Scoping review examines ML prediction of intraoperative bleeding

A scoping review published in JMIR Medical Informatics surveys machine learning (ML) approaches for predicting intraoperative bleeding in surgical patients. Intraoperative bleeding - blood loss occurring during surgery - is a critical patient safety event: severe cases can trigger hemodynamic instability, require emergency transfusion, and significantly increase morbidity and mortality risk. Accurate prediction models could help surgical teams stratify patient risk before or during procedures, enabling earlier intervention and more targeted perioperative planning.

ML-based prediction of intraoperative bleeding spans a methodologically varied landscape. Prior work in adjacent areas - such as published scoping reviews on ML models for predicting transfusion requirements in surgery - has found substantial heterogeneity in data sources, model architectures, and outcome definitions, making cross-study comparison difficult. Reviews of this type are designed to map that landscape: identifying what models have been tested, which patient populations and surgical settings are represented, what input features are commonly used, and where validation gaps remain.

Industry pattern: predictive clinical ML faces well-documented adoption barriers including limited external validation, lack of prospective deployment studies, and integration challenges with existing perioperative workflows. A 2026 JMIR AI scoping review on AI models for preventing surgical complications found that most studied models lacked clinical readiness evidence, with few validated outside their development dataset. The intraoperative bleeding review adds consolidated evidence to a field where research is active but clinical translation remains early-stage.

For data scientists and ML researchers working in health informatics, this type of review serves as a structured literature map - useful for identifying which feature engineering patterns, model families, and validation approaches have been tried, and where methodological work is still needed. For clinical researchers, it may inform prospective study design and model selection criteria for perioperative risk assessment.