Using claims data to predict dependency in activities of daily living as a proxy for frailty

Estimating drug effectiveness and safety among older adults in population-based studies using administrative health care claims can be hampered by unmeasured confounding as a result of frailty. A claims-based algorithm that identifies patients likely to be dependent, a proxy for frailty, may improve confounding control. Our objective was to develop an algorithm to predict dependency in activities of daily living (ADL) in a sample of Medicare beneficiaries.

Community-dwelling respondents to the 2006 Medicare Current Beneficiary Survey, >65 years old, with Medicare Part A, B, home health, and hospice claims were included. ADL dependency was defined as needing help with bathing, eating, walking, dressing, toileting, or transferring. Potential predictors were demographics, International Classification of Diseases, Ninth Revision Clinical Modification diagnosis/procedure and durable medical equipment codes for frailty-associated conditions. Multivariable logistic regression was used to predict ADL dependency. Cox models estimated hazard ratios for death as a function of observed and predicted ADL dependency.

Of 6391 respondents, 57% were female, 88% white, and 38% were ≥80. The prevalence of ADL dependency was 9.5%. Strong predictors of ADL dependency were charges for a home hospital bed (OR = 5.44, 95%CI = 3.28-9.03) and wheelchair (OR = 3.91, 95%CI = 2.78-5.51). The c-statistic of the final model was 0.845. Model-predicted ADL dependency of 20% or greater was associated with a hazard ratio for death of 3.19 (95%CI: 2.78, 3.68).

An algorithm for predicting ADL dependency using health care claims was developed to measure some aspects of frailty. Accounting for variation in frailty among older adults could lead to more valid conclusions about treatment use, safety, and effectiveness.