A scoping review of 190 studies examining machine learning (ML) approaches for surgical risk stratification and outcome prediction revealed significant methodological gaps. Most studies utilized single-center, private datasets with limited data modalities, and the lack of open-access surgical datasets hindered reproducibility. Key preprocessing steps like missing data handling and class imbalance were often incompletely reported. While conventional ML models predominated, deep learning and multimodal approaches were uncommon, and standardized evaluation protocols and benchmark datasets were largely absent, impeding cross-study comparisons. Furthermore, only about a third of studies incorporated explainability methods, limiting the development of clinically robust ML tools for perioperative care. AI
IMPACT Identifies critical gaps in data availability and methodology that hinder the clinical adoption of AI tools for surgical risk prediction.
RANK_REASON The cluster is based on an academic paper detailing a scoping review of machine learning approaches in a specific domain. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
