Potential Cost-Effectiveness of Machine Learning-Enabled Primary Care Identification of Hepatitis C Virus Patients in the US.

Machine learning (ML) algorithms may be effective at improving the HCV care cascade. One ML algorithm, developed using U.S. ambulatory electronic medical records (EMR), demonstrated the ability to identify people infected with HCV earlier than conventional testing strategies among those with indications for screening. We evaluated the potential cost-effectiveness of ML-enabled screening for the early identification of undiagnosed HCV among people in care in the U.S. An HCV natural history Markov model was developed to evaluate the cost-effectiveness of the ML algorithm-enabled screening compared to conventional testing over the training data period. Based on the training data, the ML algorithm identified patients on average 6.5 months earlier than conventional testing strategies. We compared the status quo to intervention scenarios using the ML algorithm at different recall levels (proportion of HCV patients identified, 5-100%). We identified the optimal algorithm recall level, which maximized health (measured in quality-adjusted life years, QALYs) while staying under a willingness-to-pay threshold of USD$100,000/QALY gained. ML-enabled screening was cost-effective (ICER < $100 k/QALY gained) in identifying undiagnosed HCV patients for recall levels up to 30%. The optimal recall level was 30% (Precision 0.27%), which resulted in a mean ICER of $94,022/QALY gained. ML-enabled screening for the early identification of undiagnosed HCV patients could be cost-effective in the U.S. Prospective evaluation of real-world effectiveness is warranted.