Predicting hepatitis C infection via machine learning.

Area of Science:

• Hepatology
• Infectious Diseases
• Medical Diagnostics

Background:

• Hepatitis C Virus (HCV) infection diagnosis often relies on costly HCV RNA testing.
• Current diagnostic markers like HCV-Antibody (HCV-Ab) and biochemical markers have suboptimal accuracy.
• The potential of machine learning to improve diagnostic accuracy for HCV using these markers is not well-established.

Purpose of the Study:

• To investigate the relationship between HCV-Antibody (HCV-Ab) levels, biochemical indicators, and HCV infection.
• To evaluate the efficacy of machine learning in enhancing the diagnostic accuracy of these markers for early HCV detection.
• To explore nonlinear associations and threshold effects of diagnostic markers in relation to HCV infection.

Main Methods:

• Retrospective study of 179 patients with elevated HCV-Ab levels.
• Utilized univariate logistic regression and restricted cubic splines (RCS) to analyze associations.
• Developed a machine learning model integrating HCV-Ab and biochemical indicators for early HCV infection prediction, validated with ROC analysis.

Main Results:

• Nonlinear relationships and threshold effects were identified for HCV-Ab, ALT, AST, mAST, and GGT.
• HCV-Ab showed an inflection point at 11.17 S/CO.
• The integrated machine learning model demonstrated excellent predictive performance with an AUC of 0.977.

Conclusions:

• HCV-Ab, ALT, AST, mAST, and GGT exhibit nonlinear associations with distinct threshold effects in HCV infection.
• Machine learning integration of these markers provides a highly accurate and cost-effective strategy for early HCV infection prediction.
• This approach offers a promising alternative to traditional diagnostic methods.