TARGETED LITERATURE REVIEW TO IDENTIFY US CLAIMS-BASED ALGORITHMS FOR ALZHEIMER’S DISEASE
Author(s)
Amos Bugri, MSc, PharmD, April N. Naegeli, DrPH, MPH, Margaret Hoyt, PhD, MPH, MS, Julie Beyrer, DrPH, MPH, MTSC;
Eli Lilly and Company, Indianapolis, IN, USA
Eli Lilly and Company, Indianapolis, IN, USA
OBJECTIVES: Healthcare claims data are used in Alzheimer’s disease (AD) research, yet coding algorithms may not be fit for purpose in different research contexts. This review aimed to identify and describe studies evaluating the validity of AD coding algorithms in United States (US) claims data.
METHODS: A targeted literature review was conducted using a natural language processing (NLP) search strategy for studies evaluating US claims-based algorithms for AD against a reference standard and reporting validation metric(s) (e.g. sensitivity, specificity, etc.) from January 1, 2015 to September 12, 2025 (Medline(R)) and October 14, 2025 (Embase(R)). An artificial intelligence tool (Microsoft Copilot(R)) abstracted data, and two independent reviewers analyzed and summarized the data.
RESULTS: Of the 274 articles screened, 17 met the inclusion criteria. Most studies (15) evaluated algorithms for identifying any type of dementia, while one study included an algorithm for identifying mild cognitive impairment (MCI) due to any cause. A single study assessed an algorithm for MCI due to Alzheimer’s disease, and another evaluated an algorithm for early symptomatic AD and MCI due to AD. ICD-9-CM codes were used in 11 studies and ICD-10-CM in 10 studies, with 5 studies including medications (acetylcholinesterase inhibitors and/or memantine) in the algorithm. The Chronic Conditions Warehouse algorithms for dementia was evaluated in 12 studies, and the Bynum dementia algorithm in 4 studies. Dementia algorithm sensitivity varied from 10-89%, while specificity was in the 80-90% range for all except one study. Dementia algorithm PPV varied from 37-89%. For MCI, sensitivity was 0.03%, specificity >99%, and PPV 8-18%. For early symptomatic AD, sensitivity was 82%.
CONCLUSIONS: Nearly all algorithms were for dementia and not specific to AD. While dementia algorithm specificities were frequently high (>80%), sensitivities and PPVs had wider variation and lower ranges. Algorithms are needed for healthcare claims data research on AD, especially in pre-dementia stages.
METHODS: A targeted literature review was conducted using a natural language processing (NLP) search strategy for studies evaluating US claims-based algorithms for AD against a reference standard and reporting validation metric(s) (e.g. sensitivity, specificity, etc.) from January 1, 2015 to September 12, 2025 (Medline(R)) and October 14, 2025 (Embase(R)). An artificial intelligence tool (Microsoft Copilot(R)) abstracted data, and two independent reviewers analyzed and summarized the data.
RESULTS: Of the 274 articles screened, 17 met the inclusion criteria. Most studies (15) evaluated algorithms for identifying any type of dementia, while one study included an algorithm for identifying mild cognitive impairment (MCI) due to any cause. A single study assessed an algorithm for MCI due to Alzheimer’s disease, and another evaluated an algorithm for early symptomatic AD and MCI due to AD. ICD-9-CM codes were used in 11 studies and ICD-10-CM in 10 studies, with 5 studies including medications (acetylcholinesterase inhibitors and/or memantine) in the algorithm. The Chronic Conditions Warehouse algorithms for dementia was evaluated in 12 studies, and the Bynum dementia algorithm in 4 studies. Dementia algorithm sensitivity varied from 10-89%, while specificity was in the 80-90% range for all except one study. Dementia algorithm PPV varied from 37-89%. For MCI, sensitivity was 0.03%, specificity >99%, and PPV 8-18%. For early symptomatic AD, sensitivity was 82%.
CONCLUSIONS: Nearly all algorithms were for dementia and not specific to AD. While dementia algorithm specificities were frequently high (>80%), sensitivities and PPVs had wider variation and lower ranges. Algorithms are needed for healthcare claims data research on AD, especially in pre-dementia stages.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
EPH21
Topic
Epidemiology & Public Health
Topic Subcategory
Disease Classification & Coding
Disease
SDC: Neurological Disorders