Development and Validation of a web-based Postoperative Clostridioides difficile infection risk prediction model

Abstract

Background: Clostridioides difficile infection is associated with significant morbidity, mortality and increased costs. Assessment of the postoperative C. difficile infection risk is necessary to improve the outcome of surgical patients. Objective: To develop and validate a risk prediction tool for C. difficile infection after surgery. Methods: In this retrospective cohort study, 2,451,169 surgical patients from the American College of Surgeons National Surgical Quality Improvement Program Database (ACS-NSQIP) over 2015-2017 were included. Nine predictors were selected for the model: age, preoperative leukocytosis >12,000 (cells/μL), hematocrit (≤30%), chronic dialysis, insulin dependent diabetes, weight loss, steroid use, presence of preoperative sepsis, and surgery type. A second model included hospital length of stay as a predictor. A predictive model was developed using ACS-NSQIP 2015-2016 training cohort (n=1,435,157) and tested using 2017 validation cohort (n=1,016,012). Multivariate logistic regression was used for the model. Main outcome: The primary outcome was postoperative 30-day C. difficile infection (CDI). Results: 0.39% of the patients (n=9,675) developed CDI and 42.3% (n=4,091) of CDI occurred post-discharge. The Clostridioides difficile risk prediction model had excellent AUC (area under the receiver operating characteristic curve) for postoperative C. difficile infection (training cohort=0.804, test cohort= 0.803). The model that includes hospital length of stay has a high AUC (training cohort=0.841, test cohort=0.838). Conclusion: The C. difficile prediction model provides a robust predictive tool for postoperative C. difficile infection.