All analyses were based on a priori hypotheses, with functional
status and nursing home admissions as the primary outcomes.
Base-line characteristics of the participants were added to the intention-to-treat models of the effects of the intervention. Proportionalhazards models were used for survival data. For functional status at
three years, we used repeated-measures regression analyses, adding
functional status at one and two years to the models after ascertaining
that there was no interaction between time and treatment effect. In
addition, standard and polychotomous logistic-regression techniques
were used. The effects of the intervention on the number of hospital
admissions for acute care, short-term nursing home admissions, and
visits to physicians were based on multivariate Poisson regression
models corrected for overdispersion