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

Le résumé fourni par la source

Abstract Objective For many medical conditions, rapid, reliable, and affordable diagnostic tests are not available, which leads clinicians to base treatment decisions on patient symptoms and history. Although prediction models can estimate disease risk, they typically do not account for the downstream health or cost consequences of acting on their predictions. We developed and evaluated methods that integrate risk prediction with decision modeling to inform personalized, cost-effective treatment recommendations. Materials and Methods We considered two integration methods to maximize the population net monetary benefit (NMB), which summarizes both health and cost outcomes of available actions. In the probability-based method, the predicted probability of disease from a risk prediction model is used as input to a decision model. In the classification-based method, the decision model relies on the risk prediction model’s binary disease classification. We applied these methods to select between two treatment regimens for patients with rifampicin-resistant tuberculosis in Moldova, while accounting for cost, toxicity, and efficacy associated with each regimen. Results Both integration methods yielded higher population NMB than standard of care and approaches based on fixed classification thresholds (e.g., 50% or the threshold that maximizes the Youden’s index). However, the classification-based approach was less sensitive to whether the model predictions were properly calibrated. Conclusion Integrating risk predictions with decision models offers a principled framework for making personalized, value-based treatment decisions. These methods explicitly account for health and cost consequences of treatment choices informed by risk prediction models, improving care quality and resource use in settings with diagnostic uncertainty.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.