University of Southern California Dissertations and Theses

A methodology to identify high -risk patients with diabetes in the California Medicaid populations (Medi -Cal)

(USC Thesis Other)

A methodology to identify high -risk patients with diabetes in the California Medicaid populations (Medi -Cal)

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content A METHODOLOGY TO IDENTIFY HIGH-RISK PATIENTS WITH DIABETES
IN THE CALIFORNIA MEDICAID POPULATIONS (MEDI-CAL)
by
Usa Chaikledkaew
A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements of the Degree
DOCTOR OF PHILOPHOPHY
(PHARMACEUTICAL ECONOMICS AND POLICY)
August 2004
Copyright 2004 Usa Chaikledkaew
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
UMI Number: 3140446
Copyright 2004 by
Chaikledkaew, Usa
All rights reserved.
INFORMATION TO USERS
The quality of this reproduction is dependent upon the quality of the copy
submitted. Broken or indistinct print, colored or poor quality illustrations and
photographs, print bleed-through, substandard margins, and improper
alignment can adversely affect reproduction.
In the unlikely event that the author did not send a complete manuscript
and there are missing pages, these will be noted. Also, if unauthorized
copyright material had to be removed, a note will indicate the deletion.
UMI
UMI Microform 3140446
Copyright 2004 by ProQuest Information and Learning Company.
All rights reserved. This microform edition is protected against
unauthorized copying under Title 17, United States Code.
ProQuest Information and Learning Company
300 North Zeeb Road
P.O. Box 1346
Ann Arbor, Ml 48106-1346
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
DEDICATION
I dedicate this work to my parents, Opas Chaikledkaew and Praphai Phongprasatsuk
and my brothers, Amnuay Chaikledkaew and Amphon Chaikledkaew.
1 1
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ACKNOWLEDGEMENTS
I would like to express my deep gratitude to Dr. Kathleen A. Johnson, my Ph.D.
advisor, for her precious advice and continual encouragement during my dissertation
research, for making me realize my capability and the importance of staying focused,
for giving me the opportunity to work on numerous projects that helped me develop
my research and organizational skills, and for her warmth and support during my
academic career.
I am sincerely grateful to Dr. Denise Globe, Dr. Elizabeth Graddy, and Dr.
Joonghoon Ahn for their invaluable suggestions and comments towards my
dissertation. Also, I gratefully acknowledge the assistance of Dr. Jeffrey McCombs
and Dr. Mike Nichol for their support and encouragement that helped me through all
difficult times.
I would like to thank the Royal Thai Government for their generosity in providing
me with a scholarship to accomplish my doctoral degree. Also, I would like to thank
Dr. Petcharat Phongchareonsuk for her support and encouragement throughout my
study.
Ill
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
I also would like to thank Shaunna Thomas and Tina Bogan for making my life at
school lively and happier and for their support and encouragement. I would like to
thank Eric Wu, my friend and colleague for his help with my dissertation research.
I would like to mention a special thank to my best friend, Busadee Rattanaprechavej
for always making me see the lighter side of things, encouraging me to do the right
thing, and always standing by me either good or bad times.
I especially would like to express my gratitude to my beloved parents, Opas
Chaikledkaew and Prapai Pongprasartsuk, and my brothers, Amnuay and Amphon
Chaikledkaew for believing in me, for encouraging me to dream, for their love,
caring, and support.
IV
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
TABLE OF CONTENTS
PAGE
DEDICATION
ACKNOWLEDGEMENTS
LIST OF TABLES
LIST OF FIGURES
ABSTRACT
CHAPTER 1: INTRODUCTION
CHAPTER 2: BACKGROUND
2.1 Assessment of Patient Risk
2.2 Methodology Used to Identify Patient Risk
2.3 Methodology Used in this Research
2.4 Policy Issues Related to Identify High-Risk Patients
2.5 Rational Compared to Other Studies
CHAPTERS: METHODS
3.1 Data Source
3.2 Selection of Patients
3.3 Research Design and Methods
11
iii
vii
X
xi
1
5
5
9
13
17
19
24
24
24
26
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
PAGE
CHAPTER 4: RESULTS 45
4.1 Overall Simple Statistics 45
4.2 Results for Each Research Question 55
Research Question #1: What are the factors identifying 57
patient risk?
Researeh Question #2: Can the risk models be validated? 83
Research Question #3: Which is the most appropriate 83
Methodological approach to identify patient risk?
CHAPTER 5: DISCUSSION AND CONCLUSIONS 96
5.1 What are the Factors Identifying Patient Risk? 98
5.2 Can the Risk Models be Validated? 105
5.3 Which is the Most Appropriate Methodological Approach to 106
Identify Patient Risk?
5.4 Significant of Findings of MediCal Policy Makers and Providers 108
High-Risk Identification 108
Healthcare Cost Savings to MediCal 112
Significant MediCal Health Policy Options 113
5.5 Limitations o f the Research 116
5.6 Future Research 117
5.7 Conclusions 118
REFERENCES 121
APPENDIX 127
VI
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
LIST OF TABLES
PAGE
Table 2.1 Applicable Statistical Methods Used in 14
Methodologies for Identifying High-Risk Patients
Table 3.1 Dependent Variables Used in this Research 29
Table 3.2 Healthcare Costs and Utilization in the Past Used 29
in this Research
Table 3.3 Demographic Factors Used in this Research 29
Table 3.4 Healthcare Use Factors Used in this Research 30
Table 3.5 Type of Drug Variables Used in this Research 30
Table 3.6 Increasing Dose Variables Used in this Research 31
Table 3.7 Adding Drug Variables Used in this Research 32
Table 3.8 Changing Drug Variable Used in this Research 32
Table 3.9 Office Visit Variable Used in this Research 33
Table 3.10 Self-Monitoring Blood Glucose Variable Used 33
in this Research
Table 3.11 Lab Testes by Healthcare Provider Variables Used 34
in this Research
Table 3.12 Medication Compliance Variable Used in this Research 34
Table 3.13 Comorbidity Variables Used in this Research 35
Table 3.14 Complication Related to Target Organ Diseases Used 35
in this Research
Table 3.15 Complication Related to Infectious Diseases 36
Used in this Research
Vll
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 3.16 Layout for Comparing Actual and Predicted
High-Risk and Low-Risk Groups
PAGE
43
Table 4.1 General Demographic Characteristics of the Sample 46
Table 4.2 Cost and Healthcare Utilization in Each Six-Month Period 51
Table 4.3 Frequency of All independent Variables 52
Table 4.4 Demographic Statistics of Development and Validation 55
Data for Model #1 and #2
Table 4.5 Demographic Statistics of Development and Validation 56
Data for Model #3
Table 4.6 Regression Analyses Results (Model #1: Total Healthcare 57
Cost)
Table 4.7 Regression Analyses Results (Model #2: The Occurrence 61
of Hospitalization or ER visits)
Table 4.8 Regression Analyses Results (Model #3; Time to 66
Hospitalization or ER visits
Table 4.9 Comparisons of Factors Associated with Patient Risk 70
Among Three Models
Table 4.10 Factors Associated with High-Risk Patients 75
Table 4.11 High-Risk Patient Selection Criteria (Age Factor Included) 77
Tahle 4.12 High-Risk Patient Selection Criteria (Age Factor Not 80
Included)
Table 4.13 Validation of Predictive Risk Models 83
Table 4.14 Performance Measures of Model #1
(Total Healthcare Costs)
Table 4.15 Performance Measures of Model #2
(The Occurrence of Hospitalizations or ER visits)
84
85
vin
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
PAGE
Table 4.16 Performance Measures of Model #3
(Time to Hospitalizations or ER visits)
Table 4.17 Means of Performance Measures for Each Model
86
88
Table 4.18 Model Disagreements on Individual Patients Using Actual 89
Total Healthcare Cost Criteria
Table 4.19 Model Disagreements on Individual Patients Using Actual 90
Occurrence of Hospitalizations or ER Event Criteria
Table 4.20 Model Disagreements on Individual Patients Using Actual 91
Probability of Survival from Not Having Hospitalizations
or ER Event Criteria
Table A1 Summary of Previous Published Studies in Diabetes 131
Table A2 Oral Hypoglycemic Drugs 134
Table A3 Insulin 135
Table A4 Glucose Elevating Drugs 136
Table A5 High-Risk Patient Selection Criteria (Age Factor Included) 136
Table A6 High-Risk Patient Selection Criteria (Age Factor 145
Not Included)
Table A7 Comparison Between High-Risk and Low-Risk Groups 154
Identified by Cost Criteria
Table A8 Comparison Between High-Risk and Low-Risk Groups 156
Identified by Hospitalization or ER visits Criteria
Table A9 Comparison Between High-Risk and Low-Risk Groups 158
Identified by Probability of Survival from Not Having
Hospitalization or ER visits Criteria
Table AlO Association of Risk Factors with Healthcare Costs 160
IX
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
LIST OF FIGURES
PAGE
Figure 4.1 Percentage of Patients in Counties with Managed Care
Plan and FFS plan in California
47
Figure 4.2 Distribution of Patients in Counties with FFS plan only in
California
47
Figure 4.3 Trend of Healthcare Costs ($) 48
Figure 4.4 Trend on the Number of Hospitalization 49
Figure 4.5 Trend on the Number of FR Visits 50
Figure 4.6 Comparison of Total Healthcare Costs of High-Risk
Patient Among Nine High-Risk Selection Criteria
with Age Factor Included
79
Figure 4.7 Comparison of Total Healthcare Costs of High-Risk
Patients Among Nine High-Risk Selection Criteria
82
without Age Factor Included
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
ABSTRACT
OBJECTIVES: To investigate factors associated with high-risk diabetic patients, to
develop three econometric models [i.e., cost model (model #1); the occurrence of
hospitalization or ER event model (model #2); time to hospitalization or ER event
model (model #3)] that can be used to identify high-risk patients, and to evaluate
whether risk models are valid based on claims data from the California Medicaid
(MediCal) diabetic patients. METHODS: A retrospective study was conducted by
using claims data from January 1995 to December 2000. Dependent variables were
total healthcare cost, the occurrence of event, and time to event. Event included
hospitalization or ER visits. Historical data including demographic factors,
healthcare cost and utilization, diabetes drug treatment, follow-up services based on
diabetic guidelines, medication compliance, complications, and comorbidities were
used as independent variables. The generalized estimating equation and the fixed
effect partial likelihood methods were used. The split sample validation method was
applied to validate the models. RESULTS: Healthcare cost and utilization in the
previous period, patients taking both insulin and oral hypoglycemic drugs and
patients having oral hypoglycemic or anti-hypertensive drugs added into the regimen
were positively associated with an increase in patient risk. However, medication
compliance and follow-up services based on diabetic guidelines factors were
negatively associated with an increase in patient risk. Comorbidities (i.e.,
hypertension and hyperlipidemia) and complications (i.e., retinopathy,
xi
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
nephropathy and foot infection) were positively associated with higher patient risk.
Moreover, if high-risk patients were identified by high healthcare costs, model #1
was the most appropriate to use since it yielded the highest percentage of correct
predictions. If high-risk patients were defined as patient who had the occurrence of
hospitalization or ER event, model #2 was the most suitable to apply. Similarly, if
high-risk patients were indicated by shorter time to hospitalization or ER event,
model #3 was the most proper to utilize. Three models were valid.
CONCLUSIONS: Factors associated with high-risk diabetic patients and high-risk
patient identification mean healthcare providers and health plans could intervene to
improve patient management and possibly reduce healthcare costs.
Xll
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
CHAPTER 1: INTRODUCTION
In the United States, the quality of care, access to health care services, and cost of
health care are important to healthcare systems. The growing population with
chronic diseases, the importance of health economic considerations in caring for
patients, and the appropriate role of managed care organizations to provide cost-
effective care of high quality are particularly significant. Diabetes, a chronic disease,
is regularly used as the model to study the effect of health care policies on the
quality, access, and cost of health care services (Vinicor, F. 1998).
Diabetes is a common, serious, and chronic disease. It may occur at any time during
life, but many patients develop diabetes age 30 years. In the United States, more
than 16 million people or 6% of the national’s population is diagnosed with the
disease (Jiwa, 1997). Generally mortality rates for diabetic patients are two to three
times higher than those in the general population (Reiber et al, 1991). Most patients
with diabetes eventually suffer from late stage complications such as retinopathy,
nephropathy, neuropathy, and macrovascular disease. In addition, diabetic patients
often have coronary and cerebrovascular disease two to three times higher than those
without diabetes (WHO, 1994). Consequently mortality and morbidity of diabetes
leads to enormous healthcare costs to patients and society.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Diabetes is very expensive. The total costs of diabetes and its complications are
estimated to be more than $90 billion per year, approximately 15% of the total U.S.
healthcare costs (Diabetes 1996 Vital Statistics, 1996), and 80% of the costs are due
to hospitalization costs (Weinberger et al, 1990). The lifetime direct costs of
diabetes ($11 million) are the highest as compared to other diseases such as acute
lympholytic leukemia, breast cancer, rheumatoid arthritis, and stoke (Berry et al,
1981). Healthcare expenditures per capita among patients with diabetes are two to
three times greater than in the general population (Jiwa, 1997). These costs are
expected to continue rising as the number of diabetic patients increases. The disease
causes medical and economic problems for the United States as well as the world.
Although diabetes is a chronic disease, it is also accompanied by acute events such
as diabetic ketoacidosis and hypoglycemia (Musey et al. 1995). These acute
problems may occur at any time during the natural history of the disease or at the
initial diagnosis of the condition (Wagner et al, 1996). Diabetic patients are not
equally at risk experiencing the same natural history or developing these acute
problems and uncontrolled diabetes may cause long-term complications or acute
exacerbation accounting for an increase in hospitalization, risk of death, and
increased health care costs. Good diabetes control and identification of problems
before becoming severe can improve better long-term outcomes and decrease future
costs of caring for diabetes complications (Dagogo, 2002). Moreover identification
of diabetic patients under poor control could mean healthcare providers and health
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
plan administrators could help intervene to improve patient management and reduce
the risk of complications in the future.
Prevention strategies have been suggested to reduce the probability of an acute
exacerbation and prevent long-term complications. These strategies include diabetes
updates for primary care physician, diabetes self-management education intervention
(DSME), and drug therapy monitoring in specialized clinics. An educational
intervention targeting primary care physicians resulted in an improvement in the
quality of care for patients with diabetes (Oosthuizen et al, 2002). Since the 1930s,
DSME has been considered as an important part of clinical management of diabetes
in order to improve quality of life, prevent acute and chronic complications, and
decrease costs (Norris et al, 2002). The goals of the DMSE intervention are to
provide diabetes education on the significance of the HbAlc test, self glucose
monitoring test, medication, appropriate diet, physical activity, prevention of
complications and the importance of regular examinations such as dilated eye and
foot check-up for patients with diabetes (Norris et al, 2002). Moreover, the
implementation of a drug therapy monitoring clinic responsible for monitoring
treatment plans, implementing clinical guidelines, and providing educational
programs dramatically improves patient compliance with diabetes medication
(Yanchick, 2000).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
From health plan administrators’ perspective, they rely on healthcare providers to
manage patients but that also means that patients have to seek needed care because
diabetes is often without symptoms until it is more severe. Patients may not seek
medical help as often as needed. In addition, medical providers may not know when
patients do not show up for regular check ups. Therefore, health plan administrators,
who ultimately pay the bills, may have an interest in determining which patients are
not in good control as part of a quality assurance program and in order to stimulate
healthcare provider or patient action.
The purpose of this research is to develop various methodologies that can be used to
identify patients at high-risk, to determine what factors best predict “risk”, and to
evaluate the validity of the risk models. This research is conducted based on claims
data collected from patients with diabetes enrolled in the California Medicaid
(MediCal) fee-for-service (FFS) program. The development of a methodology to
identify high-risk patients and an investigation of factors associated with an increase
risk for hospitalization and costs helps plan to evaluate what healthcare services need
to be provided and to whom intervention and treatment could be targeted.
Specifically, this research aims to answer the following questions:
1. What are the factors identifying patient risk?
2. Can the risk models be validated?
3. Which is the most appropriate methodological approach to identify patient risk?
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
CHAPTER 2: BACKGROUND
This chapter of the dissertation contains a literature review that provides background
information and the rationale for the dissertation. The various literature reviews
include patient risk assessment literature, literature on the methodologies used to
identify high-risk patients, statistical background literature, rationale for the
dissertation, and policy issues related to identification of high-risk patients. This
chapter is organized in the following way. First assessment of patient risk is
presented. Secondly methodologies used to identify high-risk patients are
considered. Thirdly methodologies used in this research and the rationale for this
research compared to other studies are described. Finally policy issues related to
identification high-risk patients are presented.
2.1 Assessment of Patient Risk
“Patient Risk” has been defined in various ways. In the clinical literature, the
determination of patients with high risk for increased healthcare utilization for
patients with diabetes is indicated by family history of diabetes, obesity,
race/ethnicity, older age, hypertension, and hyperlipidemia (American Diabetes
Association, 1999). Other health service literature describes high-risk patients as
the patients with an increase in total healthcare cost and utilization based on
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
administrative claims data (Bhattacharyya, 1998). In the economics literature, high-
risk patients were defined as high-cost patients and determined by using the
predicted number of hospitalization events and predicted costs in the future
(Bhattacharyya, 1998). In each of these approaches, patient risk was indicated by
healthcare cost and utilization and factors associated with an increase in patient risk
are described in the following paragraphs.
Since this research aims to investigate various factors identifying patient risk, the
literature on various factors related to an increase in cost and healthcare utilization of
patients with diabetes was reviewed. Although economic analyses of diabetes have
been studied for two decades, only six studies preformed in the US estimated
healthcare use and cost for diabetics and investigated the associated factors (Guo, et
al, 1998; Krop, et al, 1998; Krop, et al, 1999; Bhattacharyya, 1998; Bhattacharyya,
et al, 1999; Brown, et al, 1999). Previously published articles in diabetes are
summarized in the appendix (Table Al). Based on the literature review, the
similarities and differences in the results related to factors associated with an
increase in cost and healthcare utilization are summarized in the following themes:
(1) Demographic factors (2) Diabetes complication and comorbidity factors (3) Prior
healthcare utilization factors (4) Payment system factors.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Demographic Factors
The results from all studies on the impact of demographic factors were mixed. All
studies showed that age was positively related to total direct costs and
hospitalization, indicating that older patients were more likely to consume higher
costs to the system and there was no statistically significant difference in race. Male
patients had higher expenditures than female patients (Krop et al, 1998). In contrast,
Bhattacharyya (1998) found that female patients were more likely to use hospital
services than male patients. However for the costs of diabetic complications, there
was no significant difference in gender and age (Brown et al, 1999). Previous
studies showed different findings related to the impact of diabetic complications and
comorbidities on costs.
Diabetes Complication and Comorbiditv Factors
The repeated impact of diabetic complications and comorbidities has differed
between studies on costs. Bhattacharyya et al has found that the number of diabetic
complications and comorbidities were positively associated with expenditures
(Bhattacharyya et al, 1998; Bhattacharyya et al, 1999). On the contrary, Krop et al
(1998) found that the number of diabetic complications was not significantly related
to the costs, but the score of Charlson Comorbidity Index significantly affected costs.
Complications and comorbidities led to an increase in healthcare utilization.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Prior Healthcare Utilization Factors
Prior drug and service use were the important factors associated with costs. Previous
researeh shows that prior healthcare utilization had a positive significant effect on
costs. For instance, use of anti-diabetie medication (i.e., insulin and oral medication)
was significantly related to eosts (Guo, et al, 1998; Bhattacharyya, 1999). Higher
eosts were associated with more hospitalization and diabetic related services such as
dialysis and hemoglobin AlC analysis (Bhattacharyya, et al, 1998; Bhattacharyya,
1999). In addition, the number of emergency room visits and longer average of
length of stay were also positively related to total expenditures (Krop et al, 1999).
An increase in healtheare costs might be associated with payment system.
Pavment Svstem Factors
Payment system may or may not play an important role in healthcare use and costs
for diabetes. Previous studies showed that there was no impact of payment system
factor on costs. Whether patients were enrolled in the FFS or capitated systems did
not have any significant effect on the total direct costs of diabetes (Bhattacharyya et
al, 1999). There was no statistieally significant difference in patients under HMO or
FFS plan on hospitalization use (Bhattacharyya, et al, 1998). In addition, a presence
or absence of Medicaid eligibility was not significantly associated with total
expenditures (Krop et al, 1999).
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.2 Methodologies Used to Identify Patient Risk
Since one of the objectives in this research is to develop the methodologies to
identify patient risk, a literature on methodologies used to identify patient risk was
reviewed. Based on literature review, all studies related to investigating factors
associated with high-risk patients with diabetes were performed on a cross-sectional
basis. This research was performed based on a longitudinal basis, hence the
literature on statistical methods applied to longitudinal data was also reviewed. The
literature review on methodologies to identify patient risk is summarized in the
following themes: (1) Dependent variable (2) Independent variable (3) Statistical
methods for cross-sectional data (4) Statistical methods for longitudinal or panel
data.
Dependent Variable
Based on a literature review, either total healthcare cost or utilization was used as a
dependent variable in all studies. Most studies calculated total direct costs, the
summation of medication and healthcare service costs and utilized total direct costs as
a dependent variable (Guo, et al, 1998; Krop, et al, 1998; Krop, et al, 1999; Brown,
et al, 1999). The cost of medication for comorhodity treatments was also
incorporated into the calculation of total direct cost. (Bhattacharyya, 1999).
However indirect costs were not considered in any study. Instead of using total direct
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
cost, the occurrence of hospitalization event was also used as a discrete dependent
variable (Bhattacharyya et al, 1998).
Independent Variable
Several studies investigated various factors associated with healthcare use and cost of
diabetes. Most common various factors evaluated were demographic factors,
complication, comorbidity, type of medication, healthcare utilization, and payment
system. A variety of methodology was applied to investigate the impact of various
factors on healthcare costs and utilization in diabetes.
Statistical Methods for Cross-Sectional Data
Varieties of methodologies were applied. Multiple linear regression models were
mostly applied to determine the association of costs with various factors on a cross-
sectional basis (Guo, et al, 1998; Krop, et al, 1998; Krop, et al, 1999). A log
transformation linear regression model was also used to analyze costs since the
relationship between dependent and independent variables were observed to be
exponentially distributed (Bhattacharyya, et al, 1999). Moreover logistic regression
was used to predict hospitalization events measured on a binary scale
(Bhattacharyya, et al, 1998). However three different regression techniques (i.e., a
two-part model, a transformation model and an ordinary least square model) were
10
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
investigated to determine which method was the most valid. The two-part model (2
OLS models) was the most efficient in terms of both predictive accuracy and
interpretability to determine the association of various factors with costs (Brown, et
al, 1999).
Statistical Methods for Longitudinal or Panel Data
The GEE method for the analysis of longitudinal data has been proposed by Zeger
and Liang (Liang and Zeger, 1986). They used the GEE approach to fit models for
discrete and continuous outcomes (Zeger and Liang, 1986). The GEE method is
closely related to quasi-likelihood methods and is specified by the marginal
distribution. Hence it can handle both normal and mainly non-normal outcomes.
Unlike maximum likelihood (ML) estimation, the joint multivariate normal
distribution of outcome variables must be specified. The GEE method yields
consistent estimators which may differ from the ML method (Part T., 1993).
Recently the GEE approach has been considered as a practical method to use in
longitudinal studies. The GEE method was used to evaluate the change in cognitive
function before and after development of an acquired immune deficiency syndrome
(AIDS) (Seines et al, 1995). In another study of Seines et al, the effects of duration
of follow-up, decline in CD4-I- count, development of clinical symptoms,
antiretroviral use, and diagnosis of AIDS on changes in cognitive performance over
11
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
time were estimated with regression models using the GEE approach (Seines et al,
1997). In addition, the GEE method was used to investigate the effect of gender on
the occurrence of HIV-related oral conditions such as hairy leukoplakia and
candidiasis (Shiboski et al, 1996) and to determine the association of the natural
history of bacterial vaginosis in women at risk for human immunodeficiency virus
(HIV) (Jamieson et al, 2001). It was also applied to investigate the effect of
potential risk factors such as human papillomavirus (HPV) type 16 infection and
nutritional status on the progression of cervical dysplasia (Liu et al, 1995). Kahn et
al used the GEE approach to assess the relationship between risk behaviors/partner
characteristics and either age of first sexual intercourse or HPV infection (Kahn et al,
2002). The GEE approach was also applied to determine the predictors of chlamydia
trachomatis infection among female adolescents in a longitudinal analysis (Mosure et
al, 1996). Moreover the relationship between the use of antihypertensive
medications, the prevalence of high blood pressure and the presence of left
ventricular hypertrophy was examined by using the GEE approach (Mosterd et al,
1999). In the study of Baker et al, the GEE method was applied to evaluate the
effect of the use of a Web-based Diabetes Care Management Support System
(DCMSS) by primary care physicians on the likelihood of a patient’s receipt of
glycated hemoglobin testing, lipid profile testing, and retinal examinations (Baker et
al, 2001). In this research, the GEE was used to investigate the association of costs
and healthcare utilization with various risk factors in a longitudinal data since it is an
appropriate method to handle both normal and non-normal outcome variables.
12
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2.3 Methodology Used in this Research and Rationale Compared to Other
Studies
In this research, panel or longitudinal data were eonstructed based on the elaims data
obtained from MediCal patients with diabetes, and appropriate statistic/eeonometric
methods applicable to panel or longitudinal data were utilized. Panel or longitudinal
data is a data set containing a given sample of individuals over time and multiple
observations on eaeh individual in the sample. Panel or longitudinal data has several
advantages compared to cross-sectional data. For example, panel data provides a
large number of data points, increases degrees of freedom, and reduees
multicollinearity, thus it improves the efficiency of the estimates (Hsiao, 1986). In
addition, panel data can be used to construet and test more complicated behavioral
models (Hsiao, 1986). Based on these advantages over eross-seetional data, panel
data were used in this study. Moreover the main objectives of this research were to
develop the appropriate methodologies that could be used to identify patient risk in
the future and to investigate the current factors identifying future patient risk,
therefore applicable econometries methods used in panel data were applied. In this
researeh, three methodologies for high-risk identification were conducted and
applieable statistic/econometric methods were used for eaeh methodology as
presented in Table 2.1. The model justification of each econometric method is
described.
13
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 2.1: Applicable Statistical Methods Used in Methodologies for Identifying
High-Risk Patients
Methodologies Applicable Statistical Methods
Total Healthcare cost (Model #1) Generalized Estimating Equation (GEE)
model for continuous outcomes
The oecurrence of event (Model #2)
(e.g., hospitalizations and ER visits
related to diabetes)
Generalized Estimating Equation (GEE)
model for discrete outcomes
Time to event (Model #3)
(e.g., hospitalizations and ER visits
related to diabetes)
The fixed effects partial likelihood (FEPL)
model
2.3.1 Generalized Estimating Equations (GEE)
