OBJECTIVES: This analysis aimed to investigate potential differences in estimated survival outcomes between partitioned survival models and state transition (Markov) models in the common three-state model of pre-progression, post-progression, and death by using nivolumab trial data in advanced melanoma. METHODS: Each approach was applied separately to patient-level data from a phase 1b trial of nivolumab in 304 patients with previously treated advanced solid tumors (107 melanoma patients). All patients were included in parametric survival analyses (with a parameter identifying melanoma patients) to model overall survival (OS), progression free survival (PFS), and post-progression survival (PPS) to extrapolate (10 years) beyond maximum trial follow-up for melanoma patients specifically. Alternative fits to the data were compared on the basis of Akaike Information Criterion (AIC) and the plausibility of the long-term extrapolation. RESULTS: The Weibull parametric model was judged to be the most conservative in terms of extrapolation and forms the basis of these results. Log-normal and log-logistic generally had better fit in terms of AIC, but fatter tails were less suitable for extrapolation. The area under the curve (AUC) for extrapolated PFS and OS was 23.5 and 41.5 months, respectively, suggesting 18 months of PPS using the partitioned survival approach. The AUC for PPS was 12.5 months, which gives an estimated OS of 36 months under the Markovian assumption. Relaxing this assumption to allow time dependency for transitioning between states with a Weibull model for PPS gives an estimated PPS of 17.1 months and an OS of 40.6 months. CONCLUSIONS: Partitioned survival modeling and Markov modeling make intrinsically different assumptions; nevertheless, by relaxing the Markovian assumption and allowing time dependency, the two modeling forms are shown to be functionally equivalent. The selection of approach should be driven by what best represents the disease, treatment effect, and available clinical data.