Individualized Treatment Rules Based on Cost-Effectiveness Criteria in Microsimulations

Area of Science:

• Health economics
• Statistical modeling
• Decision science

Background:

• Cost-effectiveness analysis (CEA) typically uses population-level data.
• Individualized treatment rules (ITRs) can enhance CEA by accounting for patient heterogeneity.
• Statistical learning offers novel methods for designing ITRs from simulated data.

Purpose of the Study:

• To propose and evaluate a statistical learning approach for designing individualized treatment rules (ITRs).
• To optimize cost-effectiveness by tailoring treatments to individual patient characteristics using microsimulation models.
• To compare the performance of different statistical learning techniques for ITR derivation.

Main Methods:

• Defined optimal ITRs and their value in cost-effectiveness.
• Applied LASSO regression, classification trees, and policy trees to simulated patient data.
• Compared methods based on incremental net monetary benefit (NMB), computational speed, and interpretability.
• Assessed the impact of parameter and stochastic uncertainty on ITRs.
• Illustrated methods using a microsimulation model for haemophilia B.

Main Results:

• A two-layer classification tree was found to be the most suitable method based on defined criteria.
• The optimal classification tree allocated treatments based on thresholds for annualized bleeding rate and age.
• Threshold values exhibited uncertainty, with 95% credible ranges identified.
• Stochastic uncertainty was shown to impact the incremental value of ITRs.

Conclusions:

• Classification trees are expected to be superior for ITR derivation in similar microsimulation models.
• Accurate patient pathway representation is crucial due to the significant impact of stochastic uncertainty on ITRs.
• Future research should explore further empirical models and real-world application of ITRs.