Longitudinal data sets consist of repeated observations of a dependent variable and a
set of independent variables for each individual. Since repeated observations are
created from each individual, correlation is expected among measurements of each
individual. Therefore, based on quasi-likelihood theory, a method of generalized
estimating equations (GEE) for the regression parameters is proposed to describe the
marginal expectation of the dependent variable as a function of independent
variables considering the correlation among the repeated observations for a given
individual (Zeger and Liang, 1986). It is specified that a known function of the
marginal expectation of the dependent variable [|a,y= E(Y,y)] is a linear function of
the independent variables as follows:
14
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
g (lly) = XyP
• \iij is the response for individual i at time j
• Xy is a p X 1 vector of covariates
• P is a p X 1 vector of unknown regression coefficients
• g (.) is the link function
It is assumed that the variance is a known function of the mean.
V (Yy) = V ( P y )
• ( j ) is a possibly unknown scale parameter
• V (.) is the variance function
The GEE provides the consistent and asymptotically Gaussian estimators of the
regression coefficients and of their variances under weak assumptions about the
actual correlation among an individual’s observations (Zeger and Liang, 1992). The
GEE can handle intra-individual correlations as nuisance parameters among repeated
measurements on the same individual. The correlations are indicated as a working
correlation matrix form that has various possible structures. In addition, the GEE
approach can handle both normal and non-normal outcome variables such as Poisson
or binary outcomes. By iteratively solving equations based on quasi-likelihood
distributional assumptions, the GEE yields maximum likelihood estimators of the
model parameters. Without full specification of the joint distribution of an
individual’s observations, the variety of distribution forms (e.g., logistic, linear, or
log-linear) can be selected by specifying the link function.
15
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Moreover the GEE yields the relationship between the expeeted value of a dependent
variable and the independent variables by considering any intra-individual response
correlation when making statistical inferences about regression coefficients. This
leads to consistent estimates of regression coefficients and their variances, even with
misspecification of the covariance matrix structure. Both continuous and discrete
outcomes can be analyzed by using the GEE.
2.3.2 Fixed Effects Partial Likelihood Model (FEPL)
The fixed effects partial likelihood (FEPL) model is used to handle repeatable events
since the Cox proportional hazard model provides the hazard at time t until the first
event occurred and it is assumed that no individual experiences more than one event.
Because the events can be repeatable and the second event can be dependent on the
first event, therefore this dependence among events needs to be considered. The
fixed effects partial likelihood (FEPL) model is applied in this study, since it
accounts for the dependence and also helps correct some or all of the bias in the
model coefficients caused by unobserved heterogeneity. The fixed effects partial
likelihood (FEPL) model is
log hij(t) = a(t) + Pxij(t) + 8 /
where hij(t) is the hazard for theyth event for the zth individual at time t and S i is the
unobserved heterogeneity.
16
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The FEPL method assumes that s, is a set of fixed constants rather than a random
variable and absorbing this into the baseline log-hazard above function, we get
log hij(t) = a(t)+ Pxij(t)
Moreover the FEPL method is the preferred method for repeated event if the interest
is focused on covariates that vary across intervals for each individual.
2.4 Policy Issues Related to Identification of High-Risk Patients
Until now the FFS health care delivery system has mostly been used to provide care
to its eligible MediCal population. Under FFS system, qualified providers provide
healthcare services, drugs, and equipment to eligible MediCal patients. Then,
MediCal pays the allowed amount to the providers. Recently the MediCal program
has encouraged some patients to join managed care plans. Due to the various
managed care programs, the percentage of patients under the FFS MediCal program
differs across counties. One hundred percent of patients in rural counties such as
Monterey and Butte are under the MediCal FFS system, whereas there are only 45%
of all patients under the FFS system in urban counties such as Los Angeles and San
Diego. In addition, patients in urban counties who are under the MediCal FFS
program may be more complicated and sicker than patients in rural counties, since
healthier patients tend to enroll in managed care plans (LoSassa et al, 2000). In the
study of LoSassa et al (2000), it is shown that patients enrolled in the Health Plans of
San Mateo, a MediCal managed care plan, in California have fewer ambulatory
17
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
visits, lower expenditures and hospitalization compared to patients enrolled in
Ventura County’s FFS plan. This may imply that patients under managed care plans
are healthier and have lower risk for hospitalizations than patients under FFS
program or that managed care plans provide an increased emphasis on prevention in
order to keep people well. The managed care plans provide health care combining
clinical services and administrative procedures within a coordinated system.
Individual providers are linked together into a managed care system. In addition, the
managed care plan also emphasizes access to preventive or primary care and other
necessary services. This system is designed to increase the utilization of clinical
preventive services, and therefore reduce unnecessary hospitalizations and
emergency room visits.
MediCal program administrators have expressed an interest in developing methods
to intervene on certain “high risk” FFS patient populations. Preliminary work by the
Department of Health Services (DHS) identified 7 counties as being representative of
urban and rural counties and the respective FFS patient populations. Preliminary
DHS analysis also indicates that diabetes, an important disease impacting a relatively
large number of patients, is costly to MediCal, and has a literature supporting patient
behavioral factors and drug therapy components suitable for intervention to improve
patient outcomes and reduce costs to the MediCal program. MediCal may be
interested in utilizing pharmacists in community pharmacies to intervene to improve
patient behaviors and drug use in the high-risk population. The purpose of this study
18
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
is to develop an appropriate methodology to identify patient risk that would be
applicable to the MediCal program in determining patient referral into intervention
programs. The goal of high-risk patient identification is to help healthcare providers
and MediCal serving the FFS MediCal patients selectively intervene with those
patients who are predicted to be high-risk patients, so that early interventions can be
implemented towards these high-risk patients. This may potentially improve the
overall health outcomes and reduce the aggregate healthcare costs to the system.
2.5 Rationale for the Research
The policy and methodological considerations and previous published studies present
important results. However in this research, there are some different issues that
address some limitations of previous published studies.
1. MediCal is a FFS government health insurance plan including patients who
are poor and disabled. MediCal spends an enormously high amount in
healthcare expenditures for these patients every year. A primary concern for
MediCal is to be able to control healthcare utilization and costs. In addition,
in counties with managed care programs, managed care organizations tend to
obtain healthier patients in their plans (LoSassa et al, 2000). Hence the
patients tending to be left out from managed care organizations but remain in
FFS MediCal are possibly sicker and have higher risk for the increase in
healthcare utilization and costs. To address these concerns, MediCal performs
1 9
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Utilization Review (UR) activities to identify gross problems in patient
treatment patterns and also to recognize any fraud or abuse of the programs,
as a means of identifying areas where some costs can be controlled.
However these programs are not able to capture all patients who may benefit
from direct provider interventions. Therefore, there is still sufficient room
for identifying patients at high risk for increased in health care utilization and
costs. By identifying these high-risk patients, MediCal can devise selective
interventions such as a pharmacist compliance intervention to reduce drug
therapy problems and subsequently improve health care services to patients.
This study is the first to utilize the administrative/claims database provided
by MediCal to identify high-risk patients and factors associated with the
increase in risk. The results of this study may aid physicians, pharmacists,
and health plan administrators to identify high-risk patients and subsequently
develop effective interventions to influence care in a manner that provides
optimal healthcare services along with controlling healthcare costs.
2. Based on the literature review, all of these studies have addressed the costs of
diabetes on a cross-sectional basis. Cross-sectional data capture only a point
in time and miss individual changes in costs. Although diabetes is a chronic
disease, it is also accompanied by acute exacerbation. These acute events
may happen at any time and result in higher hospitalization use and increase
costs. Yet, no study has captured what occurs to costs and utilization over
time. This study is the first to consider a time series approach using six-
20
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
month period data to capture all events and costs throughout the four-year
study period in order to identify patient risk. Moreover this study determines
what risk factors in the previous period lead to a “spike” in costs and
healthcare utilization on a periodic basis.
3. Until recently, based on the administrative claims data, there is little
information regarding which factors are associated with future higher
expenditures with diabetes. Generally the factors incorporated into most
studies were demographic factors, complication, comorbidity, type of
medication, prior healthcare utilization, and payment system. Although
follow-up services based on diabetic guidelines (e.g., having HbAlc test
every six month, having glucose monitoring strip, office visit every six
months, dilated eye and feet check up every year, etc.) and diabetes treatment
(e.g., adding drugs, changing drugs, dose increasing, medication compliance,
and etc.) are important factors associated with high-risk patients, until now,
there is no study considering them as independent variables. In addition,
although health care use factors such as visits to the gynecologist,
cardiologist, nephrologist, and neurologist have an impact on healthcare
utilization and costs, there is yet no diabetes study evaluating them as
independent variables or as a proxy for an increase in severity. Since
administrative claims data were used in this study, there was no clinical
information, but there was prescription data and lab claims data. Therefore,
clinical indicators of poor control, follow-up services based on diabetic
21
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
guidelines, prescription drug use, and health care use factors were included,
in addition to demographic, comorbidity, complication of the diseases, type
of drugs, and prior healthcare utilization factors. Since patients in counties
that have proportion of FFS patients may tend to be at higher risk than
patients in rural cities (Bruno and Gilbert, 1998), percentage of managed
care plan in each county was also used as one of controlling variables in this
study. Moreover previous cost and previous hospitalization or ER event were
also included as an independent variable since tbe probability of the
occurrence of events in the future or future costs tends to depend on the
occurrence or non-occurrence of events in the past or previous costs,
respectively. Patients can switch in or out of the gap having higher
healthcare costs and the events in the future, hence by including previous
costs or hospitalization/ER event as one of independent variables may help
develop predictive risk models correctly.
4. There is little information about how to identify “risk”. Most of the studies
identified it by using healthcare utilization, costs, or clinical factors. In this
study, in addition to healthcare cost, there were two methods to identify
patient risk. First, “risk” was identified by time to event (i.e., time to
hospitalization and ER visits) since patients who had longer time to event
tend to be at lower risk for increasing healthcare costs and utilization.
Patients who have ever been hospitalized or visited the ER department tend
to have higher healthcare costs than those who have not. Therefore, the other
22
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
method of identifying “risk” is the occurrence of the events such as
hospitalization and ER. In addition, no study has investigated which
methodological approaches were the most appropriate to identify “risk”,
hence this study was the first to determine which methodology should be
used.
5. This study was the first to use the GEE approach to investigate the
association of costs and healthcare utilization with other risk factors. In this
study, panel data or longitudinal data were used. The data consists of
repeated observations of a dependent variable and a set of independent
variables for each individual. Correlation is expected among measurements
of each individual. Therefore, a GEE method for the regression parameters
was used because it provides the consistent and asymptotically Gaussian
estimators of the regression coefficients and of their variances under weak
assumptions about the actual correlation among an individual’s observations
(Zeger and Liang, 1992). In addition, cost data are usually skewed or not
normally distributed. The GEE approach is appropriate to apply since it is
able to handle both normal and non-normal outcome variables of which
distribution forms such as logistic, linear, or log-linear can be selected by
specifying the linking function.
23
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
CHAPTERS: METHODS
In this chapter, the data source, patient selection criteria, research design, and
methods are presented. Research design and methods include data management,
variable definition, and statistical methods.
3.1 Data Source
A retrospective study was condueted using the Medi-Cal database. MediCal is
responsible for financing a wide range of healthcare services for the poor and the
disabled population in California. This dataset was extracted from a random 20%
sample of the Medi-Cal 35-file long paid elaims during January 1995 to Deeember
2000.
3.2 Selection of Patients
Patients were selected based on the diagnosis of diabetes (i.e., 1CD9 codes=250.xx)
or diabetes related prescription drug use. The period of January 1995 through
December 1997 was used to select patients who met the criteria. The study period
was the period January 1997 through December 2000.
24
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Jan 95 Jan 96 JaJ97 Ian 98 Ian 99 jin 00 Dec 30
1 ^ —Selection Period— ________________ Study Period
Patient selection criteria are as follows:
1. Patients must be MediCal eligible under fee-for-service throughout the study
period.
2. Patients who have diabetes must have at least one claim with diabetes diagnosis,
diabetes related Rx medication, or diabetes related medical equipment
throughout the study period.
3. Patients must not be in a long-term care, nursing, or any other type of
institutional care facilities at the time of the selection period in database.
4. Patients must not have a Certified Hospice Service and California Children’s
Service (CCS) claims.
Based on the selection criteria, there were 14,040,096 claims of the patients
continually enrolled into the program from January 1997 to December 2000. All
claims were included in the study. This dataset contains data on patient
demographics, type of service received, date of service, amount billed, amount paid,
and days of service. In addition, prescription drug claims contain information on the
specific drug dispensed, quantity, strength, and prescription refill data. From those
claim files, two data sets were created (i.e., one with panel data and another with
cross-sectional data with repeated events).
25
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3.3 Research design and methods
3.3.1 Data management
Based on the MediCal database, there were two data sets created in this study. To
develop two risk models (i.e., total healthcare cost model and the occurrence of
hospitalization or ER event model), panel or longitudinal data set was created by
dividing the four-year study period into 8, six-month periods. The other data set
(i.e., cross-sectional data with multiple events) was created for the time to
hospitalization or ER visit event model. The SAS System for Windows, Version 8
was used for all data management.
3.3.1.1 Data format
By dividing the four-year study period into 8 six-month periods, a panel data set was
created. Thus, all variables were created individually at time f (i = 1,2,3,.. .8)
To T, l 2 Is T4 Ts To Ty Tg
Jan 95 Jan 96 Jan 97
— Selection Period — I
Jan 98 Jan 99 Jan 00 DecOO
_________Study Period____________ ^
26
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Patient Period Y X , X; X X4 X5 X,
Xj
1
2
3
4
5
6
7
8
1
8
Y= dependent variable (i.e., total healthcare costs or the occurrence of
hospitalization or ER visit)
Xi Xj = Independent variables
To develop the other risk model (i.e., time to hospitalization or ER visit related to
diabetes), the patient level data with multiple events based on hospitalization and ER
visit were created. Also, patients who had no hospitalization or ER visits during the
study period were included.
Key date event 2"‘ ^ event event
Jan 95 Jan 96 Jan 97
— Selection Period — ►
Jan 98 Jan 99 Jan 00 DecOO
________ Study Period____________ ^
27
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
“Key date” was the date of hospitalization or ER visit related to diabetes utilized
before and the most adjacent to January 1997. The period between key date and the
next service and the period between dates of event occurred were calculated.
Patient EVENT J DURATION X, X2 X3
1 1 1 1.50
1 1 2 2.75
1 1 3 3.75
i
EVENT = Having hospitalization or ER visit event (Yes=l)
J = = Event number
DURATION = the time period between key date and the next service or the time
period between dates of event occurred in days
X = Independent variables
3.3.1.2 Variable deflnitions
For the panel data analyses, dependent variables included 1) total healthcare costs
and 2) having a hospitalization or ER visit related to diabetes. Total costs were
calculated from the summation of claim costs in six-month periods. Moreover,
having a hospitalization or ER visit variable related to diabetes was created from
CPT codes from hospitalization or ER visits and ICD-9 codes for diabetes. For
cross-sectional data with multiple events, a dependent variable was the duration in
days between the key date and the next service or the duration between dates of
events which occurred.
28
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 3.1: Dependent Variables Used in this Research
Variables Label Identifiers
TOTCOST Total healthcare costs in six-month
period ($)
Claim costs
HOSER Having hospitalization or ER visit
related to diabetes (yes=l)
CPT codes and 1CD9
for diabetes
DUR Duration between key date and the
date that the event occurs (days)
Date of service and
CPT codes
Independent variables used in panel data and cross-sectional data with multiple
events are as follows:
1. Previous Healthcare costs and utilization variables
Table 3.2: Previous Healthcare Costs and Utilization Variables Used in this
Research
Variables Label Identifiers
PRECOST Total healthcare costs in previous
period ($)
Claim costs
PREHOSER Having hospitalization or ER visit in
previous period (yes=l)
CPT codes and 1CD9
for diabetes
2. Demographic Variables
Table 3.3: Demographic Variables Used in this Research
Variables Label Identifiers
AGE Age of patient (years) Date of birth
FEMALE Female patients =1 Gender
WHITE White patients=l Race
FFSCNTY County having FFS plan only
(yes= 1)
County
NORCNTY County in northern California
(yes=l)
County
MNPCNTY % of patients enrolling in
managed care plan in each county
County
29
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
3. Healthcare Use Variables:
Table 3.4: Healthcare Use Variables Used in this Research
Variables Label Identifiers
GYNEGY Having gynecologist visit in
previous period (yes=l)
Physician specialty
NEPHGY Having nephrologist visit in
previous period (yes=l)
Physician specialty
NEURGY Having neurologist visit in
previous period (yes=l)
Physician specialty
CARDGY Having cardiologist visit in
previous period (yes=l)
Physician specialty
INFECT Having infectious disease
specialist visit in previous period
(yes=l)
Physician specialty
4. Diabetes Treatment Factors
4.1 Type of drugs
Table 3.5: Type of Drugs Variables Used in this Research
Variables Label Identifiers
ORALDM Receiving oral hypoglycemic
drugs in previous period (yes=I)
NDC codes
INSULIN Receiving insulin in previous
period (yes=I)
NDC codes
GLUCOSE Receiving glucose elevating
drugs in previous period (yes=I)
NDC codes
HTNYES Receiving anti-hypertensive
drugs in previous period (yes=I)
NDC codes
LIPID YES Receiving lipid-lowering drugs in
previous period (yes=I)
NDC codes
30
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4.2 Increasing dose variables (dose/day = strength x quantity)
drug days supplied
Table 3.6: Increasing Dose Variable Used in this Research
Variables Label Identifiers
INCDOSORAL Having oral hypoglycemic drug
dose increased within the same
class in previous period (yes=l)
NDC codes, units,
date of service,
quantity
4.3 Adding drugs
4.3.1 Adding hypoglycemic drugs
The criteria for considering that patients having drugs related to diabetes
added are as follows.
-Patients who had oral hypoglycemic drugs:
-Having different classes of oral hypoglycemic drugs added
-Having insulin added
-Having glucose elevating drugs added
-Patients who had insulin:
-Having different classes of insulin added
-Having oral hypoglycemic drugs added
-Having glucose elevating drugs added
4.3.2 Adding anti-hypertensive
-Patients who had oral hypoglycemic drugs/insulin drugs:
-Having anti-hypertensive drugs added
31
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4.3.3 Adding lipid lowering drugs
-Patients who had oral hypoglycemic drugs/insulin drugs:
-Having lipid-lowering drugs added
Table 3.7: Adding Drug Variables Used in this Research
Variables Label Identifiers
ADDDM Adding any hypoglycemic drugs
in previous period (yes=l)
NDC codes
ADDHTN Adding anti-hypertensive in
previous period (yes=l)
NDC codes
ADDLIPID Adding lipid lowering drugs in
previous period (yes=l)
NDC codes
4.4 Changing drugs
Patients who had oral hypoglycemic drugs:
-Changing to different classes of oral hypoglycemic drugs or
-Changing to insulin
Table 3.8: Changing Drug Variable Used in this Research
Variables Label Identifiers
CHANGEDG Having drug changed in the
regimen in previous period
(yes=l)
NDC codes
5. Follow-up service based on the guideline for patients with diabetes
(American Diabetes Association, 2000)
32
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Key date Ti T 2 T 3 T 4 T 5 T g Tj T g
Jan 95 Jan 96 Jan 97
1 ^ — Selection Period — ►
Jan 98 Jan 99 Jan 00 DecOO
_________Study Period____________ ^
“Key date” was the date of the service utilized before and the most adjacent to
January 1997. The period between key date and the next service and the period
between services were calculated.
5.1 Office visits
Table 3.9: Office Visit Variable Used in this Research
Identifiers Label Identifiers
OFVISIT Having office visits following the
guidelines in previous period (i.e., every 3
months for patients taking insulin and every
6 months for patients taking only oral anti
diabetic drugs) (yes=l) (American Diabetes
Association, 1999)
CPT codes
5.2 Lab tests
5.2.1 Self-monitoring blood glucose test
Table 3.10: Self-Monitoring Blood Glucose Variable Used in this Research
Variable Label Identifiers
STRIP Having glucose monitoring strip in
previous period (yes=l)
CPT codes
33
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
5.2.2 Lab tests by healthcare providers
Table 3.11: Lab Tests by Healthcare Provider Variables Used in this Research
Variable Label Identifiers
HBAIC Having HbAlc test every 6 months in
previous period (yes= 1)
CPT codes
CHOLES Having cholesterol test every year in
previous period (yes=l)
CPT codes
KIDNEY Having kidney check-up every year in
previous period
CPT codes
EYEEXAM Having dilated eye examination every
year in previous period
CPT codes
FOOTEXAM Having foot check-up every year in
previous period
CPT codes
6 . Medication Compliance
Medication possession ratio (%MPR) was calculated.
%MPR = Total (mean) compliance on oral hypoglycemic drugs over all refill
intervals (i.e., from first to last day of observation)
= Sum of davs’ supplv dispensed x 100
Sum of days in all refill intervals
Table 3.12: Medication Compliance Variable Used in this Research
Variable Label Identifiers
MPRORALDM Medication possession ratio of Oral
hypoglycemic drugs in previous
period (%)
NDC codes, drug
day supply, refilled
date
34
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
7. Comorbidity
Table 3.13: Comorbidity Variables Used in this Research
Variable Label Identifiers
COMHTN Having hypertension in previous period
(yes=l)
ICD-9 codes
COMLIPID Having hyperlipidemia in previous period
(yes=l)
ICD-9 codes
COMCHRONIC Having systemic chronic disease in previous
period (e.g., asthma, COPD, epilepsy,
rheumatoid arthritis, and renal disease)
(yes= 1)
ICD-9 codes
COMCVS Having cardiovascular diseases in previous
period (e.g., angina, coronary disease, heart
disease, MI, and stroke) (yes= 1)
ICD-9 codes
COMCANCER Having cancer/severe disease in previous
period (e.g., HIV, liver disease, malignancy,
cystic fibrosis, and transplant) (yes= I)
ICD-9 codes
COMPSYCH Having psychiatric disease (e.g., depression,
bipolar, Parkinson, and psychotic) (yes=I)
ICD-9 codes
8. Complication
8.1 Target organ disease:
Table 3.14: Complication Related to Target Organ Diseases Used in this
Research
Variable Label Identifiers
RETINO Having retinopathy in previous period
(yes=l)
ICD-9 codes
NEPHRO Having nephropathy in previous period
(yes=l)
ICD-9 codes
NEURO Having neuropathy in previous period
(yes-1)
ICD-9 codes
35
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
8.2 Infectious disease:
Table 3.15: Complication Related to Infectious Diseases Used in this Research
Variable Label Identiflers
FTINFECT Having foot infection in previous period
(yes=l)
ICD-9 codes
SKINFECT Having skin infection in previous period
(yes=l)
lCD-9 codes
VGINFECT Having vagina infection in previous
period (yes=l)
lCD-9 codes
3.3.2 Splitting sample for model validation
To investigate whether the risk models were valid, the split sample validation
method was applied. Hence the dataset was divided randomly into two halves. The
outcomes of interests such as costs and healthcare utilization acute events were
equally distributed in each half. One half of data set was used to develop the models
and the other half was used to validate the models. The procedures (i.e., PROG
RANK and RANUNI function) in the SAS System for Windows, Version 8 were
used to split the sample for model validation.
3.3.3 Test for heterogeneity assumptions
Usually observations of outcome variables are assumed to be random outcomes of
some experiment with a probability distribution conditional on explanatory variables
36
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
and coefficients (Hsiao, 1986). When panel data are utilized, one of the major goals
is to use all available data to make inferences on coefficients. Therefore, to prevent
the exploitation of the data, test for heterogeneity assumptions was used in order to
test whether or not parameters characterizing the random outcome variable stay
constant across all i and t. Analysis of covariance was used for this purpose.
3.3.4 Statistical method
In this study, “risk” was defined as total healthcare costs, the occurrence of a
healthcare utilization event, and time to event. The events included hospitalization
or ER visits. Various factors identifying patient risk are presented as above.
Appropriate statistical methods (Table 2.1) were applied to answer each research
question.
Research question #1: What are the factors identifying patient risk?
Model #1: Total healthcare cost
From MediCal’ s point of view, patients who consume higher healthcare costs tend
to have higher risk for increased healthcare utilization and costs in the future. In this
study, total healthcare costs included the costs of medication, hospitalizations, ER
visits, physician visits, and etc. Since panel data were used in this study, the GEE
approach was applied to evaluate the association between costs and the factors
37
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
obtained from claims data. The GEE provides a practical method to analyze
correlated data from repeated measurements and violating the normality assumption
(Zeger and Liang, 1986). The GENMOD procedure along with repeated statement,
one of the SAS procedures. Version 8 was used to fit models with the correlation
structures hy using GEE to obtain parameter estimates and their significances.
LFsually, heath care cost is used as a dependent variable to estimate regression
models. Cost data are rarely normally distributed and they are usually skewed.
Skewed cost data may violate the assumptions required for regression analyses. The
natural log transformation is mostly used to make cost data more normally
distributed and obtain parameter estimates less likely to be biased. Cost regression
models are usually used to simulate or predict costs. However if the log of cost is
used as a dependent variable, predicted costs cannot be calculated by taking the anti
log of the fitted value since this value is related to a retransformation bias. To
correct for the bias, the smearing estimator approach was also applied (Duan, 1983).
By considering a regression model of the form,
Lncost = Xp + e
Since it is related to a non-linear transformation, the expected value of cost should be
E(cost) = E(e +
= l/nS(e^^P^^'=)
= (e ^^*^^l/nl(e
38
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
1/n S (e is the smearing estimator for models with a log-transformed dependent
variable. It is the mean of the anti-log of the residuals and usually yields a value
between 1 and 2. To correct for the retransformation bias, the fitted values are
multiplied by the smearing estimator.
Model #2: The occurrence of healthcare utilization events
Generally patients who ever had a hospitalization or ER visit in the past are more
likely to experience it in the future and these patients tend to be at high-risk for
increased healthcare utilization and costs. The occurrence of hospitalization or ER
visits related to diabetes was used as a discrete dependent variable in this method.
The GENMOD procedure based on GEE approach was also applied to evaluate the
relationship between the occurrence of the events and various independent variables.
Model #3: Time to event
Patients who have shorter time to event such as a hospitalization or ER visit are more
likely to have higher risk for increased healthcare utilization and costs. Time to
event is the period from the key date to the date when the event occurs. The unit of
time to event is days. Since the events (e.g., hospitalization or ER visits) are
repeatable and it is possible to have more than one of those events. The fixed effects
partial likelihood (FEPL) method was appropriate to use with repeated events.
Therefore the FEPL method was applied to investigate the association of various
factors on time to event.
39
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
After running regression analyses for each method and checking for
multicollinearlity, only significant parameter estimates were used to develop models
from the development data set. Furthermore the values of parameter estimate for
each independent variable and its significance were obtained to evaluate the impact
of various factors on these outcome variables.
Research question #2: Can the risk models be validated?
Predictive model validity shows how well a model predicts an outcome (lezzoni,
1997). To evaluate the accuracy of the predictive risk models for predicting
healthcare costs, the occurrence of events, and time to event, the split-sample
validation approach was applied. Since a split-sample method was used and the data
set was divided randomly into two halves (i.e., development and validation data
sets). One half of data set was used to develop the model. By running regression
analyses on development data set, significant parameter estimates were used to
develop the predictive risk models. The model was used to calculate predicted
outcomes in the validation data set. By using R^, the summary measure of model fit
calculated from observed and predicted outcomes, its predictions were compared to
actual outcomes in the validation data set.
40
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Moreoever, R^, the coefficient of determination, is used as a standard summary
measure of performance of risk models. is calculated by
R^=l- [Zi (actualY, - predictedY,)^ / Si (actualY, - meanY/)^]
where Y, is a continuous or discrete outcome. In the term subtracted from 1, the
numerator is called the sum of squared errors (SSE) and the denominator is called
the sum of squares total (SST). SST measures the variability of the outcome Y and
is determined by the data alone not by the model. On the other hand, SSE measures
the variability in Y that the model can predict. R^ is the fraction of total variability in
the dependent variable explained by differences in risk among cases included in the
model. R^ calculated from development data was compared to that obtained from
validation data.
Research question #3: Which is the most appropriate methodological approach to
identify high-risk patients?
In this study, there are three methodological approaches identifying high-risk
patients (i.e., total healthcare cost, model #1; the occurrence of hospitalization or ER
visit, model #2; and time to event, model #3). Based on the results from research
question #1, three predictive risk models were developed. For model #1, high-risk
patients were defined as high-cost patients. Predicted costs calculated from
41
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
predictive cost model were used to obtain the distribution of healthcare costs among
the data sample, individual aimual total healthcare costs were cumulated and plotted
against the cumulative number of patients with diabetes. From the cumulative plot,
diabetic patients who utilized more than 80% of healthcare costs were defined as
high-cost patients (Smith et al, 1997; Herman, et al, 1985; and Franklin, 1990). For
model #2 , high-risk patients were indicated by patients who had predicted
probability of hospitalization greater than 0.5. For model #3, high-risk patients were
identified by patients who had probability of survival from not having a
hospitalization or ER visit less than 0.5 (lezzoni, 1997). These high-risk patient
selection criteria were used with actual and predicted outcomes. Patients were
classified into two groups (i.e., high-risk and low-risk groups). Two-by-two
classification tables were used to evaluate diagnosis rules and the performance
statistics (e.g., total predictive accuracy, sensitivity, specificity, positive and negative
predictive values) in each period for each model. These performance statistics were
calculated and compared among the three models. Table 3.1 shows the layout of the
two by two classification table for comparing actual and predicted high-risk and low-
risk groups.
42
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 3.16: Layout for Comparing Actual and Predicted High-risk and Low-
risk Groups
Risk Score Prediction
Actual Outcomes
All cases High-Risk Low-Risk
High-Risk A
(True positive)
B
(False positive)
A+B
Low-Risk C
(False negative)
D
(True negative)
C+D
All cases A+C B+D A+B+C+D
Sensitivity (i.e., A/A+C) is the probability of predicting bigb-risk given that bigb-
risk patients occurred, whereas specificity (i.e., D/B+D) is the probability of
predicting low-risk given that patients are low-risk. Predictive value positive (i.e.,
A/A+B) is the probability of patients being bigb-risk given prediction of bigb-risk
and predictive value negative (i.e., D/C+D) is the probability of patients being low-
risk given low-risk prediction.
To investigate which methodological approach is the most appropriate to identify
bigb-risk patients, three risk model performance measures were also compared with
respect to disagreements among models on the prognosis of bigb-risk versus low-risk
for individual patients. “Disagreed” patients were those for whom the two models
(i.e.. Model #1 versus Model #2, Model#l versus Model #3, and Model #2 versus
Model #3) being compared disagree on bigb-risk and low-risk predictions. Usually
there were three criteria to identify actual bigb-risk patients (i.e., bigb-risk patients
43
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
are patients who consume 80% of total healthcare costs, patients who have ever had
hospitalization or ER visit related to diabetes, or patients whose probability of
survival from not having hospitalization or ER visit related to diabetes was less than
0.5). Based on these criteria, actual high-risk patients were evaluated and used to
match with predicted high-risk patients obtained from the three predictive models. If
they were matched, it means that the model yielded the right prediction. The
percentage of the correct prediction among models was calculated and compared.
44
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
CHAPTER 4: RESULTS
In this chapter, the results are divided into two parts. The first part presents overall
deseriptive statistics (i.e., demographic characteristics of all samples, trends of total
healthcare costs and hospitalization or ER visits, and frequencies of all independent
variables). The second part describes the results of eaeh of the three researeh
questions.
4.1 Overall Simple Statistics
4.1.1 Demographic Characteristics of AH Samples
Based on the patient selection criteria described previously, there were 14,040,096
claims of the patients eontinually enrolled in the program from January 1997 to
December 2000. This MediCal data set consisted of 13,832 patients with diabetes.
Table 4.1 shows the general demographic characteristics of the diabetic patients in
this sample. The mean age of the sample was 68.43 years and approximately 6 8% of
the sample was female. About 56% of the sample was white and nearly 15% of the
patients lived in a county with a FFS plan.
45
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.1: General Demographic Characteristics of the Sample (N=13,820)
Demographic Characteristics (N=13,820) Statistic Values
Mean age (years) (SD) 68.21 (12.56)
(Min-Max) (6.52-100.50)
Age group (years)
>0 - <20 0.07%
>20 - <40 2.32%
>40 - <60 20.41%
>60 - < 80 61.95%
>80 - < too 15.19%
>100- < 120
0.06%
Age greater than 65 years 66.43%
Female 67.55%
Race: White 56.32%
Black 17.30%
Hispanic 6.56%
Asian 19.51%
Other 0.39%
Patients living in county with FFS plan only 14.55%
Figure 4.1 shows the distribution of patients in counties with managed care plans and
FFS plans. In addition, the distribution of patients in counties with only FFS plans is
presented (Figure 4.1). Of all counties with managed care and FFS plans in the
sample, most patients lived in Los Angeles (41%), whereas among counties with
FFS plan only, most patients lived in Tulare (2.3%) (Figure 4.2).
46
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 4.1: Percentage of Patients in Counties with Managed Care Plan and
FFS Plan in California
% Patient
(/> o o O IQ
0> O) c c ■D
d>
Oi
0)
o
(A
0)
m
O
E
c IQ
<
c
(Q o <
ifi
O
(0 C Q
-1 c
(Q
( 0
Figure 4,2: Percentage of Patients in Counties with FFS Plan Only in California
% Patient
2 . 5 - 1
47
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
4.1.2 Total Healthcare Cost and Utilization
The first goal of this study was to develop appropriate methodologies to identify
high-risk patients. In terms of “risk”, it was indicated by three methods (i.e., total
healthcare costs, the occurrence of event, and time to event). The event included a
hospitalization or ER visit. It would be very helpful to understand the trend of
healthcare cost and utilization of the sample in this data set, hence a time series
forecasting method was applied. Figure 4.3^ demonstrates the trend of total monthly
cost converted to dollar value in year 2000 by using Consumer Pricing Indexes
(CPI).
Figure 4.3: Trend of Healthcare Cost ($)
F D r e c o s t s f o r COS T
B5D
J A N 9 7 M AY 9 7 S E P 9 7 J A N 9 S M A Y 9 a S E P 9 8 J A N 9 9 M A Y 9 9 S E P 9 9 J ANOO MAYOO S E P O O J A N O l MA Y 0 1 S E P 0 1 J A N 0 2
“ The dotted line is the cutoff point discriminated future period from the end of study period
(December 2000). Three graphs after the dotted line show the trend and its confidence interval in the
future.
48
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The figure 4.3 shows that total healtheare eosts have been significantly increasing
since 1998. Patients will most likely tend to consume higher costs in the future. In
addition, figure 4.4^ and 4.5 show the trends of the number of hospitalizations and
the number of ER visits, respectively.
Figure 4.4: Trend on Number of Hospitalization
F d r B c n s t s f o r H O S P I F AL I Z A F I O N
+ D -
J A N 9 7 J U L 3 7 J A N 9 B J U L 9 8 J A N 9 9 J U L 9 9 J A N O O J U L O Q J A N O l J U L 0 1 J A N 0 2
“ The dotted line is the cutoff point discriminated future period from the end of study period
(December 2000). Three graphs after the dotted line show the trend and its confidence interval in the
future.
49
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 4.5: Trend on the Number of ER Visits
F o f e c p s t s f o r ER_VISITS
55
50
+ 5
+ 0
35
3D
25
2D
15
ID
J A N 9 7 M A Y 9 7 S E P 9 7 J A H S f i M A Y 9 8 S E P 9 8 J A N 9 9 H A Y 9 9 S E P 9 9 M N O O MAYDQ S E P O O J A N O l MA Y 0 1 S E P 0 1 J A N 0 2
Based on the graphs above the numbers of hospitalization or ER visits have been
dramatically decreasing from 1997 through 1998 due to the change in healthcare
utilization pricing policy. The trends seem to be steady in the future. Table 4.2
shows the mean total healthcare costs converted to dollar value in year 2000,
hospitalization, and ER visits in each period.
“ The dotted line is the cutoff point discriminated future period from the end of study period
(December 2000). Three graphs after the dotted line show the trend and its confidence interval in the
future.
50
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.2: Healthcare Cost and Utilization in Each Six-Month Period
Period Mean Cost ($) (Range)
(N=13,820)
Hospitalization
N (%)
(N=13,820)
ER visits
N (%)
(N=13,820)
1
(1/1/97-6/30/97)
3,695 (86-87,212) 495 (3.58) 335 (2.42)
2
(7/1/97-12/31/97)
3,569 (80-96,515) 320 (2.32) 270(1.95)
3
(1/1/98-6/30/98)
3,675 (110-98,661) 192(1.40) 211 (1.53)
4
(7/1/98-12/31/98)
4,052 (80-86,758) 194(1.40) 221(1.45)
5
(1/1/99-6/30/99)
4,537(118-87,335) 220(1.59) 230(1.66)
6
(7/1/99-12/31/99)
4,828(108-81,164) 201 (1.45) 222(1.61)
7
(1/1/00-6/30/00)
5,210(159-91,980) 232 (1.68) 235 (1.70)
8
(7/1/00-12/31/00)
5,509 (126-92,954) 219(1.58) 252 (1.82)
4.1.3 Independent Variables
Based on claims data, various factors were associated with total healthcare cost, the
occurrence of hospitalization or ER visits, and the time to hospitalization or ER visit
were created. Frequencies of all factors used in this study are presented in Table 4.3.
51
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.3: Frequency of Ail Independent Variables
Independent Variables (N=13,820) N (%)
Diabetes treatment factors:
1. Type of drugs
-Oral hypoglycemic drugs
-Insulin
-Anti-hypertensive drugs
-lipid-lowering drugs
13820(100.00)
5000 (36.18)
12157 (87.97)
7320 (52.97)
2. Increasing dose of oral hypoglycemic drugs
6411 (46.39)
3. Adding Drugs
-Adding oral hypoglycemic drugs or insulin
-Adding anti-hypertensive drugs
-Adding lipid-lowering drugs
11245 (81.37)
12048 (87.18)
7179 (51.95)
4. Changing drug to different classes or to insulin
2050(14.83)
5. Oral hypoglycemic medication compliance
-Mean medication possession ratio (%MPR) (Range)
87.01 (0-100)
Follow-up service factors based on diabetic guidelines:
I . Having office visits (every 3 months for patients taking
insulin and every 6 months for patients taking only oral
hypoglycemic drugs)
11800 (85.38)
2. Having glucose monitoring strip
6170 (44.61)
3. Having lab tests by healthcare providers
-HbAlc every 6 months
-Cholesterol check-up every year
-Kidney check-up every year
-Ketone test every year
-Dilated eye check-up every year
-Foot check-up every year
3202 (23.17)
3545 (25.65)
0 (0.00)
0 (0.00)
3260 (23.59)
0 (0.00)
52
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.3: Frequency of All Independent Variables (Contd.)
Independent Variables (N=13,820) N (%)
Healthcare use factors:
-Gynecologist visit 1560(11.29)
-Cardiologist visit 6300 (45.59)
-Nephrologist visit 585 (4.23)
-Neurologist visit 1828 (13.23)
-Ophthalmologist visit 7880 (57.01)
-Infectious disease specialist visit 142(1.03)
Comorbidity factors:
-Hypertension
11279 (81.61)
-Hyperlipidemia
5154 (37.29)
-Systemic chronic disease
5376 (38.90)
-Cardiovascular disease
8540 (61.79)
') A A 'y A 1 \
-Cancer/severe disease
3442 (24.91)
1796(13.00)
-Psychiatric disease
Complication factors:
1. Target organ disease
-Retinopathy
3365 (24.35)
-Nephropathy
394 (2.85)
-Neuropathy
1888 (13.66)
2. Infection
-Foot
3065 (22.18)
-Skin
84 (0.61)
-Vagina
62 (0.45)
Based on the results from Table 4.3, for diabetes treatment factors, all patients in the
sample were dispensed oral bypoglycemic drugs, and there were 30% of patients
receiving insulin. There w ere about 87% o f patients taking anti-hypertensive drugs
and approximately 51% of patients taking lipid-lowering drugs. About 15% of
53
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
patients had drug changed to different classes of oral hypoglycemic drugs or to
insulin, and 46% of patients had oral hypoglycemic drug dose increased. The results
show that patients had drugs such as oral hypoglycemic drugs different classes or
insulin (81%), anti-hypertensive drugs (87%), and lipid lowering drugs (52%) added
into the regimen. Furthermore mean oral hypoglycemic medication compliance
calculated by medication possession ratio yielded 87%.
Based on diabetic guidelines, important follow-up services were recommended for
patients with diabetes. In this sample, about 85% of patients had physician office
visits followed the guidelines (i.e., every three months for patients taking insulin or
every six months for patients taking oral hypoglycemic drugs only) and 45% of
patients had glucose monitoring strip. Only 23% of patients had glycated
hemoglobin AlC (HbAlc) tests every six months, and there were 26% of patients
having cholesterol a check-up every year. There were 24% of patients having dilated
eye examination every year and 0.12% of patients receiving a kidney check-up every
year. Surprisingly no patient had ketone test, kidney checkup, and foot examination
in this sample.
For healthcare use factors, patients in this sample had specialist visits such as
ophthalmologists (57%), cardiologists (46%), neurologists (13%), gynecologists
(11%), nephrologists (4%), and infectious disease specialists (1%). The comorbidity
diseases found in this population were hj^ertension (82%), cardiovascular disease
54
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(62%), systemic chronic disease (39%), hyperlipidemia (37%), cancer/severe disease
(25%), and psychiatric disease (13%). It is found that the most complications on
target organ for patients with diabetes were retinopathy (25%), neuropathy (14%),
and nephropathy (3%).
4.2 Results for Each Research Question
After creating the data and all variables for analyses, the data were randomly divided
into two halves (i.e., development and validation data). Table 4.4 shows
demographic statistics of development and validation data used in models #1 and #2.
Table 4.4: Demographic Statistics of Development and Validation Data for
Model #1 and #2
Variables Development
Data
(N=6,910)
Validation
Data
(N=6,910)
Demographics:
-Mean age (Range) (years)
-Female gender, N (%)
-White race, N (%)
-Living in eounty with FFS plan only, N (%)
66.48 (9-100)
4670 (67.58%)
3891 (56.31%)
1025 (14.83%)
66.75 (3-100)
4680 (67.73%)
3925 (56.82%)
1011 (14.63%)
Mean costs per year ($) (Range) 4379.50
(1-98,661)
4384.21
(1-92,954)
Mean number of hospitalization per year (Range) 0.09 (0-58) 0.09 (0-55)
Mean number of ER visit per year (Range) 0.02 (0-31) 0.02 (0-15)
Mean medication compliance (%) (Range) 87.15 (0-100) 87.22 (0-100)
55
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Based on the results from chi-square and t-test analyses, it is shown that there were
no statistically significant differences in the mean age, female gender, white race,
living in county with FFS plan only, mean costs, mean number of hospitalization or
ER visit, and mean medication compliance between development and validation data
sets. Table 4.5 presents demographic statistics results from development data
compared to those from validation data used for model #3.
Table 4.5: Demographic Statistics of Development and Validation Data for
Model #3
Variables Development
Data
(N=3,212)
Validation
Data
(N=3,456)
Demographics:
-Mean age (years) (Range)
-Female gender, N (%)
-White race, N (%)
-Living in county with FFS plan only, N (%)
64.41 (9-102)
2171 (67.59%)
1816 (56.55%)
466 (14.50%)
64.74 (3-104)
2333 (67.50%)
1962 (56.77%)
511 (14.80%)
Mean time to event (days) (Range) 1093 (10-1,460) 1096 (9-1,460)
Mean costs ($) (Range) 13,237.08
(40-489,638)
13,816.80
(54-614,529)
Mean medication compliance (%) (Range) 84.85 (2.4-100) 85.36(0.7-100)
Above results from Table 4.5 shows that compared development data to validation
data, there were no statistically significant differences in mean age, female gender,
white race, living in county with FFS plan only, mean time to event, mean costs, and
mean medication compliance.
56
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Research Question #1; What are the factors identifying patient risk?
Model #1: Total Healthcare Cost
The development panel data set v v ^ a s used to perform the analysis. The dependent
variable of this model was total healthcare cost. Since the dependent variable (i.e.,
total health care cost) was not normally distributed, a log-transformation of the cost
was used to achieve normality. Table 4.6 summarizes the results from the regression
analysis of various factors associated with total healthcare cost. The first column
illustrates all independent variables used in this analysis (i.e., demographic factors,
previous healthcare cost, diabetes treatment factors, follow-up service based on
diabetic guidelines factors, healthcare use factors, comorbidity, and complication
factors) and the other column shows the parameter estimates of each variable based
on GEE analysis.
Table 4.6: Regression Analyses Results (Model #1: Total Health care cost)
Independent Variables
(N=6,910)
Parameter
estimates
Demographic factors:
-Age
-Female gender
-White
-% managed care plan in county
0.0028***
-0.0240*
-0.0463***
-0.0001
Total healthcare cost in previous period 0.0001***
Hospitalization or ER event in previous period 0.1863***
Significance Levels: ***p<0.005, **p<0.05, *p<0.1
57
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.6: Regression Analyses Results (Model #1: Total Healthcare cost)
(Contd.)
Independent Variables
(N=6,910)
Parameter
estimates
Diabetes treatment factors:
1. Type of drugs
-Both insulin and oral hypoglycemic dmgs
0.2760***
2. Increasing dose of oral hypoglycemic dmgs
0.0321*
3. Adding Drugs
-Adding oral hypoglycemic dmgs or insulin
-Adding anti-hypertensive dmgs
-Adding lipid-lowering dmgs
0.0602***
0.1204***
0.0453**
4. Changing dmg to different classes or to insulin
0.2321***
5. Compliance (%MPR)
-0.0016***
Follow-up service based on diabetic guidelines factors:
1. Having office visits (every 3 months for patients taking
insulin and every 6 months for patients taking only oral
hypoglycemic dmgs)
-0.1665***
2. Having glucose monitoring strip -0.0588***
3. Having lab tests by healthcare providers
-HBAIC every 6 months
-Cholesterol check-up every year
-Dilated eye check-up every year
-0.0276**
-0.1076***
-0.0593**
Health care use factors:
-Gynecologist visit
-Cardiologist visit
-Nephrologist visit
-Neurologist visit
-Ophthalmologist visit
-Infectious disease specialist visit
0.0829**
0.5489***
0.5978***
0.4105***
0.0889***
0.8305***
Significance Levels: ***p<0.005, **p<0.05, *p<0.01
58
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.6: Regression Analyses Results (Model #1: Total Healtheare cost)
(Contd.)
Independent Variables Parameter
(N=6,910) estimates
Comorbidity:
-Hypertension
0.0675***
-Hyperlipidemia
0.1285***
-Systemic chronic disease
0.3557***
-Cardiovascular disease
0.3705***
-Cancer/severe disease
0.4027***
-Psychiatric disease
0.2542***
Complication:
1. Target organ disease
-Retinopathy
0.1228***
-Nephropathy
0.2979***
-Neuropathy
0.0316
2. Infection
-Foot
0.1545***
-Skin
0.2730
-Vagina
0.1720
Significance Levels: ***p<0.005, **p<0.05, *p<0.01
Model significance: Model #1— Pearson Chi-Square = 6529.15 (p<0.0001)
The results show that various factors were statistically significantly associated with
total healthcare costs. Age was positively related to higher healthcare costs, whereas
white was negatively related to higher healthcare expenditures. Total healthcare cost
in the previous period had significant impact on the cost in the next period. For
diabetes treatment factors, patients taking both insulin and oral hypoglycemic drugs,
patients having drugs (e.g., oral hypoglycemic, anti-hypertensive, or lipid lowering
dmgs) added into the regimen, and patients having oral hypoglycemic dmgs changed
to different classes or insulin were positively associated with higher cost, whereas
medication compliance was negatively related to higher cost. In addition,
59
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
follow-up services based on diabetic guidelines factors (i.e., having office visit every
three months for patients taking both insulin and oral hypoglycemic drugs or every
six months for patients taking only hypoglycemic drugs, having glucose monitoring
strip, having HbAlc test every six months, having cholesterol test every year, and
having dilated eye examination every year) were negatively associated with total
healthcare cost. Healthcare use factors such as gynecologist, cardiologist,
nephrologist, neurologist, or ophthalmologist visits had a positive impact on
healthcare cost. Moreover it is found that comorbidity factors (i.e., hypertension,
hyperlipidemia, systemic chronic disease, cardiovascular disease, cancer/severe
disease, and psychiatric disease) and three complication factors (i.e., retinopathy,
nephropathy and foot infection) were positively related to healthcare costs.
Based on significant parameter estimates (p < 0.05) obtained from the GEE analysis,
the predictive model #1 was developed as follows.
Log cost = 5.9920 + 0.003*age - 0.0463*white + 0.000l*precost - i- 0.2760*both +
0.0602*adddm + 0.1204*addhtn + 0.0453*addlipid - i- 0.2321 *changedg -
0.0016*mprdm - 0.1665*ofvisit - 0.0588*strip -0.0276*HbAlc -
0.1076*choles - 0.0593*eyeexam + 0.0829*gynegy + 0.5489*cardgy +
0.5978*nephgy + 0.4105*neugy + 0.0889*ophthgy - i- 0.8305*infect +
0.0675*htn + 0.1285*lipid + 0.3557*chronic + 0.3705*cvs +
0.4027*cancer + 0.2542*psych + 0.1228*retino + 0.2979*nephro +
0.0316*neuro + 0.1545*foot + error.
60
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Therefore, predicted costs were calculated by multiplying the exponential of
logarithm of costs and the smearing estimator.
Model #2: The occurrence of hospitalization or ER visit events
The dependent variable of this model was the occurrence of hospitalization or ER
visit. Table 4.7 shows the regression analyses of various factors associated with
hospitalization or ER visit.
Table 4.7: Regression Analyses Results (Model #2: The occurrence of
hospitalization or ER visit)
Independent Variables Parameter Odds ratio
(N=6,916) estimates
Demographic factors:
-Age
0,0414*** 1.043
-Female gender
-0.1623** 0.920
-White
-0.1254 0.920
-% managed care plan in county
0.0054*** 0.995
Total Healthcare costs in previous period
0.0000 1.000
Hospitalization or ER visits in previous period
1.4479***
2.912
Significance Levels: ***p<0.005, **p<0.05, *p<0.1
61
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4,7: Regression Analyses Results (Model #2: The occurrence of
hospitalization or ER visit) (Contd.)
Independent Variables
(N=6,916)
Parameter
estimates
Odds ratio
Diabetes treatment factors:
I. Type of drugs
0.4486*** 1.717
-Both insulin and oral hypoglycemic drugs
2. Increasing dose of oral hypoglycemic dmgs
0.0250 1.146
3. Adding Drugs
0.4525*** 1.384
-Adding oral hypoglycemic dmgs or insulin
-Adding anti-hypertensive dmgs
0.1402** 1.173
-Adding lipid-lowering dmgs
0.2753 1.227
4. Changing dmg to different classes or to insulin
0.4052*** 1.479
5. Compliance (%MPR)
-0.0029** 0.997
Follow-up service based on diabetic guidelines factors:
1. Having office visits (every 3 months for patients taking
insulin and every 6 months for patients taking only oral
-0.4591***
0.506
hypoglycemic dmgs)
2. Having glucose monitoring strip
-0.3780***
0.502
3. Having lab tests by healthcare providers
-HBAIC test every 6 months
-Cholesterol check-up every year
-0.1390**
0.629
-Dilated eye check-up every year
-0.0630
0.935
-0.0376
0.976
Health care use factors:
-Gynecologist visit
0.0950 1.115
-Cardiologist visit
1.0628*** 3.125
-Nephrologist visit
0.6729*** 2.196
-Neurologist visit
0.7431*** 2.235
-Ophthalmologist visit
0.1354 1.017
-Infectious disease specialist visit
1.5169*** 5.304
Significance Levels: ***p<0.005, **p<0.05, *p<0.1
62
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.7: Regression Analyses Results (Model #2: The occurrence of
hospitalization or ER visit) (Contd.)
Independent Variables Parameter Odds ratio
(N=6,916) estimates
Comorhidity:
-Hypertension
0.0776 1.024
-Hyperlipidemia
-0.1052 0.756
-Systemic chronic disease
0.2800*** 1.304
-Cardiovascular disease
0.2298*** 1.260
-Cancer/severe disease
0.1321** 1.008
-Psychiatric disease
0.2417*** 1.396
Complication:
1. Target organ disease
-Retinopathy
0.2075*** 1.037
-Nephropathy
0.2917** 1.279
-Neuropathy
0.1199** 1.158
2. Infection
-Foot
0.2931*** 1.197
-Skin
-0.5696 0.901
-Vagina
0.3161 1.897
Significance Levels; ***p<0.005, **p<0.05, *p<0.1
Model significance: Model #2 — Pearson Chi-Square = 50606.02 (p<0.0001)
Based on the results obtained from the GEE analysis between the occurrence of
hospitalization or ER visit related to diabetes and a variety of factors (Table 4.7), it is
found that age and male gender were positively related to the occurrence of
hospitalization or ER visit related to diabetes. In addition, patients living in the
counties with higher percentage of people under managed care plan had a significant
higher probability of hospitalization or ER visit. Patients who had ever had
hospitalization or ER visit in previous period were more likely to have
63
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
hospitalization or ER visit in the next period. Similarly to the results obtained from
model #1, patients taking both oral hypoglycemic drugs and insulin, patients having
different classes of oral hypoglycemic drugs or insulin or having anti-hypertensive
dmgs added into the regimen, or patients having their drugs changed to different
classes or to insulin were positively associated with the incidence of hospitalization
or ER visit. However patients compliant to oral hypoglycemic medication were
negatively related to the occurrence of hospitalization or ER visit. Patients who had
follow-up services such as having office visits, having glucose monitoring strip, or
having HBAIC test based on diabetic guidelines were negatively associated with the
incidence of hospitalization or ER visit. Moreover having cardiologist, nephrologist,
neurologist, or infective disease specialist visits was positively related to having
hospitalization or ER event. For comorbidity factors, having systemic chronic
disease, cardiovascular disease, cancer/severe disease, or psychiatric disease was
positively related to the incidence of hospitalization or ER visit. Four complication
factors (i.e., retinopathy, nephropathy, neuropathy, and foot infection) had a positive
impact on the occurrence of hospitalization or ER visit. To develop the predictive
model #2, only significant parameter estimates (p<0.05) were incorporated.
64
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Log (odds of having hospitalization or ER visit)
= -2.2587 + 0.0414*age - 0.1623*female + 0.0054*mnpcnty + 1.4479*prehoser +
0.4486*both + 0.4525*adddm + 0.1402*addhtn - i- 0.4052*changedg -
0.0029*mprdm - 0.9591*ofvisit -I-- 0.3780*strip - 0.1390*HbAlc + 1.0628*cardgy
+ 0.6729*nephgy + 0.743 l*neugy + 1.5169*infect + 0.2800*chronic + 0.2298*cvs +
0.1321*cancer + 0.2417*psych + 0.2075*retino + 0.2917*nephro + 0.1199*neuro +
0.293 l*foot + error.
The odds of having hospitalization or ER visit was the exponential of logarithm of
the odds. The probability that a patient would have hospitalization or ER visit was
equal to the odds divided by one plus the odds.
Method #3: Time to hospitalization or ER visit
The dependent variable of this model was time to event (i.e., hospitalization or ER
visit). Table 4.8 shows the regression analyses of various factors associated with
hospitalization or ER visit.
65
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.8: Regression Analyses Results (Model #3: Time to hospitalization or
ER visit)
Independent Variables
(N=3,212)
Parameter
estimates
Hazard
Ratio
Demographic factors:
-Age
-Female gender
-White
-% managed care plan in county
0,0744***
-0.0944
0.1385
0.0025
1.072
0.910
1.149
1.003
Total healthcare cost
0.000002* 1.000
Diabetes treatment factors:
1. Type of drags
-Both insulin and oral hypoglycemic drags
0.2686*** 1.246
2. Increasing dose of oral hypoglycemic drags
0.1222 0.135
3. Adding Drags
-Adding oral hypoglycemic drags or insulin
-Adding anti-hypertensive drags
-Adding lipid-lowering drags
0.5889***
0.5255***
0.3400***
1.802
1.691
1.405
4. Changing drag to different classes or to insulin
0.0421 1.043
5. Compliance (%MPR)
-0.0041** 0.996
Follow-up service based on diabetic guidelines factors:
1. Having office visits (every 3 months for patients taking
insulin and every 6 months for patients taking only oral
hypoglycemic drags)
-0.0885** 0.907
2. Having glucose monitoring strip
-0.0766** 0.926
3. Having lab tests by healthcare providers
-HBAIC test every 6 months
-Cholesterol check-up every year
-Dilated eye check-up every year
-0.0941**
-0.0478
-0.2018**
0.910
0.953
0.817
Significance Levels: ***p<0.005, **p<0.05, *p<0.1
66
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.8: Regression Analyses Results (Model #3; Time to hospitalization or
ER visit)
Independent Variables Parameter Hazard
(N=3,212) estimates Ratio
Health care use factors:
-Gynecologist visit
0.0443 1.045
-Cardiologist visit
0.0274 1.025
-Nephrologist visit
0.0366 1.037
-Neurologist visit
0.0188 1.019
-Ophthalmologist visit
0.2329 1.208
