Derivation and validation of a novel preoperative risk prediction model for surgical site infection in pancreaticoduodenectomy and comparison of preoperative antibiotics with different risk stratifications in retrospective cohort

Surgical site infections (SSIs) are common postoperative complications of pancreaticoduodenectomy.To develop a model for preoperative identification of the risk of SSI that may improve outcomes and guide preoperative antibiotics.The prediction model was built by meta-analysis. After literature search and inclusion, data extraction, and quantitative synthesis, the prediction model was established based on the pooled odds ratio of predictors. A single-centre retrospective cohort was the validation cohort. Receiver operating characteristic curves and area under the curve were used to assess the model's ability. We also created a decision curve and a calibration plot to assess the nomogram. The effects of prophylactic antibiotics on SSI were compared between groups by multivariable logistic regression with different risk stratifications.Twenty-eight studies were included in the meta-analysis, 17 studies in the derivation cohort. Age, male gender, body mass index, pancreatic duct diameter, high-risk diagnosis, and preoperative biliary drainage were selected to build the prediction model. The model was validated in an external cohort. The cut-off value was 3.5 and area under the curve (AUC) was 0.76 in open pancreaticoduodenectomy (OPD). In laparoscopic pancreaticoduodenectomy, the cut-off value was 4.5 and AUC was 0.69. Decision curve and calibration plot showed good usability of the model, especially in OPD. Multivariable logistic regression did not indicate differences between broad- and narrow-spectrum antibiotics for SSI in different risk stratifications.The model can identify patients with a high risk of SSI preoperatively. The choice of prophylactic antibiotics under different risk stratifications should be investigated further.