Machine Learning Outperforms AIMS65 and Glasgow-Blatchford Score for Mortality Prediction in GI Bleeding: Adv GastroHep | August 2026

Gastrointestinal bleeding (GIB) is a common medical emergency associated with significant morbidity and mortality. Early identification of high-risk patients is critical for decisions regarding hospitalization, intensive care, urgent endoscopy, and resource allocation. Although the AIMS65 and Glasgow-Blatchford Score (GBS) are widely used, their ability to capture complex clinical interactions is limited. This study developed and externally validated a clustering-based machine learning (ML) model to improve prediction of 30-day mortality in patients presenting with GIB.

Why was this study needed?

Gastrointestinal bleeding requires rapid and accurate risk stratification.

Traditional scores such as AIMS65 and GBS have only moderate predictive accuracy.

Conventional scoring systems cannot adequately model nonlinear interactions between multiple clinical variables.

Machine learning offers the potential for more individualized and precise risk prediction.

Externally validated ML models are needed before clinical implementation.

Results:

The ML model was developed using 5,453 patients and externally validated in 7,166 patients, demonstrating excellent reproducibility.

The model significantly outperformed both AIMS65 and the Glasgow-Blatchford Score in predicting 30-day mortality.

It achieved high sensitivity while maintaining substantially better specificity, allowing more accurate identification of high-risk patients without unnecessarily overclassifying low-risk individuals.

A clustering approach identified 24 distinct clinical phenotypes, with mortality risk varying widely across different patient groups, supporting personalized risk assessment.

The most influential predictors of mortality included age, serum albumin, hemodynamic status, hemoglobin level, and platelet count.

The model demonstrated strong potential for integration into electronic health record (EHR) systems, enabling automated real-time risk prediction at the bedside.

Clinical Impact:

This study highlights how machine learning can substantially improve risk stratification in gastrointestinal bleeding compared with conventional clinical scoring systems. Automated ML models embedded within hospital EHRs could help clinicians rapidly identify patients requiring intensive monitoring or urgent intervention while safely recognizing those suitable for less intensive management. Such precision-based risk assessment has the potential to improve patient outcomes while optimizing healthcare resource utilization.

Bottom Line:

A clustering-based machine learning model predicted 30-day mortality in gastrointestinal bleeding more accurately than both AIMS65 and the Glasgow-Blatchford Score. By identifying distinct clinical risk phenotypes and providing superior prognostic performance, this AI-driven approach represents a promising next-generation tool for personalized risk stratification in patients with gastrointestinal bleeding.