University of Southern California Dissertations and Theses

Disease modifying treatments for Alzeimer's disease: modeling, value assessment and eligible patients

(USC Thesis Other)

Disease modifying treatments for Alzeimer's disease: modeling, value assessment and eligible patients

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content DISEASE MODIFYING TREATMENTS FOR ALZHEIMER’S DISEASE:
MODELING, VALUE ASSESSMENT AND ELIGIBLE PATIENTS
by
Yifan Wei
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(HEALTH ECONOMICS)
December 2023

ii
Acknowledgements
I would like to express great appreciation for my advisor Dr. Dana Goldman, who
provided great support and guidance for me throughout my doctoral study. I would also like to
thank Dr. Bryan Tysinger, Dr. Hanke Heun-Johnson, Dr. Jakub Hlávka, Dr. Darius Lakdawalla,
Dr. Geoffrey Joyce and Dr. Seth Seabury who have all provided important mentorship to me.
Special thanks to my family and friends for their consistent support and belief in me over the
years.

iii
Table of Contents
Acknowledgements………………………………………………………………………………. ii
List of Tables…………………………………………………………………………………….. v
List of Figures…………………………………………………………………………………….vi
Abstract…………………………………………………………………………………………..vii
Introduction………………………………………………………………………………………. 1
Overview of Alzheimer’s Disease and Disease Modifying Treatments………………………. 1
Value Assessment of Alzheimer’s Disease Modifying Treatments…………………………….3
Access to AD DMT and Specialist Capacity…………………………………………………...5
Research Aims………………………………………………………………………………….7
Chapter 1: Using Dynamic Microsimulation to Project Cognitive Function in the Elderly
Population…………………………………………………………………………………………9
Introduction……………………………………………………………………………………..9
Method………………………………………………………………………………………...11
The Future Elderly Model (FEM) Overview……………………………………………….11
Data and Measures………………………………………………………………………….11
Key Transition Models……………………………………………………………………...14
Model Validation Approach………………………………………………………………...14
Results…………………………………………………………………………………………17
Sample Characteristics……………………………………………………………………...18
Population-level Predictions………………………………………………………………..18
Individual-level Predictions………………………………………………………………...19
External Comparison……………………………………………………………………….20
Discussion……………………………………………………………………………………..21
Chapter 2: Extrapolating AD DMT Clinical Trial Results to Longer-term Value Assessment…..39
Introduction……………………………………………………………………………………39
Methods………………………………………………………………………………………..41
The Future Elderly Model (FEM) Overview……………………………………………….41
Data and Measures………………………………………………………………………….43
Models for CDR-SB………………………………………………………………………..44
Downstream Outcome Models……………………………………………………………..45
Key Model Assumptions and Accounting…………………………………………………..47
Results…………………………………………………………………………………………48
Discussion……………………………………………………………………………………..49
Appendix………………………………………………………………………………………52
Description of Downstream Outcomes……………………………………………………..52
ARIA adverse events assumptions………………………………………………………….55
Chapter 3: Estimating the Number of Patients Eligible for AD DMTs and Patient
Accessibility under Specialist Capacity Constraints………………………………...…………...76
Introduction……………………………………………………………………………………76
Data and Methods……………………………………………………………………………..78
Results…………………………………………………………………………………………81
Estimating Eligible Treatment Population………………………………………………….81
Estimating Prescription Given Specialist Capacity Constraints……………………………83
Discussion……………………………………………………………………………………..84
Appendix………………………………………………………………………………………89

iv
CDR Modeling……………………………………………………………………………...89
Calculation of Diagnosis Rates in MCI and Mild Dementia Groups………………………91
Conclusions……………………………………………………………………………………..106
Concluding Remarks…………………………………………………………………………109
References………………………………………………………………………………………111

v
List of Tables
Table 1.1. Marginal effect from FEM TICS27 and mortality transition model………………….25
Table 1.2. Characteristics of the 2006 Health and Retirement Survey (HRS) respondents……...26
Table 1.3. Distribution comparison between HRS respondents and FEM simulation, 2006-
2016………………………………………………………………………………………………27
Table 1.4. 10-year change in distributions of cognitive status based on TICS score, HRS and
FEM……………………………………………………………………………………………...28
Table 1.5. Area under the receiver operating characteristics curve (AUROC) for predicting
dementia from 5-fold cross-validation…………………………………………………………...29
Table 1.6. Area under the receiver operating characteristics curve (AUROC) for predicting
dementia or dead with dementia for 2, 4, 6, 8 and 10 years……………………………………..31
Table 1.7. Two-year TICS27 significant decline in MCI subjects and comparable results from
COMPASS……………………………………………………………………………………….32
Appendix Table 1.1A. Proxy interview cognitive impairment rating from Health and
Retirement Study, 2006-2016………………………………………………………………..…..36
Appendix Table 1.2A. TICS27 and mortality transition model………………………………….37
Table 2.1. Descriptive statistics of simulation starting cohort…………………………………...56
Table 2.2. Downstream outcomes………………………………………………………………..57
Table 2.3. Three simulation scenarios……………………………………………………………58
Table 2.4. Levels and differences in cumulative outcomes between control and the 48-Tx,
18-Tx, and upper boundary 48-Tx scenarios………………………………………………….....59
Appendix Table 2.1A. Parameters for ARIA adverse events…………………………………….66
Appendix Table 2.2A. Coefficients of CDR-SB transition model……………...………………..67
Appendix Table 2.3A. Coefficients of CDR-SB imputation model……………………………..68
Appendix Table 2.4A. Downstream outcomes…………………………………………………..69
Appendix Table 2.5A. Disease-stage dependent medical costs and caregiver disutility inputs…72
Appendix Table 2.6A. Trajectory for outcomes from control, 18-Tx, 48-Tx, and upper
boundary 48-Tx scenario…………………………………………………….…………………..73
Table 3.1. Descriptive statistics of 2016 HRS respondents by age and cognitive status………...93
Table 3.2. Estimated patient funnel for AD DMT in 2022………………………………………94
Appendix Table 3.1A. CDR models……………………………………………………………..98

vi
List of Figures
Figure 1. Change in amyloid accumulation, cognitive performance and clinical function along
clinical stage of AD………………………………………………………………………………..8
Figure 1.1. Distribution comparison between HRS and FEM, 2006-2016………………………33
Figure 1.2. Receiver operating characteristics curve for predicting dementia in 10 years………34
Figure 2.1. Trajectory for outcomes from control and the 48-Tx scenario………………………60
Figure 2.2. Patient disposition of cognitive status under control and the 48-Tx scenario ……....65
Figure 3.1. Treatment funnel for people eligible for Alzheimer’s DMTs………………………..95
Figure 3.2. Patients treated and untreated in the first 5 years based on the number of
follow-up annual visits…………………………………………………………………………...96
Figure 3.3. Number of years to initial DMT treatment for diagnosed patients by levels of
specialist capacity, number of annual follow-ups and treatment discontinue rate………………97
Appendix Figure 3.1A. Performance of predictive models for CDR, among participants
without proxy and participants with proxy……………………………………………...……...103
Appendix Figure 3.2A. Insurance status by cognition and age………………………………...104
Appendix Figure 3.3A. Lower-bound Projections of Patients Eligible for AD DMT by Age
and Cognitive Status……………………………………………………………………………105

vii
Abstract
Alzheimer’s disease (AD) is the most common cause of dementia, accounting for an
estimated 60% to 80% of cases. It is a neurodegenerative disease characterized by symptoms like
loss of cognitive function, functional impairment, and neuropsychological symptoms. Without
effective treatment, the number of Americans aged 65 and older living with AD dementia is
estimated to grow from 6.7 million in 2023 to 13.9 million in 2060. US FDA recently approved
two disease modifying therapies (DMTs) for AD, aducanumab and lecanemab. The emergence of
AD DMTs brings opportunities and challenges, as knowledge on DMTs’ long-term value and
eligible patient population size plays important role in policy decision making.
This dissertation encompasses three aims: (1) to develop and validate a population-level
microsimulation model to project cognitive trajectories across the full AD continuum; (2) to
extrapolate AD DMT’s clinical trial results to policy-relevant outcomes with a longer follow-up
period with microsimulation; and (3) to estimate the size of treatment-eligible patient population
in the US for AD DMTs and evaluate patient accessibility under specialist capacity constraints.
We developed and validated a microsimulation model to project trajectories in cognition,
based on nationally representative longitudinal data of the US population aged 51 and older. We
applied this model to extrapolate lecanemab’s clinical trial results to policy-relevant outcomes
with a longer follow up period. Lastly, we estimated the size of treatment-eligible patient
population in the US for AD DMTs based on the prevalence of cognitive impairment, diagnosis
rates, and specialist access constraints, using aducanumab as a meaningful example.

1
Introduction
Overview of Alzheimer’s Disease and Disease Modifying Treatments
Alzheimer’s disease (AD) is a type of brain disease. AD is best conceptualized as a
biological and clinical continuum covering both the preclinical and clinical phases. In the
preclinical phase of AD, individuals are clinically asymptomatic with evidence of AD pathology
measured by biomarkers like amyloid-beta and tau. In the clinical phase, clinical symptoms like
cognitive decline and functional limitations are shown. Traditionally, two key clinical stages
have been considered, mild cognitive impairment (MCI) due to AD and AD dementia, with AD
dementia further divided into mild, moderate and severe. The distinction between AD and
dementia is an important distinction between the neuropathological and syndromic diagnoses.
Dementia is an overall term for a group of symptoms like difficulties with memory, language,
problem-solving and other thinking skills. Dementia has several causes like AD, cerebrovascular
disease, frontotemporal degeneration, hippocampal sclerosis, Lewy body disease, Parkinson’s
disease, etc. While AD is one cause of dementia, the brain changes of AD include the
accumulation of the abnormal proteins beta-amyloid and phosphorylated tau, as well as the
degeneration of neurons. AD dementia refers to dementia that is caused by, or believed to be
caused by, the brain changes of AD.[1,2] Aisen et al illustrated this continuum in Figure 1[1],
where trajectories of amyloid accumulation, cognitive performance, and clinical function are
shown along clinical stage of the disease.
AD is the most common cause of dementia, accounting for an estimated 60% to 80% of
dementia cases. AD is characterized by symptoms like loss of cognitive function, functional
impairment, and neuropsychological symptoms.[2] As the sixth leading cause of death in the US,
people age 65 and older survive an average of four to eight years after a diagnosis of AD

2
dementia.[3] An estimate of 6.7 million Americans aged 65 and older, or about 1 in 9 people in
this age range, are living with AD dementia in 2023. The percentage of people with AD increases
with age. As the population of Americans aged 65 and older, the age group that is at elevated risk
of AD, is projected to grow rapidly, the absolute number of people with AD is expected to
continue to grow.[2] The number of people living with AD dementia is projected to grow from
6.07 million in 2020 to 13.85 million in 2060, assuming no effective treatment.[4]
AD imposes considerable humanistic burden on patients, as it has a set of complex
symptoms affecting cognition, functional status, social interaction, and psychological well-being.
Its total societal burden, which include direct costs (both medical and non-medical), indirect
costs (informal care and lost productivity costs) and intangible costs, is estimated at more than
$307 billion in the US in 2010 and is projected to increase by approximately 4.9-fold (to $1.5
trillion) in the US by 2050. It is notable that 41% of the 2010 total societal burden is due to
informal care costs.[5]
There is a recent emergence of disease-modifying treatment (DMT) for AD. DMTs are
treatments that produce an enduring change in the clinical progression of AD by interfering with
the underlying pathophysiological mechanisms of the disease process that lead to neuronal
death.[6] Aduhelm (aducanumab) is the first-ever AD DMT approved by the US Food and Drug
Administration (FDA). It was approved in June 2021 using an accelerated approval pathway.[7]
Leqembi (lecanemab), which was approved by US FDA in July 2023, is the first traditionally
approved AD DMT.[8] Centers for Medicare & Medicaid Services (CMS) also provides broader
Medicare coverage of lecanemab following FDA’s traditional approval, requiring eligible
Medicare patients to have a physician who participates in a qualifying registry to receive

3
Medicare coverage for lecanemab.[9] More potential DMTs are in the pipeline; in 2022, DMTs
represent 83.2% of the total number of AD drugs in trials.[10]
With the emergence of AD DMTs, challenges arise together with opportunities. While
DMTs demonstrate their effectiveness by showing statistically significant outcomes on clinical
endpoints, health policy makers are interested in the long-term value of AD DMTs and their
potential impact on the healthcare system.
Value Assessment of Alzheimer’s Disease Modifying Treatments
Value assessment traditionally centers on patients’ quality-adjusted life-years (QALYs)
and net costs, for net costs being the difference between costs to acquire this new intervention
and saving in medical spending. Due to the nature of AD, additional value components, in terms
of both QALYs and costs, should be considered within this value assessment framework. AD is a
progressive disease that may deteriorate over a long period of time; it affects patients in multiple
aspects like cognition, function, and psychological well-being; it also places immense burden on
patient caregivers’ and family members’ health, finances, and productivity; and it generates
direct and indirect costs for health and long-term care systems, economies, and entire society.[11]
For QALYs, many AD interventions yield benefits to both AD patients and their caregivers.
Value assessment should take both patients’ and caregivers’ utility into consideration. For costs,
El-Hayek et al found that the frequently measured costs of AD are only the “tip of the
iceberg.[12] Informal caregiving costs accounts for over 40% of the AD’s total societal burden,
and it should be considered in AD intervention value assessment.[5] Productivity loss should also
be considered, as shown in the 2016 Health and Retirement Study data, about 45% of patients
with mild cognitive impairment under 65 are working for pay (about 13% over 65 still do).

4
A systematic review examining 34 relevant studies to better understand the priorities of
AD patients, caregivers, and healthcare providers found that they are primarily concerned with
AD’s observable effects on their daily life, including its impact on maintaining a patient’s
independence and mental health. These priorities were consistently found across different
studies, geographies and stakeholders. Outcomes like memory decline, maintaining patient
independence and identity, mental health issues, patients’ quality of life, caregiver burden were
raised by all three stakeholders.[13] Therefore, a DMT’s ability to address these meaningful realworld outcomes that patients, caregivers and healthcare providers value the most should be
evaluated alongside its impact on clinical endpoints and QALYs. This will provide additional
angles for important stakeholders to evaluate such treatments. For example, activities of daily
living (ADLs) could explicitly measure patients’ ability to perform everyday personal care
activities like walking, getting in and out of bed, eating, etc., which greatly relates to patients’
level of independence.
As a neurodegenerative disease, people can live many years with AD and effective AD
DMTs may have long-term value for patients, their caregivers and society. However, AD DMTs’
clinical trials are only conducted for a limited period (one to two years). Even if a broad range of
outcomes measuring QALYs, cognition, functional status, psychological well-being, and
caregiver burden are included as clinical trial endpoints, there still remains an important evidence
gap for long term impacts of AD DMTs if only clinical trial results are considered.[14] Barbarino
et al emphasized the importance of long-term value demonstration in AD, as decades of scientific
research may soon lead to safe and effective AD DMTs that will likely deliver value over the
course of many years and multiple stakeholders.[15] This study outlined a two-step approach for
AD DMTs long-term value demonstration. In the long-run, it called for collaborative data

5
collection efforts to generate real-world evidence that shows a therapy’s long-term effects across
numerous stages of AD, including impacts on patient outcomes, caregivers, and healthcare
systems.[16] In the near-term, modeling may be the necessary suboptimal approach to
extrapolate short-term clinical trial results and provide meaningful estimates of AD DMTs’ longterm value on the population-level.[15]
There is room for improvement for the existing AD models. Models should be
comprehensive by including multiple stakeholders, not only focusing on patients and healthcare
systems, but also including caregivers and the broader society. It should also be granular enough
to represent the entire continuum of AD, from preclinical stages to MCI due to AD then to
different stages of AD dementia (mild, moderate and severe). Models should also be
representative of the patient population of AD, especially acknowledging the heterogeneity of the
AD population by considering the impact on different subgroups.[15] Models should report
rigorous validation of itself by comparing simulated output against either aggregate statistics or
real-world data, though this is a general challenge for AD models due to the lack of publicly
available data.[17] Finally, models should have a long-term perspective covering outcomes from
multi-domains that relate to AD, while making evidence-based and cautious extrapolation of
short-term clinical trial results.
Access to AD DMT and Specialist Capacity
It is the hope of many that with safe and effective AD DMTs available on the market, the
life of AD patients and their family will be dramatically and positively changed. However,
availability of AD DMTs on the market is only the beginning of successful AD disease
management, as many issues need to be addressed to ensure that the healthcare system is

6
prepared for treating AD with these novel DMTs and patients in need are able to access these
therapies.
Treating AD patients with DMTs is a complex process, as it requires initial screening,
diagnosis, treatment initiation and continuous treatment monitoring. As illustrated in Liu et al,
for patients treated with AD DMTs, their patient journey includes four phases. First, during the
Screening Phase, individuals above a certain age would undergo cognitive screening via a short
instrument in the primary care setting. Those with positive screening results will then undergo
the Diagnostic Phase and be referred to specialists for further evaluation, potentially with
imaging and/or amyloid deposits examination. For those diagnosed with AD, they would be
referred to routine infusion treatment with a DMT if appropriate during the Treatment Phase.
Specialists will provide ongoing monitoring for patients’AD disease progression and adverse
events in the Outcomes Phase. Patients’ access to AD DMTs depends on the capacity of primary
care doctors and specialist, availability of screening and diagnostic tools, and infusion capacity.
Insurance coverage for AD DMTs, required screening and diagnostics examination are also
critical to ensure eligible patients’ access.[18]
Assessment of whether the US healthcare system is ready for AD DMTs from the above
aspects is limited. Liu et al developed a Markov model to simulate the effect of capacity
constraints on access to care for patients with suspected AD for a hypothetical AD DMT. This
study finds the US healthcare system is ill-prepared to handle the potentially high volume of
patients who would be eligible for DMT treatment, with specialist shortage being the most urgent
issue. The estimated average wait time for the diagnostic and treatment phase are 18.6 months
when a DMT first becomes available, and it would take 14 years to eliminate waiting times given
the backlog of prevalent MCI due to AD patients.[18]

7
To address this specialist capacity constraint, policy solutions that allow a stepwise
introduction of AD DMTs or a prioritization of patients most likely to benefit might be necessary.
Such policy would also alleviate the enormous budget impact on payers from a costly DMT for
early AD, as experiences from novel treatments for hepatitis C, multiple sclerosis, and some
cancers show.[19]
Research Aims
In this dissertation, I try to address some of the concerns that have been highlighted
above: (1) to develop and validate a population-level microsimulation model to project cognitive
trajectories across the full AD disease continuum; (2) to extrapolate AD DMT’s clinical trial
results to policy-relevant outcomes and with a longer follow-up period with microsimulation;
and (3) to estimate the potential sizes of treatment-eligible patients population in the US for AD
DMT and evaluate patient accessibility under specialist capacity constraints.
In Chapter 1, I developed and validated a microsimulation model to project trajectories in
cognition across the full AD continuum, based on nationally representative data of the US
population aged 51 and older. This study demonstrated its usefulness as a model for long-term
economic evaluation for AD. Models from this chapter covers the full AD continuum.
In Chapter 2, I applied a microsimulation model to extrapolate lecanemab’s Clarity AD
trial results to policy-relevant outcomes and a longer follow up period. The emergence of AD
DMTs raise the questions of such DMTs impact beyond short-term clinical endpoints, and this
study provides a feasible evaluation framework using lecanemab as a meaningful example. This
chapter focuses on two clinical stages of AD, MCI due to AD and mild AD dementia.

8
In Chapter 3, I estimated the potential sizes of treatment-eligible patient populations in
the US for AD DMT based on the prevalence of cognitive impairment, diagnosis rates, and
specialist access constraints, using aducanumab as a meaningful example. This chapter focuses
on two clinical stages of AD, MCI due to AD and mild AD dementia.
With AD DMTs on the US market, there is opportunity together with challenges. It is
important to understand the long-term value of AD DMTs to patients, caregiver, and society, and
assess whether our healthcare system is ready to treat patients eligible for such novel treatments.
Figure 1. Change in amyloid accumulation, cognitive performance and clinical function
along clinical stage of AD.

9
Chapter 1: Using Dynamic Microsimulation to Project Cognitive Function in
the Elderly Population
Introduction
Alzheimer’s disease(AD) imposes an increasing burden on the United States society and
health care system. According to the Alzheimer’s Association, the number of Americans 65 and
older living with AD dementia is estimated to grow from 5.8 million in 2019 to 13.8 million by
2050, as the baby boom generation ages. On the other hand, the US Food and Drug
Administration recently approved aducanumab, the first disease-modifying treatment (DMT) for
AD, with more potential DMTs in the pipeline.[20,21] With the approval of aducanumab,
discussion arises about the therapies’ real-world value and their potential costs to the healthcare
system. As treatments shift in focus from dementia to earlier disease phases like mild cognitive
impairment (MCI) due to AD, concerns are rising about the high upfront costs for initial
screening and diagnostics together with late-occurring and uncertain benefits.[19,22]
Facing great opportunities and challenges in AD therapy development, it is important that
we have the proper analysis tools to evaluate the economic impact of cognitive impairment and
potential therapies. A cognitive model for use in projecting the US population should model all
stages of cognitive decline for a nationally representative sample. Additionally, incorporating
predictors and risk factors will make the model useful for assessing counterfactual scenarios and
interventions. We identified six AD economic evaluation models for the US in the literature, and
none of the existing models were based on data nationally representative of the US
population.[17,23–27] Four models used the Uniform Data Set from the US National
Alzheimer’s Coordinating Center.[17,23,25,27] The Uniform Data Set contains data from the
Alzheimer disease centers across the United States, but it is not considered a population-based

10
sample since the enrollment of patients by participating Alzheimer disease centers is not
random.[23] One model was based on the Alzheimer’s Disease Neuroimaging Initiative (ADNI)
database, which is a research cohort of participants in cognitively normal, MCI and dementia
states and also not commonly regarded as typical of the current clinical AD population.[17,24]
One model was not based on a selected sample of participants but used model input parameters
from multiple data sources.[26] Among the six models that used US data, three models tracked
individuals across the full AD continuum.[24,26,27] We also identified three models for other
countries that tracked individuals across the full AD continuum.[28–30] Among them, one model
was based on six longitudinal cohort studies from different countries; one model for the UK and
one model for Spain used model input parameters from multiple data sources. The Spain model
represented the Spanish population aged 40 years or older from 2010 to 2050 and was validated
against published life expectancy and incidence and prevalence of the dementia stages in
Spain.[29]
Among the AD economic evaluation models we identified, only one model reported
rigorous validation. The Alzheimer’s Disease Archimedes Condition-Event Simulator was
validated by comparison of risk of mortality, institutionalization, and transition to Alzheimer’s
dementia predictions to external data from patient registries, clinical trials and literature. In each
validation, the simulation cohort’s baseline characteristics were matched to the study population
in external data.[24] Several other models reported comparison of estimated transition
probabilities, cognitive trajectories, incidence of dementia and survival to published
literature.[17,23,25–27,29] However, none of the AD economic evaluation models we identified
in the literature validated their models by comparing simulated output against population-based

11
samples. Model validation is a general challenge in the area of AD and many models suffer from
this limitation, due to the lack of publicly available data.[17]
In this paper, we introduce and validate a microsimulation model to project trajectories in
cognitive test scores, FEM TICS27, across the full Alzheimer’s disease continuum, based on
nationally representative data of the US population aged 51 and older from the Health and
Retirement Study (HRS). We compare our model’s 10-year predictive performance against
longitudinal HRS data, using area under the receiver operating characteristics curve (AUROC)
[31] and five-fold cross validation. We also introduce our validation framework which we
believe is valuable to future microsimulation validation studies.
Methods
The Future Elderly Model (FEM) Overview
The Future Elderly Model (FEM) is a microsimulation model of health and economic
outcomes for the US population aged 51 and older. Here we summarize FEM’s core functions;
technical details are described in a technical appendix.[32] FEM uses first-order Markov
transition models to simulate individuals’ aging progress. It captures outcomes including health
conditions, functional status, earnings and employment status, participation in government
benefit programs, and mortality. FEM has been used in many studies answering important policy
questions in aging and dementia.[33,34] In this study, we developed and validated a new model
for cognition measurement, TICS27 score, in FEM.

