BACKGROUND: Decisions about government funding for new medicines often rely on statistical models to predict how long patients will live (overall survival, OS) based on early trial data. This study compared the accuracy of these predictions for the cancer drug pembrolizumab, using models preferred by pharmaceutical companies (Sponsors) and the Pharmaceutical Benefits Advisory Committee (PBAC), compared to real-world long-term follow-up (LTFU) data. RESEARCH DESIGN AND METHODS: We reviewed publicly available PBAC summary documents (PSDs) for all funding decisions on pembrolizumab up to November 2022. We included cases with at least three years of follow-up data and where at least 350 patients per year would be treated. We then compared survival predictions from PBAC and Sponsor models to actual survival data at two time points. RESULTS: A total over 38 PSDs covering 15 indications, five met our criteria. Sponsor-preferred models underestimated real survival by 0.54% to 16.45%, while PBAC-preferred models underestimated survival by 1.20% to 24.21%. CONCLUSION: Results demonstrate that OS extrapolation methods used by both the Sponsor and PBAC tend to underestimate long-term survival outcomes for pembrolizumab indications, with PBAC-preferred methods being more conservative.