Skip to content

Abstract Search

Aging

Evaluating Linked ICD-10 Medicare Claims Data as a Method of Dementia Case Ascertainment in Research Settings Joya Bhattacharyya* Joya Bhattacharyya Yi Chen Kan Z. Gianattasio Francine Grodstein Bryan D. James Ali Moghtaderi Christina Prather David Rein Raj C. Shah Emma K. Stapp Melinda C. Power

Background:  The United States transitioned to use of ICD-10 in 2015.  Limited previous research validating use of Medicare claims data as a case-finding strategy for identifying persons living with dementia has considered ICD-10 code definitions.   Here, we identified six ICD-10 code definitions for dementia from recent high-impact work and aimed to compare their sensitivity, specificity, and accuracy relative to research-based dementia ascertainment.

Methods: Eligible participants from 5 Rush Alzheimer’s Disease Center (RADC) cohort studies were aged 65 years or older, enrolled in Medicare fee-for-service, and consented to Medicare claims data linkage.  We compared dementia status at each annual RADC cohort study visit between 2015 and 2019 to dementia status based on presence of ICD-10 codes denoting dementia within a 12 month period centered on the RADC cohort study visit for six ICD-10 code definitions for dementia identified from the literature.

Results: All six code definitions had high accuracy (>85%).  Five of the six code definitions had high specificity (>90%) but low sensitivity (<60%), while one code definition had greater sensitivity (80%) at the expense of slightly lower specificity (87%).  For all code definitions, Black participants were more likely to meet research-based but not claims-based criteria for dementia (i.e., be underdiagnosed in clinical settings) than their White peers. Age was negatively associated with accuracy.

Conclusions: ICD-10 Medicare claims data can be used to ascertain dementia cases, but selection of code definition impacts sensitivity, specificity, and accuracy of results, and performance varied by participant characteristics.  Understanding the measurement properties of the code definition used may help inform bias mitigation strategies in studies using Medicare claims data to ascertain dementia.