Identifying postop complications using EHR data and machine learning

Editor's Note

Using machine learning on electronic health record (EHR) postoperative data linked to the American College of Surgeons National Surgical Quality Improvement Program (ACS NSQIP) outcomes data, researchers developed a model with 163 predictors of postoperative complications at the University of Colorado Hospital.

Of 6,840 patients analyzed with the model, 13.5% had at least one of the 18 complications tracked by ACS NSQIP. The model had 88% specificity, 83% sensitivity, and an area under the curve of 0.93.

This model may be useful for electronic surveillance of postoperative complications, the authors say.