Using risk prediction models to inform personalized, cost-effective treatment recommendations

Area of Science:

• Health economics
• Clinical decision-making
• Biostatistics

Background:

• Diagnostic uncertainty necessitates reliance on clinical judgment and prediction models.
• Existing prediction models often overlook downstream health and cost implications.
• Personalized treatment requires integrating risk assessment with decision analysis.

Purpose of the Study:

• To develop and evaluate methods for integrating risk prediction with decision modeling.
• To inform personalized and cost-effective treatment recommendations.
• To maximize population net monetary benefit (NMB) by considering health and cost outcomes.

Main Methods:

• Two integration methods were developed: probability-based and classification-based.
• These methods were applied to optimize treatment selection for rifampicin-resistant tuberculosis.
• The analysis accounted for regimen costs, toxicity, and efficacy.

Main Results:

• Both integration methods improved population NMB compared to standard care and fixed thresholds.
• The classification-based approach demonstrated robustness to prediction model calibration.
• The study highlights the value of integrated models in resource-constrained settings.

Conclusions:

• Integrating risk prediction with decision models provides a framework for value-based treatment decisions.
• These methods enhance care quality by accounting for health and cost consequences.
• The approach is particularly beneficial in situations with diagnostic uncertainty.