Our paper on our experience running a risk model trained in the lab (retrospectively) in real-world clinical systems (prospectively) is now out on JAMA Network Open.
In this study, we investigate how well electronic health record–based suicide risk models perform in the clinical setting and whether performance is generalizable. This cohort study of 30-day suicide attempt risk among 77,973 patients showed good performance in nonpsychiatric clinical settings at scale and in real time. Numbers needed to screen were reasonable for an algorithmic screening test that required no additional data collection or face-to-face screening to calculate. We show that suicide attempt risk models can be implemented with accurate performance at scale, but performance is not equal in all clinical settings, which requires model recalibration and updating prior to deployment in new settings.
Many thanks to coauthors Kevin B. Johnson, Michael Ripperger, Sarah Sperry, Joyce Harris, Nathaniel Clark, Elliot Fielstein, Laurie Novak, Katelyn Robinson, and Bill Stead for their roles in this work as well as dozens of Vanderbilt collaborators in the Department of Biomedical Informatics, the Department of Psychiatry and Behavioral Sciences, Vanderbilt Health, and VUMC HealthIT.
The full paper is available at this link, and the VUMC Reporter covers the study here.