BACKGROUND

Predicting the mortality from post-operative sepsis remains a continuing problem. We built a statistical model using national data to predict mortality in patients who developed post-operative sepsis.

METHODS

This is a retrospective study using the American College of Surgeons National Quality Surgical Improvement Program database, in which we gathered data from adult patients between 2011 and 2016 who experienced postoperative sepsis. We designed a predictive model using multivariable logistic regression on a training set and validated the model on a separate test set.

RESULTS

There were 128,325 patients included in the final dataset, in which 18,499 (14.4%) died within 30-days of surgery. The model consisted of 10 covariates: American Society of Anesthesiologists Physical Status classification score, preoperative sepsis, age, chronic obstructive pulmonary disease, postoperative myocardial infarction, postoperative stroke, postoperative acute renal failure, transfusion requirement, and infection type. A point-based risk calculator was developed, which had an area under the receiver operating characteristics curve of 0.819 (95% confidence interval 0.814-0.823).

CONCLUSION

Although further work is needed to confirm and validate our model on external datasets, our scoring system provides a novel way to measure mortality in septic post-operative patients.