We then took variables with a log-rank test p-value of less than .10 and entered them into a multivariate Cox Proportional Hazards regression model.

Adjusted hazard ratios and their 95% confidence intervals were examined to identify significant independent predictors of retention in care.

These analyses were conducted using SUDAAN statistical software so as to account for any intra-cluster correlation of client data within sites.