Machine Learning Versus Traditional Methods for Predicting Venous Thromboembolism After Surgery: A Systematic Review and Bayesian Network Meta‐Analysis
ANZ Journal of Surgery · 20.11.2026
DOI: 10.1111/ans.71079 ↗Kurz gefasst
Übersichtsarbeit in der Zeitschrift ANZ Journal of Surgery, online erschienen am 20.11.2026. Verfasst von Shichao Chen und 8 weiteren Autor:innen. Beteiligte Einrichtung: Goulburn Valley Health, Shepparton, Australia. Die Arbeit nennt 55 Literaturangaben.
Themen: Thromboseprophylaxe, Künstliche Intelligenz in der Pflege
Abstract (Originalsprache)
Background
Postoperative venous thromboembolism remains a major cause of preventable morbidity after surgery, and established risk scores may perform inconsistently across surgical populations.
Methods
MEDLINE, Embase and the Cochrane Library were searched for studies comparing machine learning with traditional risk models for postoperative venous thromboembolism prediction. Risk‐of‐bias was assessed with PROBAST + AI. The primary outcome was discrimination measured by area under the receiver operating characteristic curve. Bayesian random‐effects network meta‐analysis estimated mean differences versus Caprini with 95% credible intervals. Best and worst‐version analyses explored the impact of within‐study multiplicity, with subgroup analyses by cancer surgery and procedure type.
Results
Eighteen studies including 177 656 validation patients were included. Meta‐ensembles showed the largest improvement compared with Caprini in the best‐version analysis (mean difference in AUROC, 0.232; 95% credible interval, 0.129–0.334), with a similar finding in the worst‐version analysis (0.237; 0.123–0.351). Several other machine learning classes also had higher AUROC estimates than Caprini, whereas most traditional scores and surgeon judgement did not. Geneva showed lower performance than Caprini. Meta‐ensembles ranked first overall, while Caprini and Geneva ranked 14th and last, respectively. Meta‐ensembles and boosting models had higher AUROC than Caprini in both cancer surgery analyses; in general surgery, no model class showed higher performance than Caprini in the worst‐version analysis.
Conclusions
Machine learning models, particularly meta‐ensembles, showed higher reported discrimination than traditional scores, but evidence remains limited by heterogeneity, inconsistent calibration reporting, sparse decision‐analytic evaluation and infrequent external validation. Clinical adoption requires external validation and demonstration of real‐world net benefit.
Schlagwörter
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Goulburn Valley Health, Shepparton, Australia.
- Gold Coast Hospital and Health Service, Southport, Australia.
- Department of Surgery, Prometheus Research Collaborative, University of Melbourne, Austin Health, Melbourne, Australia.
- University of Melbourne, Melbourne, Australia.
- Division of Cancer Surgery, Peter MacCallum Cancer Centre, Melbourne, Australia.
- Upper Gastrointestinal Surgery Unit, Division of Surgery, Anaesthesia and Procedural Medicine, Austin Health, Heidelberg, Australia.
ORCID-Profile: Shichao Chen, Saavan Dhaliwal, Chris Papas, Ian Hughes, Marcos V Perini, David S Liu, Justin M Yeung
Zitieren
APA
Chen, S., Dhaliwal, S., Papas, C., Hughes, I., Goh, S. K., Perini, M. V., Liu, D. S., Yeung, J. M., & Muralidharan, V. (2026). Machine Learning Versus Traditional Methods for Predicting Venous Thromboembolism After Surgery: A Systematic Review and Bayesian Network Meta‐Analysis. ANZ Journal of Surgery. https://doi.org/10.1111/ans.71079
Vancouver
Chen S, Dhaliwal S, Papas C, Hughes I, Goh SK, Perini MV, et al. Machine Learning Versus Traditional Methods for Predicting Venous Thromboembolism After Surgery: A Systematic Review and Bayesian Network Meta‐Analysis. ANZ Journal of Surgery. 2026. doi:10.1111/ans.71079
BibTeX
@article{Chen2026,
author = {Chen, Shichao and Dhaliwal, Saavan and Papas, Chris and Hughes, Ian and Goh, Su Kah and Perini, Marcos V and Liu, David S and Yeung, Justin M and Muralidharan, Vijayaragavan},
title = {Machine Learning Versus Traditional Methods for Predicting Venous Thromboembolism After Surgery: A Systematic Review and Bayesian Network Meta‐Analysis},
journal = {ANZ Journal of Surgery},
year = {2026},
doi = {10.1111/ans.71079}
}
EUVIMED