-Infectious disease specialist visit
0.0966 1.092
Comorbidity:
-Hypertension
0.3183*** 1.273
-Hyperlipidemia
0.2202*** 1.199
-Systemic chronic disease
0.0655 1.063
-Cardiovascular disease
0.0717 1.074
-Cancer/severe disease
0.0357 1.126
-Psychiatric disease
0.0181 1.018
Complication:
1. Target organ disease
-Retinopathy
0.1347 1.106
-Nephropathy
0.0476 1.049
-Neuropathy
0.0446 1.044
2. Infection
-Foot
0.6037 1.829
-Skin
0.0446 1.044
-Vagina
0.0292 0.971
Significance Levels: ***p<0.005, **p<0.05, *p<0.1
Model significance: Model #3 -W ald Chi-Square = 751.52 (p<0.001)
Table 4.8 summarizes the results from regression analyses and hazard ratio of each
variable is also presented. Hazard ratio gives us the estimated percent change in the
hazard or risk for each one-unit increase in the covariate. It is shown that there was a
positive significant effect of age on the risk of hospitalization or ER visit. For each
one-year increase in age, the risk of hospitalization or ER visit went up by 4.5%.
67
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Total healthcare cost had slightly positive significance impact on the risk of
hospitalization or ER visit. For diabetes treatment factors, patients taking both oral
hypoglycemic drugs and insulin, or patients having drugs such as oral hypoglycemic,
insulin, antihypertensive or lipid lowering drugs added into the regimen were
positively related to the risk of hospitalization or ER visit. This means that patients
receiving both oral hypoglycemic drugs and insulin had 24 percent higher risk of
hospitalization or ER visit than those who did not. Similarly, patients having oral
hypoglycemic, antihypertensive, or lipid-lowering drugs added into the regimen had
80%, 69% or 41% higher risk of hospitalization or ER visit than those who did not,
respectively. In addition, the risk of hospitalization or ER visit for patients
compliant to oral hypoglycemic medication decreased by 0.4% compared to non-
compliant patients. For follow-up service based on diabetic guideline factors, there
were four factors (i.e., having office visit, having glucose monitoring strip, having
HBAIC test, and having dilated eye-examination) negatively associated with the risk
of hospitalization or ER visits. Based on diabetic guidelines, the risk of
hospitalization or ER visit for patients having office visit, having glucose monitoring
strip, having HBAIC test, or having dilated eye-examination decreased by 10%, 7%,
9%, or 18%, respectively. However there was no healthcare use factor significantly
affecting the risk of hospitalization or ER visit. Two comorbidity factors (i.e.,
hypertension and hyperlipidemia) had positive effects on the risk of hospitalization
or ER visit. Diabetic patients who had hypertension or hyperlipidemia increase the
risk of hospitalization or ER visit by 27% or 20%, respectively. There was no
68
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
significant impact of complication factors on the risk of hospitalization or ER visit.
The predictive model #3 was developed based on significant parameter estimates
(p<0.05).
Predictive Indexes (PI) = 0.0744*age + 0.2686*both + 0.5889*adddm +
0.5255*addhtn + 0.3400*addlipid - 0.0885*ofvisit -
0.0766*strip - 0.0941*HbAlc - 0.2018*eyeexam +
0.3183*htn + 0.2202*lipid
The predicted survival for an individual patient was calculated as Y=S (t)® ’ ‘ '’ for
that patient. The value of S(t) was determined as the overall survival rate when each
of independent variables are zero.
Moreover, Table 4.9 demonstrates the comparisons of factors associated with patient
risk among three models.
69
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
CD
■ D
O
Q .
C
o
CD
Q .
■ D
CD
C/)
( /)
Table 4.9: Comparisons of Factors Associated with Patient Risk among Three Models
O
O
■ D
c q '
CD
■ D
O
Q .
C
a
o
■ o
o
CD
Q .
■ D
CD
C/)
C/)
Independent Variables Model #1
(N=6910)
Model #2 (N=6916) Model #3 (N=3212)
Parameter
estimate
Parameter
estimate
Odds Ratio Parameter
estimate
Hazard Ratio
Comorbidity:
-Hypertension 0.0675*** 0.0776 1.024 0.3183*** 1.273
-Hyperlipidemia 0.1285*** -0.1052 0.756 0.2202*** 1.199
-Systemic chronic disease 0.3557*** 0.2800*** 1.304 0.0655 1.063
-Cardiovascular disease 0.3705*** 0.2298*** 1.260 0.0717 1.074
-Cancer/severe disease 0.4027*** 0.1321** 1.008 0.0357 1.126
-Psychiatric disease 0.2542*** 0.2417*** 1.396 0.0181 1.018
Complication:
1. Target organ disease
-Retinopathy 0.1228*** 0.2075*** 1.037 0.1347 1.106
-Nephropathy 0.2979*** 0.2917** 1.279 0.0476 1.049
-Neuropathy 0.0316 0.1199** 1.158 0.0446 1.044
2. Infection
-Foot 0.1545*** 0.2931*** 1.197 0.6037 1.829
-Skin 0.2730 -0.5696 0.901 0.0446 1.044
-Vagina 0.1720 0.3161 1.897 0.0292 0.971
Significance Levels: ***p<0.005, **p<0.05, *p<0.01
Model significance: Model #l--Pearson Chi-Square = 6529.15 (p<0.0001)
Model #2 — Pearson Chi-Square = 50606.02 (p<0.0001)
^ Model #3 -W ald Chi-Square = 751.52 (p<0.001)
CD
■ D
O
Q .
C
o
CD
Q .
■ D
CD
C/)
W
o'
o
o
Table 4.9: Comparisons of Factors Associated with Patient Risk among Three Models
3 "
C D
O
O
T 3
» <
c 5 '
Independent Variables Model #1
(N=6910)
Model #2 (N=6916) Model #3 (N=3212)
Parameter
estimate
Parameter
estimate
Odds Ratio Parameter
estimate
Hazard Ratio
o
s
Diabetes treatment factors;
3
1 . Type of drugs
■ n
3
-Both insulin and oral hypoglycemic drugs 0.2760*** 0.4486*** 1.717 0.2686*** 1.246
3 "
2. Increasing dose of oral hypoglycemic drugs 0.0321* 0.0250 1.146 0.1222 0.135
■ §
O
3. Adding Drugs
3 -Adding oral hypoglycemic drugs or insulin 0.0602*** 0.4525*** 1.384 0.5889*** 1.802
2 -
o'
-Adding anti-hypertensive drugs 0.1204*** 0.1402** 1.173 0.5255*** 1.691
3
■ a
-Adding lipid-lowering drugs 0.0453** 0.2753 1.227 0.3400*** 1.405
3 "
g; 4. Changing drug to different classes or to 0.2321*** 0.4052*** 1.479 0.0421 1.043
C D
Q .
insulin
S
5
o
3
5. Compliance (%MPR) -0.0016*** -0.0029** 0.997 -0.0041** 0.996
■ D
CD
(/)
(/)
CD
■ D
O
Q .
C
o
CD
Q .
■ D
CD
C/)
( /)
Table 4.9: Comparisons of Factors Associated with Patient Risk among Three Models
o
o
■ D
c q '
O ’
Q
CD
■ D
O
Q .
C
a
o
■ o
o
CD
Q .
■ D
CD
(/)
(/)
Independent Variables Model #1
(N=6910)
Model #2 (N=6916) Model #3 (N=3212)
Parameter
estimate
Parameter
estimate
Odds Ratio Parameter
estimate
Hazard Ratio
Follow-up service based on diabetic guidelines
factors:
1 . Having office visits (every 3 months for -0.1665*** -0.4591*** 0.506 -0.0885** 0.907
patients taking insulin and every 6 months for
patients taking only oral hypoglycemic drugs)
2. Having glucose monitoring strip -0.0588*** -0.3780*** 0.502 -0.0766** 0.926
3. Having lab tests by healthcare providers
-HBAIC every 6 months -0.0276** -0.1390** 0.629 -0.0941** 0.910
-Cholesterol check-up every year -0.1076*** -0.0630 0.935 -0.0478 0.953
-Dilated eye check-up every year -0.0593** -0.0376 0.976 -0.2018** 0.817
Health care use factors:
-Gynecologist visit 0.0829** 0.0950 1.115 0.0443 1.045
-Cardiologist visit 0.5489*** 1.0628*** 3.125 0.0274 1.025
-Nephrologist visit 0.5978*** 0.6729*** 2.196 0.0366 1.037
-Neurologist visit 0.4105*** 0.7431*** 2.235 0.0188 1.019
-Ophthalmologist visit 0.0889*** 0.1354 1.017 0.2329 1.208
-Infectious disease specialist visit 0.8305*** 1.5169*** 5.304 0.0966 1.092
Significance Levels: ***p<0.005, **p<0.05, *p<0.01
CD
■ D
O
Q .
C
o
CD
Q .
■ D
CD
C/)
( /)
Table 4.9: Comparisons of Factors Associated with Patient Risk among Three Models
O
O
■ D
c q '
O ’
Q
CD
■ D
O
Q .
C
a
o
■ o
o
CD
Q .
■ D
CD
(/)
(/)
Independent Variables Model #1
(N=6910)
Model #2 (N=6916) Model #3 (N=3212)
Parameter
estimate
Parameter
estimate
Odds Ratio Parameter
estimate
Hazard Ratio
Comorbidity:
-Hypertension 0.0675*** 0.0776 1.024 0.3183*** 1.273
-Hyperlipidemia 0.1285*** -0.1052 0.756 0.2202*** 1.199
-Systemic chronic disease 0.3557*** 0.2800*** 1.304 0.0655 1.063
-Cardiovascular disease 0.3705*** 0.2298*** 1.260 0.0717 1.074
-Cancer/severe disease 0.4027*** 0.1321** 1.008 0.0357 1.126
-Psychiatric disease 0.2542*** 0.2417*** 1.396 0.0181 1.018
Complication:
1 . Target organ disease
-Retinopathy 0.1228*** 0.2075*** 1.037 0.1347 1.106
-Nephropathy 0.2979*** 0.2917** 1.279 0.0476 1.049
-Neuropathy 0.0316 0.1199** 1.158 0.0446 1.044
2. Infection
-Foot 0.1545*** 0.2931*** 1.197 0.6037 1.829
-Skin 0.2730 -0.5696 0.901 0.0446 1.044
-Vagina 0.1720 0.3161 1.897 0.0292 0.971
Significance Levels: ***p<0.005, **p<0.05, *p<0.01
Model significance: Model #l--Pearson Chi-Square = 6529.15 (p<0.0001)
Model #2 — Pearson Chi-Square = 50606.02 (p<0.0001)
^ Model #3 -Wald Chi-Square = 751.52 (p<0.001)
Based on the eomparisons of factors associated with patient risk among the three
models (Table 4.9), all three risk models show that factors associated with patient
risk ranked by the extent of the standard coefficients from highest to lowest were as
follows.
1. Having Hospitalization/ER visit in the past
2. Having both oral hypoglycemic drugs and insulin
3. Having oral hypoglycemic, insulin, anti-hypertensive, or lipid lowering drugs
added
4. Lower compliance (Medication possession ratio less than 80%)
5. No office visits (every three months for patient taking insulin and every six
months for patients taking only oral hypoglycemic drugs)
6. No glucose monitoring test
7. No HbAle test every six months
8. Having comorbidities such as hypertension or hyperlipidemia
9. Having complications such as retinopathy, nephropathy, or foot infection
74
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.10: Summary of the Factors Associated With Patient Risk
Factors Associated With Patients Risk
(N=13,820)
Impact
Number of
Patients (%)
Prior Hospitalizations/ER visits t 2 times
6546 (47.37)
Patients having visits to
-Nephrologist
-Neurologist
-Infectious disease specialist
1 1 -2 times
392 (2.84)
1888 (13.66)
141 (1.02)
Patients taking both oral hypoglycemic drugs and
insulin
t 72%
4996 (36.15)
Patients having drugs changed to different classes t 50%
2050(14.83)
Patients having office visits based on diabetic
guidelines
i 50%
11804 (85.41)
Patients having glucose monitoring strip i 50%
6175 (44.68)
Patients having HbAlc test every six months 4 40%
3202 (23.17)
Patients having drug added to the regimen
-Adding oral hypoglycemic drugs or insulin
-Adding anti-hypertensive drugs
-Adding lipid-lowering drugs
t 30%
11245 (81.37)
12048 (87.18)
7179 (51.95)
Patients with comorbidities
-Hypertension
-Hyperlipidemia
-Systemic chronic disease
-Cardiovascular disease
-Cancer/severe disease
-Psychiatric disease
t 30-40%
11279 (81.61)
5154 (37.29)
5376 (38.90)
8540 (61.79)
3442 (24.91)
1796(13.00)
Patients with complications
-Retinopathy
-Nephropathy
-Foot infection
t 20%
3365 (24.35)
394 (2.85)
3065 (22.18)
75
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Based on above factors associated with patient risk and their impacts on patient risk
(Table 4.10), nine criteria for selecting high-risk patients were presented (Table
4.11). In each criterion, high-risk patients were selected based on the standard
coefficients of factors associated with patient risk in the previous period. Total
health care costs in the next period of these high-risk patients selected in the previous
period were calculated and compared among the nine criteria. Figure 4.6 shows the
comparison of total health care costs in the next period of high-risk and low-risk
patients selected based on factors associated with patient risk in the previous period
among nine criteria included age factor. However, 66% of patients in the data set
were older than 65 years old and that might lead to have a false positive selection of
high-risk patients. Table 4.12 shows nine criteria for selecting high-risk patients
without age factor. In addition, figure 4.7 demonstrates the comparison of total
health care costs in the next period of high-risk and low-risk patients selected based
on factors associated with patient risk in the previous period among nine criteria
without age factor. Tables A5 and A6 show the number of high-risk patients, mean
healthcare costs, mean age, number of female patients, mean medication compliance,
number of hospitalizations, and number of ER visits in each high-risk selection
criteria with and without age factor, respectively.
76
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.11: High-Risk Patient Selection Criteria (Age Factor Included)
Criteria Factors Associated with Patient Risk
1 - Age greater than 65 years
- Hospitalization/ER in the past
2 - Age greater than 65 years
- Hospitalization/ER in the past
- Having both drugs (i.e., oral hypoglycemic drugs and insulin)
3 - Age greater than 65 years
- Hospitalization/ER in the past
- Having both drugs (i.e., oral hypoglycemic drugs and insulin)
- No office visit in the previous six-month period
4 - Age greater than 65 years
- Hospitalization/ER in the past
- Having both drugs (i.e., oral hypoglycemic drugs and insulin)
- No office visit in the previous six-month period
- No glucose monitoring test in the previous six-month period
5 - Age greater than 65 years
- Hospitalization/ER in the past
- Having both drugs (i.e., oral hypoglycemic drugs and insulin)
- No office visit in the previous six-month period
- No glucose monitoring test in the previous six-month period
- No HbAlc test in the previous six-month period
6 - Age greater than 65 years
- Hospitalization/ER in the past
- Having both drugs (i.e., oral hypoglycemic drugs and insulin)
- No office visit in the previous six-month period
- No glucose monitoring test in the previous six-month period
- No HbAlc test in the previous six-month period
- Having oral hypoglycemic, insulin, anti-hypertensive, lipid
lowering drugs added
77
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.11: High-Risk Patient Selection Criteria (Age Factor Included)
(Contd.)
Criteria Factors Associated with Patient Risk
7 - Age greater than 65 years
- Hospitalization/ER in the past
- Having both drugs (i.e., oral hypoglycemic drugs and insulin)
- No office visit in the previous six-month period
- No glucose monitoring test in the previous six-month period
- No HbAlc test in the previous six-month period
- Having oral hypoglycemic, insulin, anti-hypertensive, lipid
lowering drugs added
- Having Comorbidities such as hypertension or hyperlipidemia
8 - Age greater than 65 years
- Hospitalization/ER in the past
- Having both drugs (i.e., oral hypoglycemic drugs and insulin)
- No office visit in the previous six-month period
- No glucose monitoring test in the previous six-month period
- No HbAlc test in the previous six-month period
- Having oral hypoglycemic, insulin, anti-hypertensive, lipid
lowering drugs added
- Having comorbidities such as hypertension or hyperlipidemia
- Having complications such as retinopathy, nephropathy, or
infection
9 - Age greater than 65 years
- Hospitalization/ER in the past
- Having both drugs (i.e., oral hypoglycemic drugs and insulin)
- No office visit in the previous six-month period
- No glucose monitoring test in the previous six-month period
- No HbAlc test in the previous six-month period
- Having oral hypoglycemic, insulin, anti-hypertensive, lipid
lowering drugs added
- Having comorbidities such as hypertension or hyperlipidemia
- Having complications such as retinopathy, nephropathy, or
infection
- Lower compliance (MPR < 80%)
78
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 4.6: Comparison of Total Healthcare Costs of High-Risk Patients Among
Nine High-Risk Selection Criteria with Age Factor Included
Cost ($)
7000 n
6000
5000
4000
3000
2000
1000
4 5 6 7 1 2 3
— High-Risk Criteria 1
— ■— High-Risk Criteria 2
High-Risk Criteria 3
— - High-Risk Criteria 4
— 3 K — High-Risk Criteria 5
— • — High-Risk Criteria 6
1 High-Risk Criteria 7
High-Risk Criteria 8
— — — High-Risk Criteria 9
— ♦ - - Low-Risk Criteria 1
— - Low-Risk Criteria 2
— -i— ■ Low-Risk Criteria 3
— ■ Low-Risk Criteria 4
Low-Risk Criteria 5
- Low-Risk Criteria 6
Low-Risk Criteria 7
— - — - Low-Risk Criteria 8
_ . Low-Risk Criteria 9
Period
The above results show that total healthcare costs in the next period of high-risk
patients who were selected based on factors associated with patient risk in the
previous period in each criterion had been increasing compared to those of low-risk
patients. Total healthcare costs of high-risk patients selected among nine criteria
were almost the same since patients greater than 65 years who were approximately
66% of the sample in the data set were selected.
79
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.12: High-Risk Patient Selection Criteria (Age Factor Not Included)
Criteria Factors Associated with Patient Risk
1 - Hospitalization/ER in the past
2 - Hospitalization/ER in the past
-Having both drugs (i.e., oral hypoglycemic drugs and insulin)
3 - Hospitalization/ER in the past
- Having both drugs (i.e., oral hypoglycemic drugs and insulin)
- No office visit in the previous six-month period
4 - Hospitalization/ER in the past
- Having both drugs (i.e., oral hypoglycemic drugs and insulin)
- No office visit in the previous six-month period
- No glucose monitoring test in the previous six-month period
5 - Hospitalization/ER in the past
- Having both drugs (i.e., oral hypoglycemic drugs and insulin)
- No office visit in the previous six-month period
- No glucose monitoring test in the previous six-month period
- No HbAlc test in the previous six-month period
6 - Hospitalization/ER in the past
- Having both drugs (i.e., oral hypoglycemic drugs and insulin)
- No office visit in the previous six-month period
- No glucose monitoring test in the previous six-month period
- No HbAlc test in the previous six-month period
- Having oral hypoglycemic, insulin, anti-hypertensive, lipid
lowering drugs added
7 - Hospitalization/ER in the past
- Having both drugs (i.e., oral hypoglycemic drugs and insulin)
- No office visit in the previous six-month period
- No glucose monitoring test in the previous six-month period
- No HbAlc test in the previous six-month period
- Having oral hypoglycemic, insulin, anti-hypertensive, lipid
lowering drugs added
- Having Comorbidities such as hypertension or hyperlipidemia
80
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.12: High-Risk Patient Selection Criteria (Age Factor Not Included)
Criteria Factors Associated with Patient Risk
8 - Hospitalization/ER in the past
- Having both dmgs (i.e., oral hypoglycemic dmgs and insulin)
- No office visit in the previous six-month period
- No glucose monitoring test in the previous six-month period
- No HbAlc test in the previous six-month period
- Having oral hypoglycemic, insulin, anti-hypertensive, lipid
lowering dmgs added
- Having comorbidities such as hypertension or hyperlipidemia
- Having complications such as retinopathy, nephropathy, or
infection
9 - Hospitalization/ER in the past
- Having both dmgs (i.e., oral hypoglycemic dmgs and insulin)
- No office visit in the previous six-month period
- No glucose monitoring test in the previous six-month period
- No HbAlc test in the previous six-month period
- Having oral hypoglycemic, insulin, anti-hypertensive, lipid
lowering dmgs added
- Having comorbidities such as hypertension or hyperlipidemia
- Having complications such as retinopathy, nephropathy, or
infection
- Lower compliance (MPR < 80%)
81
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Figure 4.7: Comparison of Total Healthcare Costs of High-Risk Patients Among
Nine High-Risk Selection Criteria Without Age Factor Included
Cost ($)
— — High-Risk Criteria 1
• High-Risk Criteria 2
High-Risk Criteria 3
• High-Risk Criteria 4
■ High-Risk Criteria 5
— • — High-Risk Criteria 6
—I -----High-Risk Criteria 7
—-— High-Risk Criteria 8
—-— High-Risk Criteria 9
- Low-Risk Criteria 1
- Low-Risk Criteria 2
-*— ■ Low-Risk Criteria 3
- Low-Risk Criteria 4
■ S C - Low-Risk Criteria 5
- Low-Risk Criteria 6
Low-Risk Criteria 7
- - Low-Risk Criteria 8
- Low-Risk Criteria 9
12000 n
10000
8000
6000
4000 ----- »
2000
2 1 3 4 5 6 7
Period
82
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Research Question #2; Can the risk models be validated?
Table 4.13: Validation of Predictive Risk Models
Predictive Models R^
Development data
R^
Validation data
Model #1: Costs 0.608
(N=55,328)
0.605
(N=55,328)
Model #2: Occurrence of
hospitalization or ER visit
0.512
(N=55,328)
0.510
(N=55,328)
Model #3: Time to event 0.123
(N=3212)
0.123
(N=3456)
Based on the results above, it is shown that there was no significant difference
between obtained from development (R^=0.608) and validation (R^=0.605) data
sets for model #1. Similarly for model #2 and model #3, R^ calculated from
development data set was not significantly different compare to that from validation
data set. In conclusion, these predictive risk models were valid.
Research question #3: Which is the most appropriate methodological approach to
identify patient risk?
Table 4.14-4.16 show the prevalence, sensitivity, specificity, and predictive value
positive, and predictive value negative in developm ent data set com pared to those in
validation data set in each period for three predictive risk m odels.
83
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.14: Performance Measures of Model #1 (Total Healthcare Costs)
Data Period Prevalence
(%)
Sensitivity
(%)
Specificity
(%)
Predictive
value positive
(%)
Predictive
value negative
(%)
Development 1 0.64 15.91 99.69 25.00 99.46
Validation 0.61 16.67 99.68 24.14 99.49
Development 2 0.45 29.03 99.65 27.27 99.68
Validation 0.56 28.21 99.52 13.16 99.51
Development 3 0.45 29.03 99.69 30.00 99.68
Validation 0.46 25.00 99.52 19.51 99.65
Development 4 0.07 40.00 99.49 5.41 99.96
Validation 0.09 33.33 99.49 5.41 99.94
Development 5 0.71 24.49 99.55 27.91 99.46
Validation 0.55 22.42 99.56 18.92 99.55
Development 6 0.49 32.35 99.16 15.94 99.66
Validation 0.38 19.23 99.32 9.62 99.69
Development 7 0.36 48.00 98.98 14.63 99.81
Validation 0.40 39.29 99.10 15.07 99.75
Development 8 0.43 33.33 98.95 12.20 99.71
Validation 0.26 33.33 98.90 7.32 99.82
84
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.15: Performance Measures of Model #2 (The Occurrence of
Hospitalization or ER visit)
Data Period Prevalence
(%)
Sensitivity
(%)
Specificity
(%)
Predictive
value positive
(%)
Predictive
value negative
(%)
Development 1 5.41 8.02 99.65 56.60 94.99
Validation 5.57 4.42 99.72 48.57 94.65
Development 2 3.90 6.67 99.50 35.29 96.33
Validation 3.76 5.77 99.59 35.71 96.44
Development 3 2.76 11.52 99.78 59.46 97.54
Validation 2.56 5.65 99.63 28.57 97.57
Development 4 2.44 12.43 99.61 44.68 97.85
Validation 2.54 7.95 99.70 41.18 97.65
Development 5 2.73 11.11 99.54 40.38 97.55
Validation 2.98 11.65 99.63 48.98 97.35
Development 6 2.81 14.95 99.60 51.79 97.59
Validation 2.59 15.64 99.50 45.16 97.80
Development 7 3.08 15.49 99.48 48.53 97.37
Validation 2.69 13.44 99.63 50.00 97.66
Development 8 3.18 19.09 99.67 65.63 97.40
Validation 2.75 13.16 99.70 55.56 97.60
85
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.16: Performance Measures of Model #3 (Time to Hospitalization or ER
visit)
Data Event Prevalence Sensitivity Specificity Predictive
positive value
Predictive
negative value
Development 1 1.35 26.67 95.35 7.27 98.96
Validation 1.61 17.65 94.59 5.08 98.59
Development 2 1.79 23.08 95.10 7.89 98.55
Validation 2.52 17.65 96.04 10.34 97.83
Development 3 3.92 20.00 95.51 15.38 96.69
Validation 3.16 26.67 95.86 17.39 97.56
Development 4 4.18 13.33 94.19 9.09 96.14
Validation 4.71 31.25 95.06 23.81 96.55
Development 5 8.02 28.57 97.51 50.00 94.00
Validation 9.21 31.82 95.39 41.18 93.24
Development 6 5.82 9.09 96.07 12.50 94.48
Validation 7.47 0.00 91.93 00.00 91.93
Development 7 11.36 20.00 95.73 37.50 90.32
Validation 8.40 40.00 95.41 44.44 94.55
Development 8 9.89 33.33 96.34 50.00 92.94
Validation 6.67 60.00 95.71 50.00 97.10
Development 9 12.00 66.67 93.18 57.14 95.35
Validation 15.38 16.67 87.88 20.00 85.29
In each period, patients were divided into two groups (i.e., high-risk and low-risk
groups) based on high-risk patient selection criteria. For model #1, patients were
predicted to be high-risk if they consumed total healthcare costs greater than 80%
and patients were predicted to be low-risk if they had total healthcare cost less than
or equal 80%. For model #2, patients were predicted to be high-risk if their
probability of having hospitalization or ER visit was greater than 0.5 and patients
were predicted to be low-risk if they had probability of having hospitalization or ER
visit less than or equal 0.5. For model #3, patients were predicted to be high-
86
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
risk if they had the probability of survival from not having hospitalization or ER visit
less than 0.5, whereas if their probability of the survival greater than 0.5, they were
predicted to be low-risk. Table A7, A8, and A9 show the comparison between high-
risk and low-risk groups identified by cost, hospitalization or ER event, time to event
criteria, respectively. Table 4.14, 4.15, and 4.16 summarize the results of
performance measures of model #1, model #2, and model #3 for both development
and validation data sets in each period, respectively. The third column illustrates the
prevalence of actual high-risk patients. The fourth column and the fifth column
show the sensitivity (i.e., the percent of high-risk patients correctly classified by the
predication rule) and the specificity (i.e., the percent of low-risk patients correctly
classified), respectively. The sixth column and the seventh column demonstrate the
predictive value positive (i.e., the percent of patients predicted to be high-risk who
are classified correctly) and the predictive value negative (i.e., the percent of patients
predicted to be low-risk who are classified correctly). For example, for development
data set in period one, the classification identified 0.64% of patients as high-risk and
16% of all cases who were high-risk while misclassifjdng only 0.31% of those who
were low-risk (100%-99.69%). Of those patients predicted to be high-risk, 25%
were actually high-risk, whereas of those predicted to be low-risk, 99% were actually
low-risk. To compare three predictive models. Table 4.17 shows the means of
performance measures for each model.
87
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table 4.17: Means of Performance Measures for Each Model.
Model Data Prevalence
(%)
Sensitivity
(%)
Specificity
(%)
Predictive
positive value
(%)
Predictive
negative value
(%)
Model #1 Development 0.45 31.52 99.40 19.80 99.68
Validation 0.41 27.18 99.39 14.14 99.68
Model #2 Development 3.29 12.41 99.60 50.30 97.08
Validation 3.18 9.71 99.64 44.22 97.09
Model #3 Development 7.29 26.74 95.44 27.41 95.27
Validation 7.39 26.85 94.20 23.58 94.73
Based on the results summarized in Table 4.17, in the development data set, there
were a higher percentage of high-risk patients correctly classified by the prediction
rule in model #1 compared to that in model #2 and model #3. In addition, the
percentage of misclassifying for those low-risk patients in model #1 and model #2
was less than that in model #3. Of those patients predicted to be high-risk, model #2
yielded higher percentage of actual high-risk patients compared to that in model #1
and model #3, whereas of those predicted to be low-risk, model #1 yielded higher
percentage of actual low-risk patients compared to that in model #2 and model #3.
The results from validation data set were similar to those in development data set.
Table 4.18, 4.19, and 4.20 demonstrate the model disagreements on individual
patients.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
CD
■ D
O
Q .
C
o
CD
O.
■ D
CD
C/)
W
o'
o
o
o
o
■ D
c q '
o
CD
■ D
O
Q .
C
a
o
o
■ O
o
CD
Q .
■ D
CD
(/)
(/)
Table 4.18: Model Disagreements on Individual Patients Using Actual Total Healthcare Cost Criteria
Model compared Data Set Comparison group:
A predicts "high-risk"
B predicts "low-risk"
# of Total
Patients
Comparison group:
A predicts "low-risk"
B predicts "high-risk"
# of Total
Patients
#of
Disagreed
patients
%
of Pts
correctly
predicted
from A
%
of Pts
correctly
predicted
from B
# of Pts correctly predicted # of Pts correctly predicted
From A From B A+B From A From B A -H B
A:-Cost Developm ent 154 67 221 186 41 111 448 75.89 24.11
B :-H ospitalization V alidation 155 88 243 174 26 200 443 74.27 25.73
A:-Cost Developm ent 154 99 253 109 4 113 366 71.86 28.14
B:-Tim e to event V alidation 155 109 264 110 1 111 375 70.67 29.33
A ;-H ospitalization Developm ent 41 196 237 95 4 99 336 40.48 59.52
B:-Tim e to event V alidation 26 183 209 98 1 99 308 40.26 59.74
Note: The top entry in each cell o f the table is the relavant percentage or num ber in the developm ent data set, and the bottom entry in italic style is the corresponding
percentage or num ber in the validation data set.
00
V O
CD
■ D
O
Q .
C
o
CD
Q .
■ D
CD
C/)
W
o'
o
o
o
o
■ D
c q '
O ’
o
CD
■ D
O
Q .
c
a
o
Q
■ O
o
CD
Q .
Table 4.19: Model Disagreements on Individual Patients Using Actual Occurrence of Hospitalization or ER Event Criteria
Model compared Data Set Comparison group:
A predicts "high-risk"
B predicts "low-risk"
# of Total
Patients
Comparison group:
A predicts "low-risk"
B predicts "high-risk"
# of Total
Patients
# o f
Disagreed
patients
%
of Pts
correctly
predicted
from A
%
of Pts
correctly
predicted
fromB
# of Pts correctly predicted # of Pts correctly predicted
From A From B A+B From A From B A+B
A >H ospitalization D evelopm ent 231 18 249 128 115 243 492 72.97 27.03
B:-Cost Validation 205 12 217 139 121 260 477 72.12 27.88
A:-H ospitalization D evelopm ent 231 15 246 83 30 113 359 87.47 12.53
B:-Tim e to event V alidation 205 16 221 85 26 111 332 87.35 12.65
A;-Cost D evelopm ent 115 139 254 84 30 114 368 54.08 45.92
B:-Tim e to event Validation 121 144 265 86 26 112 377 54.91 45.09
Note: The top entry in each cell o f the table is the relavant percentage or num ber in the developm ent data set, and the bottom entry in italic style is the con-esponding
percentage or num ber in the validation data set.
■ D
CD
(/)
(/)
V O
O
CD
■ D
O
Q .
C
o
CD
Q .
■ D
CD
C/)
W
o'
o
o
o
o
■ D
c q '
O ’
o
CD
■ D
O
Q .
c
a
o
Q
■ O
o
CD
Q .
Table 4.20: Model Disagreements on Individual Patients Using Actual Probability of Survival from Not Having Hospitalization or ER Event Criteria
Model compared Data Set Comparison group:
A predicts "high-risk"
B predicts "low-risk"
# of Total
Patients
Comparison group:
A predicts "low-risk"
B predicts "high-risk"
# of Total
Patients
#of
Disagreed
patients
%
of Pts
correctly
predicted
from A
%
of Pts
correctly
predicted
fromB
# of Pts correctly predicted # of Pts correctly predicted
From A From B A+B From A From B A+B
A:-Tim e to event Developm ent 35 75 110 242 8 250 360 76.94 23.06
B :-C ost Validation 35 77 112 250 15 265 377 75.60 24.40
A:-Tim e to event Developm ent 35 67 102 227 21 248 350 74.86 25.14
B :-H ospitalization Validation 35 69 104 199 15 214 318 73.58 26.42
A :-C ost Developm ent 8 179 187 172 21 193 380 47.37 52.63
B : -H ospitalization Validation 15 206 221 163 15 178 399 44.61 55.39
Note: The top entry in each cell o f the table is the relavant percentage or num ber in the developm ent data set, and the bottom entry in italic style is the corresponding
percentage or num ber in the validation data set.
■ D
CD
(/)
(/)
In this research, there were three criteria to identify actual high-risk patients. First,
based on actual total healthcare cost criteria, patients consuming 80% of total
healthcare costs were in high-risk group and patients consuming 20% of total
healthcare costs were in low-risk group (Table 4.18). Second, based on actual
occurrence of hospitalization or ER event criteria, patients who had at least one
hospitalization or ER event were high-risk and patients who never had
hospitalization or ER event were low-risk (Table 4.19). Last, based on actual
probability of survival from not having hospitalization or ER event criteria, patients
who had probability of survival from not having hospitalization or ER event less than
0.5 were high-risk and patients who had probability of survival from not having
hospitalization or ER event equal or greater than 0.5 were low-risk (Table 4.20).
Based on Table 4.18, model disagreements on individual patients using actual total
healthcare cost criteria are shown. In the first column, comparison group A was
compared to comparison group B. The first row was model cost compared to model
