Multivariable-adjusted cumulative incidence estimates and hazard ratios were calculated using Cox proportional hazards regression analysis. Clinical variables measured over time were treated as time-varying covariates in the regression analysis with values imputed between the mid-points of available measurement occasions. Starting from a full model that included all hypothesized risk factors and potential confounders, we employed a backwards model selection procedure where the least significant predictor variables were removed one at a time until only statistically significant variables remained. Because of its importance as a hypothesized risk factor, multiple attempt history was retained in all models regardless of statistical significance. Because the King et al. (2009a) Youth-Nominated Support Team intervention was not found to have any trend or significant effect on the outcome of suicide attempts in a randomized controlled trial, the YST intervention was not included as a variable in analyses.
