.077 7.578 8.319 8.849 54.060 57.687 62.539 12 3183 3369 3654 14.154 14.988 16.230 -5.267 -5.527 -6.060 6.474 6.957 7.415 63.591 67.260 72.967 13 3612 3770 3960 14.079 14.979 16.269 -4.047 -4.484 -5.230 9.155 9.342 9.390 74.251 77.183 80.395 14 4218 4355 4563 21.708 23.040 24.988 -10.03 -10.946 -12.317 5.402 4.999 4.445 81.370 83.415 86.474 15 2678 2842 3082 9.138 9.752 10.631 -1.714 -1.836 -2.073 7.884 8.477 9.143 56.595 59.863 64.788 16 3299 3478 3757 14.455 15.317 16.566 -5.300 -5.611 -6.150 6.718 7.107 7.640 66.254 69.761 75.212 Summary results of the effect of varying the uncertain parameter on the most relevant affected outcomes for the particular parameter 193 B.3. PATIENT COMPLIANCE MODEL Scenari o no Scenario Relevant outcomes Averted blindness / % difference Cost of screening & treatment / % difference Net benefits (medical system perspective) / % difference Net benefits (Medicare / Medicaid perspective) / % difference Net benefits (societal perspective) / % difference Modalit y Scr_fre q Insuranc e Navigato r setting Lower boun d Upper boun d Lower boun d Upper boun d Lower boun d Upper boun d Lower boun d Upper boun d Lower boun d Upper boun d 1 1 1 1 1 -40.35 21.92 -33.92 21.62 -30.01 28.48 -28.48 8.41 -43.93 23.46 2 2 1 1 1 -37.53 17.90 -31.54 20.64 -29.92 26.82 -24.23 -16.18 -41.18 18.62 3 1 2 1 1 -48.15 31.79 -35.38 23.37 -3.83 8.42 -45.54 23.81 -52.43 35.40 4 2 2 1 1 -45.73 28.36 -30.79 20.35 -13.65 14.68 -53.84 20.63 -51.24 32.37 5 1 1 2 1 -38.61 18.59 -33.12 19.87 -31.86 28.94 -25.14 5.88 -41.92 19.45 6 2 1 2 1 -35.83 15.11 -30.96 19.11 -30.75 26.04 -17.89 -19.25 -39.10 15.19 7 1 2 2 1 -46.60 27.58 -34.63 21.49 -5.55 9.79 -42.71 21.51 -50.67 30.22 8 2 2 2 1 -44.49 25.09 -30.26 18.90 -14.09 14.61 -50.47 18.85 -49.71 28.10 9 1 1 1 2 -31.59 12.99 -27.15 14.44 -26.48 21.22 -19.20 3.47 -34.42 13.46 10 2 1 1 2 -28.82 11.21 -25.51 14.04 -26.16 18.71 -4.36 -16.77 -31.76 11.27 11 1 2 1 2 -38.44 18.79 -28.53 15.61 -3.83 10.99 -34.26 13.26 -41.84 20.46 12 2 2 1 2 -35.71 18.25 -25.00 13.92 -13.42 11.76 -36.03 12.04 -40.01 20.40 13 1 1 2 2 -31.12 10.99 -26.28 13.13 -25.51 20.32 -18.45 3.44 -34.01 10.96 14 2 1 2 2 -27.63 9.77 -24.74 12.85 -25.89 17.27 -2.31 -14.92 -30.27 9.46 15 1 2 2 2 -36.90 17.73 -27.71 14.28 -6.50 8.07 -31.94 14.09 -40.09 19.14 16 2 2 2 2 -34.49 16.87 -24.40 12.74 -14.07 10.24 -34.44 12.61 -38.40 18.73 Percentage difference in outcomes obtained by using default value vs lower / upper bound in the sensitivity analysis parameter 194 Sc en ar io n o Relevant outcomes Averted blindness Costs of screening & treatment / million $ Net benefits (medical system perspective) / million $ Net benefits (Medicare / Medicaid perspective) / million $ Net benefits (societal perspective) / million $ Lowe r boun d Defa ult Uppe r boun d Lower bound Default Upper bound Lower bound Default Upper bound Lowe r boun d Defau lt Uppe r boun d Lower bound Default Upper bound 1 1969 3301 4025 8.778 13.284 16.156 -2.723 -3.891 -4.999 6.288 8.792 9.532 37.464 66.819 82.495 2 2444 3912 4613 14.071 20.554 24.797 -6.658 -9.500 -12.047 4.119 5.436 4.556 43.775 74.428 88.285 3 1241 2393 3154 5.548 8.584 10.591 -1.696 -1.763 -1.912 4.019 7.380 9.137 23.579 49.566 67.113 4 1623 2991 3839 9.481 13.699 16.487 -4.471 -5.178 -5.938 2.926 6.340 7.648 28.691 58.839 77.884 5 2112 3440 4080 9.158 13.693 16.413 -2.715 -3.985 -5.138 6.774 9.049 9.581 40.632 69.957 83.563 6 2603 4057 4669 14.603 21.151 25.192 -6.764 -9.768 -12.311 4.517 5.501 4.442 47.235 77.557 89.338 7 1346 2520 3215 5.791 8.859 10.763 -1.669 -1.767 -1.940 4.398 7.676 9.327 25.914 52.532 68.408 8 1739 3132 3918 9.816 14.075 16.735 -4.501 -5.239 -6.005 3.266 6.593 7.836 31.241 62.123 79.578 9 2493 3645 4118 10.660 14.633 16.746 -3.233 -4.398 -5.331 7.424 9.188 9.507 48.690 74.246 84.237 10 3023 4246 4722 16.788 22.536 25.700 -7.881 -10.674 -12.671 4.848 5.068 4.218 55.321 81.063 90.199 11 1693 2749 3266 6.802 9.517 11.003 -1.737 -1.807 -2.005 5.469 8.319 9.422 33.549 57.687 69.492 12 2166 3369 3984 11.240 14.988 17.074 -4.785 -5.527 -6.177 4.451 6.957 7.794 40.348 67.260 80.982 13 2597 3770 4185 11.042 14.979 16.946 -3.340 -4.484 -5.395 7.618 9.342 9.663 50.936 77.183 85.642 14 3152 4355 4780 17.339 23.040 26.001 -8.113 -10.946 -12.837 4.884 4.999 4.253 58.168 83.415 91.309 15 1793 2842 3346 7.050 9.752 11.144 -1.717 -1.836 -1.984 5.769 8.477 9.672 35.866 59.863 71.321 16 2278 3478 4064 11.579 15.317 17.268 -4.822 -5.611 -6.186 4.659 7.107 8.003 42.971 69.761 82.828 Summary results of the effect of varying the uncertain parameter on the most relevant affected outcomes for the particular parameter 195 B.4. PATIENT NAVIGATOR Sc en ar io n o Scenario Relevant outcomes Averted blindness / % difference Cost of screening & treatment / % difference Net benefits (medical system perspective) / % difference Net benefits (Medicare / Medicaid perspective) / % difference Net benefits (societal perspective) / % difference Modalit y Scr_fre q Insuranc e Navigato r setting Lower bound Upper bound Lowe r boun d Uppe r boun d Lowe r boun d Uppe r boun d Lowe r boun d Uppe r boun d Lowe r boun d Uppe r boun d 1 1 1 1 1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 2 2 1 1 1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 3 1 2 1 1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 4 2 2 1 1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 5 1 1 2 1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 6 2 1 2 1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 7 1 2 2 1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 8 2 