AUGMENTING STANDARD ALGORITHMS TO IMPROVE DEMENTIA DETECTION AND PREDICTIVE ACCURACY IN MEDICARE BENEFICIARIES

Sep 29, 2026

Upload by Bintang

Accurate identification of Alzheimer’s Disease and Related Dementias (ADRD) in Medicare populations is essential for research and care planning. This study tested whether a digital phenotype that integrates medication use and clinical diagnostic criteria improves ADRD ascertainment beyond the standard CMS Chronic Conditions Warehouse (CCW) algorithm based only on diagnosis codes. Using 205,856 home-health Medicare beneficiaries, two predictive models were built and evaluated on independent data using AUC, sensitivity, specificity, and predictive values.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages1

Reading Time3 min

AccessFree to download

Abstract

Study objective and data source

Phenotype design and modeling approach

Model evaluation metrics and results

Implications for ADRD classification

Innovation in Aging, 2025, Vol. 9, No. S2 1283 Abstract citation ID: igaf122.4280 AUGMENTING STANDARD ALGORITHMS TO IMPROVE DEMENTIA DETECTION AND PREDICTIVE ACCURACY IN MEDICARE BENEFICIARIES Sara Knox1, Janet Horn2, Kit Simpson2, and Bethany Wolf2, 1. Medical University of South Carolina, Charleston, South Carolina, United States, 2. Medical University of South Carolina, Medical University of South Carolina, South Carolina, United States Accurate identification of Alzheimer’s Disease and Related Dementias (ADRD) in Medicare populations is critical for research and care planning. This study evaluated whether incorporating medication use and clinical diagnostic criteria into a digital phenotype could improve ADRD ascertainment over the standard CMS Chronic Conditions Warehouse (CCW) algorithm, which relies solely on diagnosis codes. Using data from 205,856 Medicare beneficiaries receiving home health services, we developed two predictive models: one based on the CCW phenotype and a novel ADRD (novel) phenotype based on clinical diagnostic criteria that categorizes individuals as ADRD-positive, ADRD-negative, or uncertain. Individuals with uncertain ADRD status were excluded during model training but analyzed later. Both models incorporated demographic variables, healthcare utilization, psychiatric symptoms, cognitive function, and prescription patterns. Data imputation and stacked elastic net regression were used to develop parsimonious models. Model performance was evaluated using AUC, sensitivity, specificity, and predictive values on an independent test set. The CCW model achieved an AUC of 0.863 for predicting its phenotype, while the novel model achieved an AUC of 0.906 for its respective phenotype. When applied to individuals with uncertain ADRD status, model predictions differed, with the novel model classifying more individuals as ADRD-negative than the CCW model. Overall, our findings suggest that augmenting standard claims-based algorithms with medication and clinical diagnostic criteria improvesADRD classification, particularly in cases where diagnosis is ambiguous. The novel model may offer a more conservative and specific approach to dementia identification, enhancing population-level research and targeted interventions.

FAQs

本研究的核心目标是什么？

评估将用药信息与临床诊断标准纳入数字表型，能否提升ADRD识别与分型的预测准确性，而不仅仅依赖诊断编码的标准CCW算法。

研究使用了哪些数据和样本规模？

使用205,856名接受家庭健康服务的Medicare受益人数据来开发两种预测模型，并在独立测试集上进行评估。

与CCW模型相比，“novel”模型的主要表现差异是什么？

novel模型的AUC更高（0.906 vs 0.863），且在ADRD状态不确定的人群中，对ADRD阴性类别的判定更多，提示其在诊断模糊情况下可能更保守且更具特异性。

Sep 29, 2026

Upload by Bintang