Diagnostic Codes in AI Prediction Models and Label Leakage of Same-Admission Clinical Outcomes

Are () diagnostic codes, which are only finalized after hospital discharge, associated with inflated performance of artificial intelligence (AI) health care prediction models?In this prognostic study of 180 640 patients, 40.2% of published AI models trained to predict same-admission outcomes used codes as features. Prediction models for inpatient mortality trained on codes predicted in-hospital mortality with high accuracy, with the most important codes (eg, brain death, encounter for palliative care) not available in time for clinically useful mortality prediction.These findings suggest that to ensure that AI prediction models are both reliable and clinically deployable, greater diligence is needed in identifying and preventing label leakage.

Corresponding Author: Brett K. Beaulieu-Jones, PhD, Division of the Biological Sciences, Department of Biomedical Informatics, The University of Chicago, 5841 S Maryland Ave, M607, Chicago, IL 60637 (beaulieujones@uchicago.edu).

Link to the article in your story

We encourage you to link out to this article in your story using the link below. It includes an access token that will give free access to the article for your readers up to one year after publication. (The link will be live after the article publishes and embargo is lifted.)

Please see the article for additional information, including full author list, author contributions and affiliations, conflict of interest and financial disclosures, and funding and support.

Need more information? Contact us.