Development and validation of a novel nomogram to predict postoperative pancreatic fistula after pancreatoduodenectomy using lasso-logistic regression: an international multi-institutional observational study

Abstract

Background: Existing prediction models for clinically relevant postoperative pancreatic fistula (POPF) after pancreatoduodenectomy (PD) lack discriminatory power or are too complex. This study aimed to develop a simple nomogram that could accurately predict clinically relevant POPF after PD.

Methods: A high-volume, multicenter cohort of patients who underwent PD from the American College of Surgeons-National Surgical Quality Improvement Program database in the United States during 2014-2017 was used as the model training cohort ( n =3609), and patients who underwent PD from the Pancreatic Center of the National Cancer Center Hospital in China during 2014-2019 were used as the external validation cohort ( n =1347). The study used lasso penalized regression to screen large-scale variables, then logistic regression was performed to screen the variables and build a model. Finally, a prediction nomogram for clinically relevant POPF was established based on the logistic model, and polynomial equations were extracted. The performance of the nomogram was evaluated by receiver operating characteristic curve, calibration curve, and decision curve analysis.

Results: In the training and validation cohorts, there were 16.7% (601/3609) and 16.6% (224/1347) of patients who developed clinically relevant POPF, respectively. After screening using lasso and logistic regression, only six predictors were independently associated with clinically relevant POPF, including two preoperative indicators (weight and pancreatic duct size), one intraoperative indicator (pancreatic texture), and three postoperative indicators (deep surgical site infection, delayed gastric emptying, and pathology). The prediction of the new nomogram was accurate, with an area under the curve of 0.855 (95% CI: 0.702-0.853) in the external validation cohort, and the predictive performance was superior to three previously proposed POPF risk score models (all P <0.001, likelihood ratio test).

Conclusions: A reliable lasso-logistic method was applied to establish a novel nomogram based on six readily available indicators, achieving a sustained, dynamic, and precise POPF prediction for PD patients. With a limited number of variables and easy clinical application, this new model will enable surgeons to proactively predict, identify, and manage pancreatic fistulas to obtain better outcomes from this daunting postoperative complication.