2 2 1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 9 1 1 1 2 -2.13 1.71 -2.27 1.75 -3.23 2.24 -0.13 1.25 -2.31 1.71 10 2 1 1 2 -1.81 1.24 -2.21 1.73 -3.05 2.36 4.19 -2.30 -1.94 1.19 11 1 2 1 2 -3.44 2.37 -2.39 1.93 -0.02 0.44 -2.81 2.47 -3.82 2.49 12 2 2 1 2 -2.94 2.22 -2.16 1.73 -1.74 1.25 -1.46 2.25 -3.38 2.41 13 1 1 2 2 -1.88 1.40 -2.08 1.59 -2.95 1.88 -0.40 1.73 -1.99 1.27 14 2 1 2 2 -1.65 0.98 -2.01 1.58 -2.78 2.22 4.55 -1.58 -1.81 0.81 15 1 2 2 2 -2.84 1.90 -2.21 1.68 -1.00 0.01 -2.66 2.81 -3.00 1.85 16 2 2 2 2 -2.76 1.75 -2.00 1.53 -1.53 1.23 -1.74 2.59 -3.14 1.73 Percentage difference in outcomes obtained by using default value vs lower / upper bound in the sensitivity analysis parameter 196 Sc en ar io n o Relevant outcomes Averted blindness Costs of screening & treatment / million $ Net benefits (medical system perspective) / million $ Net benefits (Medicare / Medicaid perspective) / million $ Net benefits (societal perspective) / million $ Lowe r boun d Defa ult Uppe r boun d Lower bound Defaul t Upper bound Lower bound Default Upper bound Lowe r boun d Defaul t Uppe r boun d Lower bound Defaul t Upper bound 1 3301 3301 3301 13.284 13.284 13.284 -3.891 -3.891 -3.891 8.792 8.792 8.792 66.819 66.819 66.819 2 3912 3912 3912 20.554 20.554 20.554 -9.500 -9.500 -9.500 5.436 5.436 5.436 74.428 74.428 74.428 3 2393 2393 2393 8.584 8.584 8.584 -1.763 -1.763 -1.763 7.380 7.380 7.380 49.566 49.566 49.566 4 2991 2991 2991 13.699 13.699 13.699 -5.178 -5.178 -5.178 6.340 6.340 6.340 58.839 58.839 58.839 5 3440 3440 3440 13.693 13.693 13.693 -3.985 -3.985 -3.985 9.049 9.049 9.049 69.957 69.957 69.957 6 4057 4057 4057 21.151 21.151 21.151 -9.768 -9.768 -9.768 5.501 5.501 5.501 77.557 77.557 77.557 7 2520 2520 2520 8.859 8.859 8.859 -1.767 -1.767 -1.767 7.676 7.676 7.676 52.532 52.532 52.532 8 3132 3132 3132 14.075 14.075 14.075 -5.239 -5.239 -5.239 6.593 6.593 6.593 62.123 62.123 62.123 9 3567 3645 3707 14.301 14.633 14.890 -4.256 -4.398 -4.496 9.176 9.188 9.303 72.530 74.246 75.516 10 4169 4246 4299 22.037 22.536 22.925 -10.348 -10.674 -10.926 5.280 5.068 4.952 79.487 81.063 82.029 11 2655 2749 2815 9.290 9.517 9.701 -1.806 -1.807 -1.815 8.085 8.319 8.524 55.481 57.687 59.123 12 3270 3369 3444 14.665 14.988 15.248 -5.431 -5.527 -5.596 6.855 6.957 7.113 64.985 67.260 68.879 13 3699 3770 3823 14.668 14.979 15.217 -4.352 -4.484 -4.568 9.304 9.342 9.503 75.650 77.183 78.166 14 4283 4355 4398 22.577 23.040 23.403 -10.642 -10.946 -11.189 5.227 4.999 4.920 81.909 83.415 84.090 15 2761 2842 2896 9.536 9.752 9.916 -1.818 -1.836 -1.837 8.251 8.477 8.715 58.064 59.863 60.973 16 3382 3478 3538 15.010 15.317 15.550 -5.526 -5.611 -5.681 6.983 7.107 7.291 67.569 69.761 70.967 Summary results of the effect of varying the uncertain parameter on the most relevant affected outcomes for the particular parameter 197 B.5. CLINIC IMPLEMENTATION Scenari o no Scenario Relevant outcomes Averted blindness / % difference Cost of screening & treatment / % difference Net benefits (medical system perspective) / % difference Net benefits (Medicare / Medicaid perspective) / % difference Net benefits (societal perspective) / % difference Modalit y Scr_fre q Insuranc e Navigato r setting Lower bound Upper bound Lowe r boun d Uppe r boun d Lowe r boun d Uppe r boun d Lowe r boun d Uppe r boun d Lowe r boun d Uppe r boun d 1 1 1 1 1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 2 2 1 1 1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 3 1 2 1 1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 4 2 2 1 1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 5 1 1 2 1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 6 2 1 2 1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 7 1 2 2 1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 8 2 2 2 1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 9 1 1 1 2 -3.02 2.80 -2.80 2.65 -2.91 3.29 -2.70 1.29 -3.09 2.99 10 2 1 1 2 -2.20 2.34 -2.63 2.55 -3.46 3.05 1.48 -0.93 -2.13 2.40 11 1 2 1 2 -4.18 3.58 -3.01 2.87 -0.36 2.49 -4.04 2.02 -4.49 3.97 12 2 2 1 2 -3.55 3.31 -2.60 2.56 -1.86 2.09 -3.22 2.67 -3.93 3.64 13 1 1 2 2 -2.89 1.98 -2.61 2.47 -2.63 3.96 -2.08 0.35 -3.06 1.98 14 2 1 2 2 -1.92 2.04 -2.46 2.41 -3.34 3.08 2.98 -2.11 -1.88 2.06 15 1 2 2 2 -3.71 3.30 -2.85 2.72 -0.87 1.75 -2.85 2.35 -4.09 3.59 16 2 2 2 2 -3.08 3.04 -2.47 2.42 -2.04 2.14 -1.93 2.43 -3.44 3.30 Percentage difference in outcomes obtained by using default value vs lower / upper bound in the sensitivity analysis parameter 198 Sc en ar io n o Relevant outcomes Averted blindness Costs of screening & treatment / million $ Net benefits (medical system perspective) / million $ Net benefits (Medicare / Medicaid perspective) / million $ Net benefits (societal perspective) / million $ Lowe r boun d Defa ult Uppe r boun d Lower bound Default Upper bound Lower bound Default Upper bound Lowe r boun d Defaul t Uppe r boun d Lower bound Default Upper bound 1 3301 3301 3301 13.284 13.284 13.284 -3.891 -3.891 -3.891 8.792 8.792 8.792 66.819 66.819 66.819 2 3912 3912 3912 20.554 20.554 20.554 -9.500 -9.500 -9.500 5.436 5.436 5.436 74.428 74.428 74.428 3 2393 2393 2393 8.584 8.584 8.584 -1.763 -1.763 -1.763 7.380 7.380 7.380 49.566 49.566 49.566 4 2991 2991 2991 13.699 13.699 13.699 -5.178 -5.178 -5.178 6.340 6.340 6.340 58.839 58.839 58.839 5 3440 3440 3440 13.693 13.693 13.693 -3.985 -3.985 -3.985 9.049 9.049 9.049 69.957 69.957 69.957 6 4057 4057 4057 21.151 21.151 21.151 -9.768 -9.768 -9.768 5.501 5.501 5.501 77.557 77.557 77.557 7 2520 2520 2520 8.859 8.859 8.859 -1.767 -1.767 -1.767 