Patterns and predictors of attrition in a trial of a housing intervention for homeless people with mental illness.

PURPOSE: Participant retention is an important challenge in longitudinal research on homeless people. High attrition can threaten validity, and may represent lost opportunities to deliver interventions. In this article, we report on attrition in the At Home/Chez Soi study, a multi-site randomized controlled trial of a housing intervention for homeless people with mental illness. METHODS: We first calculate life tables, and then use clustered logistic regression to implement a discrete-time survival model. We use splines and indicator variables to capture non-linear and group-specific variation over time in the hazard function. As potential predictors, we consider study group, site, date of recruitment, age, sex, baseline substance dependence, baseline psychotic disorder, time homeless in life, community functioning, and education. RESULTS: The study recruited 2,148 homeless people with mental illness. Of these, 1,158 were randomized to the housing first intervention (HF), and 990 to treatment as usual (TAU). Excluding 79 people known to have died, attrition was 14%. This proportion was higher in TAU than in HF (21 vs. 8%, p < 0.01). Attrition was significantly lower in one site than elsewhere, and was also lower among those with substance dependence (13 vs. 18%, p < 0.01) and among those enrolled earlier in the study. The hazard varied over time in complex ways. CONCLUSIONS: Results imply that study factors are more important than participant characteristics as determinants of retention, and that the immediate period after randomization is a crucial one. The high overall retention demonstrates the effectiveness of existing techniques for retaining participants.