Presentation (CTI)

OBJECTIVES: Statistical adjustment of overall survival (OS) times in the presence of treatment switching is a common health technology assessment requirement. Treatment switching in randomised clinical trials often occurs when patients randomised to the control arm switch to the active treatment during follow up, e.g. following disease progression. The two-stage estimation (TSE) method predicts counterfactual survival times for patients who switch treatments (i.e. to estimate survival had no switching occurred). However, this can induce informative censoring and hence bias treatment effect estimates. Re-censoring can be applied to account for this but often results in significant losses of long-term information. An alternative method applying inverse probability of censoring weights (IPCW) after TSE was proposed by Latimer, et al¹. We implemented this method in a recent submission to NICE.
METHODS: In the first stage of TSE-IPCW, an acceleration factor (AF) is estimated by comparing post-progression survival times for patients who switched with those who did not. In the second stage the shrinkage factor (1/AF) is used to adjust the survival times to provide counterfactual survival times. A survival model predicting censoring is then fitted in order to estimate patient weights. Lastly, these IPCW weights are used in a weighted parametric survival analysis of OS.
RESULTS: The TSE-IPCW counterfactual OS times resided between the ‘with’ and ‘without’ re-censoring counterfactual times, whilst also preserving long term-data. This enabled more reliable estimates of the long-term survival benefit of the treatment.
CONCLUSIONS: The TSE-IPCW method is a useful (and underutilised) alternative when estimating counterfactual survival times in the presence of treatment switching.
¹ Latimer NR, Abrams KR and Siebert U. Two-stage estimation to adjust for treatment switching in randomised trials: a simulation study investigating the use of inverse probability weighting instead of re-censoring. BMC Medical Research Methodology (2019) 19:69.