OBJECTIVES: Economic evaluations (EE) of cancer treatments extrapolate observed trial overall (OS) and progression-free survival (PFS) data to a longer-term time horizon by fitting distributions (eg, exponential, Weibull, Gompertz distributions) to time-to-event data. Visual inspection is commonly used to justify parametric selections, but is subjective. We propose an algorithm for selecting and quantitatively justifying parametric model selection for OS/PFS extrapolation when only aggregate patient data are available. We illustrate the algorithm with EE of FOLFIRINOX against gemcitabine in metastatic pancreatic cancer.

METHODS: The algorithm includes seven steps: digitize treatment graphs; extract parametric functions; plot and visually inspect parametric distributions; assess goodness-of-fit; assess proportional hazard model; calculate EE; and propose results. Goodness-of-fit criteria include residual sum of squares (RSS), coefficient of determination (R), and F-test.

RESULTS: For OS, goodness-of-fit statistics were: for exponential, RSS=1.076, R=0.983, F=1554.365; for Weibull, RSS=