Presentation (CTI)

OBJECTIVES: A central goal of cost-effectiveness analysis (CEA) is to understand which uncertain parameters drive economic value under realistic joint uncertainty. While probabilistic sensitivity analysis (PSA) is routinely used to propagate uncertainty through CEA models, existing approaches largely focus on summarizing uncertainty in outcomes or supporting value-of-information metrics. There remains no systematic framework for decomposing how parameters shape variation in value across the plausible parameter space.
METHODS: We apply this approach to a CEA comparing long-acting injectable HIV pre-exposure prophylaxis (PrEP) with daily oral PrEP. Model inputs included transition probabilities, health-state utilities, and costs, and the primary outcome was incremental net monetary benefit (INMB) at a willingness-to-pay of $100,000/QALY. We generated 1,000 Monte Carlo draws of jointly uncertain inputs and computed INMB for each draw. An XGBoost emulator mapped inputs to INMB, and SHAP values quantified, for each draw, how much each parameter shifted predicted INMB above or below the mean, conditional on all other input values.
RESULTS: The emulator closely reproduced the simulated INMB surface. Mean predicted INMB was $27,691. SHAP identified three parameters as the dominant drivers of value. For HIV acquisition risk under daily oral PrEP, low realizations decreased INMB by up to $59,588 while high realizations increased it by up to $74,672 (mean $27,148). For HIV acquisition risk under long-acting PrEP, high realizations decreased INMB by up to $48,393 while low realizations increased it by up to $18,081 (mean $13,690). For long-acting PrEP cost, higher costs decreased INMB by up to $33,855 while lower costs increased it by up to $20,255 (mean $9,715).
CONCLUSIONS: SHAP values provide a transparent and scalable approach to understanding and quantify which parameters drive value in CEA. By explicitly decomposing value into parameter-level contributions, this framework clarifies which aspects of an intervention matter most for maximizing economic value for decision-makers.