Background: The field of machine learning (ML) shows great potential to enhance the process of predicting treatment results after patients undergo vascular and endovascular medical procedures. The evaluation process for new models needs to follow standardized procedures which must be rigid to establish their readiness for operational deployment.
Methods: The research used PRISMA 2020 to conduct a systematic review which combined clinical ML prediction models that forecasted surgical results. The authors used TRIPOD-AI and PROBAST-AI frameworks to assess both the reporting standards and research design of the study.
Results: The results from 50 studies showed that ML models achieved excellent discrimination ability with an average AUC value of 0.86. The research revealed two major weaknesses in the studies because calibration plots appeared in less than 36% of the studies and external validation occurred in only 40% of the research. The research used decision curve analysis as its main clinical utility assessment tool although this method appeared in less than 10% of all studies.
Conclusions: The current ML models achieve good predictive results, but their biased operation combined with insufficient disclosure of validation procedures and medical effectiveness assessment makes them inappropriate for medical use. The future development process needs to focus on three essential elements which include external validation testing and complete calibration evaluation and full compliance with TRIPOD-AI/PROBAST-AI reporting standards.
Keywords: Machine learning; vascular surgery; endovascular procedures; predictive modelling; TRIPOD-AI; PROBAST.
How to cite this article: Osman AAF, Machine learning in Vascular and Endovascular surgery: a systematic review and critical appraisal. Int J Drug Deliv Technol. 2026;16(1s): 215-230; DOI: 10.25258/ijddt.16.1.23