VALIDATING A MEDICARE CLAIMS-BASED MODEL TO CLASSIFY PHENOTYPIC FRAILTY IN OLDER ADULTS

• Published in: Innovation in aging 2017-07, Vol.1 (suppl_1), p.381-381
• Main Authors: Cuthbertson, C.C., Lund, J.L., Sturmer, T., Faurot, K.R., Bandeen-Roche, K.J., Jonsson Funk, M., Palta, P., Kucharska-Newton, A.
• Format: Article
• Language: English
• Subjects: Abstracts
• ISSN: 2399-5300; 2399-5300
• DOI: 10.1093/geroni/igx004.1383

Medicare claims are increasingly used for epidemiologic studies in older adults. Claims capture longitudinal healthcare utilization (diagnoses, procedures, and medication dispensing), but lack clinical measures including frailty, which may confound or modify observed associations. Researchers previously developed a Medicare claims-based model (MCBM, 20 diagnoses, procedures, and durable medical equipment indicators) to predict severe dependence in activities of daily living as a proxy for frailty. Using Medicare claims linked with clinical data from Atherosclerosis Risk in Communities (ARIC) cohort participants (2011–2013), we assessed the validity of the MCBM to predict phenotypic frailty (PF), derived from cohort data. We described the prevalence of MCBM indicators (e.g., skin ulcer) among participants classified as frail, pre-frail, or robust and computed measures of model discrimination (c-statistic), calibration (Hosmer-Lemeshow test), and predictive validity (mortality using Cox proportional hazards models). Of 3,146 participants (median age: 75, 60% women, 22% Black), 7% were frail. The prevalence of MCBM indicators was highest among frail participants. The mean predicted probability of MCBM-derived frailty was 6.9% (range: 0.4%-78.4%). Model discrimination for prediction of PF was fair (c-statistic=0.744), while calibration was good (p-value=0.51). The crude hazard ratio (HR) for mortality (n=71, maximum follow-up=18 months) comparing older adults with high (>20%) versus low (