Developing a novel algorithm to identify incident and prevalent dementia in Medicare claims-the ARIC Study

Area of Science:

• Pharmacoepidemiology
• Geriatric Medicine
• Health Informatics

Background:

• Accurate dementia ascertainment is vital for real-world drug safety and efficacy studies.
• Existing dementia identification algorithms may lack robustness in large healthcare datasets.
• The Atherosclerosis Risk in Communities (ARIC) Study provides a valuable cohort for validation.

Purpose of the Study:

• To develop and validate novel dementia identification algorithms using Medicare claims data.
• To compare the performance of newly developed algorithms against existing ones (Jain, Bynum, Lee).
• To evaluate the impact of algorithm selection on estimating drug effects on dementia risk.

Main Methods:

• Developed incident and prevalent dementia algorithms using Medicare inpatient, outpatient, and prescription claims.
• Validated algorithms against the ARIC Study's syndromic dementia classification.
• Assessed algorithm effectiveness in a Medicare sample evaluating liraglutide versus dipeptidyl peptidase 4 inhibitors (DPP4i) for dementia risk.

Main Results:

• The incident dementia algorithm achieved a positive predictive value (PPV) of 69.2% and specificity of 99.0%.
• Performance was comparable to existing algorithms, with PPVs ranging from 58.7% to 68.6%.
• Algorithm choice significantly influenced the estimated 3-year adjusted risk difference (aRD) for dementia between liraglutide and DPP4i.

Conclusions:

• Novel dementia identification algorithms using Medicare claims demonstrate robust performance.
• Algorithm selection critically impacts treatment effect estimates in pharmacoepidemiologic research.
• Improved dementia ascertainment methods are essential for reliable real-world drug assessment.