OBJECTIVES: In some RCTs crossover is allowed. This might bias estimates of relative treatment effect. Several crossover correction methodologies may be applied, such as the rank-preserving structural failure time model (RPSFTM), inverse probability of censoring weights (IPCW) and the two-stage method. These methods aim to adjust survival for patients who crossover to reflect clinical practice. The aim was to validate the outcomes of crossover methodologies using RWE.

METHODS: In KEYNOTE-024, pembrolizumab was compared to chemotherapy in stage IV NSCLC with PD-L1 tumor proportion score of ≥50%. The chemotherapy arm (n=151) was corrected for crossover using the two-stage method. The RWE study included n=108 stage IV NSCLC patients treated with chemotherapy. Patient characteristics of KEYNOTE-024 and RWE were compared on age, histology, sex and smoking status. Survival after crossover adjustment was compared with overall survival of stage IV NSCLC patients treated with chemotherapy in clinical practice.

RESULTS: The baseline characteristics between RWE and RCT studies were similar. PD-L1 was ≥50% in KEYNOTE-024, but not reported for RWE. Survival of RWE and RCT crossover-adjusted chemotherapy arms was almost identical up to month 10 of follow-up. After 10 months, survival for the crossover-adjusted chemotherapy arm flattened, whereas survival of chemotherapy in clinical practice did not flatten. Crossover adjusted survival was observed to provide a better prediction of RWE survival, compared to the unadjusted survival.

CONCLUSIONS: Crossover adjusted data overestimated survival compared to clinical practice. Further research is needed to determine the cause, which might be explained by differences between RWE and RCT studies. Potential differences include quality of care, subsequent treatment after crossover, inclusion and exclusion criteria (e.g., PD-L1 status), limited patient numbers in the tail, or methodological issues regarding the two-stage method. Validation against RWE remains important for prediction of clinically plausible survival.