DISPAIR vs. D-FRS - Establishing the Superiority of the Novel Distal Pancreatectomy Pancreatic Fistula Risk Prediction Models - A Multicenter International Retrospective Validation Study

Fistula Risk Score (FRS) and its derivatives have been utilized in research surrounding pancreatoduodenectomy (Callery 2013, Mungroop 2019 & 2021). Distal pancreatectomy is a rare procedure, with typically less than 20 procedures a year performed by a large hepatopancreatobiliary surgery centre. A prediction model for pancreatic fistula after distal pancreatectomy had long been warranted and Ecker et al. were the closest at acquiring this in their 2019 study with over two thousand patients (Ecker 2019). However, not enough strongly associated variables were identified in their study to develop a prediction model. Very recently, two distinct models - the DISPAIR (Bonsdorff 2022) and the D-FRS (De Pastena 2022) - which both included pancreas-specific anthropometric measurements, were developed and validated. No studies comparing the performance of these models have been conducted and thus superiority of one model over other hasn't been established.

The DISPAIR relies on three preoperative variables: pancreatic thickness at the intended transection plane measured from preoperative CT-scans, site of transection (neck vs. body/tail) and history of diabetes. It was developed in 266 patients undergoing DP in Finland and externally validated with 402 patients from Sweden. It showed good discrimination and adequate calibration upon external validation with area under the curve (AUC) of 0.80, calibration intercept of 0.19 and slope of 0.72. The D-FRS is based on PT and main pancreatic duct (MPD) diameter at the pancreatic neck, both measured from preoperative CT-scans as well. It showed a satisfactory AUC of 0.73 after an internal-external validation procedure (Steyerberg 2016) where the development cohort of 339 patients was pooled with three distinct cohorts with a total sample size of 997 patients. Pooling validation cohort with the development cohort increases the optimism of model performance parameters, and strictly speaking does not count as a full external validation. The authors claimed D-FRS to be perfectly calibrated with a calibration intercept of 0 and slope of 1. Since this is in essence impossible, the soundness of the methodology behind the study is questionable (Van Calster 2019). Nonetheless, both models have identified similar novel risk factors for pancreatic fistula and show good potential for wider utilization.

The aim of this study is to compare and externally validate the performance of the DISPAIR and the D-FRS in a fully independent cohort of DP patients. The ultimate goal is to establish the potential superiority of one model over the other and identify directions for potential model updating.

As the DISPAIR is already externally validated we expect its performance to vary little and the AUC to set in the range of 0.75 - 0.85 in external validation cohorts. We expect to identify potential avenues of DISPAIR model updating with this external validation study. As the D-FRS has been validated with a pooled internal-external procedure it is more difficult to predict its performance but an AUC of over 0.70 would be expected.

The plan is to collect 200 patients per center, as this will give approximately 40 patients with a clinically relevant pancreatic fistula, allowing external validation and comparison of the scores center-wise also (in addition to pooled external validation and comparison).