Performance metrics for machine learning solutions in laboratory medicine

Machine learning-based solutions to laboratory medicine problems have become commonplace in literature, but real-world implementations remain rare, in no small part because of the substantial investment required to incorporate such solutions into routine clinical care. A crucial step in advancing a machine learning solution from proof of concept into clinical application is a robust and comprehensive evaluation of its performance. In this review, we discuss the common methods, best practices, and potential pitfalls in evaluating machine learning-based solutions to clinical laboratory problems.