Scenarios for Survival from Randomized Trials of Immune Checkpoint Inhibitors for Advanced Non-Small Cell Lung Cancer: The Median Is Not the Message

Background Key percentiles from overall survival curves can be used to formulate worst-case, typical, and best-case scenarios for survival, and boundaries for these scenarios can be estimated for treatment with chemotherapy using simple multiples (0.25x, 0.5x, 2x, and 3x) of the median overall survival time (mOS). The suitability of this approach for treatment with immune checkpoint inhibitors (ICI) is unclear. Methods We sought randomized trials of ICIs as first-line therapy for advanced non-small cell lung cancer (NSCLC) and extracted survival times from Kaplan-Meier survival curves at the following percentiles (boundary): 90th (worst-case), 75th (lower-typical), 50th (mOS), 25th (upper-typical) and 10th (best-case). We assessed whether simple multiples of the mOS accurately estimated these percentiles in groups treated with ICI plus chemotherapy (ICI+chemo), ICI-alone, and chemotherapy-alone. Estimates were deemed accurate if they were within ± 0.33 times the actual value. Results We identified 18 trials including 14,515 participants in 40 treatment groups: 13 ICI+chemo, 9 ICI-alone, and 18 chemo-alone. The median (IQR) of the mOS in months for the ICI+chemo group was 18 months (IQR 15–21), ICI-alone was 17 months (IQR 16–21), and chemo-alone was 13 months (IQR 12–14). Simple multiples accurately estimated the 90th percentile in 7/13 (54%), the 75th in 13/13 (100%) of ICI-chemo groups, but only 1/9 (11%) and 3/9 (33%) ICI-alone groups. The 25th and 10th percentiles for most ICI-containing groups were non-evaluable due to short follow-up. Conclusion Simple multiples of the mOS reliably estimated early survival scenarios in ICI-chemo and chemo-alone groups, but tended to overestimate survival in ICI-alone groups. Longer follow-up is needed to accurately estimate survival beyond the median in ICI trials.