hospitalization, the second row was model cost compared to model time to event,
and the third row was model hospitalization compared to time to event. Based on
Table 4.19, model disagreements on individual patients using actual occurrence of
hospitalization or ER event criteria are demonstrated. In the first column, the first
row was model hospitalization compared to cost, the second row was model
hospitalization compared to time to event, and the third row was model cost
compared to time to event. Based on Table 4.20, model disagreements on individual
92
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
patients using actual probability of survival from not having hospitalization or ER
event criteria are shown. In the first column, the first row was model time to event
compared to cost, the second row was model time to event compared to
hospitalization, and the third row was model cost compared to hospitalization.
The second column was presented which data set (i.e., development or validation)
was used to estimate the results. The third column presented the number of patients
who were actual high-risk patients and also predicted to be high-risk patients by
comparison group A and the number of patients who were actual low-risk patients
and predicted to be low-risk patients by comparison group B. The fourth column
was the total number of patients who were actual high-risk and predicted to be high-
risk by comparison A plus patients who were actual low-risk and predicted to be
low-risk by comparison group B.
The fifth column presents the number of patients who were actual low-risk patients
and predicted to be low-risk patients by comparison group A and the number of
patients who were actual high-risk patients and predicted to be high-risk patients by
comparison group B. The sixth column was the total number of patients who were
actual low-risk and predicted to be low-risk by comparison A plus patients who were
actual high-risk and predicted to be high-risk by comparison group B.
93
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
The seventh column was the total number of disagreed patients (the fourth column +
the sixth column). The eighth column was the percentage of patients who were
correctly predicted by comparison group A [(the third column (From A) + the fifth
column (From A))* 100/the seventh column]. The ninth column was the percentage
of patients who were correctly predicted by comparison group B [(the third column
(From B) + the fifth column (From B))* 100/the seventh column].
To compare predicted high-risk patients from three predictive risk models with
actual high-risk patients, model disagreements on individual patients were used.
Based on the high-risk identification criteria for total healthcare costs (i.e., actual
high-risk patients were patients who consumed 80% of total healthcare costs), model
#1 predicts high-risk patients more correctly than model #2 and model #3 (Table
4.18). In addition, the percentage of correct risk prediction for model #3 was higher
than that for model #2. Based on the actual cost criteria, the ranking of model
prediction from best to worst was rated to be model #1, model #3, and model #2.
Table 4.19 demonstrates that model #2 gave a significant higher percentage of the
correct risk prediction compared to that in model #1 and model #3 based on high-risk
identification criteria regarding having hospitalization or ER visit event (i.e., actual
high-risk patients were patients who ever had hospitalization or ER visit related to
diabetes). Compared model #1 to model #3, model #1 had higher percentage of right
risk prediction. The best to worst risk model prediction was ranked to be model #2,
94
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
model #1 and model #3 based on the actual occurrence of hospitalization or ER visits
related to diabetes.
By using actual probability of survival from not having hospitalization or ER visit
criteria to identify actual high-risk patients (Table 4.20), model #3 yielded a
significant higher percentage of correct risk prediction compared to those in model
#1 and #2. It was ranked to be model #3, #2, and #1 for best to worst risk model
prediction.
95
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
CHAPTER 5: DISCUSSION AND CONCLUSIONS
Diabetes is a chronic disease and one of the most costly health problems in the US.
MediCal, a FFS government health insurance plan including patients who are poor
and disabled, has spent a high amount on health care costs for patients with diabetes.
A primary concern for MediCal is to be able to improve patients’ health outcomes
and yield long term cost saving. Prior high-risk identification may help MediCal
administrators and policy makers evaluate what healthcare services need to be
provided and to whom intervention and treatment should be targeted. However,
health plan administrators have difficulty in identifying patients for at risk programs
prior to worsened health status. Therefore, the purpose of this research is to develop
various methodologies that can be used to identify high-risk patients, to investigate
factors associated with an increase in risk, and to evaluate whether risk models were
valid based on claims data from the MediCal patients with diabetes under FFS
system.
Based on administrative claims data, large populations of patients could be used to
identify those high-risk patients to receive targeted intervention programs, hopefully
prior to worsened outcomes. In this research, “risk” was indicated by total
healthcare costs, the occurrence of hospitalization or ER visit related to diabetes, and
time to hospitalization or ER visit related to diabetes. Hence three
96
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
methodologies identifying high-risk patients were developed based on longitudinal
data and cross-sectional data with multiple events. This research used and captured
prior health care information obtained from administrative claims data over time
(i.e., longitudinal basis) in order to identify who would be high-risk patients in the
future, whereas all previous studies captured only a point in time (i.e., cross-sectional
basis) to identify who were high-risk patients at that point. The longitudinal data had
more advantages over cross-sectional data since it provided a large number of data
points, increased degrees of freedom, and reduced multicollinearity. In this research,
the GEE approach was also used to develop the risk models since it has been
extensively used to handle both normal and non-normal outcome variables with
various distribution forms in longitudinal studies and it yields consistent and
asymptotically Gaussian estimators of the regression coefficients and of their
variances under weak assumptions about the actual correlation among an
individual’s observations (Zeger and Liang, 1992). Therefore, this results in
efficient estimates to develop the risk models and accurate risk factors in the
previous period that can lead to a “spike” in costs and healthcare utilization on a
periodic basis.
The discussion section is divided into four parts. First three parts are related to the
discussions on each of the three research questions (i.e., what are the factors
identifying patient risk?, can the risk models be validated?, which is the most
appropriate methodological approach to identify patient risk?), respectively. Last,
97
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the significance of findings to MediCal policy makers and providers related to high-
risk identification and significant MediCal health policy options is presented.
5.1 W hat are the Factors Identifying Patient Risk?
To develop methodologies to identify high-risk patients, factors associated with an
increase in risk were investigated. Generally the factors incorporated into most
studies based on cross-sectional basis were demographic factors, complication,
comorbidity, type of medication, prior healthcare utilization, and payment system.
In addition to these factors, in this research, follow-up services based on diabetic
guidelines and diabetes treatment which are important factors associated with an
increase in patient risk were used to investigate the impact on patient risk.
The results indicate that there were significant association of demographic, prior
healthcare costs and use, diabetes treatment, follow-up services based on diabetic
guidelines, healthcare use, comorbidity, and complication factors with an increase in
healthcare costs and incidence of hospitalization or ER events. Even though these
factors had a significant association with an increase in patient risk, only the effect of
these factors (i.e., prior healthcare costs and use, diabetes treatment, and follow-up
services based on diabetes guidelines) would aid health plan administrators to
prevent patients becoming high-risk in the future.
98
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Prior Healthcare Cost and Utilization
The results show that prior healthcare cost and utilization such as hospitalization or
ER visit had a positive association with an increase in future healthcare cost and
utilization. Patients consuming high costs or having hospitalization or ER event in
the past were more likely to have high costs or experience the event in the future.
Similar results were also found in the studies of Guo et al (1998) and Bhattacharyya
(1999).
Diabetes Treatment
Previous research indicates that the use of anti-diabetic medication (i.e., insulin and
oral medication) was significantly associated with higher costs (Guo et al, 1998;
Bhattacharyya, 1999). In this research, for diabetes treatment factors, the results
explain that patients taking both oral hj^oglycemic drugs and insulin were probably
more severe and thus positively associated with higher total healthcare costs, the
occurrence of hospitalization or ER visit, and the risk of hospitalization or ER visit in
the future. To date, there is no study investigating the association of MediCal
treatment factors such as having drugs (e.g., the combination of oral hypoglycemic,
insulin, anti-hypertensive, lipid-lowering drugs) added into the regimen, having oral
hypoglycemic drugs changed to other classes or to insulin, having dose of oral
hypoglycemic drugs increased, medication compliance with total healthcare costs,
99
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
the occurrence of hospitalization or ER event, and the risk of hospitalization or ER
event in the future. Having drugs added into the regimen had a positive association
with costs, the occurrence of hospitalization or ER event, and the risk of
hospitalization or ER event. In addition, patients having drug changed to different
classes or to insulin were more likely to have higher costs and experience
hospitalization or ER event in the future. These are probably proxies for disease
progression, an increase in severity, or the lack of diabetes control. However having
dmg dose increased had a little positive significant association with costs (p value <
0.1). There was no significant association with having drug dose increased on the
occurrence of hospitalization or ER event or the risk of hospitalization or ER event.
Perhaps this is also as a proxy indicating provider action to better control patients
with diabetes
Medication Compliance
Although medication compliance is a very important indicator of poor control for
patients with diabetes, there is no study evaluating the effect of diabetic medication
compliance on healthcare costs and utilization. The results from three predictive risk
models confirm that patients compliant with oral hypoglycemic medication were less
likely to consume higher costs, experience hospitalization or ER event, or have
longer time to hospitalization or ER event. Patients who had lower compliance with
too
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
oral hypoglycemic drags (MPR < 80%) were high-risk patients and these patients
could be identified based on claims data. Intervention programs could be developed.
Follow-up Services based on Diabetes Guidelines
Based on guidelines for diabetes from American Diabetes Association (1996),
diabetic patients were recommended to have follow-up services such as an office
visit (every three months for patients taking insulin and every six months for patients
taking only oral hypoglycemic medications), glucose monitoring strips as a proxy for
patient self monitoring, and regular lab tests by healthcare providers. Ideal patient
care suggests an HbAIc test every six months and a cholesterol test, ketone test,
kidney check-up, dilated eye and foot examination every year. Although no study
has investigated the association of some follow-up services factors with patient risk,
these factors can be used as clinical indicators of poor control for diabetes and may
indicate patients at risk for poorer outcomes. Based on this information, healthcare
providers could prevent patients to be at risk in the future. In this research, patients
having the follow up service factors listed above had statistically significantly lower
total healthcare costs compared to those who did not have follow-up services. This
indicates that patients who were provided with follow-up services recommended by
diabetic guidelines were more likely to have lower costs in the future. This means
that those receiving these follow-up services based on guidelines for diabetes may be
able to prevent high-risk status in the future. In addition, the results also suggest that
101
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
patients having office visits, having glucose monitoring strip, and having HbAlc test
every six months were less likely to have hospitalization or ER visit and had a longer
time to hospitalization or ER visit in the future. These risk factors (i.e., the lack of
appropriate follow-up services) can be identified and intervention programs can be
developed.
Controlling Factors
Furthermore, there was a significant association of controlling factors such as
demographics, comorbidities, complications, and healthcare use such as specialist
visit with patient risk. These results may not quite aid health plan administrators to
prevent patients to be high-risk and it may be too late for MediCal to do anything
since they are already at risk, but it is still the useful information for MediCal since
these factors can be used as surrogate indicators for disease severity and MediCal
may indicate targeted patients for extra monitoring.
1) Demographic factors
For demographic factors, the results from all three predictive risk models based on
longitudinal basis suggest that patients with older age were more likely to consume
higher healthcare costs and utilization and all previous studies supported this finding.
Moreover, male patients were more likely to have higher expenditures and to use
102
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
hospital services than female patients. Krop et al (1998) found the same result,
whereas the study of Bhattacharyya et al (1998) showed that female patients were
more likely to consume higher healthcare costs and utilization. In addition, all
previous research indicates that there was no statistically significant difference in
race. However, in this research, it is found that non-white patients were more likely
to have higher healthcare costs than white patients. The research findings indicate
that patients with older age, male gender, or non-white race were more likely
associated with an increase in total healthcare cost and the risk of hospitalization/ER.
In addition, the higher the percentage of patients enrolled in a managed care plan in
counties, the more negative impact on the occurrence of hospitalization or ER visit,
indicating that patients under the FFS system and living in the counties with a higher
percentage of people under managed care plan were more likely to have
hospitalization or ER visits. The results confirm that patients being left out from the
managed care plan and being under the FFS system tended to be at higher risk for
hospitalization or ER visit related to diabetes. In general, patients under the FFS
system in urban counties where percentage of people enrolling in managed care plan
was higher were possibly more complicated and sicker than patients in rural counties
where patients were under the FFS system only.
103
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
2) Complications
Diabetic complications (e.g., retinopathy, nephropathy, and neuropathy) and
comorbidities (e.g., hypertension, hyperlipidemia, coronaty artery cerobrovascular
disease, and congestive heart failure) had a positive significant association with
higher costs (Bhattacharyya at al, 1998; Bhattacharyya et al, 1999). In this
research, comorbidities related to diabetes included hypertension, hyperlipidemia,
cardiovascular disease (e.g., angina, coronary disease, heart disease, myocardial
infarction, and stroke), cancer/severe disease (e.g., HIV, liver disease, malignancy,
cystic fibrosis, and transplant), and psychiatric disease (e.g., depression, bipolar,
Parkinson, and psychotic). The results indicate that diabetic patients with any of
these comorbidities had significantly higher costs than those without comorbidities.
In addition, diabetic patients with comorbidities (e.g., hypertension, hyperlipidemia,
systemic chronic disease, cardiovascular disease, cancer/severe disease, or
psychiatric disease) were more likely to have a hospitalization or ER visit.
3) Complications
Complications related to diabetes in this research were retinopathy, nephropathy,
neuropathy, foot infection, skin infection, and vagina infection. Diabetic patients
with complications such as retinopathy, nephropathy, and foot infection consumed
104
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
significantly higher healthcare expenditures and utilizations such as hospitalization
or ER visits than those without complications.
4) Use o f Specialists
When the serious complications occur in patients with diabetes, diabetic patients
may need to see specialists such as ophthalmologists, cardiologists, nephrologists,
neurologists, gynecologists, and infectious disease specialists. The results reveal that
having visits to these specialists were positively associated with an increase in
healthcare costs. Also, patients having visits to cardiologists, nephrologists,
neurologists, or infectious disease specialists were more likely to have higher
hospitalization or ER visit. Patients required to visit specialists were already at high-
risk for an increase in healthcare costs and utilization.
5.2 Can the Risk Models be Validated?
To develop predictive risk models, the significant parameter estimates of these
factors associated with patient risk were used. To measure predictive validity, split
sample validation was performed because the most important test of a model was
how well it worked on data that were not used to develop the model. In this
research, a summary of model fit such as R^, computed from paired “observed and
predicted” outcomes was calculated to produce a cross-validated measure of the
105
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
model’s performance. The results indicate that three predictive models were valid
since there was no significant difference in between development data set and
validation data set. Furthermore the model performance measures (i.e., specificity,
sensitivity, predictive value positive, and predictive value negative) of three
predictive risk models support that the models were valid since these measures
obtained from development and validation data sets were similar. The results
suggest that there was a higher percentage of high-risk patients correctly classified
by using high cost identification compared to that of high-risk patients correctly
classified by the occurrence of hospitalization or ER event identification. This may
imply that patients are identified as high-risk patients even though they have never
had hospitalization or ER event before. Given prediction of being high-risk patients,
the model predicting the occurrence of hospitalization or ER event (model #2) could
significantly predict higher percentage of actual high-risk patients compared to other
two models. The research findings confirm that the risk models were valid to use for
high-risk identification.
5.3 Which is the Most Appropriate Methodological Approach to Identify
Patient Risk?
It is difficult to conclude which methodology is the most appropriate to identify
high-risk patients since patient risk can be identified in various ways. Since there is
no standard method to distinguish who is an actual high-risk patient, it will depend
106
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
on how “patient risk” is identified and whose perspective of the study is considered.
In this research, “high-risk” was indicated by healthcare costs, the occurrence of
hospitalization or ER event, and time to hospitalization or ER event. It was
necessary to set the criteria to identify actual high-risk patients in order to determine
if they were matched with predicted high-risk patients. Based on model
disagreements on individual high-risk patients compared to actual high-risk patients
indicated by above three criteria, the percentage of correct prediction was compared
among models. The results confirm that if high-risk patients were identified by high
healthcare costs, the predictive cost model (model #1) was the most appropriate to
use since it yielded the highest percentage of correct prediction. Likewise, if high-
risk patients were defined as patient who had the occurrence of hospitalization or ER
event, the predictive hospitalization or ER event model (model #2) was the most
suitable to apply. Similarly, if high-risk patients were indicated by shorter time to
hospitalization or ER event, the predictive time to hospitalization or ER event model
(model #3) was the most proper to utilize. To select the appropriate methodology for
high-risk patient selection, it depended upon risk definition and the research
perspective.
This research was conducted based on MediCal’s perspectives that were to improve
patient outcomes and reduce the aggregate healthcare costs to the system. Therefore,
the predictive cost model might be the most appropriate methodology to identify
high-risk patients. The goal of high-risk identification was to help MediCal
107
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
healthcare providers and policy makers selectively intervene with the patients who
were predicted to be high-risk patients, so that early interventions can be
implemented towards these high-risk patients.
5.4 Significance of Findings to MediCal Policy Makers and Providers
High-Risk Identification
MediCal healthcare providers may be able to select the patients who tend to be high-
risk in the future based on the information in the past obtained from claims data such
as total healthcare costs, the incidence of hospitalization or ER visit, use of
specialists, and other information (e.g., drug treatment problems, needed follow-up
services based on diabetic guidelines, and comorbidities).
The research findings support that patients consuming higher healthcare cost or
having hospitalization or ER event in the past were more likely to have an increase in
healthcare costs or incidence of hospitalization or ER visit in the future. Moreover,
patients having visits to specialists were more likely to consume higher healthcare
costs and have higher hospitalization or ER visit. Patients who needed to visit
specialists were already at high-risk for an increase in costs and hospitalization or ER
visit and the use of specialists was also costly to MediCal.
In this research, based on the incidence of hospitalization or ER visit and other
108
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
infomation related to drug treatment problems, needed follow-up services based on
diabetic guidelines, and comorbidities, patients were identified to be high-risk
patients in the future if in the past, patients had hospitalization or ER visit, had both
oral hypoglycemic drugs and insulin, had oral hypoglycemic, insulin, anti
hypertensive, or lipid lowering drugs added, had lower compliance with oral
hypoglycemic drugs (MPR < 80%), had no office visits based on diabetic guidelines,
had no glucose monitoring test, had no HbAlc test every six months, or had no
cholesterol check-up and dilated eye examination every year, had comorbidities (i.e.,
hypertension or hyperlipidemia).
Even though one major characteristic used to identify patient risk was age greater
than 65 years and previous studies and this research show that patients with older age
were more likely to consume higher healthcare costs and utilization and be high-risk
patients in the future, age might not be considered as one of the factors associated
with high-risk patient selection. To understand clearly, high-risk patient selection
criteria with age greater than 65 years was compared with the criteria without age
greater than 65 years (Table 4.11-4.12). The findings indicate that if patients older
than 65 years were selected to be high-risk patients, this might lead to obtain false
positive results since 59% of patients in the sample were older than 65 years and
some of these patients might not be high-risk but were identified to be high-risk. In
addition, based on high-risk patient selection criteria (i.e., age greater than 65 years
and having hospitalization or ER visit in the past), approximately 70% of patients
109
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
were identified as high-risk patients (Table A5), whereas only 15% of patients were
selected to be high-risk patients based on high-risk patient selection criteria (i.e.,
having hospitalization or ER visit in the past) (Table A6). The more the factors
associated with high-risk patients were included in the criteria, the more the high-risk
patients were selected. Without age factor, the high-risk patient selection criteria
presented above might be helpful information for MediCal providers and policy
makers to selectively intervene and prevent patients to be at high-risk in the future.
For MediCal policy makers, the information regarding high-risk patient selection
based on the factors and their impacts associated with an increased future healthcare
costs and hospitalizations/ER visits is summarized as follows:
Demographics Factors
• Patients with older age (4% higher risk for each year of each year of age beyond
the mean age 68 years)
• Male patients (8% higher risk compared to female patients)
• Non-white patients (5% higher risk compared to white patients)
• FFS Patients living in the counties with higher percentage of people under
managed care plan than FFS plan (0.5% higher risk compared to FFS patients living
in the counties with lower percentage of people under managed care plan than FFS
plan)
110
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Prior Healthcare Use
• Prior healthcare costs (not great impact)
• Prior hospitalizations/ER visits (2 times higher risk compared to patients without
prior hospitalizations/ER visits)
• Patients with comorbidities (i.e., hypertension, hyperlipidemia, systemic chronic
disease, cardiovascular disease, cancer/severe disease, or psychiatric disease) (30-
40% higher risk compared to patients without comorbidities)
• Patients with complications (i.e., retinopathy, nephropathy, or foot infection) (20%
higher risk compared to patients without complications)
• Patients having visits to nephrologists, neurologists, or infectious disease
specialists (1-2 times higher risk compared to patients without specialist visits)
Drug Treatment Factors
• Patients taking both oral hypoglycemic drugs and insulin (72% higher risk
compared to patients taking only oral hypoglycemic drugs)
• Patients having drugs added to the regimen (30% higher risk compared to patients
without any drug added)
• Patients having drugs changed to different classes (50% higher-risk compared to
patients without any drug changed)
1 1 1
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
• Patients compliant with oral hypoglycemic medication (only little impact on a
decrease in risk)
Follow-up Services Based on Diabetic Guidelines
• Patients having office visits based on diabetic guidelines (50% lower risk
compared to patients without office visits based on diabetic guidelines)
• Patients having glueose monitoring strip (50% lower risk eompared to patients
without glucose monitoring strip)
• Patients having HbAle test every 6 months (40% lower risk eompared to patients
without HbAlc test every 6 months
Healthcare Cost Savings to MediCal
Factors associated with patient risk also have a significant association with
healthcare cost savings to MediCal (Table AlO). For diabetes treatment factors,
patients taking both insulin and oral hypoglycemic drugs or patients having drug
dose increased had healthcare costs higher by $1,210 and $141, respectively. In
addition, patients having oral hypoglycemic or insulin, anti-hypertensive, or lipid
lowering drugs added also had healthcare costs higher by $264, $528, or $199,
respectively. Patients having drugs changed to different classes or to insulin had
112
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
healthcare costs higher by $1018. However, there are some interventions that
MediCal may provide to patients and help generate cost saving to MediCal. Patients
having one percent of medication compliance increased had healthcare costs lower
by $7 in next six-month period. Moreover, patients having office visits based on
diabetic guidelines or patients having glucose monitoring strip had healthcare costs
lower by $730 or $258 in next six-month period, respectively. In addition, patients
having lab tests [e.g., HbAlC test every six months ($121), cholesterol check up
every year ($472), or dilated eye check-up every year ($260)] could lower costs in
the future. MediCal policy makers may implement some disease management
programs or health policy on patients who have drug treatment problems and patients
without follow-up services based on diabetic guidelines in order to improve patient
outcomes and decrease healthcare costs in the future.
Significant MediCal Health Policv Options
Significant MediCal health policy options can be made for patients who have drug
treatment problems (i.e., having drugs added or changed and lower medication
compliance) and patients without follow-up services based on diabetic guidelines
(i.e., having no office visits based on diabetic guidelines, no glucose monitoring test,
no HbAlc test every six months, no cholesterol check-up and dilated eye and foot
examination every year). If patients who have drug treatment problems can be
identified and selectively intervened, they can be prevented from consuming higher
113
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
healthcare costs and utilization or to be high-risk patients in the future. This may
help improve patient outcomes and reduce an enormous cost to the MediCal.
Prevention strategies have been recommended to decrease the probability of an acute
exacerbation and prevent long-term complications. The strategies proposed to
MediCal Policy Division include diabetes updates for primary care physicians and
pharmacists and a diabetes pharmaceutical care intervention.
Recently, primary care physicians and pharmacists need frequent updates of disease-
specific knowledge such as new diabetes medication and disease self management
education. Healthcare providers such as pharmacists and primary care physicians
should be well trained and educated on updates related to diabetes in order to
effectively apply and intervene on patients with diabetes. The quality of care for
patients with diabetes was improved after an educational intervention targeting
primary care physicians were applied (Oosthuizen et al, 2002). In addition,
MediCal may utilize pharmacists in community pharmacies using a pharmaceutical
care intervention. Pharmacists in the community pharmacies can play an important
and potentially cost saving role in high-risk patient selection and prevention of an
increase in costs and utilization in the future. After high-risk patients are selected,
