Duration and Timing of Treatment Waning: Determining the Start and Stop Points Using Pembrolizumab in NSCLC As a Case Study

OBJECTIVES:

Treatment waning is an important uncertainty in health technology assessments (HTAs) when extrapolating immature data cuts of clinical trials.¹ This research explored methods for determining the appropriate start and stop points of treatment waning, using mature overall survival (OS) data from KEYNOTE-024 as a case study.

METHODS:

Using OS Kaplan–Meier data for pembrolizumab and chemotherapy from the 66-month data cut-off (DCO) of KEYNOTE-024, smoothed hazard plots were generated (using the Muhaz function in R). The plots were examined to determine the waning start (when the difference between the hazard of death for pembrolizumab versus chemotherapy starts to decrease) and stop (when the hazards of death equalize) points. These were compared with the waning assumptions accepted in the corresponding National Institute for Health and Care Excellence (NICE) appraisal based on the 25-month DCO of KEYNOTE-024 (NICE TA531), whereby the observed magnitude of treatment effect was assumed to be maintained until either Month 36 or Month 60.

RESULTS:

Smoothed hazard plots showed that the incremental hazard of death for pembrolizumab versus chemotherapy decreased (i.e., survival benefit reduced) from Month 0. Before Month 10, this decrease was minimal; from Months 10–30, the decrease was more substantial. The absolute hazard of death remained lower for pembrolizumab until approximately Month 47, where the hazards equalized.

CONCLUSIONS:

Based on this case study, treatment waning should most accurately be applied immediately at the end of the observed data period until Month 47 when extrapolating the earlier DCO of KEYNOTE-024; the full observed magnitude of treatment effect should not be maintained into the extrapolation period. Therefore, the TA531 assumption of full observed treatment effect until Month 60 was optimistic. Further research is required to elucidate any trends that can be used to inform improved assumptions when extrapolating immature data.

1. Harrington H et al. (2022). MSR34, 2022-11, ISPOR Europe, Vienna, Austria