12
Data and Measures
FEM uses data from the HRS, a biennial nationally-representative longitudinal survey in
the population with more than 37,000 respondents over age 50 in the US. Baseline interviews
with existing birth cohorts have been conducted in 1992, 1993, 1998, 2004, 2010, and 2016 with
oversampling of Hispanics and African-Americans. Every six years, the HRS enrolls a new birth
cohort in order to maintain a steadystate of the US population over age 50. Participants are
followed through the life course with the core biennial surveys and supplemental data
collections. Technical details on HRS sampling design, recruitment, and measurement are
published before.[35] In this validation study, our simulation sample consisted of HRS
respondents age 53 or older in 2006 in the 2006 HRS survey. All population-level analyses used
HRS sample weights.
With its goal of understanding the challenges and opportunities of aging, HRS includes a
section on cognition, since decline in cognitive functioning is a hallmark of aging and predictive
of mortality.[36] HRS uses two different sets of measures to assess cognitive status: for
respondents who complete the survey themselves (“self-respondents”), cognitive functioning is
assessed using an adapted version of the Telephone Interview for Cognitive Status
(TICS).[36,37] For respondents who are not able to complete the survey themselves, questions
about changes in memory in the last two years are asked to proxy respondents in the HRS.
TICS is modeled after the Mini-Mental State Exam (MMSE) for use over the telephone,
and TICS scores can be converted to MMSE scores using a validated crosswalk.[36,38] TICS
tests respondents’ cognitive impairment and dementia status, and contains test items that evaluate
memory, concentration and executive function, for example by immediate and delayed word
recall, counting back from 100 by 7’s, and counting back from 20. Composite scores using these

13
test items create a measure of cognitive functioning ranging from 0 to 27 (TICS27).[36,39]
Respondents with scores from 0 to 6 are classified as having dementia, from 7 to 11 as having
MCI, and from 12 to 27 as being cognitively normal. This approach was developed and validated
by Langa and Weir (2010) [40] and has since been used by many studies on cognitive
functioning.[39,41,42] To reduce measurement error in categorizing cognitive status based on
TICS27, we require two consecutive responses for dementia: one wave with dementia followed
by either dementia or death in the next wave. For MCI, we require either one wave with MCI
followed by MCI, dementia or death, or one wave with dementia followed by MCI. All other
cases are categorized as cognitively normal. Cognitive status is considered an absorbing state;
once a respondent has been classified with “verified” dementia or MCI, we assume their
cognitive status cannot revert to a less severe state.
Our model’s target, TICS27 score, is missing from all respondents using a proxy respondent
in HRS. Some respondents cannot participate in the interview because of cognitive problems;
others might choose to use a proxy because they were working and were thus more likely to be
cognitively normal. For example, in 2016, among the 941 proxy respondents, 41.9% did not
think the respondent had any cognitive limitations, 6.2% thought the respondent may have some
cognitive limitations and 52.0% thought the respondent had cognitive limitations that prevented
him or her from being interviewed. For detailed proxy interview cognitive impairment ratings
from 2006-2016 HRS, please see Appendix Table 1.1A. This missingness of TICS27 among
respondents using a proxy was therefore assumed to be correlated with respondents’ cognitive
functioning, depending on the reason for using a proxy. Since this missingness is not at random,
and HRS does not provide imputed TICS27 values for respondents with a proxy, we adopted a
multiple imputation strategy based on HRS’ approach for missing TICS27 among self-

14
respondents. We used a combination of relevant demographic, health, and economic variables, as
well as prior wave cognitive variables to perform the imputation using the sequential regression
method.[43] The multiple imputation was performed using the multiple imputation (mi)
command in Stata Version 16.0. Following HRS’s practice, we did not impute for participants
who were non-responsive to the survey in a given wave.
Key Transition Models
FEM transition models are a mixture of continuous, binary, and categorical outcomes,
with a timescale that mimics the two-year structure of the HRS data. The marginal effects of two
specific transition models from the FEM that are most relevant to this study, the TICS27 and
mortality transition models, are shown in Table 1.1. For model coefficients of these two
transition models, please refer to Appendix Table 1.2A. The TICS27 transition model was
estimated with an ordered probit model using HRS data from 2008 to 2016. An ordered probit
model was chosen because TICS27 is a ranked score ranging from 0 to 27. As can be seen in
Table 1.1, variables in the TICS27 model are the TICS27 score from two and four years prior,
cognitive status, demographic variables (age, gender, race and ethnicity, education), chronic
disease indicators, employment, widowhood, smoking status and body mass index. Our choice of
variables aligns with evidence on risk factors for cognitive decline in existing literature.[44]
Also, we limited variables to information that is readily available from the survey, i.e. not based
on blood samples, genetic data, or other clinical procedures. The mortality transition model was
estimated with a probit model using HRS data from 2008 to 2016. Prior validation shows that
FEM projections on mortality are generally in line with observed mortality rates.[45]

15
Model Validation Approach
We validated the FEM TICS27 model at both the population- and individual-levels. At
the population-level, we looked at two outcomes, TICS27 distribution comparisons and 10-year
changes in a composite measure of both cognition and mortality. At the individual-level, we
assessed FEM’s performance in predicting dementia in 2/4/6/8/10 years and significant decline in
TICS27 in two years using receiver operating characteristics (ROC) curves.
We validated the FEM’s TICS27 model using a five-fold cross validation approach, by
comparing 10-year simulated population- and individual-level outcomes against observed HRS
data in 2016. All validation analyses results are based on 500 Monte Carlo simulations of FEM.
A five-fold cross validation approach allows us to have separate datasets for estimation and
simulation to evaluate FEM TICS27 model’s performance in an independent dataset. Crossvalidation is a data resampling method to assess the generalizability of predictive models and to
prevent overfitting.[46] To do this, we first randomly partitioned our simulation sample into five
complementary subsets. We then saved one subset for simulation and used the other four subsets
to estimate transition models for this simulation. We repeated this process five times so that each
subset was used once for simulation. Finally, we pooled results from five simulations on subsets
together for validation analyses. We chose to use five folds since it has been shown empirically
that a five- to ten-fold cross validation is the optimal approach.[47]
Two different samples were used in the validation analyses. One was the complete 2006
HRS sample, which included HRS respondents aged 53 or older in 2006. This sample was used
in population-level distribution comparison analyses. The other was the 2006 HRS with full 10-
year follow-up sample, which was a subset of the complete 2006 HRS sample and used to
determine the population-level 10-year change in cognitive/mortality status, and individual-level

16
analyses. This full 10-year follow-up sample required individuals to respond to every wave of the
HRS survey from 2006 to either 2016 or their death.
Population-level outcomes include TICS27 distribution comparisons and 10-year changes
in a composite measure of both cognition and mortality. We adopted this composite measure
since people with dementia have high mortality rates. Individual-level outcomes include
predicting dementia status in 10 years and predicting decline larger than 3 TICS27 points within
2 years for patients with MCI.
On a population-level, the distribution of simulated TICS27 in 2016 was compared to the
2016 HRS population in the same age range (age 63 or older). We also analyzed the 10-year
change in status by comparing assignment of cognitive status or death by FEM in 2016 given the
2006 cognitive status to the observed status in HRS. Cognitive status at death was determined by
the cognitive status in the last wave before death.
On the individual level, we assessed FEM’s performance in predicting dementia in 10
years and significant decline in TICS27 in two years using receiver operating characteristics
(ROC) curves. Though more commonly used in regression-based risk prediction models, ROC
curves have been used for validation of other disease simulation models as well.[31,48] For
individual-level analyses, we ran FEM 500 times over a 10-year time horizon for every
individual in the 2006 HRS full 10-year follow-up sample. After 500 simulation iterations, we
calculated the percentages of iterations for every individual with specific outcomes for two
measures: (1) alive or dead with dementia in 2016; (2) significant decline (decline greater than or
equal to 3 points) in TICS27 in 2008. Prior research found that a 3-point decline in MMSE
indicated significant decline.[49,50] Using a crosswalk between MMSE and TICS27, a 3-point
decline in MMSE translates to a 3-point decline in TICS27 for people with MCI (MMSE from

17
24 to 27).[51] We then ranked every individual in the simulation by their FEM-based risks for
each separate measure. These ranks were compared to their actual outcome in the HRS data to
generate ROC curves. We used area under the ROC curve (AUROC), which is a commonly used
measure for predictive model performance, to evaluate FEM’s performance on these two
measures.
We also compared our model’s performance to one of the best-performing models for
predicting cognitive decline, COMPASS.[52] COMPASS used data from the ADNI database
with information on age, gender, education, APOE genotype and cognitive composite scores on
memory and executive functions to predict changes in MMSE scores over 24-months.
COMPASS evaluated its performance on predicting significant decline in MMSE scores (3
points) in MCI subjects in 2 years using AUROC. We compared FEM’s performance on
predicting significant TICS27 decline in MCI subjects in 2 years to COMPASS.
The University of Southern California IRB approved this research under UP-18-00776
(“Ensuing Access to Novel Alzheimer’s and Dementia Treatments”) on November 21, 2019. This
is a retrospective study of secondary data from the Health and Retirement Study that is deidentified and publicly available. This study uses HRS Public Release data which is fully
anonymized before researchers’ access. Prior to each interview, HRS participants are provided
with a written informed consent information document. At the start of each interview, all HRS
participants are read a confidentiality statement and give oral consent by agreeing to do the
interview. Their oral consents are documented in answers to the survey questionnaire. FEM is
programmed in C++, SAS and Stata, and all validation analyses were performed using Stata
Version 16.0.

18
Results
Sample Characteristics
Weighted baseline characteristics for the 2006 HRS and the 2006 HRS with full 10-year
follow-up samples are shown in Table 1.2. Since the full follow-up sample includes relatively
more respondents who died between 2006 and 2016, respondents are older and have more
chronic conditions compared to the 2006 HRS sample. Other variables are comparable between
the two samples, including baseline TICS27 score and confirmed cognitive status. In 2006, the
mean TICS27 score for the HRS sample and the full follow-up samples were 15.57 and 15.54,
respectively (P=0.21). In the 2006 HRS sample, 1.2% of respondents were living with dementia,
8.9% had MCI and 90.0% were cognitively normal; in the full follow-up sample, 1.2% of
respondents were living with dementia, 9.0% had MCI and 89.8% were cognitively normal.
Population-level Predictions
Figure 1.1 shows the distribution of TICS27 in 2006 (grey line) and the subsequent
decline in this cognitive measure in 2016 from FEM simulations (black line) and HRS
observations (dashed line), for HRS respondents age 53+ and 70+ in 2006. Table 1.3 shows the
mean TICS27 score, 10-year change in mean TICS27 score and TICS27 score at different
percentiles for both FEM and HRS, for HRS respondent age 53+ and 70+ in 2006. In aggregate,
the distribution of TICS scores after ten years of FEM simulation matches the 2016 HRS
distribution well, both at the mean and at specific points in the distribution, for both age groups.
For HRS respondents age 53+ in 2006, the mean ten-year change in TICS27 is -0.62, compared
to -0.64 in FEM simulation; at the 10th, 50th and 90th percentiles of 2016 TICS27 distribution, the
TICS27 score in HRS is 9, 15, 21, and the TICS27 score in FEM is 9, 15, 20, respectively. For

19
HRS respondents ages 70+ in 2006, the mean ten-year change in TICS27 is -1.51, compared to -
1.67 in FEM simulation; at the 10th, 50th and 90th percentiles of 2016 TICS27 distribution, the
TICS27 score in HRS is 6, 12, 18, and the TICS27 score in FEM is 5, 12, 18, respectively.
Table 1.4 shows the 10-year change in distributions of combined cognitive/mortality
status given a respondent’s initial status. The 2016 status has five categories: cognitively normal,
MCI, dementia, dead without dementia, and dead with dementia. Overall, compared to HRS
data, FEM assigns similar percentages of people to each cognitive/mortality category in 2016. Of
HRS respondents, 56.7% retained normal cognitive function between 2006 and 2016; FEM
assigns 58.5% of respondents to this category. In HRS in 2016, 9.9% of respondents were in the
MCI category, and 1.9% of respondents were in the dementia category; the predictions from
FEM are 9.0% and 2.5%, respectively. In HRS in 2016, 27.0% of respondents were dead without
dementia and 4.5% were dead with dementia; FEM predicts 25.5% and 4.5% of respondents to
be in these categories, respectively.
Individual-level Predictions
Table 1.5 shows AUROC results for FEM predicting (1) dementia or death with
dementia, and (2) dementia conditional on being alive in 10 years, for both the full follow-up
sample and sub-population analyses (e.g. by race and ethnicity). Figure 1.2 shows the ROC
curves for the full follow-up sample (Panels A and B) and individuals with MCI in the 2006
sample (Panels C and D). In the full follow-up sample, the AUROC for dementia or dead with
dementia in 10 years is 0.904, the AUROC for dementia conditional on being alive is 0.868.
FEM’s performance on predicting MCI or worse is comparable to that of predicting dementia
(Table 1.5). Furthermore, FEM’s predictive performance is comparable for subgroups of age,

20
race and ethnicity, education and disease status. For people aged 65 years or older in 2006, the
AUROC for dementia or dead with dementia is 0.875. For non-Hispanic Black and Hispanic
people, the AUROC for dementia or dead with dementia is 0.906 and 0.881, compared to 0.891
for non-Hispanic White people. For people without a high school degree, the AUROC for
dementia or dead with dementia is 0.866, compared to 0.856 for people with high school
education and 0.880 for people with at least some college education. For people who ever had a
stroke before 2006, the AUROC for dementia or dead with dementia is 0.875. For people with
MCI in 2006, the AUROC for dementia or dead with dementia is 0.720, and the AUROC for
dementia conditional on being alive is 0.705.
To demonstrate FEM TICS27’s performance across years, Table 1.6 shows AUROC for
predicting the main outcome, dementia or dead with dementia, in 2, 4, 6, 8 and 10 years. As
shown, FEM TICS27’s predictive performance is highest for 2-year prediction and decreases as
the prediction timeframe increases. The 10-year AUROC for the full sample is still above 0.9.
External Comparison
Table 1.7 shows FEM’s AUROC results predicting significant decline (greater than or
equal to 3 points) in TICS27 in two years and its comparison to the COMPASS model’s
performance in MCI subjects. The AUROC for FEM on predicting significant decline in TICS27
in two years is 0.722 for people with MCI in 2006. FEM’s performance is better than Base
COMPASS (AUROC 0.641), which is a machine learning model that additionally uses APOE
genotype information. Advanced COMPASS (AUROC 0.814) outperforms FEM, although
Advanced COMPASS includes information not only on APOE genotype but also on
neuropsychological tests and validated composite scores for memory and executive functions.

21
Discussion
We extended the FEM microsimulation model to include a widely used cognitive test
TICS27 based on nationally representative HRS data, using individual-level information on
demographics (age, gender, race and ethnicity, education), chronic disease indicators (heart
disease, stroke, cancer, hypertension, diabetes, lung disease, heart attack), employment, smoking
status, marital status and body mass index. TICS27 can be translated to MMSE using a validated
crosswalk[51], and MMSE is commonly used in selecting eligible patients for AD DMT clinical
trials. The FEM TICS27 model can accurately depict cognitive trajectory across the full AD
continuum, and this is an important step toward economic evaluation for AD DMTs. The FEM
TICS27 model can be used to estimate the future burden and long-term value of treatments of
cognitive decline in the US. It also provides a 10-year risk score for dementia based on
information attainable from a telephone-based survey.
To our knowledge, most disease simulation models for cognitive decline and dementia
are not validated or are not validated using an unbiased approach like five-fold cross validation.
Given the limited access to data and adoption of different cognitive function tests, validation of
modeling methods is a general challenge in the area of AD.[17] We are not aware of data sources
other than the HRS using TICS27 as cognitive function measurement that are available as
independent datasets for external validation. Adoption of five-fold cross validation is an
improvement compared to most existing economic evaluation models for AD in this situation to
validate model performance. Using the same data for model estimation and validation can lead to
an upward bias in model performance estimates due to overfitting. Although k-fold cross
validation is one of the most widely used data resampling methods to estimate the true prediction
error of models and to tune model parameters in risk prediction models, it is rarely used in

22
validation for disease simulation models. Cross validation enables us to assess the
generalizability of a model without using a new independent dataset, which is critical to
obtaining unbiased results for model prediction performance.[46,47]
The FEM TICS27 model demonstrates excellent internal validity: the TICS27
distribution and 10-year change in cognitive status generated by FEM simulation closely matches
observed HRS data, and the AUROCs are larger than 0.85 for dementia prediction. For
prediction of significant decline in MCI patients, FEM’s performance is comparable to one of the
best-performing models reported in the literature.[52]
FEM TICS27’s performance on two individual-level outcomes, long-term prediction of
dementia and short-term prediction of cognitive decline, is comparable to or exceeds the
performance of existing models. Previously published studies reported AUROCs between 0.6
and 0.78 for prediction of AD/dementia within 3-40 years[53], which is lower than the AUROC
of 0.904 reported in this study for prediction of 10-year dementia or dead with dementia. For
predictions of significant decline of cognitive test scores in two years, FEM TICS27’s
performance is comparable with one of the best-performing models, COMPASS, which won the
Dialogue for Reverse Engineering And Method (DREAM) Alzheimer’s Disease Big Data
challenge. One of the drawbacks of COMPASS is that it requires knowledge of detailed clinical
information and APOE genotype, and is based on a selective disease registry, the ADNI
database.[52] FEM TICS27 on the other hand solely relies on demographic and survey-derived
variables, and can provide nationally representative estimates. Thus, FEM TICS27 demonstrated
its predictive accuracy for both long-term dementia status and short-term cognitive decline
outcomes. The increased performance of FEM over other models is likely because it utilizes

23
information on individual characteristics and behavior, like smoking, widowhood, and disease
history.
On the other hand, Advanced COMPASS is better at predicting outcomes for people with
MCI, which is especially hard to predict because of heterogeneity in the prognosis and the
disease progression with respect to patient characteristics.[54] For this specific group, additional
clinical and genotype information significantly improves prediction performance.[52] Future
development of FEM TICS27 with genotype and biomarker variables, which are both available
for a subsample of the HRS, will possibly further improve its performance for people with MCI.
Additionally, future applications of FEM TICS27 will include analyses of differences in
cognitive trajectory by education, initial cognitive status, and race and ethnicity. The model can
also be implemented in microsimulations for other countries.
With aducanumab approved as the first AD DMT and more DMTs in the development
pipeline, the future looks promising. Though crucial, availability of DMT is only one step in
enhancing cognitive function in elderly population. Demonstrating value of treatment and
identification of people at risk of cognitive impairment are two very important components as
well. FEM microsimulation could help with these. Understanding the long-term impact of AD
DMTs beyond direct medical expenditure is crucial to its value demonstration.[15] As
randomized controlled trials can only generate short-term evidence on the efficacy of AD DMTs,
to demonstrate their long-term value, projection models are needed to estimate future benefits.
Based on nationally representative data and modeling a large spectrum of cognitive functioning,
FEM TICS27 is a useful tool to assess the long-term impact of these future changes on the US
healthcare system. TICS27 can be translated to MMSE using a validated crosswalk[51], and
MMSE is commonly used for selecting patients eligible for AD DMT clinical trials[55,56].

24
Clinical Dementia Rating (CDR) is also commonly used in AD DMT clinical trials. There is no
existing crosswalk between CDR measurements and TICS27. CDR can be used characterize
patients into mild, moderate and severe stages of AD dementia, which TICS27 is not able to. In
Chapter 2, we developed models for CDR to further improve FEM’s ability to model AD DMT’s
treatment effect and its long-term value. Besides accurately modeling cognitive decline, FEM
tracks other relevant outcomes, such as functional limitations, physical health, formal and
informal care utilization, nursing home living, and medical care costs. FEM is able to provide
much-needed evidence on long-term value of AD DMT on a broad range of outcomes. The
advantage of FEM TICS27 is its high prediction accuracy using only information from a
telephone-based survey.
We present FEM TICS27’s model structure, variables, data sources and conduct
validation of its simulation outcomes against observed HRS data. We show that FEM TICS27
model can accurately predict cognitive test scores covering the full AD disease continuum for a
nationally representative sample over a 10-year period. As TICS27 can be translated to MMSE,
which is commonly used in clinical settings for AD, this paper demonstrated FEM TICS27’s
ability to accurately predict cognitive decline, which is an important step toward long-term
modeling and economic evaluation of AD DMTs.

25
Table 1.1. Marginal effect from FEM TICS27 and mortality transition model.
Variables TICS27 Coefficient
(Std. Err)
Mortality Coefficient
(Std.Err)
Two-year lag TICS score 0.0915 (0.0015)***
Four-year lag TICS score 0.0955 (0.0017)***
Non-Hispanic Black -0.2343 (0.0116)*** 0.0028 (0.0023)
Hispanic -0.1502 (0.0144)*** -0.0094 (0.0031)**
Did not graduate high school -0.1361 (0.0131)*** -0.0002 (0.0022)
At least some college 0.1967 (0.0092)*** -0.0072 (0.0019)***
Male -0.0884 (0.0086)*** 0.0177 (0.0019)***
Slope of age spline before age 65 -0.0009 (0.0013) 0.0025 (0.0004)***
Slope of age spline ages 65-74 -0.0173 (0.0015)*** 0.0029 (0.0003)***
Slope of age spline ages 75 and older -0.0353 (0.0013)***
Slope of age spline ages 75-84 0.0040 (0.0003)***
Slope of age spline ages 85 and older 0.0067 (0.0004)***
Ever diagnosed with heart problems -0.0126 (0.0108) 0.0131 (0.0020)***
Ever diagnosed with stroke -0.1288 (0.0165)*** 0.0110 (0.0023)***
Ever diagnosed with cancer 0.0008 (0.0117) 0.0334 (0.0020)***
Ever diagnosed with hypertension -0.0355 (0.0088)*** 0.0097 (0.0019)***
Ever diagnosed with diabetes -0.0693 (0.0105)*** 0.0155 (0.0020)***
Ever diagnosed with lung disease -0.0430 (0.0147)** 0.0294 (0.0023)***
Heart attack in past 2 years -0.0750 (0.0334)* 0.0049 (0.0047)
Working for pay 0.0943 (0.0100)***
Widowed -0.0298 (0.0121)* 0.0055 (0.0022)*
Ever smoked -0.0426 (0.0083)***
Verified ADOD/MCI ever -0.7105 (0.0155)***
Delta age -0.1374 (0.0208)***
Lag log BMI below 30 0.1314 (0.0350)***
Lag log BMI above 30 -0.0090 (0.0441)
Difficulty with one IADL 0.0162 (0.0028)***
Difficulty with two or more IADLs 0.0483 (0.0031)***
Difficulty with one ADL 0.0222 (0.0025)***
Difficulty with two ADLSs 0.0342 (0.0033)***
Difficulty with three or more ADLs 0.0603 (0.0028)***
Current smoker 0.0187 (0.0027)***
Diagnosed with heart problems by age
50
0.0051 (0.0067)
Diagnosed with stroke by age 50 -0.0107 (0.0167)
Diagnosed with cancer by age 50 -0.0083 (0.0057)
Diagnosed with hypertension by age
50
0.0027 (0.0042)
Diagnosed with diabetes by age 50 0.0142 (0.0034)***
Diagnosed with lung disease by age 50 -0.0278 (0.148)
Ever smoked at age 50 0.0061 (0.0022)**
Current smoker at age 50 0.0174 (0.0024)***
Ever diagnosed with congestive heart
failure
0.0259 (0.0028)***
Notes: *, significant at a=0.05; **, significant at a=0.01; ***, significant at a=0.001.

26
Table 1.2. Characteristics of the 2006 Health and Retirement Survey (HRS) respondents.
2006 HRS 2006 HRS w/ full 10-
year follow-up
P>|t|
Characteristics Mean
N 15,764 13,106
Age 66.91 67.34 0.000***
Race and Ethnicity
Non-Hispanic White 0.811 0.814 0.041*
Non-Hispanic Black 0.091 0.093 0.056
Hispanic 0.073 0.071 0.031*
Education
Less than high school 0.181 0.183 0.168
High School 0.347 0.348 0.654
Some college and above 0.472 0.469 0.165
TICS score 15.57 15.54 0.205
Verified cognitive status
Dementia 0.012 0.013 0.203
Mild Cognitive Impairment 0.089 0.090 0.203
Normal 0.900 0.898 0.203
Disease status
Heart disease ever 0.215 0.223 0.000***
Stroke ever 0.075 0.080 0.000***
Cancer ever 0.131 0.139 0.000***
Hypertension ever 0.517 0.525 0.001***
Diabetes ever 0.179 0.186 0.000***
Lung disease ever 0.083 0.089 0.000***
Heart attack 0.017 0.016 0.493
Work for pay 0.429 0.417 0.000***
Widowed 0.177 0.186 0.000***
Smoking ever 0.571 0.577 0.006**
Notes: *, significant at a=0.05; **, significant at a=0.01; ***, significant at a=0.001.

27
Table 1.3. Distribution comparison between HRS respondents and FEM simulation, 2006-
2016.
Ages 53+ in 2006
1st 5th 10th 25th 50th 75th 90th 95th 99th Mean Diff
(Mean)
2006 3 7 9 13 16 19 21 22 25 15.57
2016 HRS
(observed)
3 7 9 12 15 18 21 22 24 14.95 -0.62
2016 FEM
(predicted)
3 6 9 12 15 18 20 22 24 14.93 -0.64
Ages 70+ in 2006
1st 5th 10th 25th 50th 75th 90th 95th 99th Mean Diff
(Mean)
2006 2 5 7 11 14 17 19 20 23 13.55
2016 HRS
(observed)
1 4 6 9 12 15 18 19 22 12.04 -1.51
2016 FEM
(predicted)
1 4 5 9 12 15 18 19 21 11.88 -1.67
Note: HRS stands for Health and Retirement Study, which is the observed survey data. FEM stands for Future
Elderly Model, which generates the simulated results.

28
Table 1.4. 10-year change in distributions of cognitive status based on TICS score, HRS and
FEM.
Cognitively
Normal
2016 (%)
MCI
2016 (%)
Dementia
2016 (%)
Dead w/o
dementia
2016 (%)
Dead w/
dementia 2016
(%)
Total
HRS FEM HRS FEM HRS FEM HRS FEM HRS FEM
Cognitively
Normal
2006
56.7 58.5 7.4 6.4 0.8 1.0 23.0 22.5 1.8 1.4 89.8
MCI
2006
0 0 2.5 2.6 0.9 1.2 4.0 3.0 1.7 2.2 9.0
Dementia
2006
0 0 0 0 0.2 0.3 0 0 1.0 0.9 1.3
Total 56.7 58.5 9.9 9.0 1.9 2.5 27.0 25.5 4.5 4.5 100
Note: HRS stands for Health and Retirement Study, which is the observed survey data. FEM stands for Future
Elderly Model, which generates the simulated result.

29
Table 1.5. Area under the receiver operating characteristics curve (AUROC) for predicting
dementia from 5-fold cross-validation.
Sample Sample Size Outcome AUROC
Full sample 13,106 Dementia or dead w/
dementia
0.904
Dementia (conditional on
alive)
0.868
MCI or worse or dead w/
MCI or worse
0.897
MCI or worse (conditional
on alive)
0.826
Verified MCI
in 2006
1,512 Dementia or dead w/
dementia
0.720
Dementia (conditional on
alive)
0.705
Non-Hispanic
White
9,922 Dementia or dead w/
dementia
0.891
Dementia (conditional on
alive)
0.838
Non-Hispanic
Black
1,824 Dementia or dead w/
dementia
0.906
Dementia (conditional on
alive)
0.814
Hispanic 1,107 Dementia or dead w/
dementia
0.881
Dementia (conditional on
alive)
0.827
Less than high
school
2,983 Dementia or dead w/
dementia
0.866
Dementia (conditional on
alive)
0.788
High School 4656 Dementia or dead w/
dementia
0.856
Dementia (conditional on
alive)
0.821
Some college or
above
5,467 Dementia or dead w/
dementia
0.880
Dementia (conditional on
alive)
0.806
Age 65+ in
2006
6,272 Dementia or dead w/
dementia
0.875
Dementia (conditional on
alive)
0.828

30
MCI or worse or dead w/
MCI or worse
0.861
MCI or worse (conditional
on alive)
0.766
Ever had
diabetes before
2006
2,681 Dementia or dead w/
dementia
0.884
Dementia (conditional on
alive)
0.825
Ever had
stroke before
2006
1,239 Dementia or dead w/
dementia
0.875
Dementia (conditional on
alive)
0.849
Ever had
hypertension
before 2006
7,445 Dementia or dead w/
dementia
0.906
Dementia (conditional on
alive)
0.859
Ever had heart
disease before
2006
3,312 Dementia or dead w/
dementia
0.8947
Dementia (conditional on
alive)
0.8326

31
Table 1.6. Area under the receiver operating characteristics curve (AUROC) for predicting
dementia or dead with dementia for 2, 4, 6, 8 and 10 years.
2 years (2008) 4 years (2010) 6 years (2012) 8 years (2014) 10 years (2016)
Full sample 0.995 0.955 0.941 0.920 0.904
Age 65+ in
2006
0.993 0.946 0.929 0.902 0.883
Verified MCI
in 2006
0.961 0.812 0.783 0.740 0.720

32
Table 1.7. Two-year TICS27 significant decline in MCI subjects and comparable results
from COMPASS.
FEM TICS27
MCI
Base COMPASS
MCI
Advanced
COMPASS MCI
AUROC
(random=0.5)
0.722 0.641 0.814
Notes: FEM TICS27 is the model developed and validated in this paper. COMPASS is one of the best-performing
models reported in the literature, which won the Dialogue for Reverse Engineering And Method (DREAM)
Alzheimer’s Disease Big Data challenge.

33
Figure 1.1. Distribution comparison between HRS and FEM, 2006-2016.
Panel a. Ages 53+ in 2006
Panel b. Ages 70+ in 2006

34
Figure 1.2. Receiver operating characteristics curve for predicting dementia in 10 years.
Panel A. 2006 HRS with full 10-year full follow-up sample.

35
Panel B. MCI in 2006 sample.

36
Appendix Table 1.1A. Proxy interview cognitive impairment rating from Health and
Retirement Study, 2006-2016.
2006 2008 2010 2012 2014 2016
No reason to think the respondent has
any cognitive limitations
608 522 636 480 429 394
The respondent may have some
cognitive limitations but could
probably do the interview
92 90 91 81 73 58
The respondent has cognitive
limitations that prevent him/her from
being interviewed
560 528 655 584 547 489

37
Appendix Table 1.2A. TICS27 and mortality transition model.
Variables TICS27 Coefficient
(Std. Err)
Mortality Coefficient
(Std.Err)
Main effects
Two-year lag TICS score 0.092 (0.001)***
Four-year lag TICS score 0.096 (0.002)***
Non-Hispanic Black -0.253 (0.014)*** 0.026 (0.036)
Hispanic -0.169 (0.019)*** -0.072 (0.050)
Did not graduate high school -0.155 (0.016)*** -0.071 (0.033)*
At least some college 0.197 (0.009)*** -0.089 (0.029)**
Male -0.109 (0.010)*** -0.419 (0.901)
Slope of age spline before age 65 -0.001 (0.001) 0.019 (0.010)
Slope of age spline ages 65-74 -0.017 (0.001)*** 0.031 (0.006)***
Slope of age spline ages 75 and older -0.035 (0.001)***
Slope of age spline ages 75-84 0.045 (0.005)***
Slope of age spline ages 85 and older 0.056 (0.005)***
Ever diagnosed with heart problems -0.013 (0.011) 0.157 (0.032)***
Ever diagnosed with stroke 0.129 (0.016)*** 0.166 (0.035)***
Ever diagnosed with cancer 0.001 (0.012) 0.372 (0.030)***
Ever diagnosed with hypertension -0.036 (0.009)*** 0.094 (0.030)**
Ever diagnosed with diabetes -0.069 (0.010)*** 0.156 (0.031)***
Ever diagnosed with lung disease -0.043 (0.015)** 0.278 (0.035)***
Heart attack in past 2 years -0.075 (0.033)* -0.108 (0.081)
Working for pay 0.094 (0.010)***
Widowed -0.030 (0.012)* 0.039 (0.028)
Ever smoked -0.043 (0.008)***
Verified ADOD/MCI ever -0.710 (0.015)***
Delta age -0.137 (0.021)***
Lag log BMI below 30 0.131 (0.035)***
Lag log BMI above 30 -0.009 (0.044)
Difficulty with one IADL 0.190 (0.043)***
Difficulty with two or more IADLs 0.534 (0.044)***
Difficulty with one ADL 0.222 (0.038)***
Difficulty with two ADLSs 0.363 (0.049)***
Difficulty with three or more ADLs 0.640 (0.040)***
Current smoker 0.179 (0.046)***
Diagnosed with heart problems by age
50
-0.002 (0.113)
Diagnosed with stroke by age 50 0.036 (0.260)
Diagnosed with cancer by age 50 -0.038 (0.063)
Diagnosed with hypertension by age
50
-0.046 (0.063)
Diagnosed with diabetes by age 50 0.097 (0.055)
Diagnosed with lung disease by age 50 -0.138 (0.180)
Ever smoked at age 50 0.083 (0.033)
Current smoker at age 50 0.183 (0.040)***
Ever diagnosed with congestive heart
failure
0.290 (0.044)***
Constant -3.755 (0.590)***
Gender interaction
Male and did not graduate high school 0.046 (0.025) 0.132 (0.050)**
Male and non-Hispanic Black 0.046 (0.023)* -0.017 (0.056)
Male and Hispanic 0.046 (0.027) -0.010 (0.075)

38
Male and at least some college 0.035 (0.043)
Change in slope of age spline for
males before age 65
0.008 (0.015)
Change in slope of age spline for
males ages 65-74
0.004 (0.008)
Change in slope of age spline for
males ages 75-84
-0.006 (0.007)
Change in slope of age spline for
males age 85 and older
0.031 (0.008)***
Male and ever diagnosed with
congestive heart failure
-0.024 (0.063)
Male and ever diagnosed with heart
problem
-0.051 (0.045)
Male and ever diagnosed with stroke -0.097 (0.052)
Male and ever diagnosed with cancer -0.084 (0.043)
Male and ever diagnosed with
hypertension
0.025 (0.043)
Male and ever diagnosed with diabetes -0.013 (0.044)
Male and ever diagnosed with lung
disease
0.095 (0.051)
Male and heart attack within past 2
years
0.198 (0.109)
Male and difficulty with one IADL -0.081 (0.065)
Male and difficulty with two or more
IADLs
-0.030 (0.072)
Male and difficulty with one ADL 0.049 (0.058)
Male and difficulty with two ADLs 0.025 (0.076)
Male and difficulty with three or more
ADLs
0.034 (0.065)
Male and current smoker 0.103 (0.065)
Male and widowed 0.027 (0.049)
Male and diagnosed with heart
problems by age 50
0.080 (0.148)
Male and diagnosed with stroke by age
50
-0.218 (0.448)
Male and diagnosed with cancer by
age 50
-0.143 (0.146)
Male and diagnosed with hypertension
by age 50
0.199 (0.108)
Male and diagnosed with diabetes by
age 50
0.070 (0.084)
Male and diagnosed with lung disease
at age 50
-0.340 (0.399)
Male and ever smoked by age 50 -0.031 (0.049)
Male and current smoker at age 50 -0.006 (0.053)
Notes: *, significant at a=0.05; **, significant at a=0.01; ***, significant at a=0.001.

39
Chapter 2 Extrapolating AD DMT Clinical Trial Results to Longer-term Value
Assessment
Introduction
Alzheimer’s disease (AD) imposes an increasing burden on the United States society.
Currently, more than 6 million Americans are living with AD dementia; by 2050, this number is
projected to rise to nearly 13 million if effective interventions are not found.[57] On the bright
side, exciting progress is happening at the therapeutics development front. Aduhelm
(aducanumab), the first disease-modifying treatment (DMT) for AD, was approved by the US
Food and Drug Administration (FDA) in Jun 2021.[21] More recently, Leqembi (lecanemab),
another DMT for AD, became the first traditionally approved treatment that addresses the
underlying biology of AD and changes the course of the disease in a meaningful way for people
in the early stages.[58] More potential DMTs are in the pipeline; in 2022, DMTs represent 83.2%
of the total number of AD drugs in trials, with the other 16.8% being symptomatic agents.[10]
While clinical trials evaluate treatments’ effectiveness in terms of clinical endpoints, AD
DMTs long-term impact on broader outcomes for patients, their caregivers and society remain
unknown. Take lecanemab’s pivotal trial Clarity AD for example, its primary end point (change
from baseline at 18 months in the score on the Clinical Dementia Rating – Sum of Boxes)
comprehensively assesses lecanemab’s effectiveness in slowing down both cognitive and
functional disability in a limited time frame, while long-term outcomes that will assist in health
policy decision making, like change in quality-adjusted life years, caregiver burden, and medical
spending, etc. are not available from clinical trial results.
The number of initiatives to collect long-term real-world data for value assessment of AD
DMTs is increasing. For example, initiatives like the US National Alzheimer’s Coordinating

40
Center (NACC) and the Swedish dementia register (Svedem) are attempting to close the
evidence gap in long-term value assessment for AD DMTs. In the near-term, modeling may be
the necessary suboptimal approach to extrapolate short-term clinical trial results and provide
meaningful estimates of AD DMTs’ longer-term value on the population-level.[15] Ideally,
models used for long-term value assessment of AD DMT will cover meaningful outcomes as
well as the cost per QALY value assessment measure. To broaden the view of what constitutes
value in health care, Lakdawalla et al developed the value flower framework. The value flower
framework includes value elements like productivity and family spillovers that are especially
relevant to AD due to its nature of disease.[59,60] Outcomes like memory, patients’
independence and identity, and mental health are found to be important to multiple stakeholders
(patients, caregivers, and healthcare provider) in AD. AD DMTs’ impact on these outcomes
should be evaluated alongside its economic value to provide important additional angles that will
help value demonstration to a broader audience. Given the nature of AD, many types of costs
significantly related to it are hidden and may not be considered in a traditional value assessment
framework’s net cost. Traditional value assessments usually focus on direct medical costs, while
in the case of AD, indirect costs like costs of informal caregiving account for a significant
portion for its total societal burden. It is estimated that 41% of AD’s total societal burden of $307
billion in the US in 2010 is due to informal care costs.[5,12]
Over the years, many studies aim to evaluate the value of a hypothetical DMT that could
slow down AD disease progression. For example, Zissimopoulos et al estimated using a
microsimulation approach that delaying AD dementia onset by one year has an economic value
of $187,227 per patient, and the value per patient of a delay by three years is $355,222.[33]
However, studies trying to assess value based on effectiveness data from real AD DMTs’ clinical

41
trials are limited. As we now have two AD DMTs on the market, aducanumab and lecanemab,
both with efficacy data shown in clinical trial results, we are able to extrapolate short-term
clinical endpoints to longer-term meaningful outcomes to assess the value of real AD DMTs on
the market.[7,8]
To provide reliable evidence on the value of novel AD DMTs, in this work we applied a
microsimulation model to extrapolate lecanemab’s Clarity AD trial results to policy-relevant
outcomes and with a longer follow up period. We chose to focus on lecanemab in this study, as it
is the first AD DMT to receive traditional approval from the US FDA. The emergence of AD
DMTs raise the questions of such DMTs impact beyond short-term clinical endpoints, and we
believe our study provides a feasible evaluation framework using lecanemab as a meaningful
example.
Methods
The Future Elderly Model (FEM) overview
The Future Elderly Model (FEM) is a microsimulation model of health and economic
outcomes for the US population aged 51 and older. Here we summarize FEM’s core functions;
technical details are described in a technical appendix.[32] FEM uses first-order Markov
transition models to simulate individuals’ aging progress. It captures outcomes including health
conditions, functional status, earnings and employment status, living in nursing home, medical
costs, informal help hours, and mortality. As demonstrated in existing studies, being a
microsimulation model based on nationally representative data for the US that covers the
complete continuum of AD, the Future Elderly Model is an ideal model for longer-term value
assessment on the population level.[61] FEM has been used in another economic evaluation

42
study for DMT for AD [62] and in many studies answering important policy questions in aging
and dementia. [33,34]
In this study, we adapted the FEM to assess the impact of lecanemab on outcomes beyond
the scope of the Clarity AD clinical trial and with a longer time horizon (20-year). We also
present results to show the largest possible effects from AD DMTs similar to lecanemab. Here we
try to assess the health and economic benefits of lecanemab, not to perform a cost-effectiveness
analysis, thus we do not consider the drug cost of lecanemab. Outcomes of our health and
economic benefits assessment include living in nursing home, activities of daily living
(ADLs)/instrumental activities of daily living (IADLs), self-reported memory, self-reported
health, working for pay, earnings, quality-adjusted life years (QALYs) (measured using the
Health Utilities Index 3, HUI3), informal help hours, medical costs, and caregiver disutility.
QALYs, informal help hours, medical costs, caregiver disutility, and earnings are reported in the
health and economic benefits calculation, other outcomes are reported as trajectory differences
between treatment and control. Functional limitation, measured by ADLs and IADLs, is an
important domain of Alzheimer’s disease [63], as it often corresponds to how much help and care
AD patients may require. ADLs refer to activities related to taking care of one’s own body. In
FEM, ADLs include bathing, dressing, eating, getting in/out of bed and walking across a room.
IADLs refer to activities to support daily life within the home and community that often require
more complex interactions than those used in ADLs. In FEM, IADLs include using the phone,
managing money, and taking medications.[64,65]
In our evaluation, we also considered amyloid-related imaging abnormalities (ARIA)
adverse event from lecanemab treatment and its impact on treatment cessation, costs and
QALYs. Treatment cessation due to ARIA will subsequently affect all outcomes mentioned

43
above, and experiencing ARIA will incur management costs and negatively impact QALYs. We
utilized the value of percentage of patients experiencing ARIA and percentage with treatment
cessation due to ARIA reported from Clarity AD trial.[55] We utilized value for ARIA’s impact
on costs and QALY from the literature.[66] Detailed assumptions and parameter values can be
found in the Appendix and Appendix Table 2.1A.
Data and Measures
The primary endpoint of Clarity AD is the change in Clinical Dementia Rating-Sum of
Boxes (CDR-SB), which is a commonly used primary outcome measure that comprehensively
assesses both cognitive and functional disability, especially in early, predementia stages of AD
patients.[55,67] CDR-SB assesses three domains of cognition (memory, orientation,
judgment/problem solving) and three domains of function (community affairs, home/hobbies,
personal care). In our simulation, CDR-SB is the only outcome directly affected by lecanemab
treatment. The scores of the six domains (from 0 to 3) can be summed to generate a CDR-SB
score, ranging from 0 to 18. A related measure, global CDR, can be derived from these six
domains and categorize disease stages into normal, mild cognitive impairment (MCI), mild,
moderate and severe dementia, respectively.[68] In our study, we used a crosswalk [69] to
generate global CDR from simulated CDR-SB, in order to calculate medical costs, caregiver
disutility, and earnings outcomes that depend on disease stages.
FEM uses data from the Health and Retirement Study (HRS), a biennial nationallyrepresentative longitudinal survey in the population with more than 37,000 respondents over age
50 in the US.[35] In this study, our simulation cohort was from HRS in 2016, selected based on
inclusion and exclusion criteria in the Clarity AD clinical trial where possible.[55] The inclusion

44
criteria are: (1) age 50 to 90; (2) a Mini-Mental State Examination (MMSE)1 score between 22
and 30; (3) a Body Mass Index (BMI) between 17 and 35; and (4) a predicted global Clinical
Dementia Rating (CDR) of 0.5 or 1. Exclusion criteria are: (1) recent stroke; (2) missing
Telephone Interview for Cognitive Status (TICS) score. We then balanced the selected cohort to
more closely resemble the clinical trial cohort based on age (mean and variance), CDR-SB
(mean and variance), MMSE (mean), gender, race and ethnicity, and a global CDR mix (80%
global CDR of 0.5 and 20% global CDR of 1). Descriptive statistics of the simulation starting
cohort can be found in Table 2.1.
Models for CDR-SB
We estimated (1) an imputation model and (2) a transition model for CDR-SB, using
Alzheimer’s Disease Neuroimaging Initiative (ADNI) data.[70,71] ADNI is a registry of patients
with different disease stages (cognitively normal, MCI and AD dementia), and includes CDR-SB
scores and subjects’ amyloid-beta positivity status. Amyloid-beta positivity is an important
characteristic for selecting subjects with a similar CDR-SB trajectory as the Clarity AD clinical
trial cohort. We required ADNI subjects (1) with a minimum of 24 months of follow-up, (2) with
an initial global CDR score of 0.5 or 1 and (3) with positive initial amyloid-beta. This resulted in
a sample of 267 ADNI subjects that we used to estimate the CDR-SB transition and imputation
models for our FEM simulation.
1 MMSE is derived from the Telephone Interview for Cognitive Status (TICS) score available from HRS, using a
crosswalk between MMSE and TICS27 developed by Hlávka et al. (2022).[51]

45
(1) Imputation model:
Because CDR-SB scores are not available in our simulation cohort from the HRS, we
estimated an imputation model for CDR-SB using ADNI data. Imputed CDR-SB scores were
subsequently used as a covariate in downstream outcomes model estimations. We also used this
CDR-SB imputation model to assign CDR-SB accordingly with their global CDR category to the
simulation starting cohort. The CDR-SB imputation model was estimated with OLS regression,
including covariates age, age squared, male, race, education, TICS, marital status. The regression
coefficients are shown in Appendix Table 2.3A.
(2) Transition model:
To estimate cognitive decline in the FEM simulation, we developed a transition model for
CDR-SB using ADNI data. FEM transition models generally are a mixture of continuous, binary,
and categorical outcomes, with a timescale that mimics the two-year structure of the HRS.
Transition models are used to derive the status of CDR-SB and other target outcomes at the end
of the two-year cycle. The CDR-SB transition model was estimated with ordinary least squares
(OLS) regression, and includes splines of lag CDR-SB, age, gender and race as covariates. The
regression coefficients are shown in Appendix Table 2.2A.
Downstream Outcome Models
Downstream outcomes include living in nursing home, ADLs/IADLs, self-reported
memory, self-reported health, working for pay, earnings, QALYs, informal help hours, medical
costs, and caregiver disutility. In our study, these outcomes are determined by the CDR-SB score
and other characteristics of the respondents from the current cycle of the simulation. In our
model, mortality is also indirectly impacted by CDR-SB through its impact on ADLs/IADLs.

46
In Table 2.2 we show all the downstream outcomes and their corresponding models.
CDR-SB is only observed in ADNI data while other downstream outcomes are only observed in
HRS data. Thus, we first imputed for CDR-SB in HRS, then estimated models of observed
downstream outcomes as a function of imputed CDR-SB.
With the exception of QALYs, earnings, medical costs, and caregiver disutility, we used
full HRS data from 1998 until 2018 to estimate downstream outcomes. QALYs are measured in
HUI3, and since the latter is only observed in 2000 in HRS, we used data from this year only.
Earnings are dependent on simulated global CDR and working for pay status. For those with
simulated global CDR scores of 1, 2, or 3 (dementia) or those not working for pay, we assumed
no earnings. For others, we assigned earnings dependent on disease stage (normal or MCI),
based on HRS data from 1998 to 2016. For medical costs and caregiver disutility, we conducted
post-simulation calculations based on simulated global CDR and data from the literature.[72,73]
Covariates include imputed CDR-SB, lag of the outcome itself, lag of age, gender, marital status,
race and ethnicity, and education. A more detailed description of all downstream outcomes and
their regression coefficients and stage-dependent input values are shown in Appendix and
Appendix Table 2.4A.
We report results for three scenarios (Table 2.3). All three scenarios have 20 years of
observation. The 48-Tx scenario describes our main results; it shows 48 months of treatment
followed by 16 years of observation for a total of 20 years of observation. We believe this
scenario is a reasonable extension of the Clarity AD clinical trial effects based on clinical expert
opinion and considerations for our model assumptions. We also report one scenario with 18
months of treatment, which keep the treatment duration as in Clarity AD trial. The 18-Tx
scenario describes 18 months of treatment with 20 years of total observation. The two scenarios

47
mentioned above assume treatment effect of lecanemab as a 27% reduction in the change of
CDR-SB, based on Clarity AD clinical trial outcomes. Finally, we report an upper boundary 48-
Tx scenario, where we assumed the treatment effect as a 50% reduction in the decline of CDRSB. This scenario shows a hypothetical DMT similar to lecanemab with the largest possible
effect, based on opinions from a clinical expert. We compared results from all scenarios to a
control scenario in which simulants receive standard care.
The outcomes CDR-SB, living in a nursing home, function limitations in ADLs/IADLs,
self-reported memory, self-reported health and working for pay, are reported as trajectories
compared to standard care. For QALYs, informal help hours, medical costs, caregiver disutility,
and earnings, we report cumulative results and absolute differences between treatment and
control groups and convert these differences to dollar values. QALYs are valued at $150,000 per
QALY [74], help hours at $28.16 per hour [75], and all costs are in 2022 USD and discounted at
3% annually.
Key Model Assumptions and Accounting
In the 48-Tx and 18-Tx scenario, we modeled the treatment effect of lecanemab as a 27%
reduction in the change of CDR-SB, based on Clarity AD clinical trial outcomes. In upper
boundary 48-Tx scenario, we modeled the treatment effect as a 50% reduction in the change of
CDR-SB, to show the largest possible effects from AD DMTs similar to lecanemab. Our model
sets up an indirect effect on mortality from CDR-SB via ADLs/IADLs. Our model also assumes
treatment continuation when patients progress into moderate/severe dementia, which is
consistent with the clinical trial procedure. We interpolated outcomes at each six-month period in
the two-year simulation cycle, and mortality was assigned at the middle of a two-year cycle.

48
Results
In Figure 2.1, we show outcome trajectories for the 48-Tx scenario. Treatment with
lecanemab for 48 months improves outcomes across board compared to the control scenario. For
all outcomes, we see large improvement over control during the first 10 years after treatment
initiation, with the peak percentage improvement happens after 4 years of treatment initiation,
which is the end of treatment period. Outcomes’ trajectory under treatment gradually converges
with the control trajectory during the next 10 years. We see the largest percentage reduction in
living in a nursing home, compared to the control group, the 48-Tx scenario results in a 40.0%
reduction in prevalence (0.085 vs. 0.142) after 4 years of treatment initiation. Functional status
outcomes, both those concerning basic self-care (ADLs) and those concerning independent living
in a community (IADLs), also show decent improvements. Prevalence of the most severe basic
self-care impairment (limitations in three or more ADLs) is reduced by 18.5% (0.164 vs. 0.201),
and 12.5% (0.386 vs. 0.441) more people are living free of any ADLs limitations at 48 months
with lecanemab treatment. For the ability to live independently in a community, the most severe
IADLs impairment (limitations in two or more IADLs) is reduced by 23.2% (0.253 vs. 0.329),
and 16.6% (0.413 vs. 0.495) more people are living free of any IADL impairments at 48 months
with lecanemab treatment. The results for all scenarios are listed in Appendix Table 2.6A.
In Figure 2.2, we show the patient disposition across different levels of cognitive status
or mortality in the population for control and 48-Tx scenario, observed for 20 years. Patients
treated with lecanemab were estimated to spend longer time alive during the 20-year observation
period. They are also estimated to spend longer time before transitioning to more severe stages of
dementia. The use of lecanemab increases time spent in any alive status by 0.22 years, from
10.05 in the control group to 10.27 in the treatment group. In control group, patients progress to

49
moderate or severe dementia in 4.91 years on average, while patients treated with lecanemab
takes an additional 0.68 years to progress. Patients treated with lecanemab spend 6.98 years
before progression to severe dementia stage on average, while patients in the control group spend
6.28 years.
In Table 2.4, we compare cumulative outcomes between the control and 48-Tx, 18-Tx,
and upper boundary 48-Tx scenarios. Table 2.4 Panel a shows results for the 48-Tx scenario,
taking together QALYs, earnings, informal help hours, medical costs, and caregiver disutility, the
value of DMT is estimated at $158,000 for 48 months of lecanemab treatment observed for 20
years. Saving in medical costs ($67,100) accounts for most of lecanemab’s value, followed by
the relative decrease in informal help hours ($59,700) and increase in AD patients’ QALYs
($17,300). About 40% of the economic value gained from lecanemab is accrued to caregiver. In
Table 2.4 Panel b we show the cumulative outcomes value for the 18-Tx scenario. With 18-
month of lecanemab treatment, same as in Clarity AD clinical trial, the cumulative value
observed for 20 years is $60,400. In Table 2.4 Panel c we show the cumulative outcomes value
for the hypothetical upper boundary DMT with 50% reduction in the decline of CDR-SB, instead
of the 27% reduction for lecanemab. As the treatment effect is about twice of lecanemab, the
accumulated value also nearly doubles that of lecanemab ($335,900 vs. $158,200).
Discussion
We extrapolated clinical findings from the Clarity AD trial to policy-relevant outcomes
and a longer follow up period. We found 48-month of lecanemab treatment results in meaningful
improvements in outcomes like living in nursing home, limitations in three or more ADLs,
limitations in two or more IADLs during a 20-year period. Patients treated with lecanemab were

50
estimated to spend 0.22 more years alive during the 20-year observation period. They are also
estimated to spend longer time before transitioning to more severe stages of dementia, spending
0.68 years longer before progression to moderate or severe dementia, and 0.70 years longer
before progression to severe dementia. Considering cumulative outcomes that could be valued in
monetary term, which include QALYs, earnings, informal help hours, medical costs, and
caregiver disutility, the value of 48-month of lecanemab treatment is estimated to be $158,200.
Consistent with existing studies on disease burden of AD, about 40% of the economic value
gained from lecanemab is accrued to caregiver.[5]
Existing study extrapolating AD DMTs clinical trial results to longer-term outcomes is
limited. A study by Tahami Monfared et al evaluated the lifetime benefits of lecanemab, focusing
on extended life-years and time spent in different disease stages.[76] Tahami Monfared et al
found that treatment with lecanemab increase total undiscounted life-years by 0.96, from 6.38 in
standard of care group to 7.34 in lecanemab treatment group. It also delayed the mean time to
more severe stages of disease; treatment with lecanemab delayed progression to moderate
dementia by 2.95 years, and delayed progression to severe dementia by 2.33 years. Our estimate
of lecanemab’s treatment effect is a more conservative one compared to Tahami Monfared et al’s
findings. Though both based on lecanemab’s treatment effect from Clarity AD trial, we
considered treatment discontinuation due to adverse events while Tahami Monfared et al did not
in the base case analysis. We also improve over existing studies by including outcomes that are
found to be important to multiple stakeholders involved with AD alongside monetary value
assessment.[11] We report additional outcomes of living in a nursing home, ADLs, IADLs, selfreported memory, self-reported health, and working for pay. Among these additional outcomes,
institutionalization, ADLs and IADLs measures the independence of AD patients, which is

51
reported to be one of the most important outcomes in the perspective of AD patients and
caregivers.[16] By assessing lecanemab’s value with these additional outcomes alongside
monetary value, our study provides a more comprehensive picture of the benefits from
lecanemab to AD patients, their family and society.
We considered treatment cessation due to ARIA adverse event in this evaluation. We
adopted the 6.9% treatment cessation rate reported in Clarity AD clinical trial; as clinical trial is
a closely monitored setting, this treatment cessation rate should be considered a conservative
estimation of the treatment cessation rate in the real world. We also considered the monitoring
and management costs incurred by ARIA and its negative impact on patients’ QALY. As a
handful of deaths have occurred from brain bleeding in patients treated with lecanemab, there are
growing concerns over the risk of lecanemab’s adverse events.[77,78] More real-world evidence
on lecanemab’s adverse events profile is required for more refined modeling.
Our study has limitations. First, as in this study we only tried to assess the health and
economic benefits of lecanemab, not to perform a cost-effectiveness analysis, we did not
consider the drug cost of lecanemab nor its actual uptake rate. However, it is highly likely that
the uptake rate for a novel and expensive treatment like lecanemab will differ in different patient
subgroups, and this will subsequently affect its value for patients in minority populations. Racialethnic minority groups usually have lower uptake of novel treatments, due to the larger financial
obstacle they face and their lower level of trust in the healthcare system.[79] For example, study
has found racial-ethnic disparities in uptake of new Hepatitis C drug in the Medicare
population.[80] At the same time, racial-ethnic minority groups bear higher burden of AD, as
Blacks and Hispanics have the highest prevalence of Alzheimer’s disease and related dementias
in the US, based on Medicare fee-for-service data.[81] Extension of this work should consider

52
the heterogeneity of value by patients racial-ethnic profile. Second, though our study includes
value elements like caregivers’ utility, productivity loss and informal caregiving costs in addition
to the traditional patients’ QALYs and medical costs framework, there are still value elements
important to AD that are not considered. To broaden the view of what constitutes value in health
care, Lakdawalla et al reviewed a number of alternative frameworks and synthesized an
overarching approach, often referred to as the value flower.[59] Besides two core elements,
QALYs and net costs, that are traditionally considered, the value flower also include elements
like productivity, family spillovers, adherence improvement, value of knowing, insurance value,
fear of contagion & disease, severity of disease, value of hope, real option value, equity and
scientific spillovers.[60] Since aducanumab and lecanemab are the first two AD DMTs that could
alter the underlying disease progression, they could also present value from elements like
scientific spillovers and insurance value.[11] Since racial-ethnic minority groups bear higher
burden of AD but usually have lower initial uptake of novel treatments, the element of equity
should also be emphasized in the case of AD DMTs.[80,81]
This work extrapolates clinical findings from the Clarity AD trial to policy-relevant
outcomes and a longer follow up period for value assessment of lecanemab. We find lecanemab
shows meaningful positive impact on outcomes like living in nursing home, patients’
independence and memory. The cumulative economic value of 48-month of lecanemab treatment
is $158,200 evaluated at the end of 20 years. Future work will focus on including broader value
elements in evaluation and value assessment in race and ethnicity subpopulation.

53
Appendix
Description of downstream outcomes
1. QALYs measured by HUI3
The quality-adjusted life year (QALY) is a generic measure of disease burden, including both the
quality and the quantity of life lived. One QALY equates to one year in perfect health, and zero
in QALY equates to death. The Health Utilities Index Mark 3 (HUI3) is a system for health status
classification and health related quality of-life scoring. The 33-item HUI3 questionnaire was
included as an experimental module in the 2000 HRS wave.[82]
2. Informal help hours
In HRS, for respondents who indicated any difficulty with ADLs, IADLs and finances, the
respondent is asked if anyone ever helped them with that specific ADL, IADL or finances. A
helper may be a spouse, child, child’s spouse, grandchild, other relatives, other individual, or
member of an organization. The informal help hours measure does not count help from employee
of an institution. Subsequently, the amount of help from each helper is asked and the total is
recorded as number of hours of informal help per year.[64] In our study, we cap informal help
hours per year at 8760 hours per year, which equals 24 hours per day for 365 days.
3. Nursing home residency
HRS reports whether the respondent lives in a nursing home or other health care facility at the
time of the interview.[64]
4. ADL status
ADLs refer to activities related to personal care and mobility. In HRS, ADLs include bathing,
dressing, eating, getting in/out of bed and walking across a room. Our target variable, ADL

54
status, is a four-level categorical variable: no limitations in ADLs, or limitations in one, two, or
three or more ADLs.[64]
5. IADL status
IADLs refer to activities to support daily life within the home and community that often require
more complex interactions than those used in ADLs. In HRS, IADLs include using the phone,
managing money, and taking medications. Our target variable, IADL status, is a three-level
categorical variable: no limitations in IADLs, or limitations in one, two or more IADLs.[64]
6. Self-reported memory
HRS provides a self-reported general rating of memory variable, ranging from 1 for excellent to
5 for poor memory. Our target variable, self-reported memory, is a three-level recoded version,
with the categories: excellent/very good/good, fair, poor.[64]
7. Fair or poor self-reported health
HRS provides a self-reported general health status variable, ranging from 1 for excellent to 5 for
poor health. Our target variable, fair or poor self-reported health, is a recoded binary version of
the two lowest categories, indicating fair or poor self-reported health.[64]
8. Working for pay
HRS provides a binary indicator for whether the respondent is currently working for pay.[64]
9. Earnings
In our study, earnings are dependent on simulated global CDR and working for pay status. For
those with simulated global CDR of 1, 2, or 3 (dementia) or those not working for pay, we
assume no earning. For others, we assign earnings dependent on global CDR (normal or MCI),
based on HRS from 1998 to 2016. Adjusting for inflation with CPI to 2022 USD, the weighted

55
average annual earning for those in normal cognitive status is $68,780, for those in MCI is
$39,450.
10. Medical costs
In our study, medical costs include costs from medical care (AD medication, hospitalizations,
emergency care visits, and outpatient care) and community care (accommodation, and
community care services), based on a study by Gustavsson et al.[72] US-specific costs estimates
from this paper are converted from GBP to USD using real exchange rates provided (0.65103),
then converted from 2007 USD to 2022 USD using medical CPI. As described in Green et al., we
assume the costs of care for MCI patients are half of that of mild AD patients.[83] Medical costs
are dependent on disease-stage and nursing home residency status for those in AD disease stages.
Table 2.4A panel a shows the medical cost estimates that we use as inputs.
11. Caregiver disutility
In our study, we assume a disease-stage dependent caregiver disutility. For those received more
than 12 hours of informal care per day, we assume two caregivers. Otherwise, we assume one
caregiver. The disease-stage dependent caregiver disutility is shown in Table 2.4A panel b. [73]
ARIA adverse events assumptions
We assumed 6.9% of patients will discontinue lecanemab treatment due to adverse events. We
assumed treatment cessation happens halfway through the first year, as evidence shows that the
most serious adverse events occur within the first six months of treatment initiation. Patients
discontinue lecanemab treatment will have CDR-SB trajectory as in control group. We assumed
an annual disutility of -0.14 for patient experiencing symptomatic ARIA. This disutility was
applied for 12 weeks, which is the average duration of ARIA. This disutility represents the

56
disutility of headache, which is the most common symptom among those with symptomatic
ARIA. We assumed patients on lecanemab to receive one brain magnetic resonance imaging
(MRI) every three months during the first year on lecanemab treatment, for monitoring of ARIA.
For patients experienced ARIA, we assumed such patients to receive one brain MRI every 4
weeks during the 12 weeks duration of ARIA adverse event. These assumptions align with
lecanemab clinical trial Clarity AD and existing literature.[55,66] For detailed parameters for
ARIA adverse events, please refer to Appendix Table 2.1A.

57
Table 2.1. Descriptive statistics of simulation starting cohort.
Mean
Age 71.68
Male 0.48
Race and Ethnicity
Non-Hispanic White 0.77
Non-Hispanic Black 0.10
Hispanic 0.12
Education
Less than high school 0.17
Some college and above 0.34
Global CDR
Mild Cognitive Impairment 0.80
Mild Dementia 0.20
Work for pay 0.20
Widowed 0.21
ADL
1 ADL 0.09
2 ADLs 0.04
3+ ADLs 0.06
IADL
1 IADL 0.11
2+ IADL+ 0.11
Disease status
Stroke ever 0.12
Cancer ever 0.18
Hypertension ever 0.72
Diabetes ever 0.34
Lung disease ever 0.21
Heart attack 0.02
Living in nursing home 0.01

58
Table 2.2. Downstream outcomes.
Outcomes Variables Model
QALY QALY measured by HUI3 OLS
ADLs Count of ADLs (0, 1, 2, 3+) Ordered probit
IADLs Count of IADLs (0, 1, 2+) Ordered probit
Self-reported memory Poor, fair, good. Ordered probit
Self-reported health Fair/poor health or not. Probit
Living in nursing home Binary.
Annual informal help
hours
Hours capped at 24 hours per day. Staged, receiving any help (probit),
receiving max help (probit), receiving
in-between help (OLS)
Working for pay Binary. Probit
Medical costs No modeling; based on simulated
global CDR disease stage.
Earning No modeling; based on simulated
global CDR disease stage.
Mortality Binary. Probit. Indirect impact from CDR-SB
via ADLs/IADLs variables.

59
Table 2.3. Three simulation scenarios.
Scenario
Name
Treatment period Observation
period (total)
Treatment effect Compare to clinical trial
18-Tx 18 months 20 years 27% reduction in
the change of CDRSB
Comparable to clinical
trial but with longer
tracking.
48-Tx 48 months 20 years 27% reduction in
the change of CDRSB
Extension of clinical trial
effects with longer
tracking.
Upper
boundary
48-Tx
48 months 20 years 50% reduction in
the change of CDRSB
Extension of clinical trial
effects, for a hypothetical
DMT similar to
lecanemab with the
largest possible effect

60
Table 2.4. Levels and differences in cumulative outcomes between control and the 48-Tx,
18-Tx, and upper boundary 48-Tx scenarios.
Panel a. 48-Tx scenario.
Control 48-month
treatment
Absolute difference
(discounted)
Economics evaluation
(2022 US$, discounted
3%)
QALY (HUI3) 9.124 9.258 0.115 17,300
Medical costs ($) 1,770,930 1,685,496 -67,100 67,100
Earnings 17,198 20,794 3,200 3,200
Informal help hours 56,170 53,622 -2,100 59,700
Caregiver disutility -1.814 -1.727 0.087 11,000
Total ($) 158,200
Panel b. 18-Tx scenario.
Control 18-month
treatment
Absolute difference
(discounted)
Economics evaluation
(2022 US$, discounted
3%)
QALY (HUI3) 9,124 9.183 0.052 7,800
Medical costs ($) 1,770,930 1,740,753 -23,800 23,800
Earnings 17,198 19,019 1,700 1,700
Informal help hours 56,170 55,224 -800 22,600
Caregiver disutility -1.814 -1.779 0.030 4,500
Total ($) 60,400
Panel c. Upper boundary 48-Tx scenario.
Control 48-month
treatment
Absolute difference
(discounted)
Economics evaluation
(2022 US$, discounted
3%)
QALY (HUI3) 9,124 9.443 0.273 41,000
Medical costs ($) 1,770,930 1,593,968 -137,900 137,900
Earnings 17,198 26,994 8,800 8,800
Informal help hours 56,170 50,690 -4,500 126,000
Caregiver disutility -1.814 -1.634 0.148 22,200
Total ($) 335,900
Note: 2022 USD, rounded to the hundredths place.

61
Figure 2.1. Trajectory for outcomes from control and the 48-Tx scenario.
Panel a. CDR-SB.
Panel b. Living in nursing home.
0.00
2.00
4.00
6.00
8.00
10.00
12.00
14.00
16.00
18.00
20.00
0 year 2 year 4 year 6 year 8 year 10 year 12 year 14 year 16 year 18 year 20 year
CDRSB
control 48-Tx
0.000
0.100
0.200
0.300
0.400
0.500
0.600
0.700
0.800
0.900
1.000
0 year 2 year 4 year 6 year 8 year 10 year 12 year 14 year 16 year 18 year 20 year
Nursing Home
control 48-Tx

62
Panel c. ADLs.
0.000
0.100
0.200
0.300
0.400
0.500
0.600
0.700
0.800
0.900
1.000
0 year 2 year 4 year 6 year 8 year 10 year 12 year 14 year 16 year 18 year 20 year
Any ADL
control 48-Tx
0.000
0.100
0.200
0.300
0.400
0.500
0.600
0.700
0.800
0.900
1.000
0 year 2 year 4 year 6 year 8 year 10 year 12 year 14 year 16 year 18 year 20 year
ADL 3+
control 48-Tx

63
Panel d. IADLs.
0.000
0.200
0.400
0.600
0.800
1.000
1.200
0 year 2 year 4 year 6 year 8 year 10 year 12 year 14 year 16 year 18 year 20 year
Any IADL
control 48-Tx
0.000
0.100
0.200
0.300
0.400
0.500
0.600
0.700
0.800
0.900
1.000
0 year 2 year 4 year 6 year 8 year 10 year 12 year 14 year 16 year 18 year 20 year
IADL 2+
control 48-Tx

64
Panel e. Self-reported memory.
Panel f. Self-reported health.
0.000
0.050
0.100
0.150
0.200
0.250
0.300
0.350
0.400
0 year 2 year 4 year 6 year 8 year 10 year 12 year 14 year 16 year 18 year 20 year
Poor Self-reported Memory
control 48-Tx
0.000
0.100
0.200
0.300
0.400
0.500
0.600
0.700
0.800
0.900
1.000
0 year 2 year 4 year 6 year 8 year 10 year 12 year 14 year 16 year 18 year 20 year
Fair or poor self-reported health
control 48-Tx

65
Panel g. Working for pay.
0.000
0.050
0.100
0.150
0.200
0.250
0 year 2 year 4 year 6 year 8 year 10 year 12 year 14 year 16 year 18 year 20 year
Working for pay
control 48-Tx

66
Figure 2.2. Patient disposition of cognitive status under control and the 48-Tx scenario.
0.0000
0.1000
0.2000
0.3000
0.4000
0.5000
0.6000
0.7000
0.8000
0.9000
1.0000
Year
0
Year
1
Year
2
Year
3
Year
4
Year
5
Year
6
Year
7
Year
8
Year
9
Year
10
Year
11
Year
12
Year
13
Year
14
Year
15
Year
16
Year
17
Year
18
Year
19
Year
20
Patient Disposition
MCI control mild control moderate control severe control
dead control MCI Tx mild Tx moderate Tx
severe Tx dead Tx

67
Appendix Table 2.1A. Parameters for ARIA adverse events.
Parameter Value Source
Probability of any ARIA 21.5% van Dyck et al.[55]
Probability of symptomatic ARIA 3.5% van Dyck et al.[55]
Treatment cessation 6.9% van Dyck et al.[55]
Patient disutility -0.14 (annually) Lin et al. [66]
Costs of brain MRI $261.10 Lin et al.[66]

68
Appendix Table 2.2A. Coefficient of CDR-SB transition model.
Coefficient (Std. Err.)
Lag CDR-SB spline – knot at 4* 1.529 (0.134)
Lag CDR-SB spline – knot at 9* 1.801 (0.238)
Age 0.012 (0.019)
Male -0.304 (0.301)
White 0.952 (0.565)
Constant -0.798 (1.590)
* We placed two knots in the lag CDR-SB score, one at the cutoff between normal/MCI and mild AD (4), another at
the cutoff between mild and moderate AD (9), based on the crosswalk from CDR-SB to global CDR, to account for
the observed trajectory of CDR-SB in selected ADNI data. [69]

69
Appendix Table 2.3A. Coefficients of CDR-SB imputation model.
Coefficient (Std. Err.)
Age -0.545 (0.320)
Age squared 0.004 (0.002)
Male -0.501 (0.348)
White -0.602 (0.602)
College 0.381 (0.327)
TICS -0.459 (0.032)
Widowed 0.250 (0.612)
Constant 29.3 (11.9)

70
Appendix Table 2.4A. Downstream outcomes.
Panel a. Regression coefficients for QALYs measured by HUI3 model (OLS).
Coefficient (Std. Err.)
CDR-SB spline – knot at 4* -0.043 (0.007)
CDR-SB spline – knot at 9* -0.041 (0.018)
Age -0.002 (0.001)
Male 0.010 (0.016)
Black -0.022 (0.027)
Hispanic -0.102 (0.031)
Constant 0.958 (0.057)
* We placed two knots in the lag CDR-SB score, one at the cutoff between normal/MCI and mild AD (4), another at
the cutoff between mild and moderate AD (9), based on the crosswalk from CDR-SB to global CDR, to account for
the observed trajectory of CDR-SB in selected ADNI data. [69]
Panel b. Regression coefficients for annual informal help hours models.
b1. Receiving any help (Probit).
Coefficients (Std. Err.)
CDR-SB 0.152 (0.003)
Lag of annual informal help hours 0.000 (0.000)
Lag of age spline – less than 65 -0.006 (0.002)
Lag of age spline – 65 to 74 0.008 (0.002)
Lag of age spline – 75 and over 0.034 (0.002)
Male -0.078 (0.011)
Married 0.038 (0.011)
Black 0.170 (0.014)
Hispanic 0.143 (0.017)
Education – less than high school 0.146 (0.014)
Education – college and above -0.114 (0.012)
Living in nursing home 0.295 (0.049)
Constant -1.389 (0.115)
b2. Receiving maximum help (24 hours per day for 365 days) (Probit).
Coefficients (Std. Err.)
CDR-SB 0.076 (0.015)
Lag of annual informal help hours 0.000 (0.000)
Lag of age spline – less than 65 -0.001 (0.013)
Lag of age spline – 65 to 74 0.016 (0.011)
Lag of age spline – 75 and over 0.015 (0.008)
Male -0.112 (0.063)
Married 0.200 (0.063)
Black 0.266 (0.069)
Hispanic 0.190 (0.084)
Education – less than high school 0.016 (0.070)
Education – college and above 0.001 (0.075)
Living in nursing home -0.186 (0.198)
Constant -2.809 (0.753)
b3. Number of informal help hours received if receiving any help and not receiving maximum help (OLS).
Coefficients (Std. Err.)
CDR-SB 89.9 (8.13)
Lag of annual informal help hours 0.288 (0.009)
Lag of age spline – less than 65 -13.6 (6.05)
Lag of age spline – 65 to 74 5.59 (5.55)
Lag of age spline – 75 and over 18.5 (4.48)
Male -31.9 (32.8)
Married 317.2 (33.1)
Black 181.9 (38.8)

71
Hispanic 234.7 (47.2)
Education
– less than high school 27.5 (37.8)
Education
– college and above
-11.6 (37.3)
Living in nursing home
-685.3 (104.3)
Constant
-455.6 (357.2)
Panel c. Regression coefficients for nursing home residency model (Probit).
Coefficients (Std. Err.)
CDR
-SB 0.155 (0.008)
Lag of nursing home residency 2.524 (0.060)
Black
-0.169 (0.052)
Hispanic
-0.362 (0.085)
Education
– less than high school
-0.137 (0.044)
Education
– college and above 0.001 (0.041)
Male
-0.035 (0.052)
Male * Education (less than high school)
-0.085 (0.078)
Male * Education (college and above) 0.012 (0.068)
Male * Black 0.217 (0.084)
Male * Hispanic 0.310 (0.134)
Lag of age spline
– less than 65 0.010 (0.008)
Lag of age spline
– 65 to 74 0.013 (0.006)
Lag of age spline
– 75 and over 0.049 (0.003)
Log of delta age 0.473 (0.085)
Heart diseases ever 0.066 (0.030)
Stroke ever 0.303 (0.034)
Cancer ever 0.030 (0.035)
Hypertension ever
-0.017 (0.031)
Diabetes ever 0.179 (0.032)
Lung diseases ever 0.100 (0.041)
Widowed 0.193 (0.033)
Constant
-4.342 (0.505)
Panel d. Regression coefficients for ADL status model (Ordered Probit).
Coefficients (Std. Err.)
CDR
-SB 0.086 (0.003)
Lag of ADL status 0.839 (0.005)
Lag of age spline
– less than 65 0.002 (0.001)
Lag of age spline
– 65 to 74 0.009 (0.001)
Lag of age spline
– 75 and over 0.023 (0.001)
Male
-0.023 (0.009)
Married
-0.113 (0.009)
Black 0.122 (0.011)
Hispanic 0.132 (0.014)
Education
– less than high school 0.065 (0.011)
Education
– college and above
-0.081 (0.010)
Cut 1 2.482
Cut 2 3.080
Cut 3 3.529
Panel e. Regression coefficients for IADL status model (Ordered Probit).
Coefficients (Std. Err.)
CDR
-SB 0.131 (0.003)
Lag of IADL status 1.068 (0.007)
Lag of age spline
– less than 65
-0.005 (0.002)
Lag of age spline
– 65 to 74 0.008 (0.002)
Lag of age spline
– 75 and over 0.031 (0.001)

72
Male -0.025 (0.010)
Married -0.085 (0.010)
Black 0.105 (0.012)
Hispanic 0.099 (0.015)
Education – less than high school 0.076 (0.012)
Education – college and above -0.075 (0.011)
Cut 1 2.493
Cut 2 3.164
Panel f. Regression coefficients for self-reported memory model (Ordered Probit).
Coefficients (Std. Err.)
CDR-SB 0.097 (0.002)
Lag of self-reported memory 1.066 (0.006)
Lag of age spline – less than 65 0.006 (0.001)
Lag of age spline – 65 to 74 0.004 (0.001)
Lag of age spline – 75 and over -0.002 (0.001)
Male 0.099 (0.008)
Married 0.018 (0.008)
Black 0.147 (0.010)
Hispanic 0.227 (0.012)
Education – less than high school 0.084 (0.010)
Education – college and above -0.207 (0.008)
Cut 1 2.625
Cut 2 4.109
Panel g. Regression coefficients for fair/poor self-reported health model (Probit).
Coefficients (Std. Err.)
CDR-SB 0.121 (0.002)
Lag of age spline – less than 65 0.000 (0.001)
Lag of age spline – 65 to 74 0.002 (0.001)
Lag of age spline – 75 and over -0.004 (0.001)
Male 0.047 (0.008)
Married -0.173 (0.008)
Black 0.234 (0.010)
Hispanic 0.445 (0.012)
Education – less than high school 0.321 (0.010)
Education – college and above -0.275 (0.008)
Constant -0.809 (0.069)
Panel h. Regression coefficients for working for pay model (Probit).
Coefficients (Std. Err.)
CDR-SB -0.069 (0.003)
Lag of working for pay 2.175 (0.010)
Lag of age spline – less than 65 -0.055 (0.001)
Lag of age spline – 65 to 74 -0.033 (0.002)
Lag of age spline – 75 and over -0.046 (0.003)
Male 0.148 (0.010)
Married -0.013 (0.011)
Black -0.019 (0.013)
Hispanic -0.019 (0.016)
Education – less than high school -0.062 (0.015)
Education – college and above 0.116 (0.010)
Constant 2.019 (0.082)

73
Appendix Table 2.5A. Disease-stage dependent medical costs and caregiver disutility inputs.
Panel a. Medical costs
Medical costs (2022 US$)
Normal 8,561.9
MCI 8,561.9
Mild AD (community dwelling) 17,123.8
Moderate AD (community
dwelling)
17,780.1
Severe AD (community dwelling) 19,957.9
Mild or moderate AD (nursing
home)
137,259.1
Severe AD (nursing home) 1412,56.6
Panel b. Caregiver disutility
Caregiver disutility
MCI -0.03
Mild AD -0.05
Moderate AD -0.08
Severe AD -0.10

74
Appendix Table 2.6A. Trajectory for outcomes from control, 18-Tx, 48-Tx, and upper
boundary 48-Tx scenario.
Panel a. CDR-SB.
Control 18-Tx 48-Tx Upper boundary 48-
Tx
Year 0 3.20 3.20 3.20 3.20
Year 2 5.66 5.16 4.99 4.42
Year 4 9.78 9.03 7.87 6.33
Year 6 13.69 13.24 12.29 10.46
Year 8 15.63 15.46 15.05 14.14
Year 10 16.46 16.39 16.21 15.81
Year 12 16.87 16.82 16.73 16.52
Year 14 17.08 17.05 17.00 16.86
Year 16 17.23 17.21 17.16 17.06
Year 18 17.33 17.31 17.27 17.19
Year 20 17.42 17.40 17.36 17.30
Panel b. Living in nursing home.
Control 18-Tx 48-Tx Upper boundary 48-
Tx
Year 0 0.010 0.010 0.010 0.010
Year 2 0.033 0.027 0.025 0.021
Year 4 0.142 0.118 0.085 0.053
Year 6 0.332 0.304 0.255 0.179
Year 8 0.511 0.490 0.449 0.378
Year 10 0.643 0.630 0.603 0.554
Year 12 0.735 0.727 0.709 0.676
Year 14 0.794 0.789 0.777 0.756
Year 16 0.842 0.837 0.830 0.816
Year 18 0.871 0.868 0.863 0.851
Year 20 0.893 0.891 0.886 0.877
Panel c. ADLs.
Any ADL.
Control 18-Tx 48-Tx Upper boundary 48-
Tx
Year 0 0.210 0.210 0.210 0.210
Year 2 0.294 0.283 0.279 0.266
Year 4 0.441 0.418 0.386 0.341
Year 6 0.611 0.591 0.552 0.487
Year 8 0.740 0.726 0.699 0.647
Year 10 0.823 0.815 0.797 0.764
Year 12 0.871 0.865 0.855 0.835
Year 14 0.903 0.900 0.893 0.880
Year 16 0.925 0.924 0.918 0.909
Year 18 0.941 0.940 0.936 0.930
Year 20 0.950 0.949 0.946 0.943
Three or more ADLs.
Control 18-Tx 48-Tx Upper boundary 48-
Tx
Year 0 0.065 0.065 0.065 0.065
Year 2 0.104 0.099 0.097 0.091
Year 4 0.201 0.182 0.164 0.136

75
Year 6 0.359 0.336 0.301 0.243
Year 8 0.519 0.500 0.465 0.404
Year 10 0.643 0.629 0.604 0.554
Year 12 0.728 0.720 0.703 0.669
Year 14 0.785 0.780 0.768 0.744
Year 16 0.821 0.818 0.809 0.794
Year 18 0.850 0.848 0.842 0.831
Year 20 0.870 0.869 0.864 0.855
Panel d. IADLs.
Any IADL.
Control 18-Tx 48-Tx Upper boundary 48-
Tx
Year 0 0.185 0.185 0.185 0.185
Year 2 0.291 0.272 0.265 0.244
Year 4 0.495 0.460 0.413 0.345
Year 6 0.703 0.679 0.630 0.546
Year 8 0.827 0.815 0.789 0.734
Year 10 0.892 0.886 0.873 0.845
Year 12 0.927 0.923 0.917 0.902
Year 14 0.943 0.941 0.938 0.929
Year 16 0.955 0.953 0.950 0.944
Year 18 0.963 0.961 0.959 0.955
Year 20 0.968 0.966 0.964 0.961
Two or more IADLs.
Control 18-Tx 48-Tx Upper boundary 48-
Tx
Year 0 0.085 0.085 0.085 0.085
Year 2 0.151 0.138 0.134 0.122
Year 4 0.329 0.294 0.253 0.199
Year 6 0.564 0.530 0.476 0.382
Year 8 0.737 0.717 0.680 0.608
Year 10 0.834 0.825 0.805 0.764
Year 12 0.887 0.881 0.871 0.852
Year 14 0.913 0.910 0.906 0.893
Year 16 0.929 0.927 0.923 0.915
Year 18 0.941 0.939 0.937 0.930
Year 20 0.949 0.947 0.944 0.940
Panel e. Self-reported memory.
Poor memory.
Control 18-Tx 48-Tx Upper boundary 48-
Tx
Year 0 0.081 0.081 0.081 0.081
Year 2 0.112 0.104 0.101 0.093
Year 4 0.185 0.169 0.145 0.119
Year 6 0.270 0.260 0.237 0.198
Year 8 0.317 0.312 0.300 0.277
Year 10 0.341 0.338 0.334 0.324
Year 12 0.348 0.347 0.344 0.339
Year 14 0.354 0.353 0.352 0.349
Year 16 0.354 0.354 0.353 0.351
Year 18 0.357 0.357 0.357 0.356
Year 20 0.362 0.361 0.361 0.358

76
Panel f. Self-reported health.
Fair or poor health.
Control 18-Tx 48-Tx Upper boundary 48-
Tx
Year 0 0.339 0.339 0.339 0.339
Year 2 0.469 0.445 0.438 0.412
Year 4 0.627 0.601 0.556 0.492
Year 6 0.759 0.743 0.712 0.648
Year 8 0.828 0.822 0.807 0.773
Year 10 0.860 0.857 0.851 0.835
Year 12 0.879 0.877 0.872 0.865
Year 14 0.886 0.886 0.883 0.879
Year 16 0.891 0.891 0.889 0.886
Year 18 0.891 0.890 0.888 0.885
Year 20 0.892 0.892 0.891 0.888
Panel g. Working for pay.
Control 18-Tx 48-Tx Upper boundary 48-
Tx
Year 0 0.212 0.212 0.212 0.212
Year 2 0.152 0.156 0.157 0.161
Year 4 0.096 0.101 0.107 0.117
Year 6 0.054 0.058 0.063 0.074
Year 8 0.030 0.032 0.035 0.041
Year 10 0.018 0.018 0.020 0.023
Year 12 0.011 0.012 0.013 0.014
Year 14 0.009 0.009 0.010 0.011
Year 16 0.006 0.006 0.006 0.007
Year 18 0.004 0.004 0.004 0.005
Year 20 0.004 0.004 0.004 0.004

77
Chapter 3 Estimating the Number of Patients Eligible for AD DMTs and
Patient Accessibility under Specialist Capacity Constraints
Introduction
Alzheimer’s disease (AD) imposes increasing burden on the US society and healthcare
system as the baby boomer generation enters the age range of elevated AD risk. It is estimated
that in 2022, more than 6 million Americans aged 65 years and older are living with AD
dementia and by 2050, this number is expected to reach 12.7 million in the absence of diseasemodifying therapies (DMTs).[57] Historically, treatments for Alzheimer’s disease have provided
symptomatic relief only, but the recent approval of AD DMTs by the US Food and Drug
Administration (FDA) provides hope. The first ever AD DMT, Aduhelm (aducanumab), was
approved by the US FDA in June 2021 with the accelerated approval pathway.[84] Then in July
2023, Leqembi (lecanemab) was approved by US FDA, it is the first traditionally approved AD
DMT.[8] Centers for Medicare & Medicaid Services (CMS) also provides broader Medicare
coverage of Leqembi following FDA traditional approval, requiring eligible Medicare patients to
have a physician who participates in a qualifying registry to receive Medicare coverage for
Leqembi.[9] Many other DMT candidates are in different stages of clinical trials, including
several late-stage biologics.[85]
It is the hope of many that safe and effective AD DMTs available on the market will
dramatically and positively change the life of AD patients and their family, at the same time
alleviate the burden of AD for our society. However, availability of AD DMT on the market is
only the beginning of successful AD disease management, as the process of providing AD DMT
to patients in need requires resources in multiple sectors in the healthcare system. Treating AD

78
patients with DMTs is a complex process, as it requires initial screening, diagnosis, treatment
initiation and continuous treatment monitoring. Patients’ access to AD DMTs depend on the
capacity of primary care doctor and specialist, availability of screening and diagnostic tools, and
infusion capacity. Insurance coverage for DMTs, screening and diagnostics examination are also
critical to ensure eligible patients’ access.[18]
Assessment of whether the US healthcare system is ready for AD DMTs from the above
aspects is limited. Liu et al developed a Markov model to simulate the effect of capacity
constraints on access to care for patients with suspected AD for a hypothetical DMT. This study
found the US healthcare system is ill-prepared to handle the potentially high volume of patients
who would be eligible for treatment, with specialist shortage being the most urgent issue. The
estimated average wait time for the diagnostic and treatment phase are 18.6 months when a DMT
first becomes available, and it would take 14 years to eliminate waiting times given the backlog
of prevalent MCI patients.[18]
It is important to understand the potential size of the patient population that may present
for real AD DMTs on the market in order to develop appropriate policy responses. These policies
will require that stakeholders estimate potential access to such therapies and possible wait times,
the cost of such treatments at the population level (and the immediate budget impact for
Medicare, taxpayers and private insurance plans), and new diagnosis from advances in cognitive
and biomarker screening.
In this study, we estimated the potential sizes of treatment-eligible patient populations in
the US based on the prevalence of cognitive impairment, diagnosis rates, and specialist access
constraints. We also conducted sensitivity analysis to show how the number of years required to
initiate DMT treatment for all diagnosed and eligible patients will vary under different

79
assumptions for external factors like specialist capacity. This analysis used aducanumab as our
example AD DMT to create a realistic representation of target patient population profile.
However, as other AD DMTs are similar to aducanumab in terms of eligible disease stages and
exclusion criteria, findings from this analysis are generalizable.
Data and Methods
First, we drew on existing data for prevalence of cognitive impairment in the US
population, segmenting by age and impairment level. We then developed a treatment funnel that
consists of estimates of patients with an existing diagnosis, estimates of amyloid burden in the
cognitively impaired populations, and finally, patients with access to dementia specialists.
To estimate the number of cognitively impaired patients entering the treatment funnel, we
used the Health and Retirement Study (HRS). The HRS is a biennial nationally representative
longitudinal survey in the population with more than 37,000 respondents age 51 and above in the
U.S.[35] Table 3.1 shows the demographic and other descriptive characteristics of the 2016 HRS
sample we utilized. We differentiate by age (51-64, and 65+ years of age) and cognitive status as
measured by the adapted Telephone Interview for Cognitive Status (TICS) and the Clinical
Dementia Rating (CDR). TICS is a cognitive scale, similar to the Mini-Mental State Exam
(MMSE), which is used to assess cognitive status in HRS for all self-respondents [36,37] and
allows us to differentiate between cognitively normal individuals (TICS of 12-27), people with
mild cognitive impairment (TICS 7-11), and people with dementia (TICS <7) (staging validation
based on Crimmins et al. [39], and Langa et al. [40]).2 CDR is a widely-used cognitivefunctional scale which classifies patients as 0 = normal, 0.5 = very mild dementia, 1 = mild
2 For respondents who are not able to complete the survey themselves, questions about changes in memory in the
last two years are asked to proxy respondents in the HRS.

80
dementia, 2 = moderate dementia, and 3 = severe dementia.[86] CDR is often utilized in clinical
trials in conjunction with MMSE to assess patients’ eligibility for further screening: aducanumab
trials used an MMSE score of 24-30 and CDR score of 0.5 to identify eligible patients.[87] In
our modeling, we use a TICS score of 7-11 as a proxy for mild cognitive impairment (MCI), and
a TICS of <7 and CDR = 0.5 or 1 as a proxy for mild dementia. These two disease states are
most commonly targeted in trials of advanced disease-modification candidates. CDR was
administrated to a subset of HRS respondents in the 2002 Aging, Demographics, and Memory
Study (ADAMS), and we developed a model to estimate CDR for our 2016 HRS sample. A
detailed description of our CDR model can be found in the Appendix, detailed model
coefficients are shown in Appendix Table 3.1A and CDR model predictive performance is
shown in Appendix Figure 3.1A. Data on insurance status by cognition and age are shown in
Appendix Figure 3.2A.
As shown in Table 3.1, patients with cognitive impairment differ from the general
population in several ways. For instance, patients with mild dementia (both under and over 65)
tend to have more chronic conditions, are more likely to have 3 or more functional limitations
(measured using both Activities of Daily Living and Instrumental Activities of Daily Living
scales), a higher probability of a hospital stay in the previous 2 years and are much less likely to
work for pay. Notably, only small numbers of patients in the HRS have received an Alzheimer’s
diagnosis even when they have cognitive impairment (about 8% of those under 65 with likely
mild dementia, and 14% of those over 65 with likely mild dementia). Notably, about 45% of
patients with MCI under 65 are working for pay (about 13% over 65 still do).
APOE e4 allele represents a major risk factor for AD.[88] Given a smaller sample size of
patients with likely mild dementia under 65, estimates of nursing home residency and APOE e4

81
carrier status are relatively unreliable in that population. However, the probability of carrying the
APOE e4 gene is higher in more cognitively impaired individuals over the age of 65 (39% of
likely mild dementia patients vs. 32% of likely MCI vs. 26% of cognitively normal patients).
The treatment funnel we estimate is notionally shown in Figure 3.1. First, we estimated
percentage of cognitively impaired population that most closely correspond to MCI and early
dementia in the HRS population. Then, we used published evidence about the share of patients
with a cognitive impairment diagnosis (differentiating between MCI (30.10%) and early
dementia (51.10%)), drawing on previous work by Chen et al. and Lee et al. [42,89] (a more
detailed description of method can be found in the Appendix). We then applied estimates of
amyloid burden in the diagnosed MCI and dementia populations from Rabinovici et al [90].
Using positron emission tomography (PET), they showed 55.3% of MCI and 70.1% of dementia
patients had positive amyloid PET results in a study of 11,400 participants ages 65 and older. We
assume the same share of patients ages 51-64 have amyloid deposits as data for younger patients
were not available. With the filtering of the three above criteria, this will be our estimate of the
number of patients eligible for AD DMTs with existing diagnosis. To estimate the number of
patients eligible for AD DMTs regardless of diagnosis status, we divide the previous number by
the diagnosis rate (MCI: 30.10%, early dementia: 51.10%).
We assume that after the second filtration of the above treatment funnel, MCI or mild
dementia patients with existing diagnosis may require specialist evaluation prior to receiving AD
DMT therapy. This is a reasonable assumption, as these are patients who have existing
interaction with the healthcare system. Using expert input by a clinical neurologist and the FDA
label for aducanumab, we estimated 2 diagnostic visits to a specialist would be required prior to
therapy (one for the initial exam, another to review the results of biomarker tests), followed by 3

82
additional visits during the first year on therapy (this may be higher for aducanumab than future
DMTs due to its safety profile and required brain magnetic resonance imaging (MRI) exams
prior to the 7th and 12th infusion [91]). We varied the number of additional visits in subsequent
years from 1 to 3 annually to show the implications of different levels of demand for specialist
care by treated patients. Our model draws on the number of available dementia specialist visits in
the U.S. calculated by Liu et al (2017) in the medium scenario [18] (3.8 million visits in 2020,
3.9 million visits in 2021, 4.0 million visits in 2025, 4.1 million visits in 2030, and 4.2 million
visits in 2040). Liu et al’s estimate for dementia specialist workforce consists of neurologists,
geriatricians, and geriatric psychiatrists. They also assume that these specialists devote between
2.5 to 7.5 percent of their capacity to conduct AD management with DMT.
We also report the outcome of the number of years to initiate treatment for all eligible
patients with diagnosis and sensitivity analysis results on this outcome. In the sensitivity
analysis, we varied the value of three variables, specialist capacity, number of annual follow-ups
and treatment discontinue rate, and evaluated their impact on the number of years to initiate
treatment for all eligible patients with diagnosis.
Results
Estimating Eligible Treatment Populations
Table 3.2 shows the estimated patient funnel for AD DMTs by age and cognitive status in
2022. We show lower and upper bound estimates – lower bound shows the number of patients
presenting for care based on an existing diagnosis, upper bound shows the potential patient pools
in case all patients with cognitive impairment and amyloid deposits presented for care, regardless
of existing diagnosis status.

83
In the general population, we estimate a mean prevalence of mild cognitive impairment
(MCI) of 9.38% and 17.60% in patients 51-64 and 65+ years old, respectively. After applying
diagnosis and amyloid prevalence estimates, among MCI due to AD patients, we project 1.56%
of individuals aged 51 to 64 and 2.93% of individuals over 65 may be eligible for AD DMTs
with an existing diagnosis. Using U.S. Census projections for 2022, we estimate 918,000 of MCI
due to AD patients between age 51 and 64 and 1,749,000 of MCI due to AD patients over 65
years old may be eligible for AD DMTs with an existing diagnosis in the US, for a total of
2,668,000 treatment-eligible Americans with MCI due to AD. This would increase up to
8,863,000 if all patients with an underlying illness presented for care, regardless of diagnosis
status.
Similarly, we estimate the mean prevalence of mild dementia of 0.34% and 4.24% in
patients 51-64 and 65+ years old, respectively. After applying diagnosis and amyloid prevalence
estimates, among mild AD dementia patients, we project 0.12% of individuals under 65 and
1.52% of individuals over 65 may present for care. Using U.S. Census projections for 2022, we
estimate 72,000 mild AD dementia patients under 65 and 907,000 mild AD dementia patients
over 65 may be eligible for AD DMTs with an existing diagnosis in the US, for a total of 979,000
treatment-eligible Americans with mild AD dementia. This would increase to 1,915,000 if all
patients with an underlying illness presented for care, regardless of diagnosis status.
Combining MCI and mild dementia patients, for patients with an existing diagnosis,
990,000 patients aged 51 to 64 and 2,656,000 patients over age 65 may be eligible for AD
DMTs. If all patients with an underlying illness presented for care regardless of existing
diagnosis, these numbers will increase to 3,191,000 for patients aged 51 to 64 and 7,586,000 for
patients over age 65. This sums up to 3,646,000 patients with existing diagnosis and 10,777,000

84
patients with underlying illness regardless of diagnosis status that may be eligible for AD DMTs.
All above estimates are based on U.S. Census projections for 2022.
In Appendix Figure 3.3A, we show projections of treatment-eligible adults under and
over 65 years of age between 2021 and 2050 based on median U.S. Census projections.
Estimating Prescriptions Given Specialist Capacity Constraints
Drawing on available dementia specialist capacity, we project 3.9 million dementia
specialist visits would be available starting in 2021.[18] Based on the standard of care in
amyloid-based biologic therapy, which requires 2 initial diagnostics visits and 3 follow-on visits
to monitor for side effects (5 initial visits in the first 12 months), we show the expected US
healthcare system capacity for handling patients presenting for care conditional on the number of
follow-up visits after 12 months (1, 2, or 3 follow-up visits) in Figure 3.2. In all scenarios,
780,000 patients could be screened and initiated on therapy in the first year. However, due to
additional follow-up visits required, the number of new patients who could be initiated on
therapy would decrease in following years to allow for both new and existing patients to receive
adequate care. Using a simplified demand model, we estimate that under the most optimistic
scenario (1 follow-up visit after the first 12 months), 71.9% of all eligible patients with an
existing diagnosis would be able to receive specialist access in the first 5 years. Should two
annual follow-up visits be required, 48.6% of eligible diagnosed patients would be seen by a
specialist, and should three visits be required, just 35.3% of eligible patients would be seen over
the same time period. The distribution of patients over time (by year of initiating therapy starting
in 2022) and the share of patients receiving specialist referral and treatment by scenario are
shown in Figure 3.2. We also indicate the number of patients receiving specialist access as a

85
proportion of all possibly eligible patients (10,777,000) for DMTs in the US. We do not account
for constraints in primary care which may contribute to additional delays and limitations in
access to cognitive and biomarker screening, and to specialists.
In Figure 3.3, we show sensitivity analysis results based on the outcome of number of
years to initiate DMT treatment for all eligible patients with diagnosis (lower bound scenario).
We check the robustness of three variables on this outcome, specialist capacity (low: 2.0 million,
medium: 3.9 million, high: 5.9 million, in 2021) [18], number of annual follow-up visits (1, 2 or
3) and treatment discontinue rate (0.1, 0.2 and 0.3). Considering a base case scenario with
medium capacity of 3.9 million specialist visits available in 2021, with 2 annual follow-up visits
required and 0.2 treatment discontinue rate, it will take 8.8 years to initiate DMT treatments for
eligible patients with diagnosis (6.9 years for MCI patients and 1.9 years for mild dementia
patients). As can be seen in Figure 3.3, specialist capacity is the most influential variable on the
number of years to initial DMT treatment for all patients. Considering the high-capacity scenario,
it only takes 4.9 years to initiate treatment for all eligible patients with diagnosis; however, in the
low capacity scenario, it takes 19.8 years.
Discussion
Our work has shown that the emergence of DMTs in Alzheimer’s disease has the
potential to benefit millions of patients with mild cognitive impairment and mild dementia due to
Alzheimer’s disease. Should only those with an existing diagnosis of cognitive impairment be
referred to specialists, approximately 2.668 millions of patients with MCI due to AD and
979,000 patients with mild AD dementia may be eligible for treatment in 2022. If all patients
with the underlying disease pathology in the US population were considered, 8.863 million

86
patients with MCI due to AD and 1.915 million patients with mild AD dementia may be eligible
for treatment. However, as we estimate, only a fraction of those patients would be able to receive
access given dementia specialist constraints – ranging from 35.3% to 71.9% of diagnosed
patients under three specialist demand scenarios, and just 11.9% to 24.3% of all patients
clinically eligible – over a period of 5 years. We find specialist capacity to be the most influential
factor affecting the number of years to initial DMT treatment for all eligible and diagnosed
patients, this finding is consistent with existing literature.[18] Assuming an annual specialist
capacity of 3.9 million in 2021, the number of annual follow-up visits of 2 and a treatment
discontinue rate of 0.2, it would take 8.8 years to initial DMT treatments for all eligible patients
with diagnosis. These findings suggest that if more biologic DMTs are approved and prove to
have a clinically meaningful treatment effect, the US healthcare system will struggle to provide
efficient access to many patients with the disease and may turn away a large share of patients
who present for care that already have a cognitive impairment diagnosis.
Our estimated result is comparable with existing studies. Liu et al. reported it would take
14 years to eliminate treatment waiting times for AD DMTs, while we estimate it would take 8.8
years to initial DMT treatments for all eligible patients with diagnosis in the average scenario,
and this number would range from 5 to 20 years given variation in specialist capacity, number of
annual follow-ups required and treatment discontinuation rate.[18]
Compare with direct acting antiviral (DAA) drugs for Hepatitis C, another novel yet
expensive treatment that became available in recent history, our estimate shows better
accessibility for AD DMTs. Study conducted by Centers for Disease Control and Prevention
shows that only 1 of 3 adults diagnosed with Hepatitis C have been cured during the nine years
period from 2013 to 2022 in the US. As DAA agents only requires an 8-12 week short-course

87
treatment and results in a cure in over 95% of cases, it is reasonable to assume about one-third of
adults diagnosed with Hepatitis C initiated DAA treatment during a nine-year period.[92] While
we estimate that at least 35.3% of eligible patients with existing diagnosis will initiate AD DMT
treatment in a five-year period. This is reasonable as people eligible for AD DMTs are more
likely to be over the age of 65 and on Medicare, thus they face smaller financial obstacle when
accessing these novel yet expensive treatments. However, we estimate that 27% of eligible AD
patients are under age 65 and thus do not have access to Medicare, they may face high financial
obstacle to access AD DMTs at an annual price tag of over $20,000 without insurance.[93,94]
We did not consider demand-side access barriers like affordability in this analysis.
Based on our estimate, AD DMTs could have large impact on Medicare budget as there
are over 2.6 million patients above the age of 65 that are eligible for such treatments. Based on
our estimate of 780,000 patients initiating DMT treatment in the first year under specialist
capacity constraints, this will incur costs of $20.7 billion in that first year for DMT treatment.
Such costs would account for half of all Medicare Part B drugs spending in 2021 ($40.1
billion).[94] As our analysis focuses on patients eligible for AD DMT treatment under supplyside constraints like specialist capacity, this is likely to be an upper boundary estimate as we did
not consider demand-side barriers like partial uptake of diagnostic examination and treatment
that will happen in real world settings. However, our findings still show the need to balance
novel treatment access for patients in need and affordablility for our society. Evidence-based
health policy solutions are required to make difficult decisions.
The demand for specialist visits in our model is conditional on having received cognitive
and biomarker screening prior to a specialist visit. In the event of a specialist visit being required
before such screening, relative capacity would decrease even further as many more patients with

88
unconfirmed Alzheimer’s pathology may present for care. These findings are consistent with
earlier work that suggests the need to combine cognitive and biomarker testing to reduce the
number of false positives from cognitive tests alone and increase the cost-effectiveness of
screening.[95,96] However, even if those steps are performed prior to a specialist visit, the
demand for specialist assessment and monitoring of side effects of biologic therapies will surpass
the available supply of specialists by a large margin. As a result, strategies to prioritize patients
most likely to benefit from early access (e.g., those likely to lose the most by receiving delayed
access) need to be developed. In addition, reducing the need for follow-up visits by specialists,
which may be e.g., by using therapies with a better safety profile or using non-specialist
providers to monitor side effects, may help increase overall capacity.
The emergence of AD DMT may in turn affect specialist capacity. On one hand, AD
management with DMT may crowd out other services provided by neurologists, including
diagnosis and treatment for conditions like concussion, epilepsy, multiple sclerosis, etc. Though
our specialist capacity data source already assumes a relatively low excess capacity from the
neurology and geriatrics workforce (2.5% to 7.5% of capacity) to be available for AD DMT. On
the other hand, the specialist capacity available for AD management with DMT may also
increase over time given this emergence demand. Although the long training time to become a
board-certified specialist implies that neither expansion of postgraduate training programs nor
immigration would be a likely solution, it is still possible to improve productivity in the current
workforce by automating or delegating more tasks in the evaluation process and to qualify more
specialists for dementia care, especially if they have related experience.[18]
Our work has several limitations. First, we model a broad range of possible DMTs based
on one precedent only. Future therapies may have a significantly different safety profile or

89
require fewer (or more) diagnostic and follow-up steps than expected. Second, we study
prevalent cases (as of 2022) only under the assumption that inflows (incident cases) and outflows
(patients progressing to later disease stages) are roughly equal. However, this likely produces a
slight underestimate of future demand due to aging of the US population and the increasing
burden of dementia over time. Third, we draw on available data for amyloid and diagnostic rates,
which may not be fully representative of the US population with cognitive impairment due to
Alzheimer’s disease. For example, the diagnostic rates were calculated from Medicare claims,
which may not hold true for non-Medicare enrollees. We also do not adjust for patient
willingness to receive the therapy (or the willingness of their insurers to provide broad access),
which may reduce demand for treatment. However, given the high unmet need and the
significant burden of the disease in the population, we expect that demand will exceed supply
significantly for most clinically effective therapies. Finally, we do not account for infusion and
imaging capacity in the US, which may further reduce the capacity of the system to deliver new
biologic therapies for Alzheimer’s disease. However, the magnitude of those delays is likely
much smaller than the magnitude of delays due to specialist constraints, as shown previously.
[18]
We find that the possible demand for DMTs in Alzheimer’s disease is likely to exceed the
capacity of the US healthcare system to provide timely diagnosis and treatment. In light of
existing capacity constraints, policy solutions that allow for the prioritization of patients most
likely to benefit from therapies (or are most likely to lose by waiting) ought to be developed.
New solutions may include the training of other physician specialties to provide care for the
aging population. We show that the number of required follow-up visits has a large impact on the
overall capacity of the healthcare system to treat new patients. As new therapies for the disease

90
emerge, their safety profile, drug reimbursement, and the effectiveness of screening at the
primary care level will also determine how many patients can receive access to AD DMTs in a
timely manner.
Appendix
CDR Modeling
We developed a set of models to impute 5-level Clinical Dementia Rating (CDR), using
data from the 2002 ADAMS (Aging, Demographics and Memory Study), which is a nationally
representative study of older Americans 70+, which we linked to the core HRS survey. Given
that the distribution of CDR score was markedly different between proxy and non-proxy
respondents, we developed separate models for these two samples. A hurdle model was specified
for participants without proxies. We chose this specification because while an ordered probit was
able to distinguish between CDR 0 and CDR of 0.5+ well, it did not optimally classify the
remaining categories, which were CDR 0.5, 1, 2, or 3. As such, we have a first stage probit
regression for CDR 0 vs. higher, and then a second stage ordered probit for CDR 0.5, 1, 2, or 3
among those predicted for the “higher” category. This is performed under a joint estimation.
When modeling subjects with proxy, we simply used an ordered probit since it offered sufficient
fit.
In our models, independent variables included demographic characteristics,
comorbidities, individual ADLs and IADLs, as well as variables such as nursing home living and
smoking status. Subjects without proxy were also modeled with the Telephone Interview for
Cognitive Status – 27 points (TICS27), which is a cognitive score. Since subjects with proxies do
not have a value for this measure, we used their cognitive impairment rating (as rated by their

91
proxy) instead. We also incorporated quadratics and interactions where appropriate. We present
our model for participants with proxy in Appendix Table 3.1A Panel a, and our model for
participants without proxy in Appendix Table 3.1A Panel b and C.
In terms of performance, the 10-fold cross-validated accuracy for the two models ranged
between 56-58% (Appendix Figure 3.1A). This is rather good discrimination, since a random
guess for a 5-level CDR score would have an accuracy of 20%. A study by Lee et al., 2018
assessed a survey-based predictive model for 3-level CDR which had an accuracy rate of 58.6-
61.2%. Given we predict an extra 2 levels more, our accuracy rate is fair. When we combine the
predictions of the non-proxy and proxy models, the distribution between predicted CDR and
actual CDR match up well, with a difference of just 1 or 2 percentage points for the various CDR
categories (Appendix Figure 3.1A). Note that while our model has been validated for the over
70 population, it is yet to be tested in subjects under 70 years old.
Calculation of diagnosis rates in MCI and mild dementia groups
To calculate the diagnosis rate among those identified as cognitively impaired by HRS’s
cognitive test, we found the proportion identified as dementia with both cognitive test and a
diagnosis among all identified as dementia by cognitive test. This calculation arrived at the
51.1% diagnosis rate in the mild dementia group.
We draw on Chen et al.[42] who also use HRS, which they link to Medicare claims to
identify the availability of a dementia diagnosis. We used Chen’s estimate of dementia
prevalence with both cognitive test and a diagnosis (7.2%) as a share of those identified having
dementia using both measures and those identified having dementia with cognitive test only

92
(6.9%). This allowed us to estimate a 51.1% diagnosis rate in the dementia group, which we
assume holds for mild dementia.
To calculate the diagnosis rate in the MCI group, we used the 51.1% diagnosis rate in the
mild dementia group and additional data reported by Lee et al.[89] Lee et al. reported diagnosis
rate in both dementia and MCI population, though not from a representative data source. Given
that Lee et al. did not use a representative sample, we calculated an adjustment factor by dividing
Lee et al.’s dementia diagnosis rate (85.0%) by the dementia diagnosis rate in HRS linked to
Medicare claims (51.1%) from Chen et al. Then we adjusted Lee et al.’s reported diagnosis rate
in the MCI group (50.0%) with this adjustment factor (1.66 = 85.0%/51.1%) to estimate the
diagnosis rate in the MCI group of 30.1%.

93
Table 3.1. Descriptive statistics of 2016 HRS respondents by age and cognitive status.
Notes: Mean value and the 95% confidence intervals are shown in parenthesis. Data source is the 2016 Health and
Retirement Study (HRS); general population is everyone aged 51 and older in the 2016 HRS sample; ‘sum of conditions
ever had’ includes the following conditions: hypertension, diabetes, cancer, lung disease, stroke, psychiatric problem
and arthritis. Abbreviations: APOE, Apolipoprotein E; ADL, activities of daily living; IADL, instrumental activities of
daily living; OOP, out-of-pocket; SSDI, Social Security Disability Insurance.

94
Table 3.2. Estimated patient funnel for AD DMT in 2022.
Notes: 2022 US population estimates are from US Census Bureau 2017 National Population Projections Datasets for
2022. Patient number estimates (7th and 8th columns) rounded to 1,000s. Abbreviations: MCI, mild cognitive
impairment; AD, Alzheimer’s disease; HRS, Health and Retirement Study; DMT, disease-modifying therapies.

95
Figure 3.1. Treatment funnel for people eligible for Alzheimer’s DMTs.
Note: ADRD is short for Alzheimer’s disease and related dementias.

96
Figure 3.2. Patients treated and untreated in the first 5 years based on the number of
follow-up annual visits.

97
Figure 3.3. Number of years to initiate DMT treatment for all diagnosed patients by levels
of specialist capacity, number of annual follow-ups and treatment discontinue rate.
Panel a. Specialist capacity.
Panel b. Number of annual follow-ups.
Panel c. Treatment discontinue rate.

98
Appendix Table 3.1A. CDR models.
Panel a. Ordered probit model for participants with proxy. (N = 189)
Dependent Variable: CDR (0, 0.5, 1, 2, 3) Coefficient
PValue
Age -0.023 0.93
Age-Squared 0.0000678 0.95
Gender 0.99 0.00
Race (Ref = White)
Black/African American -0.27 0.34
Other -0.13 0.79
Education Years (Ref = 17+)
1 -0.39 0.67
2 0.07 0.92
3 0.10 0.87
4 -0.09 0.87
5 … 0.13 0.82
16 0.53 0.45
Nursing Home Living 0.02 0.93
Smoke Ever -0.12 0.57
Smoke Now 0.92 0.08
Difficulty Walking 0.25 0.36
Difficulty Bathing 0.24 0.44
Difficulty Eating 0.06 0.82
Difficulty Bed 0.08 0.80
Difficulty Toilet 0.86 0.01
Difficulty Meals 0.18 0.61
Difficulty Shopping -0.47 0.24
Difficulty Phone 1.36 0.00
Difficulty Money 0.82 0.02
Difficulty Map 0.71 0.10
Hypertension Ever -0.09 0.67
Diabetes Ever -0.06 0.78
Cancer Ever -0.14 0.55
Lung Disease Ever -0.43 0.19
Heart Disease Ever 0.07 0.76
Stroke Ever -0.24 0.31
Proxy Cognitive Impairment Rating (Ref = Don’t Think Respondent has any Cognitive
Limitations)
May Have Some Cognitive Limitations But Could Likely Do Interview 0.10 0.78
Cognitive Limitations Prevent Interview 1.33 0.00
Note: Ref = reference.

99
Panel b. First stage of hurdle model for participants without proxy. (N = 619)
Dependent Variable: CDR (0 vs. 0.5+) Coefficient P-Value
Age 0.44 0.75
Age-Squared 0.00 0.88
Age X TICS-27 Score (Ref = 70)
71 -0.03 0.39
72 -0.15 0.00
73 -0.13 0.00
74 -0.09 0.04
75 … -0.12 0.02
94 -0.56 0.72
Gender -0.11 0.56
Race (Ref = White)
Black/African-American 0.25 0.33
Other 0.29 0.54
Education Years
1 -0.96 0.36
3 -0.81 0.29
4 -0.16 0.88
5 -0.19 0.83
6 … -0.68 0.36
17.17+ -0.45 0.50
Number of IADL's X TICS-27 Score (Ref = 1)
2 -0.06 0.65
3 -1.34 1.00
Number of ADL's X TICS-27 Score (Ref = 1)
2 -0.11 0.17
3 -0.24 0.22
4 0.11 0.70
TICS-27 Score-Squared 0.00 0.62
TICS-27 Score -0.04 0.73
Number of ADL's X Number of IADL's (Ref = 1)
3 3 -6.53 1.00
4 2 1.03 0.43
4 3 -4.97 1.00
Number of IADL's X TICS-27 Score (Ref = 1)
2 0.22 0.88
3 16.48 1.00
Number of ADL's (Ref = 1)
2 1.54 0.12
3 3.55 0.17
4 -1.29 0.70
Nursing Home Living -0.40 0.70
Smoke Ever 0.05 0.78
Smoke Now -0.16 0.56
Hypertension Ever 0.19 0.24
Diabetes Ever 0.07 0.73
Cancer Ever -0.29 0.16
Lung Disease Ever 0.65 0.05
Heart Disease Ever 0.23 0.21
Stroke Ever 0.57 0.04

100
Difficulty Walking 0.10 0.95
Difficulty Bathing X TICS-27 Score -1.19 0.58
Difficulty Eating 7.02 0.16
Difficulty Bed 2.24 0.27
Difficulty Toilet -2.32 0.23
Difficulty Meals 1.57 0.16
Difficulty Shopping 0.33 0.74
Difficulty Phone 4.07 0.27
Difficulty Money -0.12 0.92
Difficulty Map 0.32 0.60
Difficulty Walking X TICS-27 Score -0.02 0.85
Difficulty Bathing X TICS-27 Score 0.10 0.59
Difficulty Eating X TICS-27 Score -0.61 0.14
Difficulty Bed X TICS-27 Score -0.11 0.43
Difficulty Toilet X TICS-27 Score 0.10 0.47
Difficulty Meals X TICS-27 Score -0.08 0.35
Difficulty Shopping X TICS-27 Score -0.01 0.92
Difficulty Phone X TICS-27 Score -0.20 0.49
Difficulty Money X TICS-27 Score 0.04 0.72
Difficulty Map X TICS-27 Score 0.00 1.00
Constant -23.89 0.66
Note: This model utilizes a probit specification. Ref = reference. ADL = Activities of Daily Living. IADL =
Instrumental Activites of Daily Living. TICS-27 = Telephone Interview for Cognitive Status – 27 Points.

101
Panel c. Second stage of hurdle model for participants without proxy. (N = 619)
Dependent Variable: CDR (0.5, 1, 2, 3) CoefficientP-Value
Age -0.78 0.41
Age-Squared 0.01 0.37
Age X TICS-27 Score (Ref = 70)
71 0.25 0.02
72 0.09 0.48
73 0.26 0.09
74 0.10 0.38
75 … 0.23 0.08
109 -3.51 1.00
Gender 0.05 0.85
Race (Ref = White)
Black/African-American 0.15 0.59
Other 0.38 0.59
Education Years
1 -0.10 0.93
2 -5.65 1.00
3 -0.06 0.93
4 -0.58 0.47
5 … 0.84 0.19
17.17+ 1.44 0.02
Number of IADL's X TICS-27 Score (Ref = 1)
2 0.07 0.59
3 -0.04 0.86
Number of ADL's X TICS-27 Score (Ref = 1)
2 0.11 0.41
3 0.02 0.94
4 0.07 0.84
TICS-27 Score-Squared -0.01 0.53
TICS-27 Score -0.27 0.23
Number of ADL's X Number of IADL's (Ref = 1)
2 2 0.71 0.38
2 3 1.03 0.38
3 2 0.79 0.37
3 3 2.94 0.01
4 2 9.24 0.99
4 3 9.69 0.99
Number of IADL's X TICS-27 Score (Ref = 1)
2 -0.53 0.60
3 -0.61 0.68
Number of ADL's (Ref = 1)
2 -1.04 0.34
3 -0.27 0.86
4 -9.29 0.99
Nursing Home Living 0.31 0.47
Smoke Ever 0.41 0.07
Smoke Now -2.40 0.00
Hypertension Ever -0.58 0.01
Diabetes Ever -0.44 0.14
Cancer Ever 0.58 0.09

102
Lung Disease Ever -0.03 0.94
Heart Disease Ever 0.28 0.25
Stroke Ever -0.11 0.70
Difficulty Walking 0.73 0.48
Difficulty Bathing X TICS-27 Score -1.68 0.17
Difficulty Eating 0.26 0.81
Difficulty Bed 3.28 0.01
Difficulty Toilet -1.25 0.39
Difficulty Meals 3.53 0.00
Difficulty Shopping -1.77 0.05
Difficulty Phone -0.92 0.28
Difficulty Money -1.15 0.16
Difficulty Map 0.94 0.17
Difficulty Walking X TICS-27 Score -0.08 0.60
Difficulty Bathing X TICS-27 Score 0.24 0.19
Difficulty Eating X TICS-27 Score -0.01 0.97
Difficulty Bed X TICS-27 Score -0.48 0.02
Difficulty Toilet X TICS-27 Score 0.13 0.52
Difficulty Meals X TICS-27 Score -0.43 0.00
Difficulty Shopping X TICS-27 Score 0.19 0.07
Difficulty Phone X TICS-27 Score 0.03 0.78
Difficulty Money X TICS-27 Score 0.23 0.02
Difficulty Map X TICS-27 Score -0.02 0.81
Note: This model utilizes an ordered probit specification. Ref = reference. ADL = Activities of Daily Living. IADL
= Instrumental Activities of Daily Living. TICS-27 = Telephone Interview for Cognitive Status – 27 Points.

103
Appendix Figure 3.1A. Performance of predictive models for CDR, among participants
without proxy and participants with proxy.
Note: CDR = Clinical Dementia Rating.

104
Appendix Figure 3.2A. Insurance status by cognition and age.
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
< 65
>= 65
all
< 65
>= 65
all
MCI Mild AD
primary Medicare uninsured primary Medicaid primary military primary private

105
Appendix Figure 3.3A. Lower-bound projections of patients eligible for AD DMT by age
and cognitive status.
0
1,000
2,000
3,000
4,000
5,000
6,000
2021 2025 2030 2035 2040 2045 2050
Thousands
< 65 MCI < 65 mild AD >= 65 MCI >= 65 mild AD

106
Conclusions
The arrival of AD DMTs marks a new era of AD disease management, bringing new hope
as well as new challenges. With the emergence of DMTs on the market, policy makers are
interested in long-term value assessment of these novel therapies. Microsimulation modeling
could extrapolate short-term clinical trial results and generate long-term estimates on meaningful
outcomes. To ensure access to DMTs for eligible patients, it is also important to estimate the size
of eligible patient population and length of wait time under specialist capacity constraint.
In Chapter 1, we extend the FEM microsimulation model to include a widely used
cognitive measure based on nationally representative HRS data, using individual-level
information on demographics, chronic disease indicators, employment, smoking status, marital
status and body mass index. In Chapter 1, we demonstrate this model’s ability to accurately
predict cognitive decline, and this is an important step toward estimating the future burden and
long-term value of AD DMTs in the US.
Our FEM TICS27 model addressed several aspects for improvements for AD disease
models. Our modeling target, TICS27, covers the full disease continuum of AD, which provides
granular information not only for patients in the dementia stage but also for people in the
preclinical and MCI stages. And TICS27 can be translated to MMSE via a published
crosswalk[51], MMSE is a cognitive measure commonly used in AD DMT clinical trials. This is
important for models aiming to evaluate novel AD DMTs, as many DMTs target people in the
MCI or early dementia stages of AD. Unlike most AD disease models, which are based on
selective clinical trial or registry data, FEM TICS27 is a population-level microsimulation based
on nationally representative longitudinal survey data. This ensures its ability to provide
meaningful long-term value assessment from the societal perspective.

107
We also rigorously validate FEM TICS27 simulation results against observed data using
an unbiased cross-validation approach. Given the limited access to data and adoption of different
cognitive function tests, validation of model is a general challenge in the area of AD
modeling.[17] The FEM TICS27 model demonstrates excellent internal validity: the TICS27
distribution and 10-year change in cognitive status generated by FEM simulation closely matches
observed HRS data, and the AUROCs are larger than 0.85 for dementia prediction. For
prediction of significant decline in MCI patients, FEM’s performance is comparable to one of the
best-performing models reported in the literature.[52] This chapter demonstrates FEM TICS27’s
ability to accurately predict cognitive decline, and this is an important step toward estimating the
future burden and long-term value of AD DMTs in the US.
In Chapter 2, we extrapolate clinical findings from the Clarity AD trial to policy-relevant
outcomes and with a longer follow up period. We find 48-month of lecanemab treatment results
in meaningful improvements in outcomes like living in nursing home, limitations in three or
more ADLs and limitations in two or more IADLs during a 20-year observation period. Patients
treated with lecanemab were estimated to spend 0.22 more years alive during the 20-year
observation period. They are also estimated to spend longer time before transitioning to more
severe stages of dementia, spending 0.68 years longer before progression to moderate or severe
dementia, and 0.70 years longer before progression to severe dementia. Considering cumulative
outcomes that could be valued in monetary term, which include QALYs, earnings, informal help
hours, medical costs, and caregiver disutility, the value of 48-month of lecanemab treatment is
estimated to be $158,200. Consistent with existing studies on disease burden of AD, about 40%
of the economic value gained from lecanemab is accrued to caregiver.[5]

108
Long-term value assessment of AD DMTs based on real efficacy data is limited. Our
study improves over existing studies by incorporating additional meaningful outcomes besides
QALYs and costs. Outcomes like living in nursing home, ADLs and IADLs capture AD patients’
level of independence which both patients and their caregivers value highly. We find that
lecanemab shows meaningful positive impact on outcomes like living in nursing home,
functional status measured in ADLs and IADLs limitations and memory. Lecanemab also
increase the time patients spend in earlier stages of dementia. Future work should focus on
incorporating more relevant value elements into value assessment and the heterogeneity of DMT
value by racial-ethnic profile.
In Chapter 3, we show that the emergence of DMTs in Alzheimer’s disease has the
potential to benefit millions of patients with mild cognitive impairment and mild dementia due to
AD. Should only those with an existing diagnosis of cognitive impairment be referred to
specialists, approximately 2.668 millions of patients with MCI due to AD and 979,000 patients
with mild dementia due to AD may be eligible for treatment in 2022. If all patients with the
underlying disease pathology in the US population were considered, 8.863 million patients with
MCI due to AD and 1.915 million patients with mild dementia due to AD may be eligible for
treatment. However, as we estimate, only a fraction of those patients would be able to receive
access given dementia specialist constraints – ranging from 35.3% to 71.9% of diagnosed
patients under three specialist demand scenarios, and just 11.9% to 24.3% of all patients
clinically eligible – over a period of 5 years. We find specialist capacity to be the most influential
factor affecting number of years to initial DMT treatment for all patients, this finding is
consistent with existing literature.[18] Assuming an annual specialist capacity of 3.9 million in

109
2021, the number of annual follow-up visits of 2 and a treatment discontinue rate of 0.2, it would
take 8.8 years to initial DMT treatments for all diagnosed patients.
Our findings suggest that if more biologic DMTs are approved and prove to have a
clinically meaningful treatment effect, the US healthcare system will struggle to provide efficient
access to many patients eligible for such treatments and may turn away a large share of patients
who present for care that already have a cognitive impairment diagnosis. As pointed out in other
studies, specialist capacity constraint is the most challenging obstacle to receiving AD DMTs
treatment.[18,19] Emergence of AD DMTs may crowd out other services provided by
neurologists, like diagnosis and treatment of concussion, epilepsy, multiple sclerosis, etc., while
productivity improvement of the current specialist workforce by automating or delegating more
tasks in the disease evaluation process and qualifying more specialists for dementia care may
also increase the available specialist capacity, facing the increase in demand due to AD DMT
emergence. Strategies to prioritize patients most likely to benefit from early access need to be
developed. And policy should allow for the prioritization of access to certain patient subgroups.
Concluding Remarks
The arrival of AD DMTs is the result of decades of scientific efforts. It is also a longawaited hope for patients with AD and their families, for a longer and happier life. As the babyboomer generation of US population enters the age range with elevated risk for AD, such novel
treatments could potentially improve the life of even more people. Availability of safe and
effective AD DMTs is only the beginning to successful AD disease management, continuous
health economics research on long-term value assessment framework and stepwise treatment

110
prioritization based on patient heterogeneity is critical to ensure broad patient access to such lifechanging treatments.

111
References
1. Aisen PS, Cummings J, Jack CR, Morris JC, Sperling R, Frölich L, et al. On the path to
2025: understanding the Alzheimer’s disease continuum. Alz Res Therapy. 2017 Aug
9;9(1):60.
2. Alzheimer’s Association. 2023 Alzheimer’s disease facts and figures. Alzheimer’s &
Dementia. 2023 Apr;19(4):1598–695.
3. James BD, Leurgans SE, Hebert LE, Scherr PA, Yaffe K, Bennett DA. Contribution of
Alzheimer disease to mortality in the United States. Neurology. 2014 Mar 3;82(12):1045.
4. Rajan KB, Weuve J, Barnes LL, McAninch EA, Wilson RS, Evans DA. Population estimate
of people with clinical Alzheimer’s disease and mild cognitive impairment in the United
States (2020–2060). Alzheimer’s & Dementia. 2021;17(12):1966–75.
5. Tahami Monfared AA, Byrnes MJ, White LA, Zhang Q. The Humanistic and Economic
Burden of Alzheimer’s Disease. Neurol Ther. 2022 Jun 1;11(2):525–51.
6. Cummings J, Fox N. Defining Disease Modifying Therapy for Alzheimer’s Disease. J Prev
Alzheimers Dis. 2017;4(2):109–15.
7. Commissioner O of the. FDA Grants Accelerated Approval for Alzheimer’s Drug [Internet].
FDA. FDA; 2021 [cited 2023 Aug 14]. Available from: https://www.fda.gov/newsevents/press-announcements/fda-grants-accelerated-approval-alzheimers-drug
8. Commissioner O of the. FDA Converts Novel Alzheimer’s Disease Treatment to Traditional
Approval [Internet]. FDA. FDA; 2023 [cited 2023 Aug 14]. Available from:
https://www.fda.gov/news-events/press-announcements/fda-converts-novel-alzheimersdisease-treatment-traditional-approval
9. Statement: Broader Medicare Coverage of Leqembi Available Following FDA Traditional
Approval | CMS [Internet]. [cited 2023 Aug 14]. Available from:
https://www.cms.gov/newsroom/press-releases/statement-broader-medicare-coverageleqembi-available-following-fda-traditional-approval
10. Cummings J, Lee G, Nahed P, Kambar MEZN, Zhong K, Fonseca J, et al. Alzheimer’s
disease drug development pipeline: 2022. Alzheimer’s & Dementia: Translational Research
& Clinical Interventions. 2022;8(1):e12295.
11. Garrison Jr LP, Baumgart M, El-Hayek YH, Hozapfel D, Leibman C. Defining Elements of
Value in Alzheimer’s Disease. Value & Outcomes Spotlight. 7(S1).
12. El-Hayek YH, Wiley RE, Khoury CP, Daya RP, Ballard C, Evans AR, et al. Tip of the
Iceberg: Assessing the Global Socioeconomic Costs of Alzheimer’s Disease and Related
Dementias and Strategic Implications for Stakeholders. Journal of Alzheimer’s Disease.
2019 Jan 1;70(2):323–41.

112
13. Tochel C, Smith M, Baldwin H, Gustavsson A, Ly A, Bexelius C, et al. What outcomes are
important to patients with mild cognitive impairment or Alzheimer’s disease, their
caregivers, and health-care professionals? A systematic review. Alzheimers Dement (Amst).
2019 Dec;11:231–47.
14. Jutkowitz E, Gaugler J. Perspective on Value Frameworks in Alzheimer’s Disease. Value &
Outcomes Spotlight. 7(S1).
15. Barbarino P, Gustavsson A, Neumann PJ. Long-Term Value Demonstration in Alzheimer’s
Disease: Evidence Needs. Value & Outcomes Spotlight [Internet]. 2021 Mar [cited 2023
Aug 14];7(S1). Available from: https://www.ispor.org/publications/journals/valueoutcomes-spotlight/vos-archives/issue/view/valuing-future-alzheimers-disease-treatmentsthe-need-for-a-holistic-approach/long-term-value-demonstration-in-alzheimers-diseaseevidence-needs
16. IMI Innovative Medicines Initiative | ROADMAP | Real world outcomes across the AD
spectrum for better care: multi-modal data access platform [Internet]. IMI Innovative
Medicines Initiative. 2016 [cited 2023 Aug 15]. Available from:
http://www.imi.europa.eu/projects-results/project-factsheets/roadmap
17. Green C, Zhang S. Predicting the progression of Alzheimer’s disease dementia:
A multidomain health policy model. Alzheimer’s & Dementia. 2016 Jul 1;12(7):776–85.
18. Liu JL, Hlavka JP, Hillestad R, Mattke S. Assessing the Preparedness of the U.S. Health
Care System Infrastructure for an Alzheimer’s Treatment [Internet]. RAND Corporation;
2017 Nov [cited 2022 Apr 6]. Available from:
https://www.rand.org/pubs/research_reports/RR2272.html
19. Wimo A. The End of the Beginning of the Alzheimer’s Disease Nightmare: A Devil’s
Advocate’s View. J Alzheimers Dis. 2018;64(s1):S41–6.
20. Cummings J, Lee G, Ritter A, Sabbagh M, Zhong K. Alzheimer’s disease drug development
pipeline: 2019. Alzheimer’s & Dementia: Translational Research & Clinical Interventions.
2019 Jan 1;5:272–93.
21. Dunn B, Stein P, Cavazzoni P. Approval of Aducanumab for Alzheimer Disease—the FDA’s
Perspective. JAMA Internal Medicine [Internet]. 2021 Jul 13 [cited 2021 Aug 9]; Available
from: https://doi.org/10.1001/jamainternmed.2021.4607
22. Lin PJ, Cohen JT, Neumann PJ. Preparing the health-care system to pay for new
Alzheimer’s drugs. Alzheimer’s & Dementia. 2020;16(11):1568–70.
23. Spackman DE, Kadiyala S, Neumann PJ, Veenstra DL, Sullivan SD. Measuring Alzheimer
Disease Progression with Transition Probabilities: Estimates from NACC-UDS. Curr
Alzheimer Res. 2012 Nov;9(9):1050–8.

113
24. Kansal AR, Tafazzoli A, Ishak KJ, Krotneva S. Alzheimer’s disease Archimedes conditionevent simulator: Development and validation. Alzheimer’s & Dementia: Translational
Research & Clinical Interventions. 2018 Jan 1;4:76–88.
25. Jutkowitz E, MacLehose RF, Gaugler JE, Dowd B, Kuntz KM, Kane RL. Risk Factors
Associated With Cognitive, Functional, and Behavioral Trajectories of Newly Diagnosed
Dementia Patients. J Gerontol A Biol Sci Med Sci. 2017 Feb 1;72(2):251–8.
26. Herring W, Keenan A, Mauskopf J, Michael T, Wiegand F. The potential economic value of
disease-modifying treatments in Alzheimer’s disease: patient-level simulation of
predementia symptom trajectories [Internet]. RTI-HS. 2017 [cited 2020 Nov 12]. Available
from: https://www.rtihs.org/publications/potential-economic-value-disease-modifyingtreatments-alzheimers-disease-patient-level
27. Davis M, O’Connell T, Johnson S, Cline S, Merikle E, Martenyi F, et al. Estimating
Alzheimer’s Disease Progression Rates from Normal Cognition Through Mild Cognitive
Impairment and Stages of Dementia. Current Alzheimer Research. 2018 Jul 1;15(8):777–
88.
28. Vermunt L, Sikkes SAM, van den Hout A, Handels R, Bos I, van der Flier WM, et al.
Duration of preclinical, prodromal, and dementia stages of Alzheimer’s disease in relation
to age, sex, and APOE genotype. Alzheimers Dement. 2019;15(7):888–98.
29. Soto-Gordoa M, Arrospide A, Moreno-Izco F, Martínez-Lage P, Castilla I, Mar J. Projecting
Burden of Dementia in Spain, 2010–2050: Impact of Modifying Risk Factors. Journal of
Alzheimer’s Disease. 2015 Jan 1;48(3):721–30.
30. Anderson R, Knapp M, Wittenberg R, Handels R, Schott JM. Economic Modelling of
Disease-Modifying Therapies in Alzheimer’s Disease [Internet]. London: Personal Social
Services Research Unit; 2018 Mar. Available from:
https://www.lse.ac.uk/cpec/assets/documents/EconomicmodellingAD.pdf
31. Pandya A, Sy S, Cho S, Alam S, Weinstein MC, Gaziano TA. Validation of a
Cardiovascular Disease Policy Microsimulation Model Using Both Survival and Receiver
Operating Characteristic Curves. Med Decis Making. 2017 Oct 1;37(7):802–14.
32. Goldman DP, Leaf DE, Sullivan J, Tysinger B. The Future Elderly Model: Technical
Documentation [Internet]. University of Southern California; 2018 Aug. Available from:
https://healthpolicy.app.box.com/v/FEM-Appendix
33. Zissimopoulos J, Crimmins E, St.Clair P. The Value of Delaying Alzheimer’s Disease
Onset. Forum for Health Economics and Policy. 2015 Jan 1;18(1):25–39.
34. Zissimopoulos JM, Tysinger BC, St. Clair PA, Crimmins EM. The Impact of Changes in
Population Health and Mortality on Future Prevalence of Alzheimer’s Disease and Other
Dementias in the United States. J Gerontol B Psychol Sci Soc Sci. 2018 Apr
16;73(suppl_1):S38–47.

114
35. Sonnega A, Faul JD, Ofstedal MB, Langa KM, Phillips JW, Weir DR. Cohort Profile: the
Health and Retirement Study (HRS). Int J Epidemiol. 2014 Apr 1;43(2):576–85.
36. Ofstedal M, Fisher G, Herzog AR, Herzog AR. Documentation of Cognitive Functioning
Measures in the Health and Retirement Study. Ann Arbor, Michigan: Institute for Social
Research, University of Michigan; 2005.
37. Brandt J, Spencer M, Folstein M. The Telephone Interview for Cognitive Status.
Neuropsychiatry, Neuropsychology, & Behavioral Neurology Summer 1988.
1988;1(2):111–8.
38. Fong TG, Fearing MA, Jones RN, Shi P, Marcantonio ER, Rudolph JL, et al. Telephone
Interview for Cognitive Status: Creating a crosswalk with the Mini-Mental State
Examination. Alzheimer’s & Dementia. 2009 Nov 1;5(6):492–7.
39. Crimmins EM, Kim JK, Langa KM, Weir DR. Assessment of Cognition Using Surveys and
Neuropsychological Assessment: The Health and Retirement Study and the Aging,
Demographics, and Memory Study. J Gerontol B Psychol Sci Soc Sci. 2011 Jul;66B(Suppl
1):i162–71.
40. Langa KM, Kabeto M, Weir D. Report on race and cognitive impairment using HRS in
2010 Alzheimer’s disease facts and figures. Retrieved July. 2010;12:2010.
41. Hurd MD, Martorell P, Delavande A, Mullen KJ, Langa KM. Monetary Costs of Dementia
in the United States. New England Journal of Medicine. 2013 Apr 4;368(14):1326–34.
42. Chen Y, Tysinger B, Crimmins E, Zissimopoulos JM. Analysis of dementia in the US
population using Medicare claims: Insights from linked survey and administrative claims
data. Alzheimer’s & Dementia: Translational Research & Clinical Interventions. 2019 Jan
1;5:197–207.
43. McCammon RJ, Fisher GG, Hassan H, Faul JD, Rogers W, Weir DR. Health and
Retirement Study Imputation of Cognitive Functioning Measures: 1992-2016 [Internet].
Ann Arbor, Michigan: Survey Research Center, University of Michigan; 2019. Available
from: https://hrs.isr.umich.edu/sites/default/files/biblio/COGIMP9216_dd.pdf
44. Baumgart M, Snyder HM, Carrillo MC, Fazio S, Kim H, Johns H. Summary of the
evidence on modifiable risk factors for cognitive decline and dementia: A population-based
perspective. Alzheimer’s & Dementia. 2015 Jun 1;11(6):718–26.
45. Leaf DE, Tysinger B, Goldman DP, Lakdawalla DN. Predicting quantity and quality of life
with the Future Elderly Model. Health Economics [Internet]. 2020 [cited 2020 Oct
12];n/a(n/a). Available from: https://onlinelibrary.wiley.com/doi/abs/10.1002/hec.4169
46. Berrar D. Cross-Validation. In: Encyclopedia of Bioinformatics and Computational
Biology. 2019. p. 542–5.

115
47. Kohavi R. A Study of Cross-Validation and Bootstrap for Accuracy Estimation and Model
Selection. Ijcai. 1995;14(2):1137–45.
48. Tysinger B. Validating risk factor and chronic disease projections in the Future Adult
Model. International Journal of Microsimulation. 2020;13(3):54–69.
49. Hensel A, Angermeyer MC, Riedel-Heller SG. Measuring cognitive change in older adults:
reliable change indices for the Mini-Mental State Examination. Journal of Neurology,
Neurosurgery & Psychiatry. 2007 Dec 1;78(12):1298–303.
50. Stein J, Luppa M, Maier W, Wagner M, Wolfsgruber S, Scherer M, et al. Assessing
cognitive changes in the elderly: Reliable Change Indices for the Mini-Mental State
Examination. Acta Psychiatrica Scandinavica. 2012;126(3):208–18.
51. Hlávka JP, Yu JC, Lakdawalla DN. Crosswalk between the Mini-Mental State Examination
and the Telephone Interview for Cognitive Status (TICS-27/30/40). Alzheimer’s &
Dementia. 2022;18(11):2036–41.
52. Zhu F, Panwar B, Dodge HH, Li H, Hampstead BM, Albin RL, et al. COMPASS: A
computational model to predict changes in MMSE scores 24-months after initial assessment
of Alzheimer’s disease. Scientific Reports. 2016 Oct 5;6(1):34567.
53. Albrecht JS, Hanna M, Kim D, Perfetto EM. Predicting Diagnosis of Alzheimer’s Disease
and Related Dementias Using Administrative Claims. JMCP. 2018 Oct 26;24(11):1138–45.
54. Lin PJ, Neumann PJ. The economics of mild cognitive impairment. Alzheimer’s &
Dementia. 2013 Jan 1;9(1):58–62.
55. van Dyck CH, Swanson CJ, Aisen P, Bateman RJ, Chen C, Gee M, et al. Lecanemab in
Early Alzheimer’s Disease. New England Journal of Medicine. 2023 Jan 5;388(1):9–21.
56. Budd Haeberlein S, Aisen PS, Barkhof F, Chalkias S, Chen T, Cohen S, et al. Two
Randomized Phase 3 Studies of Aducanumab in Early Alzheimer’s Disease. J Prev
Alzheimers Dis [Internet]. 2022 Mar 18 [cited 2022 Mar 28]; Available from:
https://doi.org/10.14283/jpad.2022.30
57. Alzheimer’s Association. 2022 Alzheimer’s Disease Facts and Figures. Alzheimers Dement.
2022;18.
58. Mahase E. Alzheimer’s disease: FDA approves lecanemab amid cost and safety concerns.
BMJ. 2023 Jan 11;380:p73.
59. Lakdawalla DN, Doshi JA, Garrison LP, Phelps CE, Basu A, Danzon PM. Defining
Elements of Value in Health Care—A Health Economics Approach: An ISPOR Special Task
Force Report [3]. Value in Health. 2018 Feb 1;21(2):131–9.

116
60. Neumann PJ, Garrison LP, Willke RJ. The History and Future of the “ISPOR Value
Flower”: Addressing Limitations of Conventional Cost-Effectiveness Analysis. Value in
Health. 2022 Apr 1;25(4):558–65.
61. Wei Y, Heun-Johnson H, Tysinger B. Using dynamic microsimulation to project cognitive
function in the elderly population. PLOS ONE. 2022 Sep 15;17(9):e0274417.
62. Hlávka JP, Tysinger B, Yu JC, Lakdawalla DN. Access to Disease-Modifying Alzheimer’s
Therapies: Addressing Possible Challenges Using Innovative Payment Models. Value in
Health. 2022 Nov 1;25(11):1828–36.
63. Patterson MB, Mack JL, Neundorfer MM, Martin RJ, Smyth KA, Whitehouse PJ.
Assessment of Functional Ability in Alzheimer Disease: A Review and a Preliminary
Report on the Cleveland Scale for Activities of Daily Living. Alzheimer Disease &
Associated Disorders. 1992 Jan;6(3):145.
64. Bugliari D, Carroll J, Hayden O, Hayes J, Hurd M, Lee S, et al. RAND HRS Longitudinal
File 2020 (V1) Documentation [Internet]. 2023. Available from: https://www.rand.org/wellbeing/social-and-behavioral-policy/centers/aging/dataprod/hrs-data.html
65. Pashmdarfard M, Azad A. Assessment tools to evaluate Activities of Daily Living (ADL)
and Instrumental Activities of Daily Living (IADL) in older adults: A systematic review.
Med J Islam Repub Iran. 2020 Apr 13;34:33.
66. Lin G, Whittington MD, Wright A, Agboola F, Herron-Smith S, Pearson SD, et al.
Lecanemab for Early Alzheimer’s Disease. Institute for Clinical and Economic Review;
2023 Apr.
67. Cedarbaum JM, Jaros M, Hernandez C, Coley N, Andrieu S, Grundman M, et al. Rationale
for use of the Clinical Dementia Rating Sum of Boxes as a primary outcome measure for
Alzheimer’s disease clinical trials. Alzheimer’s & Dementia. 2013 Feb 1;9(1,
Supplement):S45–55.
68. Morris JC. The Clinical Dementia Rating (CDR): Current version and scoring rules.
Neurology. 1993;43:2412.
69. O’Bryant SE, Waring SC, Cullum CM, Hall J, Lacritz L, Massman PJ, et al. Staging
Dementia Using Clinical Dementia Rating Scale Sum of Boxes Scores. Arch Neurol. 2008
Aug;65(8):1091–5.
70. Veitch DP, Weiner MW, Aisen PS, Beckett LA, Cairns NJ, Green RC, et al. Understanding
disease progression and improving Alzheimer’s disease clinical trials: Recent highlights
from the Alzheimer’s Disease Neuroimaging Initiative. Alzheimer’s & Dementia. 2019 Jan
1;15(1):106–52.
71. Petersen RC, Aisen PS, Beckett LA, Donohue MC, Gamst AC, Harvey DJ, et al.
Alzheimer’s Disease Neuroimaging Initiative (ADNI). Neurology. 2010 Jan 19;74(3):201–
9.

117
72. Gustavsson A, Brinck P, Bergvall N, Kolasa K, Wimo A, Winblad B, et al. Predictors of
costs of care in Alzheimer’s disease: A multinational sample of 1222 patients. Alzheimer’s
& Dementia. 2011 May 1;7(3):318–27.
73. Lin G, Whittington M, Wright A, Agboola F, Herron-Smith S, Pearson S, et al. BetaAmyloid Antibodies for Early Alzheimer’s Disease: Effectiveness and Value; Evidence
Report. Institute for Clinical and Economic Review; 2023 Apr.
74. Grosse SD. Assessing cost-effectiveness in healthcare: history of the $50,000 per QALY
threshold. Expert Review of Pharmacoeconomics & Outcomes Research. 2008 Apr
1;8(2):165–78.
75. CareScout. Cost of Care Survey [Internet]. Genworth; 2021. Available from:
https://www.genworth.com/aging-and-you/finances/cost-of-care.html
76. Tahami Monfared AA, Tafazzoli A, Chavan A, Ye W, Zhang Q. The Potential Economic
Value of Lecanemab in Patients with Early Alzheimer’s Disease Using Simulation
Modeling. Neurol Ther. 2022 Sep 1;11(3):1285–307.
77. Revised clinical trial form for Alzheimer’s antibody warned of fatal brain bleeds [Internet].
[cited 2023 Sep 4]. Available from: https://www.science.org/content/article/revised-clinicaltrial-form-alzheimer-s-antibody-warned-fatal-brain-bleeds
78. Reish NJ, Jamshidi P, Stamm B, Flanagan ME, Sugg E, Tang M, et al. Multiple Cerebral
Hemorrhages in a Patient Receiving Lecanemab and Treated with t-PA for Stroke. New
England Journal of Medicine. 2023 Feb 2;388(5):478–9.
79. Medlinskiene K, Tomlinson J, Marques I, Richardson S, Stirling K, Petty D. Barriers and
facilitators to the uptake of new medicines into clinical practice: a systematic review. BMC
Health Services Research. 2021 Nov 5;21(1):1198.
80. Jung J, Feldman R. Racial-Ethnic Disparities in Uptake of New Hepatitis C Drugs in
Medicare. J Racial Ethn Health Disparities. 2017 Dec;4(6):1147–58.
81. Matthews KA, Xu W, Gaglioti AH, Holt JB, Croft JB, Mack D, et al. Racial and ethnic
estimates of Alzheimer’s disease and related dementias in the United States (2015–2060) in
adults aged ≥65 years. Alzheimer’s & Dementia. 2019 Jan 1;15(1):17–24.
82. Leaf DE. Health Utilities Index Data Derived from the 2000 Health and Retirement Study.
Health and Retirement Study; 2015 Feb.
83. Green C, Handels R, Gustavsson A, Wimo A, Winblad B, Sköldunger A, et al. Assessing
cost-effectiveness of early intervention in Alzheimer’s disease: An open-source modeling
framework. Alzheimer’s & Dementia. 2019 Oct 1;15(10):1309–21.
84. Canady VA. FDA approves first drug therapy for Alzheimer’s in 18 years. Mental Health
Weekly. 2021;31(23):3–4.

118
85. Cummings J, Lee G, Zhong K, Fonseca J, Taghva K. Alzheimer’s disease drug development
pipeline: 2021. Alzheimer’s & Dementia: Translational Research & Clinical Interventions.
2021;7(1):e12179.
86. CDR® Dementia Staging Instrument | Knight Alzheimer Disease Research Center |
Washington University in St. Louis [Internet]. [cited 2022 Apr 6]. Available from:
https://knightadrc.wustl.edu/professionals-clinicians/cdr-dementia-staging-instrument/
87. Biogen. A Phase 3 Multicenter, Randomized, Double-Blind, Placebo-Controlled, ParallelGroup Study to Evaluate the Efficacy and Safety of Aducanumab (BIIB037) in Subjects
With Early Alzheimer’s Disease [Internet]. clinicaltrials.gov; 2021 Aug [cited 2022 Apr 5].
Report No.: NCT02484547. Available from:
https://clinicaltrials.gov/ct2/show/NCT02484547
88. Farrer LA, Cupples LA, Haines JL. Effects of age, sex, and ethnicity on the association
between apolipoprotein e genotype and alzheimer disease: a meta-analysis. JAMA.
1997;278(16):1349–56.
89. Lee E, Gatz M, Tseng C, Schneider LS, Pawluczyk S, Wu AH, et al. Evaluation of
Medicare Claims Data as a Tool to Identify Dementia. Journal of Alzheimer’s Disease.
2019 Jan 1;67(2):769–78.
90. Rabinovici GD, Gatsonis C, Apgar C, Chaudhary K, Gareen I, Hanna L, et al. Association
of Amyloid Positron Emission Tomography With Subsequent Change in Clinical
Management Among Medicare Beneficiaries With Mild Cognitive Impairment or
Dementia. JAMA. 2019 Apr 2;321(13):1286–94.
91. U.S. Food & Drug Administration. Aduhelm injection, highlights of prescribing information
[Internet]. 2021. Available from:
https://www.accessdata.fda.gov/drugsatfda_docs/label/2021/761178s000lbl.pdf
92. Wester C. Hepatitis C Virus Clearance Cascade — United States, 2013–2022. MMWR
Morb Mortal Wkly Rep [Internet]. 2023 [cited 2023 Sep 4];72. Available from:
https://www.cdc.gov/mmwr/volumes/72/wr/mm7226a3.htm
93. Biogen Announces Reduced Price for ADUHELM® to Improve Access for Patients with
Early Alzheimer’s Disease | Biogen [Internet]. [cited 2023 Sep 4]. Available from:
https://investors.biogen.com/news-releases/news-release-details/biogen-announcesreduced-price-aduhelmr-improve-access-patients
94. New Alzheimer’s Drugs Spark Hope for Patients and Cost Concerns for Medicare
[Internet]. KFF. 2023 [cited 2023 Sep 4]. Available from: https://www.kff.org/policywatch/new-alzheimers-drugs-spark-hope-for-patients-and-cost-concerns-for-medicare/
95. Lam J, Hlávka J, Mattke S. The Potential Emergence of Disease-Modifying Treatments for
Alzheimer Disease: The Role of Primary Care in Managing the Patient Journey. J Am
Board Fam Med. 2019 Nov 1;32(6):931–40.

119
96. Mattke S, Cho SK, Bittner T, Hlávka J, Hanson M. Blood-based biomarkers for
Alzheimer’s pathology and the diagnostic process for a disease-modifying treatment:
Projecting the impact on the cost and wait times. Alzheimer’s & Dementia: Diagnosis,
Assessment & Disease Monitoring. 2020 Jan 1;12(1):e12081.

Abstract (if available)

Abstract Alzheimer’s disease (AD) is the most common cause of dementia, accounting for an estimated 60% to 80% of cases. It is a neurodegenerative disease characterized by symptoms like loss of cognitive function, functional impairment, and neuropsychological symptoms. Without effective treatment, the number of Americans aged 65 and older living with AD dementia is estimated to grow from 6.7 million in 2023 to 13.9 million in 2060. US FDA recently approved two disease modifying therapies (DMTs) for AD, aducanumab and lecanemab. The emergence of AD DMTs brings opportunities and challenges, as knowledge on DMTs’ long-term value and eligible patient population size plays important role in policy decision making.

This dissertation encompasses three aims: (1) to develop and validate a population-level microsimulation model to project cognitive trajectories across the full AD continuum; (2) to extrapolate AD DMT’s clinical trial results to policy-relevant outcomes with a longer follow-up period with microsimulation; and (3) to estimate the size of treatment-eligible patient population in the US for AD DMTs and evaluate patient accessibility under specialist capacity constraints.

We developed and validated a microsimulation model to project trajectories in cognition, based on nationally representative longitudinal data of the US population aged 51 and older. We applied this model to extrapolate lecanemab’s clinical trial results to policy-relevant outcomes with a longer follow up period. Lastly, we estimated the size of treatment-eligible patient population in the US for AD DMTs based on the prevalence of cognitive impairment, diagnosis rates, and specialist access constraints, using aducanumab as a meaningful example.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

Pdf

Alzheimer's disease: risk factors, value, and alternative payment models

PDF

The value of novel antihyperlipidemic treatments in the U.S. healthcare system: Reducing the burden of cardiovascular diseases and filling the gap of low adherence in statins

PDF

Value in oncology care and opportunities for improvement

PDF

Dihydromyricetin as a potential therapeutic for Alzheimer's disease-like changes induced by social isolation

PDF

Comparison of participant and study partner predictions of cognitive impairment in the Alzheimer's disease neuroimaging initiative 3 study

PDF

Crohn’s disease: health outcomes and resource utilization in the biologic era

PDF

Essays in opioid use and abuse

PDF

Long-term blood pressure variability across the clinical and biomarker spectrum of Alzheimer’s disease

PDF

Insulin sensitivity in cognition, Alzheimer's disease and brain aging

PDF

Documenting the impact of under-treatment for chronic disorders

PDF

The impact of treatment decisions and adherence on outcomes in small hereditary disease populations

PDF

Assessing value defects in limb preservation care

PDF

The effects of a fasting mimicking diet (FMD) on mouse models of Alzheimer's and Parkinson's disease

PDF

Memory abnormalities in Alzheimer's disease and anxiety models

PDF

Understanding the value of hemophilia therapies

PDF

Delivering better care for children with special health care needs: analyses of patient-centered medical home and types of insurance

PDF

Age dependent modulation of synaptic plasticity and insulin mimetic effect of lipoic acid on a 3xTg-AD mouse model of Alzheimer's disease: implications as a therapeutic/nutraceutical agent

PDF

Neuroinflammation and ApoE4 genotype in at-risk female aging: implications for Alzheimer's disease

PDF

Sex differences in the TgF344-AD rat model: Investigating the behavior, pathology, and neuroanatomical structures

PDF

Phenotypic and multi-omic characterization of novel C. elegans models of Alzheimer's disease

Asset Metadata

Creator Wei, Yifan (author)

Core Title Disease modifying treatments for Alzeimer's disease: modeling, value assessment and eligible patients

School School of Pharmacy

Degree Doctor of Philosophy

Degree Program Health Economics

Degree Conferral Date 2023-12

Publication Date 10/26/2024

Defense Date 10/19/2023

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

Tag Alzheimer's disease,disease-modifying treatment,economic evaluation,microsimulation,OAI-PMH Harvest

Format theses (aat)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Goldman, Dana (committee chair), Joyce, Geoffrey (committee member), Lakdawalla, Darius (committee member), Seabury, Seth (committee member), Tysinger, Bryan (committee member)

Creator Email effieywei@gmail.com,weiyifan@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC113761203

Unique identifier UC113761203

Identifier etd-WeiYifan-12440.pdf (filename)

Legacy Identifier etd-WeiYifan-12440

Document Type Dissertation

Format theses (aat)

Rights Wei, Yifan

Internet Media Type application/pdf

Type texts

Source 20231030-usctheses-batch-1103 (batch), 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 author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email cisadmin@lib.usc.edu