Survival analyses were used to construct multivariate models that predict time-to-suicide attempt over 1 year (Kalbfleisch and Prentice 2002). We considered a continuous measurement of time until first suicide attempt by assuming that attempts occurred at mid-month. In order to make use of the partial information available for those known to have made an attempt but missing the exact month, we imputed an attempt date equal to the mid-point of the time interval between the measurement occasion at which the indication of an event was provided and the last known attempt-free measurement occasion (21% of observed attempters). Those lost to follow-up prior to conclusion of the 1-year period were censored at the last non-missing measurement.
