Risk stratification for postoperative pancreatic fistula using the pancreatic surgery registry StuDoQ|Pancreas of the German Society for General and Visceral Surgery
Introduction
Postoperative pancreatic fistula (POPF) is one of the major factors for morbidity and mortality in pancreatic surgery. The International Study Group for Pancreatic Surgery (ISGPS) published in 2005 a definition and a classification of POPF, which have been widely accepted and used [1]. According to this definition, a pancreatic fistula is defined as any amylase-rich drainage fluid (>3 times the upper limit of institutional normal serum amylase activity) starting from the third day after surgery. POPF are then classified in grade A, grade B and grade C, according to the extent of the clinical impact. Grade A POPF have no impact on the postoperative course, grade B POPF alter the normal postoperative course, while the condition of the patient stays stable, and grade C POPF lead to severe complications such as organ failure with the need of intensive care or surgical intervention. In 2016, the ISGPS narrowed the definition of POPF to the clinically relevant POPF B and C [2]. While POPF B does not lead to higher mortality, the in-hospital death rate among patients with POPF C is as high as 35% and over one third of in-hospital deaths after pancreatoduodenectomy are contributed to POPF C [3,4]. Thus, assessing risk for POPF has a pivotal role when estimating overall risk in pancreatic surgery.
There has been considerable effort to develop a measurement tool for risk of POPF [[5], [6], [7], [8]]. In a recent review, Sandini et al. described 10 clinical risk scores for POPF, published between 2008 and 2016 [5]. The most common parameters used in the development of predictive scores for POPF are the consistency of the pancreatic parenchyma and the diameter of the main pancreatic duct (MPD). While usually the consistency of the pancreatic parenchyma is subjectively assessed by the surgeon as soft or hard, the MPD is either used as a continuous or as a multicategorial parameter, measured at the resection plane of the pancreas. Since measurements of the MPD intraoperatively, especially when the pancreatic duct is small, might be impractical and imprecise, in the pancreatic surgery registry StuDoQ|Pancreas of the German Society for General and Visceral Surgery MPD has been recorded as a binary variable <3 mm versus ≥3 mm [9]. Furthermore, there is an association of soft pancreatic parenchyma and small MPD that might cause the problem of multicollinearity in the regression models used to develop the prediction scores, and thus lead to instability of the models [10].
Another issue, when developing a predictive model for POPF, is the parameter selection, which usually has been based on univariable testing for significance or some prior knowledge, which might lead to selection bias and potentiating of uncertainties due to multiple testing.
In this study, we address the problems of parameter selection and multicollinearity using multivariable logistic regression with elastic net regularization [11] to develop a predictive model for POPF, using a large sample from the German pancreatic surgery registry StuDoQ|Pancreas [9].
Section snippets
Study Design and Approval
This is a retrospective, registry-based, explorative study. It was approved by the German Society for General and Visceral Surgery and the local ethics committee of the University of Lübeck.
Software
All data processing and analysis was performed with R version 3.3.1. The following R packages were used: Hmisc, knitr, tables, xtable, stats, pROC, stringr, mice, Amelia, caret, ggplot2, gridExtra. The graphics were produced with the R packages ggplot2 and gridExtra.
Study population and parameters
The study population consisted of all
Descriptive and univariable analysis
A total of N = 2488 patients were identified. Of these, 1671 were assigned to the training set and 817 to the test set. Appendix A, Table A1 displays the descriptive statistics including missing data percentage. In the training set, POPF B rate was 8.2%, POPF C - 6.3% and total POPF B/C - 14.5%. The 30-day mortality was 3.4%, in-hospital mortality - 4.6%. Of the patients, who died during postoperative hospital stay, 34.2% had POPF C and 28.4% of the patients with POPF C died during
Discussion
In the era of precision medicine, estimation and quantification of operative risk is of major importance in surgery. Being able to estimate the operative risk allows for an individual therapy concept but also to develop and target new therapeutics and operative techniques to a specific population at risk as well as for risk adjusted quality assurance.
Postoperative pancreatic fistula has acquired much attention due to its enormous impact on morbidity and mortality in pancreatic surgery.
Conclusion
In conclusion, the model presented here is a valid measurement tool for POPF risk. It can be applied to guide the decision for surgery, drainage management, risk adjustment in research studies as well as quality assurance in surgery.
Conflicts of interest
T. Keck and U. F. Wellner are members of the steering committee of StuDoQ|Pancreas.
Acknowledgements
None.
References (26)
- et al.
Postoperative pancreatic fistula: an international study group (ISGPF) definition
Surgery
(2005) - et al.
The 2016 update of the International Study Group (ISGPS) definition and grading of postoperative pancreatic fistula: 11 Years after
Surgery
(2017) - et al.
The characterization and prediction of ISGPF grade C fistulas following pancreatoduodenectomy
J Gastrointest Surg
(2016) - et al.
Toward defining grade C pancreatic fistula following pancreaticoduodenectomy: incidence, risk factors, management and outcome
HPB
(2012) - et al.
Histomorphological features of the pancreatic remnant as independent risk factors for postoperative pancreatic fistula: a matched-pairs analysis
Pancreatology
(2011) - et al.
Using the NSQIP pancreatic demonstration project to derive a modified fistula risk score for preoperative risk stratification in patients undergoing pancreaticoduodenectomy
J Am Coll Surg
(2017) - et al.
Preoperative risk score for pancreatic fistula after distal pancreatectomy: Derivation and multicenter external validation
Pancreatology
(2017) - et al.
A preoperative predictive score of pancreatic fistula following pancreatoduodenectomy
HPB
(2014) Multivariable fractional polynomial method for regression model
Ann Transl Med
(2016)- et al.
Penalized least squares regression methods and applications to neuroimaging
Neuroimage
(2011)