OBJECTIVES

: Hepatitis C virus (HCV), the leading cause of advanced liver disease, has enormous economic burden. Effective treatments exist, yet identifying patients at risk of treatment failure could lead to interventions that improve cure rates. Our goal was to develop and evaluate a prediction model of HCV treatment failure.

METHODS

: We analyzed HCV patients initiating direct-acting antiviral therapy at four United States institutions. Non-achievement of sustained virologic response (SVR) 12 weeks post-treatment completion defined treatment failure. The cohort was divided into a derivation (67%) and validation (33%) set. From 20 patient-level candidate variables collected before treatment initiation, we identified a subset associated with treatment failure in bivariate analyses. In the derivation set, separate predictive models were developed from 100 bootstrap samples using clustered logistic regression. From the 100 models, candidate variables were ranked by frequency of selection as predictors, using cutoffs of ≥80%, ≥50%, ≥40%, and all variables. In the validation set, predictive performance was compared across models using area under the receiver operating characteristic curve (AUC).

RESULTS

: From 1,253 HCV patients, overall SVR rate was 86.1% (95% CI=84.1%, 88.0%). From the cutoffs, the number of variables included in final models were: ≥80%=three (AUC=0.576); ≥50%=five (AUC=0.605); ≥40%=nine (AUC=0.684); all=11 (AUC=0.681). The best performing model (≥40%) had significantly better predictive ability than the ≥50% (p=0.03) and ≥80% models (p=0.02) in the validation set, but poor discriminative ability. Most likely predictors of treatment failure in our dataset were older age, presence of hepatocellular carcinoma, and private versus public insurance, which appeared in 89%, 84%, and 80% of bootstrap models, respectively.

CONCLUSIONS

: This study did not result in a highly predictive model, but highlighted baseline factors associated with HCV treatment failure. Treatment failure prediction may facilitate development of data-driven clinical tools to identify patients who would benefit from interventions to improve SVR rates.