Identification of postoperative complications using electronic health record data and machine learning

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Surgical Outcomes Research

Background:

• The American College of Surgeons National Surgical Quality Improvement Program (NSQIP) tracks surgical complications.
• Accurate identification of postoperative complications is crucial for patient care and quality improvement.

Purpose of the Study:

• To develop a machine learning algorithm for identifying patients with postoperative complications.
• To utilize electronic health record (EHR) data for complication prediction.

Main Methods:

• An elastic-net model was employed for regression and variable selection.
• Predictors included International Classification of Disease (ICD-9) codes, Common Procedural Terminology (CPT) codes, medications, and CPT-specific complication rates.
• Data was sourced from the University of Colorado Hospital's EHR and linked to NSQIP outcomes.

Main Results:

• The study analyzed 6840 patients, with 922 (13.5%) experiencing at least one of 18 NSQIP-tracked complications.
• The developed model demonstrated high performance: 88% specificity, 83% sensitivity, 97% negative predictive value, and 52% positive predictive value.
• An area under the curve (AUC) of 0.93 indicated strong predictive accuracy.

Conclusions:

• Machine learning applied to EHR data can effectively identify postoperative complications.
• A model utilizing 163 EHR predictors showed strong performance in predicting complications at the institutional level.
• This approach facilitates improved surgical quality and patient safety through enhanced complication detection.