Objectives
Survival in oncology is commonly extrapolated using traditional parametric distributions or cubic splines. Bayesian model averaging (BMA) can provide an easily interpretable comparative assessment of goodness-of-fit (GOF) of individual extrapolations and combine these into a single model. The objective of this study was to explore patterns in GOF of traditional parametric extrapolations and cubic splines with BMA, applied to data for nivolumab (NIVO) and ipilimumab (IPI) across four indications.
Methods
Thirteen independent parametric extrapolations, 7 standard distributions and 6 splines, were fitted to recurrence-free survival (RFS) for NIVO and IPI in adjuvant melanoma (NCT02388906), overall survival (OS) for NIVO in squamous cell carcinoma of the head and neck (NCT02105636), NIVO, IPI, and NIVO+IPI in metastatic melanoma (NCT01844505), and NIVO in squamous non-small cell lung cancer (NCT01642004), without making assumptions regarding proportional hazards. BMA with uninformative priors was used to obtain the posterior probability associated with each individual extrapolation being true.
Results
Comparison across indications and treatment arms demonstrated that the exponential, gamma, and Weibull distributions with a survival trend towards zero, were poor fits to the analyzed survival data with estimated posterior probabilities <1%. Generalized gamma, Gompertz, log-logistic and log-normal distributions all provided better fits and mostly induced long-term survival tails. The log-normal distribution was amongst the best-fitting distributions for OS data across all indications and arms, with posterior probabilities ranging from 18%-70%. Splines provided a good fit to RFS with >98% of the posterior probability attributed to a spline. For OS, 1-knot splines provided better fits than 2-knot splines, though cumulative posterior probabilities of splines were <31%.
Conclusions
Exponential, gamma and Weibull distributions provided a poor GOF to nivolumab and ipilimumab survival across indications. For OS, a log-normal distribution provided a consistently good fit to observed data and may thus be a preferable choice for survival extrapolations.
Article info
Identification
Copyright
© 2019 Published by Elsevier Inc.
User license
Elsevier user license | How you can reuse 
Elsevier's open access license policy
Elsevier user license
Permitted
For non-commercial purposes:
- Read, print & download
- Text & data mine
- Translate the article
Not Permitted
- Reuse portions or extracts from the article in other works
- Redistribute or republish the final article
- Sell or re-use for commercial purposes
Elsevier's open access license policy
