-
- John A Harvin, Charles E Green, Claudia Pedroza, Jon E Tyson, Laura J Moore, Charles E Wade, John B Holcomb, and Lillian S Kao.
- Department of Surgery, University of Texas McGovern Medical School, Houston, TX; Center for Translational Injury Research, University of Texas McGovern Medical School, Houston, TX. Electronic address: john... more
- J. Am. Coll. Surg. 2019 Mar 1; 228 (3): 255264255-264.
BackgroundIn an earlier study, we reported the successful reduction in the use of damage control laparotomy (DCL); however, no change in the relative frequencies of specific indications was observed. In this study, we aimed to use machine learning to help identify the changes in surgical decision making that occurred.Study DesignAdult patients undergoing emergent trauma laparotomy were included: pre-quality improvement (QI): January 1, 2011 to October 31, 2013 and post-QI: November 1, 2013 to June 30, 2016. Using 72 variables before or during emergent laparotomy, random forest algorithms predicting DCL before and after a QI intervention were created. The main end point of the algorithms was the strength of individual factor significance in predicting the use of DCL, calculated by determining the mean decrease in accuracy (MDA) in the model if that variable was removed.ResultsIn the pre-QI group, 24 of 72 factors significantly predicted DCL, the strongest being bowel resection (mean MDA 16) and operating room RBC transfusions (mean MDA 15). The remaining variables were spread along the continuum of care from injury to emergent laparotomy end. In the post-QI group, 12 of 72 factors significantly predicted DCL, the strongest being last operating room lactate (mean MDA 12) and operating room RBC transfusions (mean MDA 14). In addition to having 12 fewer significant factors predictive of DCL, the predictive factors in the post-QI group were mainly intraoperative factors.ConclusionsA machine learning analysis provided novel insights into the changes in decision making achieved by a successful QI intervention and should be considered an adjunct to understanding successful pre- and post-intervention QI studies. The analysis suggested a shift toward using mostly intraoperative factors to determine the use of DCL.Copyright © 2019 American College of Surgeons. Published by Elsevier Inc. All rights reserved.
Notes