Theme: Epidemiology & Public Health Research Year: 2024
Background: Precision-medicine approaches to personalise HCV treatment durations could improve health system efficiency and cost-effectiveness. This analysis aimed to (1) develop a model to predict short duration treatment outcomes; (2) evaluate model clinical utility and treatment cost-savings.
Methods: Short duration HCV treatment data (<56-days DAA) from clinical trial and cohort studies were included. A logistic machine learning model was developed using baseline clinical factors. Nested cross-validation was used to assess performance and optimize hyperparameters. Predicted failure probabilities were used to assign treatment duration. Average cost savings for treatment were assessed using Decision Curve Analysis. Correctly assigned standard durations (true failures) cost $10000; incorrectly assigned standard durations (false failures) were penalised excess treatment costs ($2500-$5000). Correctly assigned short durations (true cures) cost $5000-$7500; incorrectly assigned short durations (false cures) were penalised retreatment costs ($10000).
Results: Of 254 receiving short duration HCV treatment (52% sofosbuvir/velpatasvir; 26% glecaprevir/pibrentasvir), 23% had treatment failure. Clinical predictors of short duration failure included higher baseline HCV RNA, ALT-AST ratio and liver fibrosis. For a standard threshold probability of failure (0.5) was used to assign duration, sensitivity and specificity were 78% and 89%, respectively. When low threshold probability for failure (0.2) was used, sensitivity increased (99%) at the cost of specificity (40%). When a high threshold probability for failure (0.8) was used, specificity increased (96%) at the cost of sensitivity (29%). Average treatment cost-savings were greater for low ($3270) vs. high ($2322) failure thresholds (Figure 1).
Conclusions: High sensitivity (low threshold) models provided greatest average treatment cost-savings. In dynamic settings such as prisons, utilising precision-medicine models to assign HCV treatment duration could be clinically effective and cost-effective. However, more short duration treatment data are needed to validate model performance and ensure generalisability. Further, comprehensive assessment of cost-benefits should be conducted.
Download abstract