Can AI Predict Potential Treatment Failure in HCV Patients?


The XGBoost machine learning model outperformed other AI and logistic regression models for predicting DAA failure, with an AUROC of 1.000 in the training dataset and 0.803 in the validation dataset.

Ming Ling-Lu, MD, PhD

Credit: Coalition for Global Hepatitis Elimination

Ming Ling-Lu, MD, PhD

Credit: Coalition for Global Hepatitis Elimination

This article originally appeared on our sister site, HCPLive.

Machine learning algorithms may be a viable tool for facilitating risk stratification in direct-acting antiviral (DAA) failure and providing clinicians with additional information about factors associated with treatment failure.

In a real-world, multicenter study conducted using the Taiwan HCV Registry (TACR) database, AI models outperformed conventional logistic regression models for predicting DAA failure, successfully detecting 69.7% of the patients who did not achieve sustained virological response (SVR) and further identifying several variables linked to treatment failure.1

“As comprehensive HCV elimination programs are advocated worldwide, an increasing number of patients with HCV infection are expected to require DAA salvage therapy. Thus, all the risk factors associated with DAA failure must be considered simultaneously to reduce the retreatment burden,” wrote investigators.1

Although the World Health Organization estimates DAAs can cure more than 95% of the 58 million cases of hepatitis C virus (HCV) worldwide, access to diagnosis and treatment remains low and continues to inhibit disease management and eradication efforts. Additionally, clinicians’ understanding of the clinical factors inhibiting certain individuals from achieving SVR remains limited. As AI and machine learning continue to become more prevalent in the medical field, their use for predicting DAA outcomes may be promising and thus merits further research.2

Ming-Lung Yu, MD, PhD, dean of the college of medicine and senior vice president of National Sun Yat-sen University in Taiwan, and a team of investigators sought to explore the risk factors associated with DAA failure by applying artificial intelligence to identify patients with HCV prone to virological failure. To do so, they examined the TACR database, a nationwide HCV-registered platform implemented by the Taiwan Association for the Study of the Liver in 2020.1

A total of 34,301 chronic HCV patients ≥ 18 years of age who received DAAs with available SVR data were enrolled in the real-world, multicenter, prospective cohort study. The primary outcome was SVR, defined as undetectable serum HCV RNA 12 weeks after the end of treatment.1

Participants were randomly assigned to a training dataset (70%; n = 23,955) and a validation dataset (30%; n = 10,346), which investigators pointed out had similar baseline demographics for age, sex, body mass index, biochemical data, cirrhosis, HCV genotypes, viral load, DAA regimens, HBV coinfection, and hepatocellular carcinoma. A total of 55 host, virological, and on-treatment features were input into the decision tree, random forest, eXtreme Gradient Boosting (XGBoost), and artificial neural network packages of R software.1

The performances of the machine-learning models were assessed using the area under the receiver operating characteristic curve (AUROC), accuracy of the confusion matrix, precision-recall curve, and F1-score.1

The overall DAA failure rate was 1.6% (n=538). Multivariate logistic regression analysis revealed liver cirrhosis, hepatocellular carcinoma, poor DAA adherence, and elevated hemoglobin A1c were significantly associated with virological failure. Investigators applied these 4 independent risk factors as components of the logistic regression model to be compared to the AI algorithms.1

Results showed XGBoost outperformed the other algorithms and logistic regression models, with an AUROC of 1.000 in the training dataset and 0.803 in the validation dataset. Specifically, the performance of XGBoost was superior to the random forest (P = .021), decision tree (P = 4.4×10-8), artificial neural network (P = 5.4×10-9), and logistic regression model (P = 2.5×10-10).1

What You Need to Know

The study suggests that machine learning algorithms can be effective tools for predicting the failure of direct-acting antiviral (DAA) treatment in patients with hepatitis C virus (HCV).

The AI models successfully identified several risk factors associated with DAA failure, including liver cirrhosis, hepatocellular carcinoma, poor DAA adherence, and elevated hemoglobin A1c.

XGBoost demonstrated superior performance compared to other machine learning algorithms and logistic regression models.

The accuracy, sensitivity, specificity, positive predictive value, and negative predictive value of the XGBoost model at a cutoff value of 0.5 were 99.5%, 69.7%, 99.9%, 97.4%, and 99.5%, respectively, for the entire dataset.1

Using Shapley Additive exPlanations (SHAP) to measure the contributions to the outcome from each feature of the XGBoost model, investigators found elevated viral load, α-fetoprotein, FIB-4 index, bilirubin, and AST levels increased the risk of DAA failure. Individuals with a lower body mass index, platelets, albumin, and younger age also had a lower probability of achieving SVR.1

Although the study’s results paint a promising future for the use of AI in predicting DAA treatment outcomes, investigators called attention to several limitations. Specifically, they pointed out the validation dataset’s inferior performance compared to the training dataset and limited model generalizability due to potential overfitting and heterogeneity in this dataset.1

“ML algorithms facilitate risk stratification in DAA failure and provide additional information on factors associated with DAA failure,” investigators concluded.1


  1. Lu MY, Huang CF, Hung CH, et al. Artificial intelligence predicts direct-acting antivirals failure among hepatitis C virus patients: A nationwide hepatitis C virus registry program. Clin Mol Hepatol.
  2. World Health Organization. Hepatitis C. Newsroom. July 18, 2023. Accessed January 19, 2024.
Related Videos
© 2024 MJH Life Sciences

All rights reserved.