Machine Learning

Intraoperative Hypotension Predictor

Evidence-based ML model that predicts IOH risk using 14 clinical features. Trained on 6,388 real surgical cases from the VitalDB open dataset.

Educational Tool Only: This ML model is designed for learning about IOH risk prediction. It is NOT intended for clinical decision-making or actual patient care. Always consult evidence-based guidelines and clinical judgment for real patient management.

Patient & Hemodynamic Parameters

Enter current intraoperative data to estimate IOH risk using our RandomForest classifier.

Patient Information

Age*

Sex*

Weight (kg)*

Height (cm)*

ASA Class*

Hemodynamic Parameters

Baseline MAP (mmHg)*

Baseline HR (bpm)*

Current MAP (mmHg)*

MAP 5 min ago*

MAP 10 min ago*

Surgical & Anesthetic Factors

Surgery Type*

Surgery Duration (min)*

Induction Agent*

Vasopressor in Use*

Emergency Surgery*

How the ML Model Works

Our RandomForest classifier provides transparent, evidence-based IOH risk predictions. Here's how it works:

Training Dataset

Trained on 6,388 real surgical cases from the VitalDB open dataset (Seoul National University Hospital, 2011-2020). Dataset includes high-resolution intraoperative vital signs across diverse surgical procedures.

14 Clinical Features

Model analyzes demographics (age, sex, BMI, ASA), hemodynamics (MAP trends, HR), and surgical factors (type, duration, induction agent, emergency status).

RandomForest Algorithm

Uses 300 decision trees with balanced class weights and feature scaling. ROC-AUC: 0.87, Test accuracy: 89%. Most important features: current MAP (18%), MAP 5-min trend (15%), MAP 10-min trend (12%).

Evidence-Based

Training data from VitalDB open dataset (Lee et al. Nature Sci Data 2022). Performance comparable to published ML models: Springer 2024 (ROC-AUC: 0.92), eClinicalMedicine 2024 (ROC-AUC: 0.93).

Interpretable Outputs

Provides 3 time-windowed predictions (5/10/20 min) with decay modeling. Identifies top risk factors and suggests evidence-based interventions for clinical context.

Clinical Validation

Trained and validated on real surgical data from 6,388 cases. Test set performance: ROC-AUC 0.87. Model can be retrained as new evidence emerges. For educational purposes only.