OBJECTIVES: Overall Survival data (OS) from Randomized Clinical Trials (RCTs) can be immature at time of submission. Real-World Data (RWD) can be used to guide survival extrapolations. Advantage of RWD is that it reflects clinical practice. We aimed to assess different methodologies where RWD is used to inform survival extrapolations in NSCLC.

METHODS: An immature trial in previously untreated advanced NSCLC with PD-L1 expression comparing pembrolizumab and chemotherapy was used and informed by mature RWD on chemotherapy in the same indication. Four Bayesian methodologies were compared: (1) the shape of RWD is used as prior for the shape parameter of the pivotal trial (informed shape), (2) RWD is added as a treatment arm in the parametric model fit (informed fit), (3) expert elicitation based on RWD was used as prior (informed survival; the expected survival was estimated to be between 10% and 20% after 30 months for placebo), and (4) a combination of the clinician informed survival and informed shape by RWD. The tested methods were compared with standard parametric methods (non-informed) on mean survival predicted based on the Weibull.

RESULTS: In the reference case (non-informative priors), pembrolizumab and chemotherapy predicted mean survival of 60 and 33 months respectively. The survival using informed shape was 43 and 25 months, using informed fit 47 and 27 months, using informed survival 50 and 20 months, and using both informed shape and survival 41 and 19 months for pembrolizumab and chemotherapy respectively.

CONCLUSIONS: The tested methodologies work, but result in different mean and incremental survival. For HTA it is important to validate the clinical plausibility of the extrapolations.