The Application Of Machine Learning To The Prediction Of Postoperative Sepsis Following Appendectomy
*Corinne Bunn1, *Sujay Kulshrestha1, *Neelam Balasubramanian2, *Jason Boyda2, *Steven Birch2, *Ibrahim Karabayir3, Marshall S Baker1, *Francois Modave4, *Oguz Akbilgic3
1Department of Surgery, Loyola University Medical Center, Chicago, IL;2Loyola University, Chicago, IL;3Department of Health Informatics and Data Science, Loyola University, Chicago, IL;4Department of Health Outcomes and Biomedical Informatics, College of Medicine, University of Florida, Gainesville, FL
Development of sepsis following appendectomy is uncommon but is associated with significant morbidity and increased costs of care. Factors associated with the development of sepsis post-appendectomy are not well defined. We use a national dataset to assess the utility of various machine learning algorithms in modeling the development of postoperative sepsis following appendectomy.
The National Surgery Quality Improvement Program database was queried to identify appendectomy cases performed in adult patients presenting without pre-operative sepsis between 2012 - 2017. Postoperative sepsis was defined as systemic sepsis/septic shock as classified by the Systemic Inflammatory Response Syndrome (SIRS) Criteria in combination with documented infection and/or end-organ ischemia. Data was split into 80% training and 20% hidden test sets with subsequent 5-fold cross validation of the training set. Logistic Regression (LR), Support Vector Machines (SVM), Random Forest Decision Trees (RFDT), and Extreme Gradient Boosting (XGB) Machines were used to model the occurrence of postoperative sepsis.
223,214 patients met inclusion criteria. 2,141 (0.96%) developed postoperative sepsis. LR, RFDT, and XGB models performed similarly, while SVM yielded significantly lower performance. The XGB model provided the highest accuracy (AUC = 0.70, 95% confidence interval [0.68-0.73]). Variable importance analyses revealed that recent 30-day exacerbation/diagnosis of congestive heart failure (CHF), transfusion of 1 or more units of red blood cells within 72-hours pre-operatively and acute renal failure within 24 hours prior to surgery were the most important predictors of postoperative sepsis (variable importance scores 100.0, 48.04, 44.81 respectively).
Machine learning methods can be used to predict the development of postoperative sepsis post-appendectomy with moderately high accuracy. In patients who initially present without sepsis, factors associated with the development of post-operative sepsis after appendectomy include recent CHF exacerbation/diagnosis, acute renal failure and preoperative transfusion. Such predictive models may ultimately allow for recognition of patients at risk for postoperative sepsis following appendectomy prior to surgical intervention potentially facilitating early risk mitigation and improve informed decision making.
Back to 2020 Abstracts