Identifying Medicare beneficiaries with dementia.

Area of Science:

• Health Informatics
• Medical Diagnosis
• Epidemiology

Background:

• No prior validation of International Classification of Disease (ICD)-10 dementia diagnoses exists within Medicare claims data against a clinician-adjudicated standard.
• Electronic Health Record (EHR) and claims data linkage enables novel validation approaches.

Purpose of the Study:

• To examine the accuracy of claims-based dementia diagnoses using ICD-10 codes.
• To compare the performance of different ICD-10 dementia definitions against expert clinician adjudication.

Main Methods:

• Retrospective observational study involving 1002 Medicare beneficiaries aged 65+ from Mass-General-Brigham Healthcare.
• Dementia status was adjudicated by four specialists using a standardized review of medical records.
• Accuracy was measured using area under the receiver operating curve (AUROC) and calibration metrics, with 10-fold cross-validation.

Main Results:

• The best-performing ICD-10 definition showed excellent accuracy (CV-AUC 0.94) and good calibration.
• This definition outperformed an updated version of a previously validated ICD-9 dementia definition.
• The adjudicated prevalence of dementia in the study population was 7%.

Conclusions:

• This study provides the first validation of ICD-10 dementia codes against a robust clinical reference standard in Medicare claims.
• The optimal ICD-10 definition, incorporating codes with strong face validity, is a reliable tool for dementia ascertainment in claims data.