A Bayesian Model Leveraging Multiple External Data Sources to Improve the Reliability of Lifetime Survival...

Area of Science:

• Oncology
• Biostatistics
• Health Economics

Background:

• Reliable long-term survival extrapolation is crucial for novel cancer therapies.
• Early trial data often lacks sufficient follow-up for definitive survival predictions.
• Registry and historical trial data can supplement early-phase study information.

Purpose of the Study:

• To extend Bayesian multiparameter evidence synthesis (B-MPES) to incorporate historical trial data.
• To assess the impact of external data sources on survival predictions from early data cuts.
• To compare B-MPES predictions with standard parametric models (SPMs) for metastatic non-small-cell lung cancer (mNSCLC).

Main Methods:

• Fitted B-MPES models to survival data from the CheckMate 9LA study (nivolumab plus ipilimumab plus chemotherapy vs. chemotherapy).
• Supplemented trial data with registry data (SEER) and historical trial data (adjusted for confounding).
• Compared B-MPES predictions with SPMs using early and 4-year data cuts.

Main Results:

• B-MPES models better captured the survival plateau observed in later data cuts compared to SPMs.
• SPMs produced overly conservative short-term survival extrapolations.
• A B-MPES model using historical NIVO+IPI data slightly overestimated survival due to unaddressed hazard confounding.

Conclusions:

• Incorporating historical trial data via B-MPES improves the plausibility of lifetime survival extrapolations for novel therapies with immature data.
• B-MPES offers a transparent method for survival extrapolation by leveraging multiple external data sources.
• Careful specification of model flexibility and prior data confidence is essential to avoid overfitting in B-MPES.