Medication-Wide Discovery of Candidate Medications Associated with Alzheimer’s Disease and Related Dementias Using Explainable Artificial Intelligence and Longitudinal Real-World Data

Objectives: Alzheimer’s disease and related dementias (ADRD) impose substantial clinical and public health burdens, while treatment options remain limited. We developed a novel mechanism-agnostic Medication-Wide Association Study (MWAS) approach to identify medications associated with ADRD using longitudinal electronic health records and explainable artificial intelligence (AI). Methods: Veterans Health Administration data from 1999–2022 were used to assemble 505,817 matched case–control pairs (1,011,634 Veterans). Cases were defined by ADRD-related diagnosis codes or ADRD-related medication prescriptions, and controls were free of ADRD diagnosis and ADRD-related medication use. Historical features were aggregated into annual intervals ending 1 year before the index date. An upgraded Hybrid Value-Aware Transformer (HVAT 2.0) jointly modeled longitudinal and nonlongitudinal data, including cumulative medication dose. To enhance interpretability, we applied a medication-specific impact score method to estimate model-derived associations between medication exposure and ADRD risk. Results: The model demonstrated stable performance across data partitions, with area under the receiver operating characteristic curve values of 0.791 in the training set, 0.772 in the validation set, and 0.775 in the testing set. Metolazone and varenicline had the most negative population-level impact scores: −0.196 per 1800 mg and −0.134 per 280 mg cumulative dose unit, respectively, suggesting potentially protective associations with new-onset ADRD. Although individual-level impact scores varied, most exposed patients had negative scores, including 12,020 of 12,480 metolazone users (96%) and 8341 of 8786 varenicline users (95%). Conclusions: MWAS combining dose-aware longitudinal modeling and explainable AI can generate medication-repurposing signals from population-scale real-world data. The hypotheses generated require further epidemiologic and clinical validation.

Authors

Institutions

Publication Details

Journal
Medical Sciences
Published
2026-11-13
DOI
https://doi.org/10.3390/medsci14070778
Primary Topic
Pharmaceutical Practices and Patient Outcomes
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Medication-Wide Discovery of Candidate Medications Associated with Alzheimer’s Disease and Related Dementias Using Explainable Artificial Intelligence and Longitudinal Real-World Data

Mark W. Logue, Debby W. Tsuang, Tracey H. Taveira, EDWARD Y. ZAMRINI et al.
4 links
Medical Sciences
Pharmaceutical Practices and Patient Outcomes
article

Medication-Wide Discovery of Candidate Medications Associated with Alzheimer’s Disease and Related Dementias Using Explainable Artificial Intelligence and Longitudinal Real-World Data

Mark W. Logue, Debby W. Tsuang, Tracey H. Taveira, EDWARD Y. ZAMRINI, Qing Zeng‐Treitler, Siamack Ayandeh, Yijun Shao, Ying Yin, John E. McGeary, Yan Cheng, Ali Ahmed
article en

Abstract

Objectives: Alzheimer’s disease and related dementias (ADRD) impose substantial clinical and public health burdens, while treatment options remain limited. We developed a novel mechanism-agnostic Medication-Wide Association Study (MWAS) approach to identify medications associated with ADRD using longitudinal electronic health records and explainable artificial intelligence (AI). Methods: Veterans Health Administration data from 1999–2022 were used to assemble 505,817 matched case–control pairs (1,011,634 Veterans). Cases were defined by ADRD-related diagnosis codes or ADRD-related medication prescriptions, and controls were free of ADRD diagnosis and ADRD-related medication use. Historical features were aggregated into annual intervals ending 1 year before the index date. An upgraded Hybrid Value-Aware Transformer (HVAT 2.0) jointly modeled longitudinal and nonlongitudinal data, including cumulative medication dose. To enhance interpretability, we applied a medication-specific impact score method to estimate model-derived associations between medication exposure and ADRD risk. Results: The model demonstrated stable performance across data partitions, with area under the receiver operating characteristic curve values of 0.791 in the training set, 0.772 in the validation set, and 0.775 in the testing set. Metolazone and varenicline had the most negative population-level impact scores: −0.196 per 1800 mg and −0.134 per 280 mg cumulative dose unit, respectively, suggesting potentially protective associations with new-onset ADRD. Although individual-level impact scores varied, most exposed patients had negative scores, including 12,020 of 12,480 metolazone users (96%) and 8341 of 8786 varenicline users (95%). Conclusions: MWAS combining dose-aware longitudinal modeling and explainable AI can generate medication-repurposing signals from population-scale real-world data. The hypotheses generated require further epidemiologic and clinical validation.

Medical SciencesVol. 14(7)
Boston University (US), University of Rhode Island (US), Georgetown University (US), George Washington University (US), University of Washington (US), University of Utah (US), Brown University (US), VA Boston Healthcare System (US), Providence VA Medical Center (US), VA Puget Sound Health Care System (US), Washington DC VA Medical Center (US)
U.S. Department of Veterans Affairs, National Institutes of Health, National Institute on Aging
Good health and well-being
Pharmaceutical Practices and Patient Outcomes
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon