A novel prediction model of pancreatic fistula after pancreaticoduodenectomy using only preoperative markers

Hiroto Matsui; Yoshitaro Shindo; Daisaku Yamada; Hiroyuki Ogihara; Yukio Tokumitsu; Masao Nakajima; Michihisa Iida; Nobuaki Suzuki; Shigeru Takeda; Yuki Nakagami; Shogo Kobayashi; Hidetoshi Eguchi; Tatsuya Ioka; Yoshihiko Hamamoto; Hiroaki Nagano

doi:10.1186/s12893-023-02213-1

A novel prediction model of pancreatic fistula after pancreaticoduodenectomy using only preoperative markers

BMC Surg. 2023 Oct 12;23(1):310. doi: 10.1186/s12893-023-02213-1.

Authors

Affiliations

¹ Department of Gastroenterological, Breast and Endocrine Surgery, Yamaguchi University Graduate School of Medicine, 1-1-1 Minami-Kogushi, Ube, Yamaguchi, Ube, 755-8505, Yamaguchi, Japan.
² Department of Gastroenterological Surgery, Graduate School of Medicine, Osaka University, Osaka, Japan.
³ Department of Computer Science and Electronic Engineering, National Institute of Technology, Tokuyama College, Shunan, Yamaguchi, Japan.
⁴ Department of Data Science, Graduate School of Economics, Shimonoseki City University, Shimonoseki, Yamaguchi, Japan.
⁵ Oncology Center, Yamaguchi University Hospital, Ube, Yamaguchi, Japan.
⁶ Division of Electrical, Electronic and Information Engineering, Graduate School of Sciences and Technology for Innovation, Yamaguchi University, Ube, Yamaguchi, Japan.
⁷ Department of Gastroenterological, Breast and Endocrine Surgery, Yamaguchi University Graduate School of Medicine, 1-1-1 Minami-Kogushi, Ube, Yamaguchi, Ube, 755-8505, Yamaguchi, Japan. hnagano@yamaguchi-u.ac.jp.

Abstract

Background: Since clinically relevant postoperative pancreatic fistula (CR-POPF) can cause intra-abdominal hemorrhage and abscesses, leading to surgery-related deaths after pancreaticoduodenectomy (PD), its preoperative prediction is important to develop strategies for surgical procedures and perioperative management. This study aimed to establish a novel prediction model for CR-POPF using preoperative markers.

Methods: On a training set of 180 patients who underwent PD at the Yamaguchi University Hospital, a combination of CR-POPF predictors were explored using the leave-one-out method with a unique discrete Bayes classifier. This predictive model was confirmed using a validation set of 366 patients who underwent PD at the Osaka University Hospital.

Results: In the training set, CR-POPF occurred in 60 (33%) of 180 patients and 130 (36%) of 366 patients in the validation set using selected markers. In patients with pancreatic ductal adenocarcinoma (PDAC), the main pancreatic duct (MPD) index showed the highest prognostic performance and could differentiate CR-POPF with 87% sensitivity and 81% specificity among 84 patients in the training set. In the validation set, the sensitivity and specificity of the MPD index-based model for 130 PDAC samples were 93% and 87%, respectively. In patients with non-PDAC, the MPD index/body mass index (BMI) combination showed the highest prognostic performance and could differentiate CR-POPF with 84% sensitivity and 57% specificity among 96 patients in the training set. In the validation set, the sensitivity and specificity of the MPD index/BMI-based model for 236 non-PDAC samples were 85% and 53%, respectively.

Conclusion: We developed a novel prediction model for pancreatic fistulas after PD using only preoperative markers. The MPD index and MPD index/BMI combination will be useful for CR-POPF assessment in PDAC and non-PDAC samples, respectively.

Keywords: Discrete Bayes classifier; Pancreaticoduodenectomy; Postoperative pancreatic fistula; Prediction model.

MeSH terms

Bayes Theorem
Carcinoma, Pancreatic Ductal* / surgery
Humans
Pancreatic Fistula / diagnosis
Pancreatic Fistula / etiology
Pancreatic Neoplasms* / complications
Pancreatic Neoplasms* / surgery
Pancreaticoduodenectomy / adverse effects
Postoperative Complications / diagnosis
Postoperative Complications / etiology
Postoperative Complications / surgery
Retrospective Studies
Risk Factors