Presentation (CTI)

OBJECTIVES: To improve survival modeling in cost-effectiveness evaluation, we explain the theoretical shortcomings of traditional parametric models, whose simple perspective often fails to capture complex population-level hazard dynamics. We then evaluate the practical challenges of newer, complex methods. Building on this, we reintroduce and validate a simpler, historically recognized approach that incorporates observable patient heterogeneity, leading to recommendations for updating current modeling guidance.
METHODS: Our methodology revisits statistical literature from the 1980s and 1990s that addressed patient heterogeneity. Applying principles from this period, we distinguished between individual-level survival and resulting population-level hazard dynamics. We then operationalized a straightforward technique from that literature to derive aggregate population-level estimates from the distribution of individual characteristics. The technique's performance was tested with a simulation model and validated against a real-world dataset.
RESULTS: Our results confirmed that traditional parametric models are often structurally unsuited for capturing complex population-level hazard dynamics. While more advanced methods like splines offer flexibility, they introduce significant new methodological challenges. In contrast, the heterogeneity-based approach directly addresses this issue by aggregating varied individual risk profiles to explain population-level outcomes. The utility and validity of this simpler technique in reproducing complex survival patterns were verified in both the simulation and the real-world data analysis.
CONCLUSIONS: Incorporating observable heterogeneity offers a robust and parsimonious solution for survival analysis in many cost-effectiveness models, bridging the gap between overly simplistic and unnecessarily complex methods. We recommend updating current modeling guidance to include this approach. Doing so can enhance the accuracy of long-term extrapolations and improve analytical accessibility, negating the need for more difficult methods in many cases.