International Multicentre Validation of the Left Pancreatectomy Pancreatic Fistula Prediction Models and Development and Validation of the Combined DISPAIR-FRS Prediction Model

Hepatobiliary System Medicine
left pancreatectomy pancreatic fistula prediction models multicentre study DISPAIR-FRS risk score postoperative complications

Academic Background

Pancreatectomy is an important treatment for pancreatic diseases, but postoperative pancreatic fistula (POPF) is a common and serious complication, with an incidence rate as high as 20%. POPF not only prolongs hospital stays and increases medical costs but is also closely associated with increased postoperative mortality. Therefore, accurate prediction of POPF risk is crucial for improving patient outcomes and optimizing treatment strategies. Currently, two independent preoperative prediction models—DISPIR and D-FRS—were published in 2022 and have undergone external validation. However, the performance of these models in practical applications remains unclear. This study aims to conduct an international multicenter validation of these two models and explore whether a new, better-performing model can be developed by combining the existing ones.

Source of the Paper

This research was conducted by scholars from several internationally renowned institutions, including Helsinki University Hospital in Finland, Oslo University Hospital in Norway, Aarhus University Hospital in Denmark, Pitié-Salpêtrière Hospital in Paris, France, and Karolinska University Hospital in Sweden, among others. The study was published in 2025 in the British Journal of Surgery (BJS) under the title “International multicentre validation of the left pancreatectomy pancreatic fistula prediction models and development and validation of the combined DISPIR-FRS prediction model.”

Research Process

Study Design

This study is an international multicenter retrospective cohort study, including adult patients who underwent left pancreatectomy after January 1, 2010, at nine high-volume pancreatic surgery centers (eight in Europe and one in North America). Inclusion criteria were age over 18 years and open or minimally invasive left pancreatectomy (with or without splenectomy). Each center had to include at least 150 patients, with a total of 2,284 patients ultimately included.

Data Collection and Processing

Patient data were obtained through medical record reviews at each center, anonymized, and encrypted before being transmitted to the coordinating team. Study variables included patient demographic information, pancreatic thickness (PT) and main pancreatic duct diameter (MPD) measured from preoperative CT images, surgical methods, postoperative complications, etc. Missing data were handled using multiple imputation methods.

Model Validation and Update

The study first externally validated the DISPIR and D-FRS models, assessing model discrimination (area under the ROC curve, AUC) and calibration (calibration curves). Subsequently, a new combined model—DISPIR-FRS—was developed by integrating stable predictors from the two models and other readily available patient demographic information. The new model was validated using an internal-external validation method and evaluated for its performance in different subgroups.

Main Results

Model Validation Results

Both the DISPIR and D-FRS models performed suboptimally in the overall validation cohort, with AUCs of 0.62. Calibration curves showed that both models significantly overestimated POPF risk when predicted probabilities exceeded 20%. This indicates that the predictive capabilities of the two models are limited in practical applications, potentially due to overfitting.

New Model Development and Validation

By combining stable predictors from the DISPIR and D-FRS models, the study developed a new combined model, DISPIR-FRS. The new model includes predictors such as pancreatic thickness (PT), main pancreatic duct diameter (MPD), age, sex, and pancreatic transection site. Internal-external validation results showed that DISPIR-FRS had an AUC of 0.72, a calibration slope of 0.93, and a calibration intercept of -0.02, significantly outperforming the original models.

Subgroup Analysis

DISPIR-FRS demonstrated good generalizability across different subgroups, including different surgical methods (open or minimally invasive), different pancreatic transection sites, and whether intra-abdominal drainage was used. Only in the subgroup of patients without intra-abdominal drainage did the model exhibit systematic overestimation of POPF risk.

Conclusions and Significance

This study, through international multicenter validation, found that the existing DISPIR and D-FRS models performed poorly in practical applications, with issues of overfitting and systematic risk overestimation. By combining stable predictors from the two models, the study developed a new combined model, DISPIR-FRS, which significantly outperformed the original models in terms of predictive performance and stability. The DISPIR-FRS model can not only be used for clinical decision support but also for research design, case-mix adjustment, and surgical decision optimization, holding significant scientific and practical value.

Research Highlights

  1. International Multicenter Validation: The study included 2,284 patients from multiple countries, ensuring broad representativeness and reliability of the data.
  2. Model Update and Optimization: By combining stable predictors from existing models, a better-performing DISPIR-FRS model was developed.
  3. Internal-External Validation: The internal-external validation method ensured the model’s stability and generalizability.
  4. Subgroup Analysis: The model demonstrated good generalizability across different subgroups, providing important references for clinical practice.

Other Valuable Information

The study also found that pancreatic thickness (PT) is a reliable predictor of POPF, particularly at the pancreatic transection site. Additionally, age has a non-linear protective effect on POPF risk, especially in patients over 50 years old. These findings provide new directions for future research, such as further exploring the relationship between pancreatic thickness and surgical techniques, as well as the mechanisms by which age influences POPF risk.

Through international multicenter validation and model updates, this study provides a more reliable tool for predicting the risk of postoperative pancreatic fistula after left pancreatectomy, holding significant clinical and research value.