Interpretable machine learning for procedure-specific mortality prediction after open surgical necrosectomy or video-assisted retroperitoneal debridement in infected pancreatic necrosis
Frontiers in Medicine · 20.11.2026 · 13
DOI: 10.3389/fmed.2026.1932030 ↗Kurz gefasst
Fachartikel in der Zeitschrift Frontiers in Medicine, online erschienen am 20.11.2026. Verfasst von Chengyu Li und 9 weiteren Autor:innen. Die Arbeit nennt 30 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Objective
Infected pancreatic necrosis (IPN) carries substantial mortality, and patients undergoing open surgical necrosectomy (OSN) or video-assisted retroperitoneal debridement (VARD) may exhibit different prognostic profiles. Given the limited clinical data available for surgically treated IPN, this study aimed to develop interpretable, procedure-specific mortality prediction models and characterize distinct prognostic patterns associated with each surgical strategy.
Methods
This retrospective study analyzed 143 patients from Ruijin Hospital after propensity-score matching on age and sex and subsequent exclusion of one VARD patient, including 96 patients treated with OSN during 2014–2018 and 47 treated with VARD during 2019–2024. The outcome was all-cause mortality within 90 days after the first necrosectomy. Separate CatBoost models were developed using preoperative clinical, laboratory, and imaging variables. Internal evaluation included repeated feature selection under fixed CatBoost settings and an additional 10-repetition nested cross-validation analysis with hyperparameter and feature selection confined to each outer training set. Calibration and conditional bootstrap intervals were evaluated using stored predictions from the earlier fixed-parameter analysis. SHapley Additive exPlanations (SHAP) characterized global and individual feature-prediction associations. Independent cohorts from the First Affiliated Hospital of Harbin Medical University were used for preliminary external evaluation.
Results
In the additional nested analysis, median out-of-fold AUROC was 0.810 for OSN and 0.726 for VARD, with ranges across repetitions of 0.773–0.833 and 0.670–0.788, respectively. Median Brier scores were 0.149 and 0.100. In the selected OSN model, platelet count, body mass index, and preoperative procalcitonin had prominent SHAP contributions; in the selected VARD model, preoperative procalcitonin and renal and respiratory status were prominent. These highlighted predictors were repeatedly selected in the resampling analyses, supporting further investigation of their procedure-specific prognostic patterns. Preliminary external evaluation yielded AUROCs of 0.891 in 66 OSN patients and 0.778 in 10 VARD patients. SHAP analysis further revealed distinct nonlinear, procedure-specific prognostic profiles, highlighting differences in the dominant risk patterns associated with the two surgical strategies.
Conclusion
Interpretable procedure-specific models identified heterogeneous prognostic patterns after OSN and VARD. These exploratory findings provide a basis for developing and prospectively evaluating individualized surgical management strategies in IPN.
Zitieren
APA
Li, C., Zhang, H., Wang, G., Dai, D., Pan, C., Mao, E., Wei, X., Zhao, B., Zhang, H., & Shen, M. (2026). Interpretable machine learning for procedure-specific mortality prediction after open surgical necrosectomy or video-assisted retroperitoneal debridement in infected pancreatic necrosis. Frontiers in Medicine, 13. https://doi.org/10.3389/fmed.2026.1932030
Vancouver
Li C, Zhang H, Wang G, Dai D, Pan C, Mao E, et al. Interpretable machine learning for procedure-specific mortality prediction after open surgical necrosectomy or video-assisted retroperitoneal debridement in infected pancreatic necrosis. Frontiers in Medicine. 2026;13. doi:10.3389/fmed.2026.1932030
BibTeX
@article{Li2026,
author = {Li, Chengyu and Zhang, Haibo and Wang, Gang and Dai, Dongbo and Pan, Chenhao and Mao, Enqiang and Wei, Xiao and Zhao, Bing and Zhang, Huiran and Shen, Meihua},
title = {Interpretable machine learning for procedure-specific mortality prediction after open surgical necrosectomy or video-assisted retroperitoneal debridement in infected pancreatic necrosis},
journal = {Frontiers in Medicine},
year = {2026},
volume = {13},
doi = {10.3389/fmed.2026.1932030}
}
EUVIMED