All Times EDT
Keywords: causal inference; time-to-event endpoint; non-proportional hazards; real-world data; propensity score; MaxCombo method.
In clinical or observational studies that utilize real-world data, time-to-event outcomes, e.g. overall survival, are ubiquitous and often germane to the scientific questions of interest, especially in areas of unmet medical needs, such as oncology. Two main obstacles when analyzing time-to-event outcomes in such studies are the presence of non-proportional hazards patterns and the confounding bias induced by the lack of randomization or other reasons. Existing literature elaborated methods that could adjust for either NPH patterns or confounding bias, but no previous work delineated the complexity of simultaneous adjustments for both of these issues. we proposed a novel method can simultaneously adjust for observed confounding bias and NPH patterns, and can be pre-specified to test the hypothesis of treatment difference under NPH when in many RWD studies when the NPH pattern is not known in advance. Our proposed method is easy to implement and has robust performance demonstrated in the simulation studies. we also applied our proposed method in a case study of non-small cell lung cancer.