Comparison of regression models for the analysis of fall risk factors in older veterans.

PURPOSE: To compare the performance of eight regression models for analyzing risk of falling, focusing on the effect of physical inactivity in older veterans.
METHODS: This study uses data from a fall risk factor screening and modification trial in community-dwelling Canadian male veterans of World War II or the Korean War, with falls ascertained prospectively using calendars and physical activity (PA) measured at baseline with a single global question. The effect of PA on falling was assessed using eight different multivariable regression models, with three models treating falling as a non-recurrent event whereas the other five models regard falls as recurrent events.
RESULTS: Recurrent event models showed that male veterans who reported being less active than their peers were 1.42 (1.02-1.97) to 2.46 (1.18-5.14) times more likely to fall than those who reported being about as or more active than their peers (n = 270; mean age +/- SD = 81.1 +/- 4.0 years). None of the non-recurrent event models detected a statistically significant association between PA and falls.
CONCLUSIONS: Risk of falling may be better analyzed using regression models for recurrent events. These results have important implications for the collection and analysis of fall outcome data.