Use of predicted vital status to improve survival analysis of multidrug-resistant tuberculosis cohorts.

BACKGROUND: Multidrug-resistant tuberculosis (MDR-TB) cohorts often lack long-term survival data, and are summarized instead by initial treatment outcomes. When using Cox proportional hazards models to analyze these cohorts, this leads to censoring subjects at the time of the initial treatment outcome, instead of them providing full survival data. This may violate the non-informative censoring assumption of the model and may produce biased effect estimates. To address this problem, we develop a tool to predict vital status at the end of a cohort period using the initial treatment outcome and assess its ability to reduce bias in treatment effect estimates.

METHODS: We derive and apply a logistic regression model to predict vital status at the end of the cohort period and modify the unobserved survival outcomes to better match the predicted survival experience of study subjects. We compare hazard ratio estimates for effect of an aggressive treatment regimen from Cox proportional hazards models using time to initial treatment outcome, predicted vital status, and true vital status at the end of the cohort period.

RESULTS: Models fit from initial treatment outcomes underestimate treatment effects by up to 22.1%, while using predicted vital status reduced this bias by 5.4%. Models utilizing the predicted vital status produce effect estimates consistently stronger and closer to the true treatment effect than estimates produced by models using the initial treatment outcome.

CONCLUSIONS: Although studies often use initial treatment outcomes to estimate treatment effects, this may violate the non-informative censoring assumption of the Cox proportional hazards model and result in biased treatment effect estimates. Using predicted vital status at the end of the cohort period may reduce this bias in the analyses of MDR-TB treatment cohorts, yielding more accurate, and likely larger, treatment effect estimates. Further, these larger effect sizes can have downstream impacts on future study design by increasing power and reducing sample size needs.