7.676 7.676 7.676 52.532 52.532 52.532 8 3132 3132 3132 14.075 14.075 14.075 -5.239 -5.239 -5.239 6.593 6.593 6.593 62.123 62.123 62.123 9 3535 3645 3747 14.223 14.633 15.020 -4.270 -4.398 -4.542 8.940 9.188 9.307 71.949 74.246 76.463 10 4153 4246 4346 21.942 22.536 23.109 -10.305 -10.674 -10.999 5.143 5.068 5.021 79.338 81.063 83.011 11 2635 2749 2848 9.231 9.517 9.790 -1.800 -1.807 -1.852 7.983 8.319 8.487 55.095 57.687 59.976 12 3249 3369 3480 14.598 14.988 15.371 -5.424 -5.527 -5.642 6.733 6.957 7.143 64.617 67.260 69.706 13 3662 3770 3845 14.588 14.979 15.349 -4.366 -4.484 -4.661 9.148 9.342 9.374 74.824 77.183 78.709 14 4271 4355 4444 22.473 23.040 23.595 -10.580 -10.946 -11.283 5.148 4.999 4.894 81.844 83.415 85.137 15 2736 2842 2935 9.474 9.752 10.017 -1.820 -1.836 -1.869 8.236 8.477 8.676 57.414 59.863 62.014 16 3371 3478 3584 14.938 15.317 15.688 -5.497 -5.611 -5.731 6.969 7.107 7.279 67.361 69.761 72.066 Summary results of the effect of varying the uncertain parameter on the most relevant affected outcomes for the particular parameter 199 B.6. CLINIC SIZE Scenari o no Scenario Relevant outcomes Averted blindness / % difference Cost of screening & treatment / % difference Net benefits (medical system perspective) / % difference Net benefits (Medicare / Medicaid perspective) / % difference Net benefits (societal perspective) / % difference Modalit y Scr_fre q Insuranc e Navigato r setting Lower bound Upper bound Lowe r boun d Uppe r boun d Lowe r boun d Uppe r boun d Lowe r boun d Uppe r boun d Lowe r boun d Uppe r boun d 1 1 1 1 1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 2 2 1 1 1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 3 1 2 1 1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 4 2 2 1 1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 5 1 1 2 1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 6 2 1 2 1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 7 1 2 2 1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 8 2 2 2 1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 9 1 1 1 2 1.23 -1.85 1.68 -1.08 3.60 -0.71 -1.06 -0.98 1.29 -2.16 10 2 1 1 2 1.45 -1.90 1.63 -0.94 2.22 -0.66 -2.50 -0.89 1.55 -2.37 11 1 2 1 2 1.33 -2.24 1.74 -1.21 5.09 0.93 -0.57 -1.60 1.43 -2.60 12 2 2 1 2 1.12 -2.38 1.54 -0.97 2.98 0.23 -1.06 -1.54 1.07 -3.01 13 1 1 2 2 1.06 -2.19 1.57 -0.98 3.13 -0.14 -0.28 -1.35 1.00 -2.62 14 2 1 2 2 1.32 -1.79 1.53 -0.86 2.04 -0.59 -1.46 -0.53 1.32 -2.26 15 1 2 2 2 1.50 -2.06 1.62 -1.10 3.53 0.21 0.46 -1.10 1.56 -2.42 16 2 2 2 2 1.28 -2.27 1.47 -0.90 2.56 0.32 -0.08 -1.34 1.22 -2.90 Percentage difference in outcomes obtained by using default value vs lower / upper bound in the sensitivity analysis parameter 200 Sc en ar io n o Relevant outcomes Averted blindness Costs of screening & treatment / million $ Net benefits (medical system perspective) / million $ Net benefits (Medicare / Medicaid perspective) / million $ Net benefits (societal perspective) / million $ Lowe r boun d Defa ult Uppe r boun d Lower bound Default Upper bound Lower bound Default Upper bound Lowe r boun d Defaul t Uppe r boun d Lower bound Default Upper bound 1 3301 3301 3301 13.284 13.284 13.284 -3.891 -3.891 -3.891 8.792 8.792 8.792 66.819 66.819 66.819 2 3912 3912 3912 20.554 20.554 20.554 -9.500 -9.500 -9.500 5.436 5.436 5.436 74.428 74.428 74.428 3 2393 2393 2393 8.584 8.584 8.584 -1.763 -1.763 -1.763 7.380 7.380 7.380 49.566 49.566 49.566 4 2991 2991 2991 13.699 13.699 13.699 -5.178 -5.178 -5.178 6.340 6.340 6.340 58.839 58.839 58.839 5 3440 3440 3440 13.693 13.693 13.693 -3.985 -3.985 -3.985 9.049 9.049 9.049 69.957 69.957 69.957 6 4057 4057 4057 21.151 21.151 21.151 -9.768 -9.768 -9.768 5.501 5.501 5.501 77.557 77.557 77.557 7 2520 2520 2520 8.859 8.859 8.859 -1.767 -1.767 -1.767 7.676 7.676 7.676 52.532 52.532 52.532 8 3132 3132 3132 14.075 14.075 14.075 -5.239 -5.239 -5.239 6.593 6.593 6.593 62.123 62.123 62.123 9 3690 3645 3577 14.878 14.633 14.475 -4.556 -4.398 -4.367 9.091 9.188 9.098 75.201 74.246 72.642 10 4308 4246 4166 22.902 22.536 22.323 -10.911 -10.674 -10.603 4.941 5.068 5.023 82.320 81.063 79.141 11 2786 2749 2688 9.682 9.517 9.402 -1.899 -1.807 -1.823 8.271 8.319 8.186 58.510 57.687 56.187 12 3407 3369 3289 15.218 14.988 14.843 -5.691 -5.527 -5.540 6.883 6.957 6.850 67.979 67.260 65.239 13 3810 3770 3688 15.214 14.979 14.832 -4.624 -4.484 -4.477 9.316 9.342 9.216 77.952 77.183 75.161 14 4412 4355 4277 23.393 23.040 22.842 -11.169 -10.946 -10.882 4.926 4.999 4.973 84.520 83.415 81.528 15 2884 2842 2783 9.910 9.752 9.645 -1.901 -1.836 -1.840 8.516 8.477 8.384 60.795 59.863 58.416 16 3522 3478 3399 15.542 15.317 15.178 -5.755 -5.611 -5.629 7.101 7.107 7.012 70.614 69.761 67.738 Summary results of the effect of varying the uncertain parameter on the most relevant affected outcomes for the particular parameter 201 B.7. DISCOUNT FACTOR Sc en ar io n o Scenario Relevant outcomes Averted blindness / % difference Cost of screening & treatment / % difference Net benefits (medical system perspective) / % difference Net benefits (Medicare / Medicaid perspective) / % difference Net benefits (societal perspective) / % difference Mo dali ty Sc r_ fr eq In su ra nc e Navi gato r setti ng Lower bound Upper bound Lowe r boun d Upper bound Lowe r boun d Upper bound Lowe r boun d Upper bound Lowe r boun d Upper bound 1 1 1 1 1 38.66 -24.94 26.25 -17.58 26.77 -15.70 51.67 -32.31 37.94 -24.88 2 2 1 1 1 38.11 -24.61 27.97 -18.38 30.99 -18.94 64.60 -40.43 37.61 -24.75 3 1 2 1 1 38.44 -24.80 25.93 -17.30 26.07 -13.71 47.86 -30.19 37.55 -24.63 