Cost-effectiveness models fully informed by real-world epidemiological parameters yield the best results, but they are costly to obtain. Model calibration using real-world data/evidence (RWD/E) on routine health indicators can provide an alternative to improve the validity and acceptability of the results. We calibrated the transition probabilities of the reference chemotherapy treatment using RWE on patient overall survival (OS) to model the survival benefit of adjuvant trastuzumab in Indonesia.

A Markov model comprising four health states was initially parameterized using the reference-treatment transition probabilities, obtained from published international evidence. We then calibrated these probabilities, targeting a 2-year OS of 86.11 percent from the RWE sourced from hospital registries. We compared projected OS duration and life-years gained (LYG) before and after calibration for the Nelder–Mead, Bound Optimization BY Quadratic Approximation, and generalized reduced gradient (GRG) nonlinear optimization methods.

The pre-calibrated transition probabilities overestimated the 2-year OS (92.25 percent). GRG nonlinear performed best and had the smallest difference with the RWD/E OS. After calibration, the projected OS duration was significantly lower than their pre-calibrated estimates across all optimization methods for both standard chemotherapy (~7.50 vs. 11.00 years) and adjuvant trastuzumab (~9.50 vs. 12.94 years). LYG measures were, however, similar (~2 years) for the pre-calibrated and calibrated models.

RWD/E calibration resulted in realistically lower survival estimates. Despite the little difference in LYG, calibration is useful to adapt external evidence commonly used to derive transition probabilities to the policy context, thereby enhancing the validity and acceptability of the modeling results.