Preoperative prediction of pancreatic fistula after distal pancreatectomy – Finally here!

A preoperative prediction model for pancreatic fistula after distal pancreatectomy, called the DISPAIR (DIStal PAcreatectomy fIstula Risk) score, was recently developed in a Finnish cohort and validated in a Swedish cohort with 668 patients undergoing distal pancreatectomy altogether¹. It utilizes preoperative variables, portion of which are measured from CT-scans, and is based on the following three: pancreatic thickness, transection site (neck vs. body/tail) and diabetes. It showed good discrimination in the external validation cohort with an AUC of 0.80, as well as adequate calibration. With DISPAIR, specific fistula risk can be calculated for each patient. The DISPAIR can be accessed at https://www.evidencio.com/models/show/2611

A risk estimation tool for pancreatic fistula after distal pancreatectomy had long been warranted. In 2019, Ecker et al. got close with their study of over two-thousand patients, but in the end their multivariable analyses provided a model with a disappointing AUC of 0.65². It was discussed that their study, although extensive, lacked the inclusion of pancreas-specific parameters and was thus incapable of predicting pancreatic fistula, a process of the pancreatic remnant. With this in mind, we conducted a study including reasonable preoperatively assessable parameters from CT-scans, such as pancreatic thickness and duct diameter at different regions, visceral fat and sarcopenia¹. We also checked how reliably these measurements could be repeated by independent observers in order to exclude variables with high interobserver variability. For example, main pancreatic duct diameter had too high interobserver variation to be included in the model. Similarly, pancreatic texture was excluded as it is difficult to assess in minimally invasive surgery prior to transecting the pancreas. After uni- and multivariable analyses the DISPAIR-score was born.

Interestingly, at the same time of the publication of the DISPAIR, a somewhat similar model, the D-FRS (Distal-Fistula Risk Score), was published ahead of print by De Pastena et al.³. It was developed in Italy and validated in an internal-external fashion in cohorts from the Netherlands and the USA. The preoperative D-FRS is based on two preoperative variables, the pancreatic thickness and duct diameter at the neck (both associated positively with the fistula in D-FRS) and showed satisfactory performance with its AUC of 0.73 in internal-external validation. However, no pure external validation has been carried out for D-FRS. Pooling the development cohort with the validation cohort (ie. internal-external validation) usually translates to more optimistic measures of performance (ie. overfit). The authors of D-FRS also presented an intraoperative D-FRS with extra variables (BMI, pancreatic texture, operative time), but this did not undergo any validation.

In addition to the scholarly aspects and utilities of the DISPAIR – for example risk adjusted analyses and risk group selection for trials – clinical applications exist. Ideally, the DISPAIR could be used to stratify patients into risk groups (low, moderate, high, extreme) before the operation, which allows effective targeted use of mitigation strategies in patients with high risk of pancreatic fistula. The operating surgeon could easily measure pancreatic thickness at a desired transection site while checking the preoperative CT-scan. As the fistula risk in the DISPAIR depends on transection site as well as thickness (which tends to vary between possible sites of transection), the operating surgeon could assess different strategies of approach, and weigh the fistula risk associated with them. In other words, the DISPAIR could be used to determine an optimal site of transection regarding pancreatic fistula risk.

A recent systematic review demonstrated that the previously published pancreatic fistula risk scores for pancreatoduodenectomy (such as FRS, A-FRS, or UA-FRS) either lack sufficient external validation or show suboptimal performance in external validation⁴. The AUC-values, representing model’s ability to classify patients with or without fistula, were on average 0.70 for the most utilized models. In this respect, the DISPAIR with its AUC of 0.80 in external validation shows a lot of promise. Interestingly, the fistula risk scores for pancreatoduodenectomy can also be used to predict postoperative pancreatitis⁵.

As there now seem to be two new kids (D-FRS and DISPAIR) in town, a multicenter international comparative validation study is surely warranted.