4 2 2 1 1 37.86 -24.49 28.05 -18.25 32.73 -18.97 50.89 -32.74 37.00 -24.43 5 1 1 2 1 38.79 -25.00 26.31 -17.61 27.17 -15.87 51.94 -32.39 38.02 -24.91 6 2 1 2 1 38.13 -24.62 28.00 -18.40 31.21 -19.06 65.15 -40.68 37.56 -24.72 7 1 2 2 1 38.86 -24.98 25.98 -17.32 25.86 -13.50 48.46 -30.40 37.97 -24.81 8 2 2 2 1 38.19 -24.64 28.04 -18.25 32.62 -18.91 51.84 -33.04 37.31 -24.56 9 1 1 1 2 42.39 -22.71 29.35 -15.25 31.14 -13.23 54.79 -31.95 41.86 -22.47 10 2 1 1 2 40.84 -22.57 30.91 -16.14 34.95 -16.58 68.32 -43.99 40.32 -22.60 11 1 2 1 2 43.56 -22.16 29.09 -14.66 27.95 -10.56 51.53 -29.00 43.14 -21.66 12 2 2 1 2 42.11 -21.82 30.72 -15.85 34.89 -16.98 55.71 -31.15 41.58 -21.43 13 1 1 2 2 41.45 -23.43 29.17 -15.41 32.02 -12.66 54.03 -32.89 40.66 -23.34 14 2 1 2 2 40.45 -22.80 30.77 -16.26 35.10 -16.63 68.07 -45.62 39.84 -22.84 15 1 2 2 2 43.31 -22.44 28.92 -14.79 26.96 -11.04 52.32 -28.90 42.74 -22.02 16 2 2 2 2 41.89 -22.08 30.54 -15.95 34.99 -16.95 55.93 -31.54 41.24 -21.74 Percentage difference in outcomes obtained by using default value vs lower / upper bound in the sensitivity analysis parameter 202 Sc en ar io n o Relevant outcomes Averted blindness Costs of screening & treatment / million $ Net benefits (medical system perspective) / million $ Net benefits (Medicare / Medicaid perspective) / million $ Net benefits (societal perspective) / million $ Lowe r boun d Defa ult Uppe r boun d Lower bound Default Upper bound Lower bound Default Upper bound Lower bound Defaul t Uppe r boun d Lower bound Default Upper bound 1 4577 3301 2478 16.771 13.284 10.948 -4.932 -3.891 -3.280 13.336 8.792 5.952 92.172 66.819 50.193 2 5403 3912 2949 26.302 20.554 16.775 -12.444 -9.500 -7.701 8.948 5.436 3.238 102.418 74.428 56.010 3 3313 2393 1800 10.811 8.584 7.100 -2.223 -1.763 -1.522 10.912 7.380 5.152 68.180 49.566 37.357 4 4123 2991 2258 17.541 13.699 11.198 -6.873 -5.178 -4.196 9.566 6.340 4.264 80.608 58.839 44.467 5 4775 3440 2580 17.296 13.693 11.282 -5.067 -3.985 -3.352 13.749 9.049 6.118 96.554 69.957 52.528 6 5603 4057 3058 27.074 21.151 17.259 -12.816 -9.768 -7.906 9.086 5.501 3.264 106.685 77.557 58.388 7 3499 2520 1890 11.161 8.859 7.325 -2.224 -1.767 -1.529 11.396 7.676 5.342 72.479 52.532 39.498 8 4328 3132 2361 18.022 14.075 11.506 -6.949 -5.239 -4.249 10.011 6.593 4.415 85.303 62.123 46.868 9 5190 3645 2817 18.928 14.633 12.402 -5.767 -4.398 -3.816 14.222 9.188 6.253 105.326 74.246 57.560 10 5981 4246 3288 29.501 22.536 18.899 -14.404 -10.674 -8.904 8.531 5.068 2.839 113.750 81.063 62.746 11 3947 2749 2140 12.285 9.517 8.122 -2.312 -1.807 -1.616 12.606 8.319 5.906 82.571 57.687 45.191 12 4787 3369 2634 19.593 14.988 12.613 -7.455 -5.527 -4.588 10.833 6.957 4.790 95.228 67.260 52.844 13 5333 3770 2887 19.349 14.979 12.671 -5.919 -4.484 -3.916 14.389 9.342 6.269 108.565 77.183 59.171 14 6117 4355 3362 30.128 23.040 19.293 -14.788 -10.946 -9.126 8.402 4.999 2.719 116.649 83.415 64.360 15 4072 2842 2204 12.572 9.752 8.310 -2.331 -1.836 -1.634 12.912 8.477 6.027 85.449 59.863 46.684 16 4934 3478 2710 19.995 15.317 12.874 -7.575 -5.611 -4.660 11.082 7.107 4.865 98.531 69.761 54.597 Summary results of the effect of varying the uncertain parameter on the most relevant affected outcomes for the particular parameter 203 APPENDIX C: SAMPLE CODES C.1. CODE FOR DR PROGRESSION SUB-MODULE function State_pat=MicrosimDiseaseProgression(State_pat, State_prob, i, k, index, RandNo) %Finds which disease state the patient by comparing random number from %RandNo %for each year i, patient k; index is for which RandNo is used Cumsum_State_prob=cumsum(State_prob(i+1,:,k),2); %Find smallest (i.e. 1st) index that satisfies rand()<cumsum_state_prob PatState = find(RandNo(index)<Cumsum_State_prob,1); State_pat(i+1,PatState,k)=1; end C.2. CODE FOR THE MAIN FUNCTION FOR ONE ITERATION OF THE DR SCREENING SIMULATION function [PV_Results, PropPatientsNs, PropPatientsScr] = MainDRSimScrFreq(model, modality, scr_freq, ins_setting, patnav_setting, no_yrs, rep, param, ... Mortality, TransitionBase, ScrStrategyChar, no_patients, PatCharRaw, PatChar, ClinicChar, ... TrScr, SensScr, SpecScr, ... modality_cost, extendedvisit_cost, treatment_cost, posttreatment_cost, benefit_breakdown, portion_benefit, scaling_factor, ... no_states, State_prob_ns, Transition_pat_ns) %This is the main method for one iteration of the DR simulation %modality: 1=opthalmoscope, 2=nonmydriatic camera, 3=mydriatic %camera %scr_freq: 1=annual, 2=biennial (every 2 years), 3=every 3 years, ... %ins_setting: 1=status quo, 2=mandatory insurance %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %State_pat_ns is the patient's actual state, other than the initial state %(i.e. at year 1), the rest are initialized as zeros. State_prob_scr=State_prob_ns; State_pat_ns=State_prob_ns; State_pat_scr=State_prob_ns; 204 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %These keep track of decision states for cost calculation purposes %This indicates if a patient in a particular year was diagnosed as a positive (1 if so, 0 otherwise). Those with positive diagnosis are followed up with gold standard test to confirm diagnosis. %Latent disease state remains the same no matter what the screening outcome is until disease state is progressed PositiveDiagnosis = zeros (no_yrs+1, no_patients); %This indicates if a patient in a particular year complies with screening (1 if so, 0 otherwise) CompScr = zeros (no_yrs+1, no_patients); %This indicates if a patient in a particular year complies with treatment (1 if so, 0 otherwise) CompFollowUp = zeros (no_yrs+1, no_patients); %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %Set the appropriate adjusted / unadjusted patient compliance based on %screening strategy component setting and model chosen %Calc. compliance based on patient characteristics and adjusted compliance %based on screening strategy component chosen [AdjPatLvlCompliance, PatLvlCompliance, BaselineCompliance, ClinicComplies] = CalcCompliance(modality,scr_freq,ins_setting,patnav_setting, ... no_patients, no_yrs, PatCharRaw, ClinicChar, ScrStrategyChar, param); %Use the appropriate patient compliance based on the model chosen (if 1, %patcompliance = baselinecompliance, if 2 = patlvlcompliance) PatCompliance = BaselineCompliance; if model==1 %%Set patient compliance to be average compliance PatCompliance(:,:) = mean(BaselineCompliance(1,:)); elseif model ==2 PatCompliance = PatLvlCompliance; elseif model == 3 PatCompliance = AdjPatLvlCompliance; end; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %Uniform number generator %Use a specific seed for random no generator (0 is the default generator %setting, so if rep=1, seed=0 or default) rng(1000*(rep-1)); 205 %Generate enough random numbers for the disease %progression module. The multiplicant 4 is indicate the number of RandNo %needed for every iteration when screening takes place, i.e. for disease %progression, diagnosis, compliance with screening and follow-up. When no %screening takes place, only 1 RandNo is needed for disease progression. %Thus this creates enough RandNo for the case where max RandNo is needed %i.e. when there's screening every year. RandNo=rand(no_yrs*no_patients*4,1); %Starting index of the RandNo, will be incremented in the disease %progression module index=1; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %Disease progression module for i=1:no_yrs for k=1:no_patients %for each patient (patient index k) %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %no screening State_prob_ns(i+1,:,k)=State_pat_ns(i,:,k)*Transition_pat_ns(:,:,i,k); State_pat_ns = MicrosimDiseaseProgression(State_pat_ns, State_prob_ns, i, k, index, RandNo); %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %with screening % Screen the first year of any given screening frequency. if scr_freq == 1 ... % Screen every year if annual screening || mod(i, scr_freq) == 1 % Screen the first year of a given frequency %if patient is in low or high risk state and complies with screening if ((find(State_pat_scr(i,:,k))<=2) && (RandNo(index+1)<PatCompliance(i,k))) CompScr(i+1,k)=1; 206 %find out if patient's diagnosis is +ve, i.e. false positive if patient in low risk (no DR), or true positive if patient in high risk if (State_pat_scr(i,1,k)==1 && (RandNo(index+2)>SpecScr(modality))) || (State_pat_scr(i,2,k)==1 && (RandNo(index+2)<SensScr(modality))) PositiveDiagnosis(i+1,k)=1; end; %if true positive and patient follows up with treatment, natural disease progression is modified if (State_pat_scr(i,2,k)==1 && (RandNo(index+2)<SensScr(modality))) && (RandNo(index+3)<PatChar.CompFollowUp) State_prob_scr(i+1,:,k)=State_pat_scr(i,:,k)*TrScr(:,:,modality)*Transition_p at_ns(:,:,i,k); State_pat_scr = MicrosimDiseaseProgression(State_pat_scr, State_prob_scr, i, k, index, RandNo); CompFollowUp(i+1,k)=1; %else (false positive or negative diagnosis), or true positive but doesn't follow up with treatment, patient follows natural disease progression else State_prob_scr(i+1,:,k)= State_pat_scr(i,:,k)*Transition_pat_ns(:,:,i,k); State_pat_scr = MicrosimDiseaseProgression(State_pat_scr, State_prob_scr, i, k, index, RandNo); end; %if patients are in (treated, blind, dead) state, or are in low / high risk state but do not comply with screening, progress through natural %disease progression else State_prob_scr(i+1,:,k)= State_pat_scr(i,:,k)*Transition_pat_ns(:,:,i,k); State_pat_scr = MicrosimDiseaseProgression(State_pat_scr, State_prob_scr, i, k, index, RandNo); end; %for non-screening years else % Adjust disease progression for non-screening years State_prob_scr(i+1,:,k)= State_pat_scr(i,:,k)*Transition_pat_ns(:,:,i,k); State_pat_scr = MicrosimDiseaseProgression(State_pat_scr, State_prob_scr, i, k, index, RandNo); 207 end; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% index=index+4; end; end; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %Outcome measures state_treated=3; state_blind=4; %Discount factor = 0.05 disc_fact = 0.05; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %Sight years %Find no of yrs of averted blindness (i.e. sight yrs saved) %Sum the number of people who are not in the blind or dead state (i.e. in states 1-3) %over all patients (i.e. dimension 3) and across the three states (i.e. dimension 2) %to find the number of total sight years %Create a matrix where the vector PatChar.Weight is duplicated by the no of %yrs+1, so we can do elementwise multiplication to find weighted results %below PatCharWeight=repmat(PatChar.Weight,1,no_yrs+1)'; SightYrsSaved = sum((State_pat_ns(:,state_blind,:)- State_pat_scr(:,state_blind,:)),3); %This is the same thing as SightYears WeightedSightYrsSaved = sum(squeeze(State_pat_ns(:,state_blind,:)- State_pat_scr(:,state_blind,:)).