Predicting hepatitis C infection via machine learning

HCV infection is frequently asymptomatic, with current diagnosis relying mainly on costly and less accessible HCV RNA testing. While HCV-Ab and biochemical markers exhibit suboptimal diagnostic performance, whether machine learning can enhance their accuracy remains unclear.

This study is a retrospective study, which included data from 179 patients whose HCV-Ab levels were greater than 1.00 S/CO to explore the relationship between HCV-Ab, biochemical indicators, and HCV infection. Univariate logistic regression and restricted cubic splines (RCS) were employed to explore these associations. Machine learning integrated HCV-Ab and biochemical indicators to predict early HCV infection (undiagnosed chronic cases), with validation conducted using receiver operating characteristic curve (ROC) analysis. The machine learning approach randomly divided study participants into training and test sets at a 5:5 ratio, with the training set being used for variable selection and model construction.

After full adjustment, TP showed no significant association with HCV infection. Restricted cubic spline (RCS) analysis revealed nonlinear relationships between HCV-Ab, ALT, AST, mAST, GGT, A/G and HCV infection. HCV-Ab exhibited an inflection point at 11.17 (below: OR = 1.04 per unit increase; above: no association). Similar threshold patterns were observed for ALT, AST, mAST and GGT. The integrated HCV-Ab and biochemical marker model achieved excellent predictive performance (AUC = 0.977).

TP exhibited a linear association with HCV infection, whereas HCV-Ab, ALT, AST, mAST and GGT showed nonlinear associations with distinct threshold effects. Early prediction of HCV infection using these indicators represents a cost-effective strategy.