サイトロゴ

Next-Day Major Bleeding Risk Prediction and Monitoring System for ECMO Patients

An AI system providing highly accurate, dynamic, short-term predictions of next-day major bleeding risk from daily EMR data.

Advantages

- Superior Predictive Accuracy and High NPV: Achieves a mean AUROC of 0.820.
- Dynamic, Short-Term (24-Hour) Prediction: Calculates next-day risk directly from daily EMR data, unlike static baseline scores or Point-of-Care tests.
- Explainable AI (XAI) and Patient Stratification: Incorporates SHAP values for risk factor transparency and UMAP for visualizing patient phenotype transitions.
- Determination of Intervention Timing: Supports decisions on the timing of interventions such as tracheostomy, drainage procedures, catheter replacement, patient transport, prone positioning, and early rehabilitation. For example, interventions may be postponed when the patient is classified as high risk.

Current Stage and Key Data

Model validated using a 516-case multicenter dataset; ongoing external validation and EMR real-time integration.
- Mean AUROC: 0.820 (maintaining high performance across all evaluation periods)
- Negative Predictive Value (NPV): ≥ 95% (ensuring safe de-escalation of monitoring)
- Random Forest Global Approach: High stability demonstrated across all evaluated

Partnaring Model

Technology licensing, joint development (EMR integration), and algorithm embedding into medical devices/systems. The technology may also be used for prognostic enrichment in clinical trials.
- Potential partners: ECMO/Oxygenator manufacturers, ICU CDSS/EMR vendors, pharmaceutical companies

Background and Technology

ECMO serves as a critical life-support system for severe cardiac and respiratory failure. However, bleeding complications occur in 30–50% of ECMO patients, with major bleeding being a leading cause of mortality. Conventional predictive models rely on static baseline data at initiation, resulting in lower precision, while point-of-care coagulation monitoring requires high equipment costs and frequent blood draws. This technology continuously analyzes daily EMR data to dynamically predict next-day risk.

Principal Investigator

Daisuke Kasugai (Nagoya University Hospital Emergency & Medical ICU)

Patents and Publications

- Patent pending

Project No:bk-05414