Machine learning approach for risk factors detection of pancreatic fistula and AI diagnostic systems development

Mikhail Potievskiy, Sergei Ivanov, Andrei Kaprin, Ruslan Moshurov, Leonid Petrov, Peter Shegai, Pavel Sokolov, Vladimir Trifanov

potievskiymikhail [at] gmail.com

Abstract

Introduction. The aim of the study was to develop a predictive ML model for postoperative pancreatic fistula and to determine the main risk factors of the complication.
Materials and Methods. We performed a single-centre retrospective clinical study. 150 patients, who underwent pancreatoduodenal resection in FSBI NMRRC, were included. We developed ML models of biochemic leak and fistula B/C development. Logistic regression, Random forest and CatBoost algorithms were employed. The risk factors were evaluated basing on the most accurate model, roc auc, and Kendall correlation, p<0.05.
Results. We detected a significant positive correlation between blood and drain amylase level increase in association with biochemical leak and fistula B/C. The CatBoost algorithm was the most accurate, roc auc 74%-86%. The main pre- and intraoperative prognostic factors of all the fistulas were tumor vascular invasion, age and BMI, roc auc 70%. Specific fistula B/C factors were the same. Basing on the 3-5 days data, biochemical leak and fistula B/C risk factors were blood and drain amylase levels, blood leukocytes, roc auc 86% and 75 %.
Conclusion: We developed sufficient quality ML models of postoperative pancreatic fistulas. Blood and drain amylase level increase, tumor vascular invasion, age and BMI were the major risk factors of further fistula B/C development.

Keywords: machine learning, precision oncology, risk factor detection, pancreatoduodenal resection, pancreatic fistula