pharmacists can selectively intervene and improve patient behaviors and drug use.
Pharmacists can provide diabetes education on the significance of the HbAlc test,
self glucose monitoring, medication compliance, appropriate diet, exercise, and
complications. Moreover, pharmacists can assure that patients receive needed
114
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
examination such as HbAlc test every six months, cholesterol test, dilated eye and
foot check-up every year. In addition, the research findings indicate that patients
under FFS system in urban counties where percentage of people enrolling in
managed care plan was higher were more complicated and sicker than patients in
rural counties where patients were under the FFS system only. Therefore, MediCal
may focus on providing prevention strategies to patients with diabetes in urban
counties where percentage of people enrolling in managed care plan was higher since
these patients tend to have higher risk than patients in rural counties. Although
demographic factors, comorbidities, complications, and visits to specialists can
identify high-risk patients, these results do not give options to MediCal policy
makers. Patients with older age, patients with comorbidities such as hypertension,
patients with complication, or patients visiting to specialists are already at high-risk
and it is too late for MediCal to prevent them to be at high-risk or to consume higher
healthcare costs and utilization in the future. However, it is still worth to selectively
intervene on them in order to prolong their lives and improve quality of life of these
patients and it is still the useful information for MediCal since these factors can be
used as surrogate indicators for disease severity.
115
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
5.5 Limitations of the Research
Generally risk factors for patients developing diabetes or diabetes complications
include age older than 45 years, ethnicity, obesity, hypertension, family history of
diabetes in first-degree relatives, low high density lipoprotein (HDL) or high
triglyceride level, history of gestational diabetes mellitus, and high fasting blood
glucose (Harris et al, 1998). The ideal method to identify high-risk patients with
diabetes is to take into account their family history and clinical information.
Unfortunately the family history and clinical information is not available on
administrative claims data. These are certainly limitations to this study, as in any
other retrospective claims data analysis. Clinical information such as blood glucose
level and other laboratory values would have been very useful in identifying if
patients actually had been under control or had really been high-risk for an increase
in healthcare costs and utilization. Without these clinical measures, perfect high-risk
prediction may not be possible. However the findings may be still useful from
MediCal’s perspective since the purpose of this research is to determine whether
high-risk patients can be identified based on administrative claims data without
clinical chart data that is generally unavailable to administrators of health plans. In
addition, this research was performed to develop valid methodologies to identify
high-risk patients and investigate the factors associated with patient risk based on
available retrospective MediCal claims data. These factors are usually available
from MediCal claims data in order to be used by MediCal administrators to predict
116
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
which patients are high-risk. Although it is not a perfect prediction, it can be used as
a tool for MediCal administrators to foresee the trend that patients may be at risk for
an increase in healthcare costs and utilization based on the best available data.
5.6 Future Research
A development of valid methodologies for identifying high-risk MediCal patients
and an investigation of factors associated with an increase risk in costs and
hospitalization help MediCal to evaluate what healthcare services need to be
provided and to whom intervention should be targeted. In addition, the MediCal
Poliey Division is also interested in a demonstration pilot project to perform
pharmaceutical care services provided by pharmacists in community pharmacies to
intervene to improve patient behaviors and drug use in the high-risk population and
in an effort to reduce costs to the MediCal program to improve patient care. Thus,
the future research would be that these methodologies to identify high-risk patients
could be used by MediCal program in the future for implementing pharmacist care
and disease management services to a sample of high-risk patients. Also, total
healthcare cost, patient outcomes, utilization of healthcare services, and net cost
saving to the MediCal program may be evaluated as a results of providing
pharmaceutical care services to these patients. In addition, a pharmacist
reimbursement system to pay pharmacists for disease management services could be
developed. The knowledge gained from the development of valid methodologies for
117
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
identifying high-risk MediCal patients may be useful to other state Medicaid
programs or other health plan data. In addition, future research might also examine
the predictive validity by using external validation with an entirely independent data
set such as data from Medicaid program in other states or other health plan data.
This may provide the strongest test of predictive validity of the models.
5.7 Conclusions
Diabetes is an important disease affecting a relatively large number of patients and is
costly to the FFS MediCal program. Especially patients in the counties with a higher
percentage enrolling in managed care plans tend to be high-risk for an increase in
healthcare costs and utilization since they may be left out from the managed care
plan. Thus, MediCal program administrators have expressed an interest in
developing methods to intervene on certain high-risk FFS patients in order to
improve patient outcomes and reduce healthcare costs in the future.
This study identifies patient risk by using costs, the occurrence of hospitalization or
ER visit and time to hospitalization or ER visit and assesses the determinants of
patient risk by applying econometric methods on longitudinal data. Factors
associated with higher patient risk are demographic (e.g., age, race, gender), county
with a higher percentage of patients enrolled un managed care plans, previous high
healthcare cost and increased utilization, diabetes treatment (e.g., type of drugs,
118
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
having dose increased, having drug added, having drug changed, and medication
compliance), lack of follow-up services based on guidelines for diabetes, increased
healthcare use, more comorbidities, and complications in patients already having
diabetes. Demographic data indicate who are at risk for higher cost and use, but do
not give options to MediCal policy makers. As patients have drugs added/changed
or have no follow-up service based on diabetic guidelines, these may be used as
proxies to indicate that patients are worsening or patients lack proper health behavior
to control their disease. Significant policy options exist for those who are identified
by having drugs changed/added or have drug dose increased and those without
follow-up services based on diabetic guidelines. MediCal may encourage
healthcare practitioners to provide proper follow-up services based on guidelines for
diabetes to diabetic patients in order to prevent them to be at risk for an increase in
healthcare costs and utilization in the future. A better understanding of the
determinants of patient risk is needed to have informed discussions of MediCal
policies.
This study develops three valid predictive methodologies using healthcare costs, the
occurrence of hospitalization or ER event, and time to hospitalization or ER event
based on the FFS Medical claims data. However we may never be able to know
perfectly in advance who will consume a lot of healthcare costs, or who will have
hospitalization or ER event in the future, it is still very useful to be able to identify a
list of risk factors that put patients at “high-risk” for an increase in healthcare costs
119
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
or utilization and to selectively intervene on them. To conclude which methodology
is the most appropriate to identify patient risk, it will be depend upon how patient
risk has been identified and from whose perspective high-risk is. For example, in the
MediCal administrators’ perspective, high-risk patients are defined as patients who
will consume higher costs in the future. In this case, patient risk may be identified
by healthcare cost, so that predictive cost model is the most appropriate to use.
Considering these valid methodologies for identifjdng high-risk patients with
diabetes in the FFS MediCal populations, these findings are worthy of future
research for MediCal program and Medicaid programs in other states or other health
plan data.
120
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
REFERENCE
American Diabetes Assoeiation. (1996). Diabetes 1996 Vital Statistics (1996).
Rockville, MD: American Diabetes Association.
American Diabetes Association (1998). Economic consequences of diabetes
mellitus in the U.S. in 1997. Diabetes Care, 21(2), 296-309.
American Diabetes Association (1999). Screening for type 2 Diabetes. Diabetes
Care, 22 (Suppl), S20-S26.
Baker, A.M., Lafata, J.E., Ward, R.E., Whitehouse, F., and Divine, G. (2001). A
Web-based diabetes care management support system. Joint Commission Journal on
Quality Improvement. 27(4), 179-90.
Bhattacharyya, S.K. (1998). Predicting Hospitalization of patients with Diabetes
mellitus. Pharmacoeconomic, 13(5 Ft 1), 519-529.
Bhattacharyya, S.K., and Else, B.A. (1999). MediCal eosts of managed care in
patients with type 2 Diabetes mellitus. Clin Ther., 21(12), 2131-42.
Berry, R., Stone, C., Thompson, M., and et al. (1981) The incidence-based
economics costs of Diabetes mellitus. In evaluation of cost o f illness ascertainment
methodology, part II: applications o f methodology o f ascertain lifetime economic
costs o f illness in an incidence cohort. Final report on a research on a research
project under contract to the Department of Health and human Services, Hyattsville,
Maryland, 1981.
Brown, J.B., Pedula, K.L., and Baskt, A.W. (1999). The progressive cost of
complications in type 2 Diabetes mellitus. Arch Intern Med.,159, 1873-1880.
Diabetes: United States, 1982-1992 (1995). M.M. W.R., 43, 952-955.
Bruno, R. and Gilbert, B.P. (1998). In California, Medi-Cal managed eare is
superior to Medi-Cal fee-for-serviee. Manag Care Q. 6(4), 7-14.
Centers for Disease Control and Prevention (1997, August 7). Diabetes
hospitalizations and readmissions among children and young adults in Wisconsin
1991-1995. M.M.W.R..
Cox, D.R. (1972). Regression models and life tables (with discussion). JRoyal Stat
Soc, B (34), 187-220.
1 2 1
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Dagogo-Jack S. (2002). Preventing diabetes-related morbidity and mortality in the
primary care setting. J Nation Med Assoc. 94(7), 549-60.
Diggle, P.J., Liang, K-Y., Zeger, S.L. (1994) Analysis of Longitudinal Data. Oxford,
Oxford Science Publications.
Dudley, R.A., Harrell, F.E. Jr, Smith, L.R., Mark, D.B., Califf, R.M., Pryor, D.B.,
Glower, D., Lipscomb, J., and Hlatky, M. (1993). Comparison of analytic models for
estimating the effect of clinical factors on the cost of coronary artery bypass graft
surgery. J Clin Epidemiol, 46(3), 261-71.
Durcharme, P.M. and Kramer, M.S. (1993). Relapse following emergency treatment
for acute Diabetes: can it be predicted or prevented. J Clin Epidemic, 46(12), 1395-
1402.
Duan, N. (1983). Smearing estimate: a nonparametric retransformation method. J
Am Stat Assoc, 78, 605-610.
Epstein, A.M. and Cumella, E.J. (1988). Using predictors of MediCal utilization to
adjust rates. Health Care Financ Rev, 10,51-69.
Etzioni, R.D., Feuer, E.J., Sullivan, S.D., Lin, D., Hu, C., Ramsey, S.D. (1999). On
the use of survival analysis techniques to estimate MediCal care costs. J Health
Econ,lS, 365-380.
Etzioni, R., Urban, N., and Baker, M. (1996). Estimating the costs attributable to a
disease with application to ovarian cancer. J Clin Epidemiol, A9, 95-103.
Fenn, P., McGuire, A., Backhouse, M., and Jones, D. (1996). Modelling programme
costs in economics evaluation. J Health Econ,\5, 115-25.
Fenn, P., McGuire, A., Phillips, V., Backhouse, M.., and Jones, D. (1995). The
analysis of censored treatment cost data in economics evaluation. Med Care, 33,
851-63.
Franlin, G.M. (1990). Sensory neuropathy in non-insulin-dependent Diabetes
mellitus: the San Luis Vally Diabetes Study. Am J Epidemiol, 131, 633-43.
Guo, J.J., Gibson, J.T., Gropper, D.M., Oswald, S.L., and Barker, K.N. (1998).
Empiric investigation on direct costs-of-illness and healthcare utilization of Medicaid
patients with Diabetes mellitus. Am J Manag Care,A{\C), 1433-46.
122
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Harris, M.I., Flegal, K.M., Cowie, C.C., et al. (1998). Prevalence of diabetes,
impaired fasting glucose, and impaired glucose tolerance in U.S. adults: the third
national health and nutrition examination survey, 1988-1994. Diab Care, 21,518-
24.
Heckman, J.J. (1978). Statistical models for discrete panel data. Structural analysis
of discrete data with econometric applications (pp. 114-78). Cambridge: MIT Press.
Herman, W.H., Teutsch, S.M. (1985/ Kidney diseases associated with Diabetes.
In: Diabetes in America, Diabetes data compiled 1984 (NIH Pub. No. 85-1468).
Bethesda, MD: National Diabetes Data Group.
Hsiao, C. (1986). Analysis o f panel data. Cambridge University Press.
Ash, A.S. and Shwartz M. (1997). Evaluating the performance of risk-adjustment
methods: dichotomous outcomes. In L.I. lezzoni (2nd ed.). Risk adjustment for
measuring healthcare outcomes (pp. 427-469). Chicago, Illinois: Health
Administration Press.
Jamieson, D.J., Duerr, A., Klein, R.S., Paramsothy, P., Brown, W., Cu-Uvin, S.,
Rompalo, A., and Sobel, J. (2001). Longitudinal analysis of bacterial vaginosis:
findings from the HIV epidemiology research study. Obstetrics & Gynecology.
98(4), 656-63.
Jiwa, F. (1997). Diabetes in the 1990s—an overview. Stat Bull Metrop insur
Co,78(l), 2-8.
Kahn, J.A., Rosenthal, S.L., Succop, P.A., Ho, G.Y., and Burk, R.D. (2002).
Mediators fo the association between age of first sexual intercourse and subsequent
human papillomavirus infection. Pediatrics. 109(1), E5.
Klein, R., Klein, B.E.K., and Moss, S.E. (1992). In S.S. Feman (Ed.). The
epidemiology of ocular problems in Diabetes mellitus. Ocular problems in Diabetes
mellitus. Boston: Blackwell.
Krop, J.S., Powe, N.R., Weller, W.E., Shaffer, T.J., Saudek, C.D., and Anderson,
G.F. (1998). Patterns of expenditures and use of services among older adults with
Diabetes. Diabetes Care, 21(5), 747-752.
Krop, J.S., Saudek, C.D., Weller, W.E., Powe, N.R., Shaffer, T., and Anderson, G.F.
(1999). Predicting expenditures for Medicare beneficiaries with Diabetes: A
prospective cohort study from 1994 to 1996. Diabetes Care, 22(10), 1660-1666.
123
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Lin, D. Y., Feuer, E.J., Etizioni, R., and Wax, Y. (1997). Estimating medical costs
from incomplete follow-up data. Biometrics,SI, 113-28.
Lipscomb, J., Ancukiewicz, M., Parmigiani, G., Hasselblad, V., Samsa, G., and
Matchar, D.B. (1998) Predicting cost of illnesses. Med Decis Making, 18 (Suppl),
S39-S56.
Liu, T., Soong, S.J., Alvarez, R.D., Butterworth, C.E.Jr. (1995). A longitudinal
analysis of human papillomavirus 16 infection, nutritional status, and cervical
dysplasia progression. Cancer Epidemiology, Biomarkeres & Prevention. 4(40:
373-80.
LoSasso, A.T., and Freund, D.A. (2000) Longitudinal evaluation of the effect of
MediCal Managed care on Supplemental security income and aid to families with
dependent children enrollees in two California Counties. Med Care, 38(9), 937-947.
Mosterd, A., D’Agostino, R.B., Silbershatz, H., Sytkowski, P.A., Kannel, W.B.,
Grobbee, D.E., and Levy, D. (1999). Trends in the prevalence of hypertension,
antuhypertensive therapy, and left ventricular hypertrophy from 1950 to 1989. N
Eng J Med. 340 (16): 1221-7.
Mosure, D.J., Berman, S., Kleinbaum, D., and Halloran, M.E. (1996). Predictors of
Chlamydia trachomatis infection among female adolescents: a longitudinal analysis.
AmJEpidemio. 144(10): 997-1003.
Musey, V., Lee, J., Crawford, R., Klatka, M.A., McAdams, D., and Phillips, L.S.
(1995) Diabetes in urban Afiican-Americans. I. Cessation of insulin therapy is the
major precipitating causes of Diabetes ketoacidosis. Diabetes Care, 18, 483-9.
Norris, S.L., Nichols, P.J., Caspersen, C.J., Glasgow, R.E., Engelgau, M.M., et al.
(2002). Increasing diabetes self-management education in community setting. Am J
PrevM. 22(4S), 39-66.
Oosthuizen, H., Riedijk, R., Nonner, J., Rheeder, P., and Ker, J.A. (2002) An
educational intervention to improve the quality of care of diabetic patients. S Afr
MedJ. 92(6), 459-64.
Orchard, T.J. (1996). The impact of gender and general risk factors on the
occurrence of atherosclerotic vascular disease in non-onsulin-dependent diabetes
mellitus. Ann Med, 28(4), 323-33.
Part, T. (1993). A comparison of the generalized estimating equation approach with
the maximum likelihood approach for repeated measurements. Stat Med. 12(18),
1723-32.
124
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Ray N., Willis S., and Thamer M. (1993). Direct and indirect costs o f Diabetes in
the United States in 1992. Alexandria, VA: American Diabetes Association.
Reiber, G.E., King, H. (1991). Guidelines for the development of a national
programme for diabetes mellitus. Geneva; World Health Organization.
Seines, O.A., Galai, N., Bacellar, H., Miller, E.N., Becker, J.T., Wesch, J., and Van
Gorp, W., McArthur, J.C. (1995). Cognitive performance after progression to AIDS:
a longitudinal study from the Multicenter AIDS Cohort Study. Neurology. 45(2),
267-75.
Seines, O.A., Galai, N., McArthur, J.C., Cohn, S., Royal, W. Esposito, D., and
Vlahov, D. (1997). Neurology. 48(1); 223-30.
Shiboski, C.H., Hilton, J.F., Neuhaus, J.M., Canchola, A., and Greenspan, D. (1996).
Human immunodeficiency virus-related oral manifestations and gender. A
longitudinal analysis. The University of California, San Francisco Oral AIDS Center
Epidemiology Collaborative Group. Arch Intern Med. 156(19); 2249-54.
Sussman, K.E., Reiber, G., Albert, S.F. The diabetic foot problem - a failed system
for healthcare? Diabetes Res Clin Pract 1992; 17; 1-8.
Smith, D.H., Malone, D.C., Lawson, K.A., Okamoto, L.J., Battista, C., Saunders,
W.B. (1997). A national estimate of the economic costs of asthma. Am J Respir Grit
Care Med. 156(3 Pt 1), 787-93.
Verweij, P.J.M. and Van Houwelingen, H.C. (1993). Cross-validation in survival
analysis. Stat Med. 12,2305-2314
Vinicor, F. (1998). Diabetes mellitus and Diabetes; “Twin” challenges for public
health and managed care systems. Am JPrev Med, 14(3S), 87-92.
Wagner, E., Austin, B., and Von Korff, M. (1996). Organizing care for patients with
chronic illness. MilbankQ, 4,511-44.
Weinberger, M., Cowper, P.A., Kirkman, M.S., and Vinicor, F. (1990). Economic
impact of diabetes mellitus in the elderly. Clin Ceriatr Med, 6(4), 959-70.
WHO Study Group on Prevention of Diabetes Mellitus. (1994). Prevention o f
diabetes mellitus (WHO Technical Report Series no. 844). Geneva; World Wealth
Organization.
125
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Wolinsky, F.D., Wyrwich, K.W., Jung, S.C., and Gumey, J.G. (1999). The risk of
hospitalization for acute myocardial infarction among older adults. J Geront. Series
A, Biological Sciences & Medical Sciences, 54(5), M254-61.
Wu, A.W., Gray, S.M., and Brookmeyer, R. (1999). Application of random effects
models and other methods to the analysis of multidimensional quality of life data in
an AIDS clinical trial. Med Care, 37(3), 249-58.
Yanchick, J.K. (2000). Implementation of a drug therapy monitoring clinic in a
primary-care setting. Am J Health Syst Pharm. 57(Suppl4): S30-4.
Zeger, S.L., Liang, K.Y. (1986). Longitudinal data analysis for discrete and
continuous outcomes. Biometrics, A2, 121-130.
Zeger, S.L., Liang, K.Y., and Albert, P.S. (1988). Models for longitudinal data: a
generalized estimating equation approach. Biometrics. 44(4), 1049-60.
Zeger, S.L., and Liang, K.Y. (1992). An overview of methods for the analysis of
longitudinal data, 11, 1825-1839.
126
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
APPENDIX
Previous published literatures are summarized as follows and presented in Table Al.
Guo et al (1998) used the multiple regression and canonical correlation methods to
analyze the direct costs of diabetic patients and the healthcare service costs used by
each patient. The data of 7,931 diabetic patients were obtained from Alabama
Medicaid program from 1992 to 1995. The four largest components of the direct
costs for diabetic patients were hospitalization (29.9%), prescription drugs (28.2%),
outpatient care (21.3%), and physician visit (14.3%). The direct costs for diabetic
patients were significantly affected by the number of comorbidities, physician visits,
insulin-dependent diabetes mellitus, and complications.
Krop et al (1998) determined healthcare use and costs for patients with diabetes, as
well as the factors associated with higher costs. They presented a treatment
approach to caring for diabetic patients under managed care. Data of 1,221,615
patients were obtained from a nationwide 5% random of Medicare beneficiaries
during 1992, and 188,470 people of these had a diagnosis of diabetes. Costs were
determined by hospital inpatient care, hospital outpatient care, and physician visit.
Factors associated with higher costs were examined by using a linear regression
model. Total costs were the outcome variable and age, sex, race, complications,
comorbidities, physician visits, hospitalizations, and average length of stay were the
127
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
independent variables. The results showed that the costs of diabetic patients were
1.5 times higher than those of non-diabetic patients. Hospitalization costs (60%)
were the majority of the total costs, whereas outpatient and physician services were 7
and 33%, respectively. In the regression model, age, male sex, number of
complications but not race were positively related to the higher expenditures.
However the model had minimal predictive power (R^ = 0.0006). By adding
comorbidities into the model, it was positively related to the costs (R^ =0.20).
A year later, B C rop et al (1999) also conducted a prospective cohort study of
healthcare costs and utilization patterns among older diabetic patients and
determined factors associated with costs over a 3-year period. The data of 169,613
diabetic patients and 968,832 non-diabetic patients were obtained from a 5% random
national sample of fee-for-service claims between 1994 and 1996. A linear
regression model was used to examine the factors affecting high costs. The
dependent variables were the total costs, physician costs, and inpatient costs. The
independent variables were age, sex, race, Medicaid eligibility, diabetic
complications, comorbidities, and healthcare utilization. They found that per capita
costs of diabetic patients ($6,525) were 1.7 times higher than those of non-diabetic
patients ($3,760). Based on the linear regression analyses, age, sex, race, and
presence/absence of Medicaid eligibility were able to explain 7% of the variation in
total costs and 10.7% of the variation in physician costs.
Bhattacharyya (1998) proposed a Bayesian discrimination model or logistic
128
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
regression model to predict the hospitalization for diabetic patients with commonly
observed comorbidities. Both medical and pharmacy claims data of 6,841 diabetic
patients were extracted from the Hawaii Medical Service Association (HMSA)
Private Business Claims (PBS) files in 1995. The outcome of the model was having
one or more hospitalization events. Independent variables were demographics (e.g.,
age and gender), drugs (e.g., insulin or oral drugs), comorbidities (e.g., hypertension,
hyperlipidemia, coronary artery cerebrovascular disease, CHF, retinoparty, renal
disorders, neurologic disorders), healthcare services (e.g., Hb Aic test, inpatient
hospitalization, eye examination, and dialysis), and payment systems (e.g., capitation
and fee-for-service). The results of analyses showed that age, gender, types of drugs,
presence of hypertension, hyperlipidemia, CHF, multiple cardiovascular disease, eye
examination, and dialysis service were highly able to predict one or more
hospitalization events and a predictive power was almost 90%.
A year later, Bhattacharyya and Else (1999) analyzed the treatment costs on patients
with type 2 diabetes mellitus in a managed care setting. They used retrospective
integrated 1995 medical and pharmacy claims data of 5,175 patients and ordinary
least square regression model to identify the principal costs. Dependent variable was
the natural log of medical costs of diabetes and its comorbidity treatments.
Demographics (e.g., age and gender), drugs (e.g., insulin or oral drugs),
comorbidities (e.g., hypertension, hyperlipidemia, coronary artery cerebrovascular
disease, CHF, retinoparty, renal disorders, neurologic disorders), healthcare services
129
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
(e.g., Hb Ale test, inpatient hospitalization, eye examination, and dialysis), and
payment systems (e.g., capitation and fee-for-service) were independent variables.
The study showed that age was positively related to these costs. Comorbidities and
healthcare services had significant impact on the costs. However payment systems
had no significant effect on the medical costs of diabetes treatment.
Brown et al (1999) also estimated the progressive cost of cardiovascular and renal
complications in type 2 diabetes mellitus by using 9 years of clinical data on 11,768
type 2 diabetic patients of Kaiser Permanente Northwest Division (BCPNW ). They
divided the complications into three stages (i.e., stage 0=no treatment, stage l=pre-
event treatment, stage 2=post-event disease, and stage 3=end stage disease). The
costs (i.e., outpatient and inpatient costs within and without KPNW facilities) were
calculated by using 2 OLS models. The first model regressed total cost per person
per year on cardiovascular and renal complication levels, age, and sex. The second
model was similar to the first but added an interaction term of renal and
cardiovascular complications. They found that there were no significant differences
in gender and age for the prevalence of complications. Costs per person increased
from $2,033 to $3,120 after initiation of cardiovascular drug therapy and to $9,385
after major cardiovascular events. Diabetes treatment costs were increased by
$1,337 due to abnormal renal function, $3,939 due to advanced renal disease, and
$15,675 due to end-stage renal disease.
130
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table A l: Summary of Previous Published Studies in Diabetes
Study Data/Setting
(Study Period)
Objective Method Results/Comments
Guo et 7,931 Dm To calculate Multiple -Four largest
al patients from the direct regression and components of
(1998) Alabama costs of DM canonical the direct costs =
Medicaid and correlation hospitalization.
program from investigate the methods drugs, outpatient
1992 to 1995 factors that -Dep.Var.= care, and
(3 years) affect the direct costs physician visit.
costs -Indep.Var.=
age, sex,race,
comorbidity,
complication,
drug, and
physician
visits
-Drugs, physician
visit,
complication, and
comorbidity
significantly
affected the costs.
Krop et 188,470 DM To estimate Linear -The cost of DM
al patients from the direct regression was 1.5 times
(1998) Medicare in costs of DM model higher than that
1992 and -Dep.Var. = of non-DM.
(1 year) investigate the direct costs -hospitalization
factors that -Indep.Var.= cost the most
affect the age, sex,race. (60%).
costs comorbidity
(Charlson
Index), and
complication
-Age, sex (male),
complication, and
comorbidity were
positively related
to the costs.
-Predictive power
of the model was
quite low
(R^=0.0006)
131
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table A l; Summary of Previous Published Studies in Diabetes (Contd.)
Study Data/Setting
(Study
Period)
Objective Method Results/Comments
Krop et al 169,163 To estimate Linear -The cost of DM
(1999) DM and the total direct regression = $6,525 or 1.7
968,832 costs. model times higher than
non-DM physician -Dep.Var. = that of non-DM
patients of costs, and direct costs. ($3,760).
Medicare inpatient costs physician -Age, sex, race.
from 1994- of DM and costs, and complication, and
1996 investigate the inpatient costs Mediciad
(3 years) factors that -Indep.Var.= eligibility were
affect the age, sex,race. able to explain
costs comorbidity,
complication,
physician
visit,
Medicaid
eligibility, and
hospitalization
7% of the
variation in the
total costs and
10% in physician
costs.
Bhattacharyya Pharmacy To develop Bayesian -Age, gender.
SK claims data model and discrimination types of drugs.
(1998) of 6,841 predict the model HTN,
DM hospitalization (Logistic hyperlipidemia.
patients events regression CVD, dialysis
from associated model) service, and eye
HMSA and with DM -Dep.Var. = examination were
PBS in Having 1 or able to predict 1
1995 >1 or >1
(1 year) hospitalization
events
-Indep.Var.=
age,sex,
comorhidity,
complication,
drug,
healthcare
services, and
payment
systems
hospitalization
events (R^ =
90%)
132
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table A l: Summary of Previous Published Studies in Diabetes (Contd.)
Study Data/Setting
(Study
Period)
Objective Method Results/Commen
ts
Bhattacharyya Pharmacy To develop Log -Age,
SK claims data model and transformation comorbidities.
(1999) of5,175 predict the of linear and health care
type 2 DM costs regression services had
patients associated model significant
from with type 2 -Dep.Var. = impact on the
HMSA and DM Medi-Cal costs costs.
PBS in of DM and its -Payment
1995 comorbidity systems had no
(1 year) treatments
-Indep.Var.=
age, sex,
comorbidity,
complication,
drug, healthcare
services, and
payment
systems
significant effect
on the costs.
Brown et al 9 years of To analyze Two-part model -No significant
(1999) clinical data the (2 OLS models) difference in
on 11,768 progressive model: gender and age
type 2 DM cost of -Dep.Var.= total for the
patients of cardiovascu costs prevalence of
KPNW lar -Indep. Var. complications
(9 years) (CVD)and =CVD and renal -Cost per person
renal complication for CVD varied
complicatio levels, age, and from $2,033 to
ns in type 2 sex $9,385
DM 2 " * ^ model:
Similar to
model but added
an interaction
term of CVD
and renal
-For renal
complication,
the cost per
person ranged
from $1,337 to
$15,675.
133
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
There are three types of drugs usually dispensed to diabetic patients. These drugs
were used to create diabetes treatment factors in this research. Type of drugs and
their classifications are presented in Table A2-A4.
Table A2: Oral Hypoglycemic Drugs
Classiflcation Generic name (Trade name) Dose (MG)
-First generation
sulfonylureas
-Second
generation
sulfonylureas
Chlorpropamide (Diabenese, Insulase)
Tolbutamide (Orinase)
Tolazamide (Tolinase)
Acetohexamide (Dymelor)
Glyburide (Diabeta, Micronase,
Glynase)
Glyburide (Micronized formulation)
Glyburide (Glycron)
Glipizide (Glucotrol, Glucotrol XL)
Glimepiride (Amaryl)
100, 250
500
100, 250, 500
250, 500
1.25.2.5, 5
1.5, 3,4.5, 6
1.5, 3,4.5, 6
5, 10
1,2, 4 ,5 ,1 0
Biguanides Metformin (Glucophage)
Glyburide and Metformin
(Glucovance)
500, 850, 1000
1.25:250,
2.5:500, 5:500
a-Glucosidase
Inhibitors
Acarbose (Precose)
Miglitol (Glyset)
25, 50, 100
25, 50, 100
Miglitinides Repaglinide (Prandin)
Nateglinide (Starlix)
0.5, 1,2
60, 120
Thiazolidinediones
(Insulin
Sensitizers)
Troglitazone (Rezulin)
Pioglitazone (Actos)
Rosiglitazone (Avandia)
200, 300, 400
