Hierarchical components of physical frailty predicted incidence of dependency in a cohort of elderly women

Objective: To identify the most important predictors of early disability incidence and devise a simple score of physical frailty. Methods: A cohort of 545 high-functioning women aged 75 years and older was followed for 7 years. Every year, the self-reported loss of at least one instrumental activity of daily living was chosen as definition of disability. An extension of the logistic regression for repeated responses, the random-effect model, was used to assess the effects of baseline predictors. The regression coefficients of the final multivariate model were scaled and rounded to create a practical score. Results: The proportion of women reporting disability increased from 22.1% to 52.1% throughout the follow-up. In the multivariate model, increasing age, lower performances in mobility and balance tests, bad perceived health, lower muscle strength, higher body mass index, lower educational level, and lower reported physical activity were strong predictors of disability. Evaluating the predictive value of the simplified predictive score on an independent cohort gave a c-statistic equal to .71. Conclusion: The use of a powerful fitting method allows to establish a hierarchy between the components of physical frailty and to provide a predictive score with substantial practical value for clinicians and public health professionals. (c) 2005 Elsevier Inc. All fights reserved.