Trends and Prediction of Surgical Site Infection After Elective Spine Surgery: An Analysis of the American College of Surgeons National Surgical Quality Improvement Project Database.

Abstract

Background: Surgical site infection (SSI) is an infrequent but costly complication after elective spine surgery. Identification of important temporal changes and predictive factors may inform targeted prevention efforts. Patients and Methods: A retrospective study of elective spine surgery patients was performed using the National Surgical Quality Improvement Programs (NSQIP) database from 2011 and 2019. Temporal changes in SSI and related factors were examined descriptively. Recursive partitioning and bootstrap forest techniques were used to inform the development of predictive models for SSI. Results: A total of 6,038 (1.66%) of 363,754 patients had an SSI recorded. Peri-operative transfusion and preoperative anemia decreased over the nine-year period, however, obesity and diabetes mellitus increased, whereas the SSI rate remained essentially unchanged. A full model including 15 variables had an area under the curve (AUC) of 0.693 (95% confidence interval [CI], 0.686-0.700) whereas a reduced model with just nine variables had an AUC of 0.690 (95% CI, 0.683-0.697). Adjusted odd ratios (aOR) greater than two were noted for only three variables; a posterior approach (aOR, 2.32; 95% CI, 2.14-2.50), body mass index (BMI) >40 kg/m2 (aOR, 2.63; 95% CI, 2.39-2.90), and surgical duration longer than 350 minutes (aOR, 2.39; 95% CI, 2.14-2.67). Remaining retained variables included albumin <3.5 g/dL, inpatient procedure, peri-operative transfusion, diabetes mellitus (both insulin/non-insulin), anemia, and smoking. Conclusions: Surgical site infection rate remained unchanged over a nine-year period despite the lower rates of allogeneic blood transfusion. Class 3 obesity, long operative times, and a posterior approach mainly for thoracic/lumbar spine procedures seemed more pragmatic, but their predictive performance was only modest in our prediction models for SSI.