Performance of Different Survival Models for Extrapolation of Immature OS Data for 2L+ NSCLC Therapies

OBJECTIVES: Non-small cell lung cancer (NSCLC) remains a significant public health burden, and early evaluation of therapeutic interventions, before overall survival (OS) data reaches maturity is needed for early patient access to novel, more effective treatment options. This study aims to assess the implications of OS data maturity and impact of survival model selection on mean OS, a key component of cost-effectiveness analyses and healthcare access decisions. Our research included different classes of therapies in 2L+ NSCLC: immunotherapy, targeted therapy and chemotherapy.

METHODS: A targeted review of clinical trials of monotherapies in 2L+ NSCLC identified 12 studies (with 20 populations) reporting OS with short and long-term follow-up. Kaplan-Meier curves were digitized to generate two datasets for each population (2x20) and 9 survival models were used to estimate mean OS, capped by UK general population mortality. Goodness-of-fit was based on the Akaike information criterion (AIC).

RESULTS: In 19/20 comparisons, mean OS increased with longer follow-up data (overall 23%). Mean OS increased for all survival models when considering the average across all populations. Log-normal (2%) and Log-logistic (5%) were less prone to underestimate OS extrapolated from shorter follow-up datasets. In turn, Gompertz showed the highest increase in mean OS (78%). In the longer follow-up datasets, Log-logistic had the best statistical fit in 8/20 studies and Log-normal in 6/20, of which 6 remained as best fits in the shorter follow-up datasets (Log-logistic: 2, Log-normal: 4).

CONCLUSIONS: This research has implications for economic evaluations and healthcare decision-making in the treatment of 2L+ NSCLC. Our results suggest that OS extrapolated from shorter follow-up tend to be underestimated, although some survival curves may generate more stable results. The differences observed with longer follow-up emphasize the need to complement survival models with external data from other trials, registries or expert elicitation when extrapolating immature data for cost-effectiveness analyses.