Risk factors for falling among community-dwelling seniors using home-care services: an extended hazards model with time-dependent covariates and multiple events.

The identification of risk factors for falls in longitudinal studies becomes difficult because of exposures that change during the follow-up and also because individual subjects may experience an event more than once. These issues have been neglected and improper statistical techniques have been used. The typical approaches have been to report the proportion of fallers or the time to first fall. Both avoid the underlying assumption of independence between events and discard pertinent data. We review the existing methods and propose a Cox hazards extension. We exemplify it in the study of potential risk factors associated with all falls in 959 seniors. Finally, we compare the results of the proposed Wei, Lin, & Weissfeld (WLW) method with those of several other techniques. Stable exposure variables measured at baseline and updated time-varying exposures include socio-demographic characteristics, BMI, nutritional risk, alcohol consumption, home hazards, gait and balance, and medications. Results demonstrate that the usual methods of analyzing risk factors for falling are inappropriate, as they produce considerable biases relative to the WLW model using time-dependent covariates. Results also show that modeling for first events may be inefficient, given that the risk of occurrence varies between falls.