OBJECTIVES: The use of subsequent therapies has the potential to confound assessment of overall survival (OS) in oncology trials, in particular for trials in early lines of therapy and for malignancies with several registered or investigational treatment options. Standard intent-to-treat analysis is biased, since treatment choices are likely to be influenced by events associated with mortality risk, such as disease progression. We review and compare available statistical methods to obtain unbiased estimates of OS effects in presence of subsequent therapies. METHODS: Marginal structural modeling methods include inverse-probability of censoring weighting (IPCW) and inverse-probability of treatment weighting (IPTW). These methods explicitly model both treatment choices and effects of treatments on mortality. Rank-preserving structural failure time models (RPSFT) instead depend on parametric assumptions regarding the effect of investigational and subsequent therapies on survival, and require non-standard estimation methods such as G-estimation or iterative parameter estimation (IPE). We compare the results with the different methods with data from the Lux Lung 1 trial of the tyrosine kinase inhibitor afatinib in non-small cell lung cancer. RESULTS: IPCW and IPTW require detailed information on covariates that influence treatment choices and are sensitive to model misspecification. RPSFT may not yield a single estimate of treatment effects due to limitations of the G-estimation procedure. All methods were consistent with a potential OS benefit from afatinib, but the hazard ratio varied from 0.583 (p=0.038) with the pre-specified IPCW method to 0.894 (0.281) with RPSFT/IPE.            CONCLUSIONS: The proposed methods for obtaining unbiased OS estimates in presence of subsequent therapies rest on assumptions that cannot be tested empirically. There is currently no accepted standard method; pre-specification of model choice is of importance as well as testing alternative methods. Care should be taken to avoid unbalance in subsequent therapy and to record specific information on administered treatments with potential OS effects.