15,30, 45
2, 4 ,8
134
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table A3: Insulin
Classiflcation Generic name (Trade name) Dose
(UNT/ML)
Rapid-Acting -Human analog (Humulog, Novolog) 100 UNT/ML
Short-Acting
-Insulin injection
-Prompt insulin
zinc suspension
-Insulin
-Regular Iletin I Purified preparations:
Regular Iletin II (beef or pork) (Actrapid
(pork), Velosulin (pork))
-Human Insulin (Humulin R, Novolin R,
Velosulin (human))
-Semilente Insulin
-Semelente Iletin I Purified preparation:
Semi-tard (pork)
100 UNT/ML
100 UNT/ML
100 UNT/ML
100 UNT/ML
100 UNT/ML
Intermediate
Acting
-Isophane insulin
suspension
-Insulin zinc
suspension
-NPH Iletin I
-NPH Insulin Purified Preparations
(Insulintard NPH, Iletin II (beef or pork),
NPH purified pork)
-Human insulin (Humulin N, Novolin N)
-Lente insulin
-Lente Iletin I Purified preparations
(Lente Iletin II (beef or pork), Lente
purified pork)
-Human insulin (Humulin L, Novolin L)
100 UNT/ML
100 UNT/ML
100 UNT/ML
100 UNT/ML
100 UNT/ML
100 UNT/ML
Long-Acting
-Extended
Insulin zinc
suspension
-Ultralente Purified preparation
(Ultratard (beef))
-Human U (ultralente)
100 UNT/ML
100 UNT/ML
Insulin
Mixtures
- NPH 50% Regular 50% (Humulin
50/50, Novolin 50/50)
-NPH 70% Regular 30% (Humulin
70/30, Novolin 70/30, Mixtard 70/30)
100 UNT/ML
135
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table A4: Glucose Elevating Drugs
Generic name (Trade name) Dose (MG)
Glucagon injection (Glucagon, Hyperstat)
Glucagon emergency kit
1, 15 MG/ML,
300MG/20ML
IMG/VIL
Diazoxide (Proglycem oral suspension) 50 MG/ML
Table A5: Healthcare Costs and Utilizations of High-Risk Patients based on
Each High-Risk Patient Selection Criteria (Age Factor Included)
Criteria 1
- Age greater than 65 years
- Hospitalization/ER in the past
Period Next 6 months Age Female
(%)
MPR
(%)
Costs
($)
Hospitaliza
tion
(%)
ER
visit
(%)
1 High-risk patients
(N=4796)
71 69 86 3800 2.50 1.79
Others (N-2120) 54 64 86 3048 2.36 2.08
2 High-risk patients
(N=4779)
71 70 87 3850 1.40 1.36
Others (N=2137) 54 64 87 3285 1.82 1.64
3 High-risk patients
(N=4803)
72 69 87 4180 1.23 1.17
Others (N=2113) 54 64 86 3764 2.04 1.85
4 High-risk patients
(N=4892)
72 70 87 4706 1.39 1.41
Others (N=2024) 55 63 87 4129 1.78 2.27
5 High-risk patients
(N=4947)
73 70 88 5015 1.15 1.31
Others (N=1969) 55 63 88 4358 2.44 2.59
6 High-risk patients
(N=5016)
73 69 89 5454 1.57 1.40
Others (N=1900) 55 63 87 4566 2.26 3.16
7 High-risk patients
(N=5124)
73 69 89 5793 1.66 1.66
Others (N=1792) 55 63 88 4699 1.67 2.79
136
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table A5: Healthcare Costs and Utilizations of High-Risk Patients based on
Each High-Risk Patient Selection Criteria (Age Factor Included) (Contd.)
Criteria 2
- Age greater than 65 years
- Hospitalization/ER in the past
- Having both drugs (oral hypoglycemic drugs and insulin)
Period Next 6 months Age Female
(%)
MPR
(%)
Costs
($)
Hospitaliza
tion
(%)
ER
visit
(%)
1 High-risk patients
(N=5054)
70 69 86 3822 2.63 1.92
Others (N=1862) 54 64 86 2884 1.99 1.77
2 High-risk patients
(N=5082)
70 69 87 3898 1.48 1.42
Others (N=1834) 54 63 87 3059 1.69 1.53
3 High-risk patients
(N=5145)
71 69 87 4242 1.24 1.22
Others (N=1771) 54 64 87 3502 2.15 1.81
4 High-risk patients
(N=5272)
71 69 87 4720 1.48 1.42
Others (N=1644) 55 63 88 3952 1.58 2.43
5 High-risk patients
(N=5355)
71 69 88 5047 1.31 1.40
Others (N=1561) 54 62 88 4078 2.24 2.63
6 High-risk patients
(N=5429)
72 69 88 5490 1.69 1.62
Others (N=1487) 55 63 87 4188 2.02 2.82
7 High-risk patients
(N=5551)
72 69 89 6064 1.64 1.82
Others (N=1365) 55 62 88 4678 1.76 2.49
137
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table A5: Healthcare Costs and Utilizations of High-Risk Patients based on
Each High-Risk Patient Selection Criteria (Age Factor Included) (Contd.)
Criteria 3
Age greater than 65 years
Hospitalization/ER in the past
Having both drugs (i.e., oral hypoglycemic drugs and insulin)
No office visit in the previous six-month period
Period Next 6 months Age Female
(%)
MPR
(%)
Costs
($)
Hospitaliza
tion
(%)
ER
visit
(%)
1 High-risk patients
(N=5567)
68 68 86 3713 2.50 1.90
Others (N=1349) 54 66 87 2974 2.30 1.78
2 High-risk patients
(N=5628)
69 68 87 3792 1.46 1.40
Others (N=1288) 55 66 87 3166 1.86 1.63
3 High-risk patients
(N=5868)
68 68 87 4168 1.18 1.18
Others (N=1048) 55 68 87 3408 3.15 2.48
4 High-risk patients
(N=6069)
69 67 87 4671 1.38 1.43
Others (N=847) 55 68 88 3580 2.36 3.31
5 High-risk patients
(N=6074)
69 69 88 5007 1.27 1.40
Others (N=842) 55 68 88 3534 3.33 3.68
6 High-risk patients
(N=6142)
70 67 88 5428 1.64 1.58
Others (N=774) 55 72 87 3481 2.71 4.26
7 High-risk patients
(N=6213)
70 68 89 5673 1.51 1.79
Others (N=703) 55 70 88 4061 2.99 3.41
138
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table A5: Healthcare Costs and Utilizations of High-Risk Patients based on
Each High-Risk Patient Selection Criteria (Age Factor Included) (Contd.)
Criteria 4
- Age greater than 65 years
- Hospitalization/ER in the past
- Having both drugs (i.e., oral hypoglycemic drugs and insulin)
- No office visit in the previous six-month period
- No glucose monitoring test in the previous six-month period
Period Next 6 months Age Female
(%)
MPR
(%)
Costs
($)
Hospitaliza
tion
(%)
ER
visit
(%)
I High-risk patients
(N=6896)
66 68 86 3568 2.47 1.89
Others (N=20) 52 50 88 2974 0.00 0.00
2 High-risk patients
(N=6893)
66 68 87 3674 1.54 1.45
Others (N=23) 51 43 88 3000 0.00 0.00
3 High-risk patients
(N=6888)
67 68 87 4045 1.48 1.38
Others (N=28) 55 50 83 3508 0.00 0.00
4 High-risk patients
(N=6854)
67 68 87 4515 1.52 1.66
Others (N=62) 54 63 88 3500 0.00 1.61
5 High-risk patients
(N=6856)
68 68 88 4809 1.53 1.69
Others (N=60) 53 55 89 3200 0.00 0.00
6 High-risk patients
(N=6851)
68 68 88 5212 1.78 1.90
Others (N=65) 53 62 84 3480 0.00 0.00
7 High-risk patients
(N=6869)
69 68 89 5518 1.67 1.95
Others (N=47) 55 53 86 4326 0.00 2.13
139
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table A5: Healthcare Costs and Utilizations of High-Risk Patients based on
Each High-Risk Patient Selection Criteria (Age Factor Included) (Contd.)
Criteria 5
- Age greater than 65 years
- Hospitalization/ER in the past
- Having both drugs (i.e., oral hypoglycemic drugs and insulin)
- No office visit in the previous six-month period
- No glucose monitoring test in the previous six-month period
- No HbAlc test in the previous six-month period
Period Next 6 months Age Female
(%)
MPR
(%)
Costs
($)
Hospitaliza
tion
(%)
ER
visit
(%)
1 High-risk patients
(N=5494)
65 68 85 3780 2.77 2.06
Others (N=1422) 67 67 92 2757 1.27 1.20
2 High-risk patients
(N=5557)
66 68 86 3843 1.66 1.62
Others (N=1359) 67 67 93 2991 1.03 0.74
3 High-risk patients
(N=6129)
67 68 86 4176 1.44 1.31
Others (N=787) 63 69 92 3091 1.78 1.91
4 High-risk patients
(N=6436)
67 68 87 4624 1.54 1.65
Others (N=480) 62 66 93 3386 1.04 1.88
5 High-risk patients
(N=6385)
68 68 87 4936 1.50 1.63
Others (N=531) 62 71 94 3533 1.69 2.26
6 High-risk patients
(N=6510)
67 68 88 5280 1.74 1.81
Others (N=406) 60 72 92 4085 2.22 2.96
7 High-risk patients
(N=6500)
69 68 88 5598 1.63 2.00
Others (N=416) 62 68 93 4139 2.16 1.20
140
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table A5: Healthcare Costs and Utilizations of High-Risk Patients based on
Each High-Risk Patient Selection Criteria (Age Factor Included) (Contd.)
Criteria 6
Age greater than 65 years
Hospitalization/ER in the past
Having both drugs (i.e., oral hypoglycemic drugs and insulin)
No office visit in the previous six-month period
No glucose monitoring test in the previous six-month period
No HhAlc test in the previous six-month period
Having oral hypoglycemic, insulin, anti-hypertensive, lipid lowering drugs added
Period Next 6 months Age Female
(%)
MPR
(%)
Costs
($)
Hospitaliza
tion
(%)
ER
visit
(%)
1 High-risk patients
(N=6915)
66 68 86 3570 2.46 1.88
Others (N=l) 42 100 70 1840 0.00 0.00
2 High-risk patients
(N=6915)
66 68 86 3570 2.46 1.88
Others (N=l) 42 100 70 1840 0.00 0.00
3 High-risk patients
(N=6915)
67 68 87 4053 1.48 1.37
Others (N=l) 64 0 60 1432 0.00 0.00
4 High-risk patients
(N=6914)
67 68 87 4538 1.50 1.66
Others (N=2) 36 100 62 2706 0.00 0.00
5 High-risk patients
(N=6915)
68 68 87 4828 1.52 1.68
Others (N=l) 63 0 100 1971 0.00 0.00
6 High-risk patients
(N=6914)
68 61 88 5210 1.76 1.74
Others (N=2) 55 70 97 2169 0.00 2.05
7 High-risk patients
(N=6914)
68 61 88 5210 1.76 1.74
Others (N=2) 55 70 97 2169 0.00 2.05
141
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table A5: Healthcare Costs and Utilizations of High-Risk Patients based on
Each High-Risk Patient Selection Criteria (Age Factor Included) (Contd.)
Criteria 7
- Age greater than 65 years
- Hospitalization/ER in the past
- Having both drugs (i.e., oral hypoglycemic drugs and insulin)
- No office visit in the previous six-month period
- No glucose monitoring test in the previous six-month period
- No HbAlc test in the previous six-month period
- Having oral hypoglycemic, insulin, anti-hypertensive, lipid lowering drugs added
- Having Comorbidities such as hypertension or hyperlipidemia
Period Next 6 months Age Female
(%)
MPR
(%)
Costs
($)
Hospitaliza
tion
(%)
ER
visit
(%)
1 High-risk patients
(N=6914)
67 68 88 3510 1.66 1.94
Others (N=2) 60 50 98 2285 0.00 50.00
2 High-risk patients
(N=6915)
65 68 86 3570 2.46 1.88
Others (N=l) 42 too 70 1830 0.00 0.00
3 High-risk patients
(N=6915)
65 68 86 3570 2.46 1.88
Others (N=I) 42 too 70 1830 0.00 0.00
4 High-risk patients
(N=6915)
67 68 87 4053 1.48 1.37
Others (N=l) 64 0 60 1432 0.00 0.00
5 High-risk patients
(N=6915)
67 68 87 4053 1.48 1.37
Others (N=l) 64 0 60 1432 0.00 0.00
6 High-risk patients
(N=6915)
67 68 87 4053 1.48 1.37
Others (N=l) 64 0 60 1432 0.00 0.00
7 High-risk patients
(N=6915)
67 68 87 4053 1.48 1.37
Others (N=l) 64 0 60 1432 0.00 0.00
142
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table A5: Healthcare Costs and Utilizations of High-Risk Patients based on
Each High-Risk Patient Selection Criteria (Age Factor Included) (Contd.)
Criteria 8
- Age greater than 65 years
- Hospitalization/ER in the past
- Having both drugs (i.e., oral hypoglycemic drugs and insulin)
- No office visit in the previous six-month period
- No glucose monitoring test in the previous six-month period
- No HbAlc test in the previous six-month period
- Having oral hypoglycemic, insulin, anti-hypertensive, lipid lowering drugs added
- Having comorbidities such as hypertension or hyperlipidemia
- Having complications such as retinopathy, nephropathy, or infection
Period Next 6 months Age Female
(%)
MPR
(%)
Costs
($)
Hospitaliza
tion
(%)
ER
visit
(%)
1 High-risk patients
(N=69I5)
65 68 86 3570 2.46 1.88
Others (N=l) 42 too 70 1830 0.00 0.00
2 High-risk patients
(N=6915)
65 68 86 3570 2.46 1.88
Others (N=l) 42 100 70 1830 0.00 0.00
3 High-risk patients
(N=6915)
67 67 87 4053 1.48 1.37
Others (N=l) 64 100 60 1432 0.00 0.00
4 High-risk patients
(N=6915)
67 67 87 4053 1.48 1.37
Others (N=l) 64 100 60 1432 0.00 0.00
5 High-risk patients
(N=6915)
67 67 87 4053 1.48 1.37
Others (N=l) 64 100 60 1432 0.00 0.00
6 High-risk patients
(N=6915)
67 67 87 4053 1.48 1.37
Others (N=l) 64 100 60 1432 0.00 0.00
7 High-risk patients
(N=6915)
67 67 87 4053 1.48 1.37
Others (N=l) 64 100 60 1432 0.00 0.00
143
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table A5: Healthcare Costs and Utilizations of High-Risk Patients based on
Each High-Risk Patient Selection Criteria (Age Factor Included) (Contd.)
Criteria 9
- Age greater than 65 years
- Hospitalization/ER in the past
- Having both drugs (i.e., oral hypoglycemic drugs and insulin)
- No office visit in the previous six-month period
- No glucose monitoring test in the previous six-month period
- No HbAlc test in the previous six-month period
- Having oral hypoglycemic, insulin, anti-hypertensive, lipid lowering drugs added
- Having comorbidities such as hypertension or hyperlipidemia
- Having complications such as retinopathy, nephropathy, or infection
- Lower compliance (MPR < 80%)
Period Next 6 months Age Female
(%)
MPR
(%)
Costs
($)
Hospitaliza
tion
(%)
ER
visit
(%)
1 High-risk patients
(N=6915)
65 68 86 3570 2.46 1.88
Others (N=l) 42 too 70 1830 0.00 0.00
2 High-risk patients
(N=6915)
65 68 86 3570 2.46 1.88
Others (N=l) 42 too 70 1830 0.00 0.00
3 High-risk patients
(N=6915)
65 68 86 3570 2.46 1.88
Others (N=l) 42 too 70 1830 0.00 0.00
4 High-risk patients
(N=6915)
65 68 86 3570 2.46 1.88
Others (N=l) 42 too 70 1830 0.00 0.00
5 High-risk patients
(N=6915)
65 68 86 3570 2.46 1.88
Others (N=l) 42 too 70 1830 0.00 0.00
6 High-risk patients
(N=6915)
65 68 86 3570 2.46 1.88
Others (N=l) 42 100 70 1830 0.00 0.00
7 High-risk patients
(N=6915)
65 68 86 3570 2.46 1.88
Others (N=l) 42 100 70 1830 0.00 0.00
144
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table A6: Healthcare Costs and Utilizations of High-Risk Patients based on
Each High-Risk Patient Selection Criteria (Age Factor Not Included)
Criteria 1
- Hospitalization/ER in the past
Period Next 6 months Age Female
(%)
MPR
(%)
Costs
($)
Hospitaliza
tion
(%)
ER
visit
(%)
1 High-risk patients
(N=1362)
62 70 85 5969 5.73 5.14
Others (N=5554) 66 67 87 2980 1.66 1.08
2 High-risk patients
(N=1025)
61 72 86 5754 4.78 4.78
Others (N=5891) 66 67 87 3314 0.97 0.87
3 High-risk patients
(N=818)
60 69 86 6793 6.36 6.11
Others (N=6098) 68 68 87 3685 0.82 0.74
4 High-risk patients
(N=746)
58 70 86 7413 7.10 7.24
Others (N=6170) 68 68 88 4190 0.83 0.99
5 High-risk patients
(N=767)
60 73 87 7891 6.26 7.69
Others (N=6149) 69 67 88 4446 0.93 0.93
6 High-risk patients
(N=725)
59 72 88 8456 9.66 8.97
Others (N=6191) 69 67 88 4830 0.84 1.05
7 High-risk patients
(N=814)
61 72 88 9997 9.95 9.09
Others (N=6102) 70 67 89 4911 0.56 1.00
145
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table A6: Healthcare Costs and Utilizations of High-Risk Patients based on
Each High-Risk Patient Selection Criteria (Age Factor Not Included) (Contd.)
Criteria 2
- Hospitalization/ER in the past
- Having both drugs (oral hypoglycemic drugs and insulin)
Period Next 6 months Age Female
(%)
MPR
(%)
Costs
($)
Hospitaliza
tion
(%)
ER
visit
(%)
1 High-risk patients
(N=1878)
63 70 85 5490 5.01 4.42
Others (N=5038) 67 67 87 2853 1.51 0.93
2 High-risk patients
(N=1696)
63 72 86 5301 3.54 3.42
Others (N=5220) 67 67 87 3147 0.88 0.80
3 High-risk patients
(N=1601)
62 70 85 6109 3.62 3.56
Others (N=5315) 68 67 87 3433 0.83 0.71
4 High-risk patients
(N=1637)
62 69 86 6537 4.03 4.03
Others (N=5279) 69 67 88 3918 0.72 0.93
5 High-risk patients
(N=1715)
63 71 86 7089 3.67 4.14
Others (N=5201) 69 67 88 4082 0.81 0.87
6 High-risk patients
(N=1759)
64 71 87 7287 4.72 4.72
Others (N=5157) 70 67 89 4502 0.76 0.91
7 High-risk patients
(N=1947)
65 71 88 8451 4.52 4.93
Others (N=4969) 70 67 89 4357 0.54 0.78
146
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table A6: Healthcare Costs and Utilizations of High-Risk Patients based on
Each High-Risk Patient Selection Criteria (Age Factor Not Included) (Contd.)
Criteria 3
Hospitalization/ER in the past
Having both drugs (i.e., oral hypoglycemic drugs and insulin)
No office visit in the previous six-month period
Period Next 6 months Age Female
(%)
MPR
(%)
Costs
($)
Hospitaliza
tion
(%)
ER
visit
(%)
1 High-risk patients
(N=3573)
64 68 86 4179 3.14 2.84
Others (N=3343) 67 67 87 3152 1.82 0.98
2 High-risk patients
(N=3454)
65 68 87 4055 1.04 0.90
Others (N=3462) 67 67 87 3297 2.03 2.00
3 High-risk patients
(N=4852)
67 67 87 4244 1.36 1.36
Others (N=2064) 65 69 88 3603 1.74 1.41
4 High-risk patients
(N=5573)
68 68 87 4723 1.44 1.51
Others (N=1343) 62 68 88 3769 1.79 2.31
5 High-risk patients
(N=5520)
69 68 87 5047 1.30 1.54
Others (N=1396) 63 69 88 3963 2.36 2.22
6 High-risk patients
(N=5757)
69 67 88 5463 1.72 1.67
Others (N = ll59) 62 70 88 3953 1.98 2.93
7 High-risk patients
(N=5750)
70 68 89 5750 1.62 1.93
Others (N=1166) 63 67 88 4325 1.89 2.06
147
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table A6: Healthcare Costs and Utilizations of High-Risk Patients based on
Each High-Risk Patient Selection Criteria (Age Factor Not Included) (Contd.)
Criteria 4
- Hospitalization/ER in the past
- Having both drugs (i.e., oral hypoglycemic drugs and insulin)
- No office visit in the previous six-month period
- No glucose monitoring test in the previous six-month period
Period Next 6 months Age Female
(%)
MPR
(%)
Costs
($)
Hospitaliza
tion
(%)
ER
visit
(%)
1 High-risk patients
(N=6854)
66 68 86 3566 2.48 1.90
Others (N=62) 67 66 87 3999 0.00 0.00
2 High-risk patients
(N=6846)
66 68 87 3674 1.55 1.45
Others (N=66) 66 60 88 3867 0.00 1.43
3 High-risk patients
(N=6764)
66 68 87 4041 1.51 1.40
Others (N=152) 70 66 87 4562 0.00 0.00
4 High-risk patients
(N=6583)
67 68 87 4496 1.58 1.72
Others (N=333) 71 69 89 3366 0.00 0.60
5 High-risk patients
(N=6588)
67 68 88 4787 1.59 1.73
Others (N=328) 71 70 89 3657 0.00 0.61
6 High-risk patients
(N=6534)
68 68 88 5193 1.85 1.97
Others (N=382) 72 71 89 3514 0.26 0.26
7 High-risk patients
(N=6567)
68 68 88 5538 1.75 2.04
Others (N=349) 73 68 90 4979 0.00 0.29
148
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table A6: Healthcare Costs and Utilizations of High-Risk Patients based on
Each High-Risk Patient Selection Criteria (Age Factor Not Included) (Contd.)
Criteria 5
- Hospitalization/ER in the past
- Having both drugs (i.e., oral hypoglycemic drugs and insulin)
- No office visit in the previous six-month period
- No glucose monitoring test in the previous six-month period
- No HbAlc test in the previous six-month period
Period Next 6 months Age Female
(%)
MPR
(%)
Costs
($)
Hospitaliza
tion
(%)
ER
visit
(%)
1 High-risk patients
(N=6915)
66 68 86 3570 2.46 1.88
Others (N=l) 42 too 70 1830 0.00 0.00
2 High-risk patients
(N=6915)
66 68 86 3570 2.46 1.88
Others (N=l) 42 too 70 1830 0.00 0.00
3 High-risk patients
(N=6915)
67 67 88 4053 1.48 1.37
Others (N=l) 64 0 60 1432 0.00 0.00
4 High-risk patients
(N=6914)
67 68 87 4538 1.50 1.66
Others (N=2) 35 100 62 2706 0.00 0.00
5 High-risk patients
(N=6912)
68 68 87 1822 1.52 1.68
Others (N=4) 68 50 95 1454 0.00 0.00
6 High-risk patients
(N=6914)
68 68 88 1210 1.76 1.88
Others (N=2) 55 50 97 1168 0.00 0.00
7 High-risk patients
(N=6912)
69 68 88 1511 1.66 1.94
Others (N=4) 62 75 89 2620 0.00 25.00
149
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table A6: Healthcare Costs and Utilizations of High-Risk Patients based on
Each High-Risk Patient Selection Criteria (Age Factor Not Included) (Contd.)
Criteria 6
Hospitalization/ER in the past
Having both drugs (i.e., oral hypoglycemic drugs and insulin)
No office visit in the previous six-month period
No glucose monitoring test in the previous six-month period
No HbAlc test in the previous six-month period
Having oral hypoglycemic, insulin, anti-hypertensive, lipid lowering drugs added
Period Next 6 months Age Female
(%)
MPR
(%)
Costs
($)
Hospitaliza
tion
(%)
ER
visit
(%)
1 High-risk patients
(N=6915)
66 68 86 3570 2.46 1.88
Others (N=l) 42 too 70 1840 0.00 0.00
2 High-risk patients
(N=6915)
66 68 86 3570 2.46 1.88
Others (N=l) 42 too 70 1840 0.00 0.00
3 High-risk patients
(N=6915)
65 68 87 4053 1.48 1.37
Others (N=l) 64 0 60 1432 0.00 0.00
4 High-risk patients
(N=6915)
65 68 87 4053 1.48 1.37
Others (N=l) 64 0 60 1432 0.00 0.00
5 High-risk patients
(N=6914)
68 68 88 4824 1.52 1.68
Others (N=2) 70 too 89 1736 0.00 0.00
6 High-risk patients
(N=6915)
65 68 87 4053 1.48 1.37
Others (N=l) 64 0 60 1432 0.00 0.00
7 High-risk patients
(N=6913)
68 68 88 5511 1.66 1.94
Others (N=3) 63 67 86 1967 0.00 33.33
150
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table A6: Healthcare Costs and Utilizations of High-Risk Patients based on
Each High-Risk Patient Selection Criteria (Age Factor Not Included) (Contd.)
Criteria 7
Hospitalization/ER in the past
Having both drugs (i.e., oral hypoglycemic drugs and insulin)
No office visit in the previous six-month period
No glucose monitoring test in the previous six-month period
No HbAlc test in the previous six-month period
Having oral hypoglycemic, insulin, anti-hypertensive, lipid lowering drugs added
Having Comorbidities such as hypertension or hyperlipidemia
Period Next 6 months Age Female
(%)
MPR
(%)
Costs
($)
Hospitaliza
tion
(%)
ER
visit
(%)
1 High-risk patients
(N=6915)
66 68 87 3570 2.46 1.88
Others (N=l) 42 too 70 1840 0.00 0.00
2 High-risk patients
(N=6915)
66 68 87 3570 2.46 1.88
Others (N=l) 42 100 70 1840 0.00 0.00
3 High-risk patients
(N=6915)
67 68 87 4053 1.48 1.37
Others (N=l) 64 0 60 1432 0.00 0.00
4 High-risk patients
(N=6915)
67 68 87 4053 1.48 1.37
Others (N=l) 64 0 60 1432 0.00 0.00
5 High-risk patients
(N=6915)
67 68 87 4053 1.48 1.37
Others (N=l) 64 0 60 1432 0.00 0.00
6 High-risk patients
(N=6915)
67 68 87 4053 1.48 1.37
Others (N=l) 64 0 60 1432 0.00 0.00
7 High-risk patients
(N=6915)
66 68 86 3570 2.46 1.88
Others (N=l) 42 100 70 1840 0.00 0.00
151
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table A6: Healthcare Costs and Utilizations of High-Risk Patients based on
Each High-Risk Patient Selection Criteria (Age Factor Not Included) (Contd.)
Criteria 8
- Hospitalization/ER in the past
- Having both drugs (i.e., oral hypoglycemic drugs and insulin)
- No office visit in the previous six-month period
- No glucose monitoring test in the previous six-month period
- No HbAlc test in the previous six-month period
- Having oral hypoglycemic, insulin, anti-hypertensive, lipid lowering drugs added
- Having comorbidities such as hypertension or hyperlipidemia
- Having complications such as retinopathy, nephropathy, or infection
Period Next 6 months Age Female
(%)
MPR
(%)
Costs
($)
Hospitaliza
tion
(%)
ER
visit
(%)
1 High-risk patients
(N=69I5)
67 68 87 4053 1.48 1.37
Others (N=l) 64 0 60 1432 0.00 0.00
2 High-risk patients
(N=6915)
67 68 87 4053 1.48 1.37
Others (N=l) 64 0 60 1432 0.00 0.00
3 High-risk patients
(N=6915)
67 68 87 4053 1.48 1.37
Others (N=l) 64 0 60 1432 0.00 0.00
4 High-risk patients
(N=6915)
67 68 87 4053 1.48 1.37
Others (N=l) 64 0 60 1432 0.00 0.00
5 High-risk patients
(N=6915)
67 68 87 4053 1.48 1.37
Others (N=l) 64 0 60 1432 0.00 0.00
6 High-risk patients
(N=6915)
67 68 87 4053 1.48 1.37
Others (N=l) 64 0 60 1432 0.00 0.00
7 High-risk patients
(N=6915)
67 68 87 4053 1.48 1.37
Others (N=l) 64 0 60 1432 0.00 0.00
152
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table A6: Healthcare Costs and Utilizations of High-Risk Patients based on
Each High-Risk Patient Selection Criteria (Age Factor Not Included) (Contd.)
Criteria 9
- Hospitalization/ER in the past
- Having both drugs (i.e., oral hypoglycemic drugs and insulin)
- No office visit in the previous six-month period
- No glucose monitoring test in the previous six-month period
- No HbAlc test in the previous six-month period
- Having oral hypoglycemic, insulin, anti-hypertensive, lipid lowering drugs added
- Having comorbidities such as hypertension or hyperlipidemia
- Having complications such as retinopathy, nephropathy, or infection
- Lower compliance (MPR < 80%)
Period Next 6 months Age Female
(%)
MPR
(%)
Costs
($)
Hospitaliza
tion
(%)
ER
visit
(%)
1 High-risk patients
(N=69I1)
66 68 87 3571 2.46 1.88
Others (N=5) 57 40 85 1838 0.00 0.00
2 High-risk patients
(N=6904)
66 68 87 3676 1.54 1.45
Others (N=12) 56 50 94 1660 0.00 0.00
3 High-risk patients
(N=6909)
67 68 87 4055 1.48 1.38
Others (N=7) 57 57 95 1778 0.00 0.00
4 High-risk patients
(N=6911)
67 68 87 4539 1.50 1.66
Others (N=5) 50 too 88 2322 0.00 0.00
5 High-risk patients
(N=6913)
68 68 88 4830 1.52 1.68
Others (N=3) 69 67 98 1470 0.00 0.00
6 High-risk patients
(N=6911)
68 68 88 5213 1.77 1.88
Others (N=5) 58 80 97 1875 0.00 0.00
7 High-risk patients
(N=6914)
69 68 88 5511 1.66 1.95
Others (N=2) 58 50 97 1221 0.00 0.00
153
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table A7: Comparison Between High-Risk and Low-Risk Groups Identified by
Cost Criteria
Independent Variables (N=6,916) High-Risk
Group
(N=263)
Low-Risk
Group
(N=6,647)
N (%) N (%)
Mean total healthcare costs ($) (Range) 20,936***
(12,665-96,515)
3,701
(54-29,436)
Having hospitalization or ER visits 220 (83)*** 3059 (46)
Diabetes treatment factors:
1 . Type of drugs
-Oral hypoglycemic drugs
-Insulin
-Anti-hypertensive drugs
-lipid-lowering drugs
263 (100)
202 (77)***
257 (97)***
0.9 (52)
6647 (100)
2301(35)
5817 (87)
3538 (53)
2. Increasing dose of oral hypoglycemic drugs
93 (35) 3132(47)***
3. Adding Drugs
-Adding oral hypoglycemic drugs or insulin
-Adding anti-hypertensive drugs
-Adding lipid-lowering drugs
231(87)**
254 (95)***
134 (50)
5420 (82)
5761 (87)
3478 (52)
4. Changing drug to different classes or to insulin
57(21)** 1008 (15)
5. Oral hypoglycemic medication compliance
-Mean medication possession ratio (%MPR)
(Range)
88 (1-100) 88 (1-100)
Follow-up service factors based on diabetic
guidelines:
1 . Having office visits (every 3 months for
patients taking insulin and every 6 months for
patients taking only oral hypoglycemic drugs)
2. Having glucose monitoring strip
240 (91)**
140 (53)***
5687(86)
2959 (45)
3. Having lab tests by healthcare providers
-HbAlc every 6 months
-Cholesterol check-up every year
70 (27)
31(12)***
1600 (24)
343 (5)
Significance Levels: ***p<0.005, **p<0.05, *p<0.1
154
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table A7: Comparison Between High-Risk and Low-Risk Groups Identified by
Cost Criteria (Contd.)
Independent Variables (N=6,916) High-Risk
Group
(N=263)
N (%)
Low-Risk
Group
(N=6,647)
N (%)
Healthcare use factors:
-Gynecologist visit
-Cardiologist visit
-Nephrologist visit
-Neurologist visit
-Infectious disease specialist visit
33 (12)
216(81)***
68(26)***
90 (34)***
19 (7)***
764(15)
2900 (44)
227 (3)
817(12)
49 (0.7)
Comorbidity factors:
-Hypertension
-Hyperlipidemia
-Systemic chronic disease
-Cardiovascular disease
-Cancer/severe disease
-Psychiatric disease
254(95)***
IC O (38)
199 (75)***
245 (92)***
170(64)***
68 (26)***
5429 (82)
2537(38)
2515 (38)
4038 (61)
1566 (24)
825 (12)
Complication factors:
1 . Target organ disease
-Retinopathy
-Nephropathy
-Neuropathy
2. Infection
-Foot
-Skin
-Vagina
106 (40)***
29(11)***
66 (25)***
108(41)***
1 (0.4)
0(0)
1640 (25)
163 (2)
915 (14)
1437 (22)
37 (0.6)
34 (0.5)
Significance Levels: ***p<0.005, **p<0.05, *p<0.1
155
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table A8: Comparison between High-Risk and Low-Risk Groups Identified by
Probability of Hospitalization or ER visits Criteria
Independent Variables (N=6,916) High-Risk
Group
(N=223)
Low-Risk
Group
(N=6,693)
N (%) N (%)
Mean total healthcare costs ($) (Range) 9 724***
(1047-53,501)
4,245
(54-65,766)
Having hospitalization or ER visits 220 (99)*** 3059 (46)
Diabetes treatment factors:
1 . Type of drugs
-Oral hypoglycemic drugs
-Insulin
-Anti-hypertensive drugs
-lipid-lowering drugs
6693 (100)
175 (78)***
203 (91)
120 (54)
223 (100)
2334 (35)
5871 (88)
3557 (53)
2. Increasing dose of oral hypoglycemic drugs
116(52) 3109 (46)
3. Adding Drugs
-Adding oral hypoglycemic drugs or insulin
-Adding anti-hypertensive drugs
-Adding lipid-lowering drugs
198 (89)**
203 (91)
117(52)
5453(81)
5812(87)
3495 (52)
4. Changing drug to different classes or to insulin
46 (21)** 1019(15)
5. Oral hypoglycemic medication compliance
-Mean medication possession ratio (%MPR)
(Range)
86
(1-100)
87
(1-100)
Follow-up service factors based on diabetic
guidelines:
1. Having office visits (every 3 months for
patients taking insulin and every 6 months for
patients taking only oral hypoglycemic drugs)
2. Having glucose monitoring strip
222 (99)***
157 (70)***
5705 (85)
2948(44)
3. Having lab tests by healthcare providers
-HbAlc every 6 months
-Cholesterol check-up every year
168 (75)***
46 (21)***
1507 (23)
328 (5)
Significance Levels: ***p<0.005, **p<0.05, *p<0.1
156
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table A8: Comparison between High-Risk and Low-Risk Groups Identified by
Probability of Hospitalization or ER visits Criteria (Contd.)
Independent Variables (N=6,916) High-Risk
Group
(N=223)
N (%)
Low-Risk
Group
(N=6,693)
N (%)
Healthcare use factors:
-Gynecologist visit
-Cardiologist visit
-Nephrologist visit
-Neurologist visit
-Infectious disease specialist visit
58(26)***
189(85)***
55 (25)***
103 (46)***
27 (12)***
739(11)
2927 (44)
240 (4)
804 (12)
041 (0.6)
Comorbidity factors:
-Hypertension
-Hyperlipidemia
-Systemic chronic disease
-Cardiovascular disease
-Cancer/severe disease
-Psychiatric disease
206 (92)***
138(62)***
186(83)***
194 (87)***
108 (48)***
81 (36)***
5477 (82)
2499(37)
2528 (38)
4089(61)
1628 (24)
812(12)
Complication factors:
1 . Target organ disease
-Retinopathy
-Nephropathy
-Neuropathy
2. Infection
-Foot
-Skin
-Vagina
97 (^44)***
34(15)***
75 (34)***
107 (48)***
4 (2)**
2 (0.9)
1649 (25)
158 (2)
906 (14)
1438 (21)
34 (0.5)
32 (0.5)
Significance Levels: ***p<0.005, **p<0.05, *p<0.1
157
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table A9: Comparison between High-Risk and Low-Risk Groups Identifled by
Probability of Survival from Not Having Hospitalization or ER visits Criteria
Independent Variables (N=6,916) High-Risk
Group
(N=108)
Low-Risk
Group
(N=6,865)
N (%) N (%)
Mean total healthcare costs ($) (Range) 6617***
(317-44,774)
4413
(54-65,766)
Having hospitalization or ER visits 106 (98)*** 3229 (47)
Diabetes treatment factors:
1 . Type of drugs
-Oral hypoglycemic drugs
-Insulin
-Anti-hypertensive drugs
-lipid-lowering drugs
108(100) 56
(52)***
77(71)
40 (37)
6865 (100)
52 (48)
6041 (88)***
3657 (53)***
2. Increasing dose of oral hypoglycemic drugs
54 (50) 3201 (47)
3. Adding Drugs
-Adding oral hypoglycemic drugs or insulin
-Adding anti-hypertensive drugs
-Adding lipid-lowering drugs
98 (91)**
77(71)
40 (37)
5606(82)
5982 (87)***
3592 (52)***
4. Changing drug to different classes or to insulin
22 (30) 1055 (15)
5. Oral hypoglycemic medication compliance
-Mean medication possession ratio (%MPR)
(Range)
(1-100)
87
(3-100)
Follow-up service factors based on diabetic
guidelines:
1 . Having office visits (every 3 months for
patients taking insulin and every 6 months for
patients taking only oral hypoglycemic drugs)
2. Having glucose monitoring strip
101 (94)**
64 (59)***
5879 (86)
3079 (45)
3. Having lab tests by healthcare providers
-HbAlc every 6 months
-Cholesterol check-up every year
59 (55)***
14(13)***
1648 (24)
370 (5)
Significance Levels: ***p<0.005, **p<0.05, *p<0.1
158
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table A9: Comparison between High-Risk and Low-Risk Groups Identifled by
Probability of Survival from Not Hospitalization or ER visits Criteria (Contd.)
Independent Variables (N=6,916) High-Risk
Group
(N=108)
N (%)
Low-Risk
Group
(N=6,865)
N (%)
Healthcare use factors:
-Gynecologist visit
-Cardiologist visit
-Nephrologist visit
-Neurologist visit
-Infectious disease specialist visit
30 (28)***
47 (44)
12 (11)***
28 (26)***
4(4)**
781 (11)
3096 (45)
292 (4)
891 (13)
68(1)
Comorbidity factors:
-Hypertension
-Hyperlipidemia
-Systemic chronic disease
-Cardiovascular disease
-Cancer/severe disease
-Psychiatric disease
69 (64)
56 (52)***
72 (67)***
51 (47)***
41 (38)***
48 (44)***
5651 (82)***
2612 (38)
2678 (39)
4257(62)
1715 (25)
867(13)
Complication factors:
1 . Target organ disease
-Retinopathy
-Nephropathy
-Neuropathy
2. Infection
-Foot
-Skin
-Vagina
26(24)
10(9)***
25 (23)**
36(33)**
1 (0.9)
1 (0.9)
1733 (25)
187 (3)
969 (14)
1530(22)
38 (0.6)
34 (0.5)
Significance Levels: ***p<0.005, **p<0.05, *p<0.1
159
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.
Table AlO: Association of Risk Factors with Healthcare Cost
Independent Variables
(N=6,910)
Impact on Healthcare
Costs ($)
Diabetes treatment factors:
1. Type of drugs
-Both insulin and oral hypoglycemic drugs
t 1210
2. Increasing dose of oral hypoglycemic drugs
t 141
3. Adding Drugs
-Adding oral hypoglycemic drugs or insulin
-Adding anti-hypertensive drugs
-Adding lipid-lowering drugs
t2 6 4
t5 2 8
t 199
4. Changing drug to different classes or to insulin
t 1018
5. Compliance (%MPR)
i l
Follow-up service based on diabetic guidelines
factors:
1. Having office visits (every 3 months for patients taking
insulin and every 6 months for patients taking only oral
hypoglycemic drugs)
f 730
2. Having glucose monitoring strip
i2 5 8
3. Having lab tests by healthcare providers
-HBAIC every 6 months
-Cholesterol check-up every year
-Dilated eye check-up every year
i 121
f 472
i2 6 0
160
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

