Sample size calculation to demonstrate effects on fall rate reduction were based on a review of the existing litera-ture.25 Using an exact test with an alpha level of 0.05 and a power of 0.80, a sample size of 250 subjects was considered to be necessary to detect a 20% reduction of falls during the follow-up year. The Wilcoxon rank sum test was used to examine differences in continuous variables and the chi-square test for categorical variables (e.g., Barthel score, MMSE, state of health) with the modification for linear trend when appropriate. Calculations assumed binomial distribution of end-points. The number of falls per person per year in each group was compared using the negative binomial distribution. 26 The incidence rate ratio (IRR) with the corresponding 95% confidence intervals (CIs) was calculated. The model takes into account all falls and adjusts for the individual follow-up time in the trial. The first five falls for
each participant were used in this analysis rather than all falls (maximum 22) to avoid over-weighting by subjects who fell more then five times. If it is assumed that everyone falls occasionally, whether the person falls once during the trial period may not be the best indication of whether the intervention is effective. Thus, a subgroup analysis of people with no or one fall versus people with two or more falls was also conducted. This analyticalstrategy was decided a priori.