*PatCharWeight,2); %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %Find which year someone first becomes blind %No of yrs needed to adjust after being blind (this is for benefit - %indirect med calculation); after 3 yrs (i.e. 2 yrs after becoming blind) we assume people have adjusted to %being blind and will not incur indirect medical costs 208 FirstBlind_ns=zeros(no_yrs+1,no_patients); FirstBlind_scr=zeros(no_yrs+1,no_patients); for k=1:no_patients YrFirstBlind_ns=find(State_pat_ns(:,state_blind,k),1,'first'); YrFirstBlind_scr=find(State_pat_scr(:,state_blind,k),1,'first'); if YrFirstBlind_ns <= (no_yrs+1-param.YrsAdjustAfterBlind) FirstBlind_ns(YrFirstBlind_ns:YrFirstBlind_ns+param.YrsAdjustAfterBlind,k)=1; else %if blind in the last few yrs of the simulation, count those yrs only until end of simulation FirstBlind_ns(YrFirstBlind_ns:no_yrs+1,k)=1; end; if YrFirstBlind_scr <= (no_yrs+1-param.YrsAdjustAfterBlind) FirstBlind_scr(YrFirstBlind_scr:YrFirstBlind_scr+param.YrsAdjustAfterBlind,k) =1; else FirstBlind_scr(YrFirstBlind_scr:no_yrs+1,k)=1; end; end; %This is the number of "adjustment period" years saved with screening, i.e. %the default is the first 3 years of blindness (no of yrs is specified in %param in the input) WeightedFirstBlind_ns=FirstBlind_ns.*PatCharWeight; WeightedFirstBlind_scr=FirstBlind_scr.*PatCharWeight; %sum across second dimension i.e. over all patients WeightedAdjustmentPeriodSaved = sum(WeightedFirstBlind_ns,2)- sum(WeightedFirstBlind_scr,2); %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %Costs (across years with index i) no_clinics = max(PatChar.Clinic); WeightedCompScr=CompScr.*PatCharWeight; WeightedPositiveDiagnosis=PositiveDiagnosis.*PatCharWeight; WeightedCompFollowUp=CompFollowUp.*PatCharWeight; WeightedTreated=squeeze(State_pat_scr(:,state_treated,:)).*PatCharWeight; if model==1 || model ==2 WeightedModalityCost=sum(WeightedCompScr,2)*modality_cost.perpatient(modality ); ProviderLvlPolicyCost=0; elseif model ==3 WeightedModalityCost=no_clinics*modality_cost.fixed(modality) + sum(WeightedCompScr,2)*modality_cost.var(modality); if patnav_setting==2 ProviderLvlPolicyCost=length(ClinicComplies)*ScrStrategyChar.NavigatorOR; else ProviderLvlPolicyCost=0; end; 209 end; WeightedExtendedVisitCost=sum(WeightedPositiveDiagnosis,2)*extendedvisit_cost ; WeightedTreatmentCost=sum(WeightedCompFollowUp,2)*treatment_cost; WeightedPostTreatmentCost=sum(WeightedTreated,2)*posttreatment_cost; Cost_scr = WeightedModalityCost + WeightedExtendedVisitCost + WeightedTreatmentCost + WeightedPostTreatmentCost + ProviderLvlPolicyCost; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %Benefits %Calculate total years of averted blindness under the age of 65 (for productivity %loss calculation) Averted_blind_under65=zeros(length(SightYrsSaved),1); WeightedAverted_blind_under65=Averted_blind_under65; for i=1:length(SightYrsSaved) for k=1:no_patients if (i<65-PatChar.Age(k)) && (State_pat_ns(i,4,k)==1) && (State_pat_scr(i,4,k)==0) Averted_blind_under65(i)=Averted_blind_under65(i)+1; WeightedAverted_blind_under65(i)=WeightedAverted_blind_under65(i)+PatChar.Wei ght(k); end; end; end; %Averted blindness over age 65 (for nursing home cost / direct nonmedical cost calculation) WeightedAverted_blind_over65=WeightedSightYrsSaved- WeightedAverted_blind_under65; %Key: Benefit=[direct_med indirect_med direct_nonmed lostwages reducedwages] Benefit=zeros(no_yrs+1,6); Benefit(:,1) = WeightedSightYrsSaved*benefit_breakdown.direct_med; Benefit(:,2) = WeightedAdjustmentPeriodSaved*benefit_breakdown.indirect_med*portion_benefit. indirect_med; Benefit(:,3) = WeightedAverted_blind_over65*benefit_breakdown.direct_nonmed*portion_benefit. direct_nonmed; Benefit(:,4) = WeightedAverted_blind_under65*benefit_breakdown.lostwages*portion_benefit.los twages; Benefit(:,5) = WeightedAverted_blind_under65*benefit_breakdown.reducedwages*portion_benefit. reducedwages; Benefit(:,6) = sum(Benefit(:,1:5),2); %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 210 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %Calculate Net-benefit lengthBenefit=size(Benefit,2); CostBenefit=zeros(no_yrs+1,lengthBenefit); PV_CostBenefit=zeros(1,lengthBenefit); PV_Benefit=PV_CostBenefit; for index=1:lengthBenefit PV_Benefit (index) = npv (Benefit(:,index), disc_fact); CostBenefit(:,index)=Benefit(:,index) - Cost_scr; PV_CostBenefit(index) = npv(CostBenefit(:,index),disc_fact); %Net-benefit end; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %Calculate present value of outcomes and costs PV_UnweightedSightYears = npv (SightYrsSaved, disc_fact); PV_AvertedBlindOver65 = npv (WeightedAverted_blind_over65, disc_fact); PV_AvertedBlindUnder65 = npv (WeightedAverted_blind_under65, disc_fact); PV_SightYearsSaved = npv (WeightedSightYrsSaved, disc_fact); PV_Cost_scr = npv (Cost_scr, disc_fact); PV_Results=[PV_SightYearsSaved PV_AvertedBlindUnder65 PV_AvertedBlindOver65 PV_Cost_scr PV_Benefit]; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %Generate summary measures %Proportion of patients in each state PropPatientsNs=sum(State_pat_ns(:,:,:),3); PropPatientsScr=sum(State_pat_scr(:,:,:),3); %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% if model==1 save DRResultsMod1; elseif model == 2 save DRResultsMod2; else save DRResultsMod3; end; end 211 C.3. CODE FOR COMPLIANCE MODULE %Calculates baseline patient compliance (PatCompliance) based on patient characteristics function [AdjPatLvlCompliance, PatLvlCompliance, BaselineCompliance, ClinicComplies] = CalcCompliance(modality,scr_freq,ins_setting, patnav_setting,... no_patients, no_yrs, PatCharRaw, ClinicChar, ScrStrategyChar, param) %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %Patient characteristis %Read indep. variables needed for calculation of baseline compliance start_age = PatCharRaw(:,1); %Dummy vars for age categories age_1 = PatCharRaw(:,12); age_2 = PatCharRaw(:,13); %poverty = less than 200% lvl poverty, insulin = whether