Evaluation of machine learning algorithms for predicting direct-acting antiviral treatment failure among patients

Haesuk Park1, Wei-Hsuan Lo-Ciganic2, James Huang2

  • 1Department of Pharmaceutical Outcomes and Policy, College of Pharmacy, University of Florida, HPNP Building Room 3325, 1225 Center Drive, Gainesville, FL, 32610, USA. hpark@cop.ufl.edu.

Scientific Reports
|October 27, 2022
PubMed

Insights

Machine learning models effectively predict direct-acting antiviral treatment failure in hepatitis C virus (HCV) patients. Identifying key risk factors can guide interventions to improve treatment outcomes and reduce retreatment burdens.

Area of Science:

  • Hepatology and Viral Hepatitis
  • Medical Informatics and Machine Learning
  • Public Health and Epidemiology

Background:

  • Hepatitis C virus (HCV) infection remains a significant global health challenge, causing substantial liver-related morbidity and mortality despite effective direct-acting antiviral (DAA) therapies.
  • Predicting DAA treatment failure is crucial for optimizing patient management and public health strategies aimed at HCV eradication.

Purpose of the Study:

  • To develop and validate machine learning (ML) algorithms for predicting treatment failure in adults undergoing all-oral DAA therapy for HCV.
  • To compare the performance of various ML models against traditional statistical methods and identify key predictors of DAA treatment failure.

Main Methods:

  • Development and validation of elastic net (EN), random forest (RF), gradient boosting machine (GBM), and feedforward neural network (FNN) ML algorithms using data from the HCV-TARGET registry.
  • Comparison of ML model performance with multivariable logistic regression (MLR) using C statistics and other evaluation metrics.
  • Identification of significant predictors of DAA treatment failure through the EN model.

Main Results:

  • ML models demonstrated comparable and superior performance in predicting DAA treatment failure (C statistics ranging from 0.72 to 0.75) compared to MLR (C statistic: 0.51).
  • The EN model, utilizing 27 predictors, achieved 58.4% sensitivity and 77.8% specificity, identifying one treatment failure for every nine patients evaluated.
  • Key predictors identified by the EN model included male sex, shorter treatment duration, treatment discontinuation due to adverse events, low albumin, high bilirubin, advanced liver disease, and substance use (tobacco, alcohol, vitamins).

Conclusions:

  • Machine learning algorithms are effective tools for predicting direct-acting antiviral treatment failure in hepatitis C virus patients.
  • Identifying and addressing modifiable risk factors, such as treatment adherence and lifestyle choices, can potentially reduce treatment failures and the need for retreatment.
  • These ML models hold promise for informing public health policies and clinical decision-making to enhance curative treatment strategies for HCV.