Evaluating predictive models in reproductive medicine
Artificial intelligence (AI) use in clinical decision- making in reproductive medicine requires trans- parent and accountable data-reporting practices, and journal referees and readers cannot evaluate AI models without standardization.
VOLUME 114, ISSUE 5, P921-926
Carol Lynn Curchoe, Ph.D., Adolfo Flores-Saiffe Farias, Ph.D., Gerardo Mendizabal-Ruiz, Ph.D., Alejandro Chavez-Badiola, M.D.
Predictive modeling has become a distinct subdiscipline of reproductive medicine, and researchers and clinicians are just learning the skills and expertise to evaluate artificial intelligence (AI) studies. Diagnostic tests and model predictions are subject to evaluation. Their use offers potential for both harm and benefit in terms of diagnosis, treatment, and prognosis. The performance of AI models and their potential clinical utility hinge on the quality and size of the databases used, the types and distribution of data, and the particular AI method applied. Additionally, when images are involved, the method of capturing, preprocessing, and treatment and accurate labeling of images becomes an important component of AI modeling. Inconsistent image treatment or inaccurate labeling of images can lead to an inconsistent database, resulting in poor AI accuracy. We discuss the critical appraisal of AI models in reproductive medicine and convey the importance of transparency and standardization in reporting AI models so that the risk of bias and the potential clinical utility of AI can be assessed.