OBJECTIVES: Partitioned survival models (PSMs) and multistate models (MSMs) are discussed in NICE TSD 19. One of the disadvantages of MSMs is that post progression survival extrapolations are driven by early progressors, especially when the data is immature. This may lead to survival predictions which underestimate the actual survival if early progressors have a higher probability to die. We compare a PSM and an MSM on different data-cuts in terms of maturity and validate and compare the predictions based on the mature survival data of a non-small-cell lung cancer trial. METHODS: Individual patient-level data of the comparator arm (gemcitabine and cisplatin) of the SQUIRE trial conducted in non-small-cell lung cancer was obtained from Project Data Sphere (an open-access data-sharing platform). Time to death was split into three transitions (time to progression, pre-progression survival, and post-progression survival) required for an MSM, and standard parametric distributions fitted based on the percentage of patients that had progressed at three data-cuts: 25%, 50%, 75%. The maximum difference in area under the curve (ΔAUC_max) and mean survival time (ΔMST) versus long-term data were compared. Subgroup analyses were performed to validate the results. RESULTS: When 50% of patients had progressed, the difference in mean survival and ΔAUC_max using an MSM (ΔMST 0.3 - 4.7 months; ΔAUC_max 0.007 - 0.067 months) was less than versus a PSM (ΔMST 0.3 – 8.0 months; ΔAUC_max 0.019 - 0.075 months). When 75% of patients had progressed, the PSM predicted long-term survival slightly better than in an MSM (MSM: ΔMST 2.0 – 5.0 months; ΔAUC_max 0.022 - 0.074 months; PSM: ΔMST 1.2 - 4.1 months; ΔAUC_max 0.017 - 0.060 months). CONCLUSIONS: MSMs may predict long-term outcomes better and have lower structural uncertainty than PSMs using immature data, yet they underestimate OS across all data-cuts.