A new paradigm to evaluate quality-adjusted life years (QALY) from secondary database: Transforming health status instrument scores to health preference

PDF

Effects of a formulary expansion of the use of SSRIs and health care services by depressed patients in the California Medicaid program

PDF

Assessing the cost implications of combined pharmacotherapy in the long term management of asthma: Theory and application of methods to control selection bias

PDF

Compliance study of second-generation antipsychotics on patients with schizophrenia

PDF

The construct validity of the Health Utilities Index in patients with chronic respiratory disease in a managed care population

PDF

Controlling for biases from measurement errors in health outcomes research: A structural equation modeling approach

PDF

Assessment of prognostic comorbidity in hospital outcomes research: Is there a role for outpatient pharmacy data?

PDF

Essays on organizational forms and performance in California hospitals

PDF

Physician adherence to national hypertension guidelines in an elderly Medicaid population

PDF

Biopharmaceutical strategic alliances: Interorganizational dynamics and factors influencing FDA regulatory outcomes

PDF

A study of employee health plan choice and medical cost: Panel data probit regression and sample selection model

PDF

Individual heterogeneity and program evaluation

PDF

Globalization and the decline of the welfare state in less developed countries

PDF

Cost analysis of three pharmacy counseling programs for diabetics in a health maintenance organization

PDF

Hospital conversions: The California experience

PDF

Drug transport mechanisms in the rabbit corneal epithelium: In vitro-in vivo correlation

PDF

Care management for the uninsured: A force field analysis of the business case

PDF

Asymmetry and risk aversion in first -price sealed bid auctions: Identification, estimation, and applications

PDF

Alteration of the in vitro and in vivo processing of a polypeptide, BBI, through conjugation with palmitic acid

PDF

Infertility and assisted reproductive technology in a pluralistic world: A development and application of a Hindu ethic

Asset Metadata

Creator Chaikledkaew, Usa (author)

Core Title A methodology to identify high -risk patients with diabetes in the California Medicaid populations (Medi -Cal)

School Graduate School

Degree Doctor of Philosophy

Degree Program Pharmaceutical Economics and Policy

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag economics, general,health sciences, health care management,Health Sciences, Pharmacy,OAI-PMH Harvest

Language English

Contributor Digitized by ProQuest (provenance)

Advisor Johnson, Kathleen A. (committee chair), Ahn, Joonghoon (committee member), Globe, Denise (committee member), Graddy, Elizabeth (committee member)

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c16-505932

Unique identifier UC11340227

Identifier 3140446.pdf (filename),usctheses-c16-505932 (legacy record id)

Legacy Identifier 3140446.pdf

Dmrecord 505932

Document Type Dissertation

Rights Chaikledkaew, Usa

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 au...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA