Predictors and Consequences of the Timing and Accuracy of Clinical Dementia Diagnosis

Ali Moghtaderi, Ph.D.

Associate Professor of Health Policy and Management 

Milken Institute School of Public Health

The George Washington University

moghtaderi@email.gwu.edu

Rush University System for Health |

MELODEM Spring Working Group Session

​

February 19, 2026

​

Bryan D. James, Ph.D.

Associate Professor,

Rush Alzheimer’s Disease Center (RADC)

Dept. of Internal Medicine,

Rush University Medical Center (RUMC)

Bryan_James@Rush.edu

+

Bon Voyage, Melinda!

Melinda Power, ScD

Associate Director (HEOR/RWE), AstraZeneca

Funding and Disclosures

Funding: NIA

• R01AG072559 (PI: James & Power)
• R01AG079226 (PI: Grodstein & Bynum)
• P30AG010161 (PI: Schneider)
• R01AG017917 (PI: Bennett)
• R01AG022018 (PI: Barnes)
• K01AG050823 (PI: James; 2015-2020)

​

Consulting:

• Alzheimer’s Association (Facts & Figures Report)

​

Journal editorial roles:

• American Journal of Epidemiology (AJE) (Associate Editor for Social Media)
• Alzheimer’s & Dementia: Journal of the Alzheimer’s Association (Reviewing Editor)
• AJE Advances: Research in Epidemiology (Editor)

Rush Alzheimer’s Disease Center (RADC)

Faculty

Puja Agarwal, PhD

Neelum T. Aggarwal, MD

Sonal Agrawal, PhD

Konstantinos Arfanakis, PhD

Zoe Arvanitakis, MD, MS

Denis Avey, PhD

Lisa L. Barnes, PhD

David A. Bennett, MD (Director)

Patricia A. Boyle, PhD

Aron Buchman, MD

Yi Chen, PhD

Lianlian Du, PhD

Matra Estrella, PhD

Jose Farfel, MD, PhD

Debra Fleischman, PhD

Francine Grodstein, ScD

Bryan James, PhD

Alifiya Kapasi, PhD

Nicola Kearns, PhD

Melissa Lamar, PhD

Brittney Lange-Maia, PhD

Sue Leurgans, PhD

Katia Lopes, PhD

David X Marquez, PhD *visiting appt

Rupal Mehta, MD

Sukriti Nag, MD, PhD

Bernard Ng, PhD

Shahram Oveisgharan, MD

Victoria Poole, PhD

Julie A. Schneider, MD, MS

Raj C. Shah, MD

Ajay Sood, MD, PhD

Shinya Tasaki, PhD

Ricardo Vialle, PhD

Maude Wagner, PhD

Tianhao Wang, PhD

Yanling Wang, MD, PhD

Robert Wilson, PhD

Jishu Xu, MA

Jingyun Yang, PhD

Lei Yu, PhD

Andrea Zammit, PhD

​

Staff

~130 staff

Medicare Linkage Team

Fran Grodstein, PhD

Yi Chen, PhD

Ana Capuano, PhD (U of C)

Brittney Lange-Maia, PhD

Raj Shah, MD

Joseph Vanghelof, PhD

​

Data Documentation & Distribution Teams

Maylene Liang

Greg Klein

John Gibbons

Yan Li

Tiffany Wu

​

Research Education Component

Puja Agarwal, PhD

Lupe Romero

​

THANK YOU TO

​

Research Participants

Cohort study participants

Clinical trial participants

​

​

Funding

National Institute on Aging

Illinois Department Public Health

​

R01 MPI

Melinda Power, PhD, George Washington University

​

​

Jennifer Schrack for the invite!

• Why link Medicare claims to cohort data to study dementia?
• R01 Aim 1: Correlates of timely diagnosis of dementia
• R01 Aims 2&3: Associations with health & healthcare utilization (preliminary)
• R01 Associated findings

5

5

How can linking Medicare claims to cohort data help us to understand dementia?

6

Advantages of Medicare Linkage

• Health & healthcare utilization info not captured in cohort assessment

- Diagnoses (ICD), procedures (CPT), hospitalizations, preventive care, specialist visits, medications, costs

• Specific dates of medical encounters
• Not reliant on self report
• Claims predate cohort enrollment 🡪 observation of “the past”
• Linkage of claims to cohort data allows comparison to “gold standard”
• Self-report (cohort) to medical billing (claims)
• Clinical observation (claims) to robust uniform assessment (cohort)

​

7

Obstacles for Medicare Linkage

• Disparate data structures (e.g., one row per study visit vs one row per claim)
• Temporality: coverage / observation windows
• Incomplete claims coverage
• Consent to Medicare access; Linkage success rate; Medicare advantage vs FFS
• Only see what is billed for
• No test results, no staging, no labs, no clinician notes, etc.

8

RADC Medicare Linkage

Rush Alzheimer’s Disease Center (RADC) cohorts

Memory and Aging Project (n=2,144)

Religious Orders Study (n=1,307)

Minority Aging Research Study (n=402)

Clinical Core (n=250)

Latino Core (n=185)

CMS: Medicare claims records

• Medicare Parts A & B: 1991 - 2022
• Medicare Part D: 2006 - 2022
• Medicare Part C: 2015 - 2022

Linkage

Health Insurance Claim (HIC) number

Medicare Beneficiary ID (MBI) number

Social Security number

>4,200 linked to claims

(Now >4,700)

(Now through 2024)

9

https://www.radc.rush.edu/

10

RADC Cohorts www.radc.rush.edu

Religious Orders Study (ROS) 

Older Catholic nuns, priests, and brothers from more than 40 groups across US

​

Rush Memory and Aging Project (MAP) 

Older adults recruited from retirement communities / senior housing facilities throughout Chicagoland

​

Minority Aging Research Study (MARS) 

Older adults who self-identify as African American

​

African American Clinical Core (AA Core) 

Older adults from Chicago and suburbs, self-identify as African American

​

Latino CORE Study (LATC) 

Older adults who self-identify as Latino/Hispanic

​

​

• Harmonized annual cognitive assessment
• Numerous substudies, some unique to each cohort
• Brain donation required for ROS & MAP, voluntary for other cohorts

​

11

RADC cohort data

12

NIA LINKAGE Enclave

NIA-sponsored secure environment to:

• House CMS Medicare claims
• Upload and link cohort data to claims
• Perform analyses (SAS, R, STATA)

Contact: LINKAGE@acumenllc.com

https://www.nia.nih.gov/research/dbsr/nia-data-linkage-program-linkage

​

Pros: privacy/security; accessibility/sharing; low costs

Cons: New environment learning curve; Import/export slowdowns; Cohort-specific folders; paperwork!

13

Predictors and Consequences of the Timing and Accuracy of Clinical Dementia Diagnosis

​

R01AG072559

​

Aim 1: Characterize predictors of underdiagnosis (late or missed diagnosis) of clinical dementia

Hypothesis 1a: More likely to have late or missed dx: less education, Black or Latino race, male, living alone depressive symptoms, disabilities, comorbidities, less frequent medical visits

Hypothesis 1b (exploratory, in autopsied subset): Those with pathologic Alzheimer’s disease more likely to have timely dx

Aim 2: Determine the impact of timing of clinical dementia diagnosis on health outcomes

Aim 3: Determine the impact of timing of clinical dementia diagnosis on healthcare utilization & cost

14

https://journals.sagepub.com/doi/10.1177/23337214231213185

15

How important is a timely diagnosis of dementia?

Questions:

• How common is delayed diagnosis of dementia?
• ~50% of people with dementia are not diagnosed by the healthcare system.
• How many have delayed diagnosis?
• What contributes to delayed diagnosis?
• What are the harms (or benefits?) of delayed diagnosis?

16

Aim 1: Correlates of timely diagnosis of dementia

17

Aim 1a

https://alz-journals.onlinelibrary.wiley.com/doi/full/10.1002/alz.14067

18

Outcome: Underdiagnosis (late or missed)

Strategy: Compare presence/timing of clinical dx in claims (ICD-9 or ICD-10 diagnostic codes) to assessment of dementia in RADC cohort study

​

• Timely diagnosis: clinical dx within window of 3 yrs before or 1 yr after the date of incident RADC dementia
• Underdiagnosis: No clinical dx within timely window

​

​

​

19

Timely clinical DX

Timely clinical DX

Late clinical DX

19

Analytic sample

Mean=12.9 (sd=5.6)

Median=12

Inclusion:

• Have Medicare data (FFS)�as of 12/31/2019

​

• Have RADC dementia

​

^a Considering disruption of the COVID-19 pandemic on seeking and delivering healthcare, we restricted the present analyses to 12/31/2019.

^b We excluded 190 participants with cohort-assessed dementia at their baseline evaluation, for whom we did not have a date of incident diagnosis.

(83%)

20

Potential correlates

21

From RADC cohort data From Medicare claims

21

Multivariable analysis – Model 1

Notes: Model included participants with non-missing values in examined variables (N=710).

Odds Ratios for “Underdiagnosis”

22

Multivariable analysis – Model 2

Notes: Model included participants with non-missing values in examined variables (N=668).

Odds Ratios for “Underdiagnosis”

23

Multivariable analysis – Model 3

Notes: Model included participants with non-missing values in examined variables (N=333).

Odds Ratios for “Underdiagnosis”

24

Conclusions: correlates of underdiagnosis

• Half were not diagnosed in a timely fashion by the healthcare system

​

• Race & SES: Black older adults and those with lower income were more likely to be underdiagnosed

​

• Few other factors explained underdiagnosis other than cognitive function and comorbidities

25

[UNDER REVIEW]:

Associations of neuropathologies with timely diagnosis of dementia in healthcare settings

Yi Chen, Annie Chen, Melinda C. Power, Francine Grodstein, Alifiya Kapasi, Ana W. Capuano, Brittney S. Lange-Maia, Ali Moghtaderi, Emma K. Stapp, Joya Bhattacharyya, Raj C. Shah, Lisa L. Barnes, David A. Bennett, Bryan D. James

​

Rush Alzheimer's Disease Center, George Washington University, University of Chicago

​

26

Percent with timely diagnosis of dementia, according to presence of each neuropathology (n=500) [UNPUBLISHED]

N=394

N=226

N=428

N=112

27

Odds ratios for timely diagnosis by pathology [UNPUBLISHED]

Adjusted for age at dx, sex, race/ethnicity, education, time between dx and death

ADNC LATE-NC Vascular LB

Predicted probabilities of timely diagnosis of dementia:

28

Predicted probability of timely diagnosis, by number of neuropathologies present [UNPUBLISHED]

29

conclusions:

• Pathologic Alzheimer’s disease and LATE-NC are associated with a greater likelihood of timely diagnosis of dementia prior to death

​

• Having mixed pathologies is associated with greater likelihood of timely diagnosis of dementia

​

• May be due to differences in clinical manifestations?
• Clinicians more likely to recognize AD as dementia? (LATE-NC mimics AD symptoms)
• Dementia screening tests more likely pick up on AD/LATE-NC symptoms?
• More rapid decline for AD/LATE-NC / mixed pathologies 🡪 timely dx?

30

Aims 2 & 3: Impact of the Timing of Dementia on Health & Healthcare Utilization (Preliminary)

31

Obstacle to comparing Timely dx to Underdx

• Aim 1 showed those with timely diagnosis are different:
• Socio-demographic: more likely to be non-Latino White; higher income
• Clinical: more cognitively impaired; sicker

​

• Direct comparison of outcomes may be biased
• Factors influencing exposure assignment (i.e. receipt of a timely diagnosis) are also associated with outcomes
• “Confounding by indication” in epi, or “endogeneity” in econ

32

Research Design

• Compare participants with and without timely diagnosis who are otherwise similar

​

< 1 year

Cohort incident dementia

Healthcare DX

Cohort incident dementia

Timely DX

Under DX

Index date (T₀)

33

To make the groups similar:

• Weight by propensity score
• #1 Estimate probability of timely diagnosis based on extensive characteristics at index date (e.g. socio-demographics, cognition, health, lifestyle, personality)
• #2 Create stabilized inverse probability weights to balance confounders

​

• Loop randomization and weighting processes 5 times
• Pool all loops to form analytic sample

​

• Compare weighted outcomes for timely diagnosed v. under diagnosed
• Mixed effects models to account for clustering at person level due to repeated observations per person

​

34

Outcomes & Preliminary Findings

35

[Ali’s presentation goes here]

36