An inverse probability weighting method for estimating the net benefit in survival analyses in observational studies

Abstract

Background: Recently, in the context of randomized trials, a measure on the difference scale called the net benefit was developed for survival analysis. As this measure does not require the assumption of proportional hazards, it is an attractive new measure of the treatment effect to apply instead of the hazard ratio calculated under this assumption. However, no method for estimating it has been presented in observational studies. Therefore, we describe a simple method to estimate the net benefit adjusted for confounding. 

Methods: We reviewed a method for estimating the net benefit in a randomized trial and extended it to a method that adjusts for confounding using inverse probability of treatment weights. 

Results: We performed Monte Carlo simulations to test the performance of our method. The results show that our method estimated adjusted net benefits in an unbiased manner regardless of whether the assumption of proportional hazards held. In addition, we illustrated our method using data from an observational study evaluating disease-free survival of Ewing’s sarcoma patients. Our method yielded an adjusted net benefit of –0.032, whereas an existing method, used to analyze data from randomized trials, yielded an unadjusted net benefit of 0.284. 

Conclusions: In observational studies with a time-to-event outcome, the net benefit adjusted for confounding can readily be estimated using inverse probability of treatment weights.