take insulin or %not, insurance = have insurance last 12 mths poverty = PatCharRaw(:,3); insulin = PatCharRaw(:,11); insurance = PatCharRaw(:,10); %whether doctor check HbA1c level and feet dr_chck_hba1c = PatCharRaw(:,4); dr_chk_feet = PatCharRaw(:,5); %dummy vars for how often check blood sugar level, freq of fastfood %consumption, confidence in managing diabetes diab_sgrchk_1 = PatCharRaw(:,14); diab_sgrchk_2 = PatCharRaw(:,15); fastfood_1 = PatCharRaw(:,16); fastfood_2 = PatCharRaw(:,17); confidence_1 = PatCharRaw(:,18); confidence_2 = PatCharRaw(:,19); calculated_compliance = PatCharRaw(:,21); PatClinic = PatCharRaw(:,32); %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %Effect coding for dummy age variables throughout simulation %initialize with values for age 50-64 currentage_1=zeros(no_yrs+1, no_patients); currentage_2=ones(no_yrs+1, no_patients); currentage=currentage_1; for k=1:no_patients for i=1:no_yrs+1 currentage(i,k)=start_age(k)+(i-1); end; end; 212 cat1=currentage<50; cat3=currentage>64; %if current age <50, then age_cat = 1, effect code 1 and 0. currentage_1(cat1)=1; currentage_2(cat1)=0; %if current age >65, then age_cat = 1, effect code 1 and 0. currentage_1(cat3)=-1; currentage_2(cat3)=-1; %otherwise if current age 50-64, then age_cat = 1, effect code 1 and 0 (remain as default). %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %Calc. Baseline Compliance %Initialize BaselineCompliance BaselineCompliance = zeros(no_yrs+1,no_patients); logitBaselineCompliance = BaselineCompliance; %Find out if baseline compliance has been overridden, i.e. if all baseline %compliances just take on one value (this is for when we have no %information about individual patient characteristics, just avg %compliance); if so, don't calculate baseline compliance based on pat %characteristics, just use inputted figure if all(calculated_compliance==calculated_compliance(1)) BaselineCompliance(:,:)=calculated_compliance(1); logitBaselineCompliance(:,:)=log(calculated_compliance(1)/(1- calculated_compliance(1))); else %Calculate baseline patient compliance based on pat characteristics %using logistic regression - see input excel file for complete %listing of coefficients with std errors & p-values for k=1:no_patients for i=1:no_yrs+1 logitBaselineCompliance (i,k) = - 0.3820 - 0.4344 *currentage_1(i,k) + 0.00947 *currentage_2(i,k) - 0.2342 *poverty(k) - 0.3084 *insulin(k) + 0.6720 *insurance(k) + 0.4277 *dr_chck_hba1c(k) + 0.5181 *dr_chk_feet(k) - 0.2332 *diab_sgrchk_1(k) + 0.0475 *diab_sgrchk_2(k) + 0.0186 *fastfood_1(k) - 0.1230 *fastfood_2(k) + 0.1917 *confidence_1(k) - 0.0351 *confidence_2(k); BaselineCompliance(i,k) = 1 / (1+exp(- (logitBaselineCompliance(i,k)))); end; end; end; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %Modify Baseline Compliance based on patient-lever screening strategy %components / policy levers %Initialize PatLvlCompliance PatLvlCompliance = BaselineCompliance; AdjPatLvlCompliance = BaselineCompliance; 213 %Change the coefficient of the insurance variable in the logistic %regression by multiplying it by a scaling factor %if insurance = 0 (don't have insurance), then the effect is like changing %insurance to 1, then multiplying by the scaling factor (in that case the %0.672*insurance = 0) %if insurance =1, then the effect is to change the original coefficient of %insurance by the scaling factor %if insurance setting = 2 (mandatory health insurance), adjust pat lvl %compliance logitPatLvlCompliance = logitBaselineCompliance; if ins_setting==2 for k=1:no_patients for i=1:no_yrs+1 logitPatLvlCompliance(i,k) = logitBaselineCompliance(i,k) - 0.6720*insurance(k) + 0.6720; end; end; end; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %Calculate probability of baseline patient compliance from the logit PatLvlCompliance = 1 ./ (1.+exp(-(logitPatLvlCompliance))); %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %Adjusts patient compliance for provider-level policy %Set Adj Pat Lvl Compliance as the same as Pat Lvl Compliance, later this %will only be changed if provider level policy is present and only for patients whose clinics comply logitAdjPatLvlCompliance=logitPatLvlCompliance; %Determine which providers comply %ClinicSzThreshold is the clinic size threshold at which providers would be %incentivized / mandated to implement the policy, i.e. hire patient navigators %ClinicPercentComply is the percentage of clinics (among those clinics that %meet the size threshold) that will actually comply (i.e. implement the %policy) ClinicMeetsThreshold=find(ClinicChar.Size>param.ClinicSzThreshold); ClinicComplies=randsample(ClinicMeetsThreshold,round(param.ClinicPercentCompl y*length(ClinicMeetsThreshold))); 214 %Determine which patients are affected by provider's compliance if patnav_setting==2 %%only if navigators setting = 2 (hire navigators) for m=1:length(ClinicComplies) WhichPatCompliantClinic=find(PatClinic==ClinicComplies(m)); for i=1:no_yrs+1 logitAdjPatLvlCompliance(i,WhichPatCompliantClinic) = logitPatLvlCompliance(i,WhichPatCompliantClinic) + log(ScrStrategyChar.NavigatorOR); end; end; end; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %Adjust patient compliance for provider level policy AdjPatLvlCompliance = 1 ./ (1.+exp(-(logitAdjPatLvlCompliance))); %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% save BaselineComp; end
Abstract (if available)
Abstract
Recent health care reform brings forth the importance of preventive strategies such as screening. However, concerns about the growing population and rising healthcare spending necessitate health plans and public health policymakers to consider and determine cost‐beneficial population‐based screening strategies. Screening strategies often vary on compliance and, thus, result in suboptimal cost‐benefit for the population. To achieve maximum cost benefit, especially for screenings that often suffer low compliance, such as Diabetic Retinopathy screening, policymakers need to consider multi‐level (e.g. both patient‐ and provider‐level) interventions to improve screening compliance. To determine which of such interventions are cost-beneficial, many different screening strategies need to be considered. Yet, it is costly, impractical, and time‐consuming to do clinical trials to test all the different screening strategies. Simulation models provide a more cost‐ and time‐effective way to help determine cost-beneficial screening strategies. ❧ Simulation models have been used extensively to address cost‐benefit of screening. However, there are shortcomings to existing models. First, they do not have a structure that enables the evaluation of strategies that include policies affecting compliance, even though screening compliance is often low. Thus, strategies that include policies targeting the improvement of compliance may be necessary to achieve maximum cost‐benefit. Additionally, even policies not specifically targeting compliance may affect compliance, and models evaluating such policies without considering their impact on compliance will under‐ or over‐estimate the policy impact. Second, current simulation models do not have a structure that can evaluate multi‐level strategies (e.g. those targeting patient, providers, and clinics) even though they are more likely to have sustained or powerful effect than those targeting only the individual‐level. This research develops a generic conceptual model for screening services that addresses the two shortcomings and then constructs the model for the case of Diabetic Retinopathy screening to illustrate the model. ❧ The first shortcoming is addressed by including compliance as a mediating variable in the model, rather than a fixed input variable as in current methods. Compliance is influenced by patient characteristics (demographics, disease severity, self‐care, health belief) and screening strategy used, and in turn compliance influences disease progression and healthcare utilization and, thus, cost and benefit of the screening strategy. The second shortcoming is addressed by using hierarchical simulation with nested design where policy effects are manifested not universally but through a hierarchical structure. This enables evaluation of the impact of policies at a higher aggregate level (e.g. policies that target providers) on individual (patient) outcomes. The multi‐level design has the benefit of taking into account behavior at several levels and enabling the incorporation of economy of scale (e.g. how clinic size may affect cost‐benefit of certain screening strategies) in the model. The developed model is then compared with models without compliance as a mediating variable and models without the hierarchical structure to illustrate its advantages in terms of policy impact assessment. In addition, the models were used to evaluate cost‐benefit of different screening strategies using multiple perspectives. As the results indicate differing decisions made by policymakers viewing through these different perspectives, a case for context‐based decision making is made.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
A series of longitudinal analyses of patient reported outcomes to further the understanding of care-management of comorbid diabetes and depression in a safety-net healthcare system
PDF
Developing an agent-based simulation model to evaluate competition in private health care markets with an assessment of accountable care organizations
PDF
Designing infectious disease models for local level policymakers
PDF
Investigation of health system performance: effects of integrated triple element method of high reliability, patient safety, and care coordination
Asset Metadata
Creator
Vidyanti, Irene
(author)
Core Title
Simulation modeling to evaluate cost-benefit of multi-level screening strategies involving behavioral components to improve compliance: the example of diabetic retinopathy
School
Viterbi School of Engineering
Degree
Doctor of Philosophy
Degree Program
Industrial and Systems Engineering
Publication Date
05/05/2014
Defense Date
03/24/2014
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
Compliance,cost‐benefit analysis,diabetic retinopathy,eye screening,multi‐level,OAI-PMH Harvest,patient behavior,patient heterogeneity,preventive services,provider behavior,screening,screening strategies,simulation modeling,telescreening
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Wu, Shinyi (
committee chair
), Chou, Chih-Ping (
committee member
), Kesselman, Carl K. (
committee member
)
Creator Email
irenevidyanti@gmail.com,vidyanti@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c3-411652
Unique identifier
UC11295389
Identifier
etd-VidyantiIr-2495.pdf (filename),usctheses-c3-411652 (legacy record id)
Legacy Identifier
etd-VidyantiIr-2495.pdf
Dmrecord
411652
Document Type
Dissertation
Format
application/pdf (imt)
Rights
Vidyanti, Irene
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
cost‐benefit analysis
diabetic retinopathy
eye screening
multi‐level
patient behavior
patient heterogeneity
preventive services
provider behavior
screening
screening strategies
simulation modeling
telescreening