Predictive factors of virological response to direct-acting antivirals in patients with HBV/HCV coinfection

Abstract Background Hepatitis B and C virus (HBV/HCV) coinfection complicates antiviral management, and predictors of direct-acting antiviral non-response remain insufficiently characterized in endemic settings. This study evaluated baseline clinical, virological, and fibrosis-related predictors of early and sustained non-response after sofosbuvir- and daclatasvir-based therapy. Methods In this retrospective cohort study, 207 adults with HBV/HCV coinfection treated at Holy Family Hospital, Rawalpindi, Pakistan, received 12 weeks of sofosbuvir and daclatasvir. SVR4 and SVR12 failure were defined as detectable HCV RNA levels at 4 and 12 weeks after treatment completion. Predictors were measured before treatment initiation. Univariable screening, exploratory multivariate logistic regression, and internally validated machine learning models were applied separately for SVR4 and SVR12 failure. Performance was assessed using five-fold cross-validation, AUROC, PR-AUC, and SHAP importance. Results SVR4 was achieved by 186 patients (89.9%), whereas 21 (10.1%) had SVR4 failure. SVR12 was achieved by 175 patients (84.5%), whereas 32 (15.5%) had SVR12 failure. In multivariable models, SVR4 failure was associated with a higher BMI, higher baseline HCV RNA levels, and male sex; IFN-regimen status was associated with a lower adjusted odds of failure. SVR12 failure was associated with older age, higher BMI, male sex, and higher baseline HCV RNA levels; IFN-regimen status again showed lower adjusted odds. XGBoost showed the highest SVR4 PR-AUC, whereas Random Forest showed the highest SVR12 PR-AUC. SHAP identified HCV RNA levels, age, BMI, and fibrosis indices as key contributors. Conclusions DAA therapy achieved high SVR rates in HBV/HCV coinfection. Baseline host and viral variables informed failure prediction; while fibrosis-related variables contributed at the descriptive and model interpretation level; however, the models remain exploratory and require external validation.