Objectives: The aim of this study was to use Microsoft Excel spreadsheet software to fit parametric survival distributions. We also explain the differences between individual patient data (IPD) and survival data reconstructed in Excel and SAS. Methods: Three sets of patient data on overall survival were compared using different methods: ‘original’ IPD, ‘reconstructed SAS’, and ‘reconstructed Excel’. The best-fit distribution was selected using visual observation, supported by linear plots of predicted probabilities, goodness-of-fit coefficients, and the sum of squared error of prediction. Outcomes included the incremental cost-effectiveness ratio (ICER), incremental net benefit (INB), incremental cost, and life-years gained over short-term and lifetime horizons. These were compared for different data sets. Results: In this example, log-normal, log-logistic, and Weibull distributions showed best-fit with the visual tests and goodness-of-fit statistics. Weibull and exponential distributions showed significant differences compared with IPD data. Data on short-term (5 years) time horizons produced by different data re-creation methods showed closeness with data reconstructed from SAS. The ICER and INB results were dependent on the time horizon and selected parametric distribution from the model. Conclusions: Different approaches used in fitting parametric survival distributions yielded predicted probabilities that substantially differed from those using original IPD. Our study highlights the importance of following guidelines for economic evaluations with a systematic approach to parametric survival analysis techniques in order to select best fitting parametric survival distributions.
ASJC Scopus subject areas
- Health Policy
- Public Health, Environmental and Occupational Health