Imputation methods for missing failure times in recurrent-event survival analysis: Application to suicide attempts in the transgender population.

Abstract

Suicide risk among transgender populations is an important public health issue. In a project evaluating association between gender affirmation and suicide attempts in the US Transgender Survey, we evaluated the relationship between gender affirmation and risk for suicide attempts. One of the challenges is that the age at suicide attempts was only collected for the first and last attempt. The initial zero-inflated negative binomial model enabled us to evaluate the association between gender affirmation and number of suicide attempts per 5 years adjusting for other covariates. However, ignoring missing failure times of recurrent events may have caused bias and loss of efficiency. In this paper, we use a recurrent-event survival analysis incorporating time-varying covariates with three approaches to impute the age at suicide attempt, estimates from three imputation approaches are similar. We were able to confirm the findings from the initial model and identify additional associations that were not detected in the initial analysis. Findings suggest the need to consider additional analytical approaches in settings with high data missingness by design. Research to validate and compare measures that ask first and last attempt to those which enumerate all attempts in this population will be important for future surveys.