Predicting deep surgical site infection in patients receiving open posterior instrumented thoracolumbar surgery: A-DOUBLE-SSI risk score - a large retrospective multicenter cohort study in China

Abstract

Background: To develop a practical prediction model to predict the risk of deep surgical site infection (SSI) in patients receiving open posterior instrumented thoracolumbar surgery.

Methods: Data of 3419 patients in four hospitals from 1 January 2012 to 30 December 2021 were evaluated. The authors used clinical knowledge-driven, data-driven, and decision tree model to identify predictive variables of deep SSI. Forty-three candidate variables were collected, including 5 demographics, 29 preoperative, 5 intraoperative, and 4 postoperative variables. According to model performance and clinical practicability, the best model was chosen to develop a risk score. Internal validation was performed by using bootstrapping methods.

Results: After open posterior instrumented thoracolumbar surgery, 158 patients (4.6%) developed deep SSI. The clinical knowledge-driven model yielded 12 predictors of deep SSI, while the data-driven and decision tree model produced 11 and 6 predictors, respectively. A knowledge-driven model, which had the best C-statistics [0.81 (95% CI: 0.78-0.85)] and superior calibration, was chosen due to its favorable model performance and clinical practicality. Moreover, 12 variables were identified in the clinical knowledge-driven model, including age, BMI, diabetes, steroid use, albumin, duration of operation, blood loss, instrumented segments, powdered vancomycin administration, duration of drainage, postoperative cerebrospinal fluid leakage, and early postoperative activities. In bootstrap internal validation, the knowledge-driven model still showed optimal C-statistics (0.79, 95% CI: 0.75-0.83) and calibration. Based on these identified predictors, a risk score for deep SSI incidence was created: the A-DOUBLE-SSI (Age, D [Diabetes, Drainage], O [duration of Operation, vancOmycin], albUmin, B [BMI, Blood loss], cerebrospinal fluid Leakage, Early activities, Steroid use, and Segmental Instrumentation) risk score. Based on the A-DOUBLE-SSI score system, the incidence of deep SSI increased in a graded fashion from 1.06% (A-DOUBLE-SSIs score ≤8) to 40.6% (A-DOUBLE-SSIs score>15).

Conclusions: The authors developed a novel and practical model, the A-DOUBLE-SSIs risk score, that integrated easily accessible demographics, preoperative, intraoperative, and postoperative variables and could be used to predict individual risk of deep SSI in patients receiving open posterior instrumented thoracolumbar surgery.

Trial registration: ClinicalTrials.gov NCT05740865.

Figure 1

The study flow diagram.

Figure 2

Calibration for deep SSI prediction in the internal validation data set. The Italian Group for the Evaluation of Intervention in Intensive Care Medicine (GiViTI) calibration belt (right) and the Hosmer—Lemeshow test (H–L) calibration plot (left). The GiViTI belt visualizes risk intervals of under- and over-prediction, respectively, for a given model as the 95% CI does not cross the red diagonal bisector line.

Figure 3

The A-DOUBLE-SSI score for deep SSI incidence and the risk groups (low ≤8, moderate >8-≤12, high >12-≤15, and very high >15).