Prediction of Adverse Events in Stable Non-Variceal Gastrointestinal Bleeding Using Machine Learning.

Dong-Woo Seo,Hahn Yi,Chang Hwan Sohn,Won Young Kim,Youn-Jung Kim,Byuk Sung Ko,Ilsang Woo,Beomhee Park,Namkug Kim,Dae Ho Jung

doi:10.3390/jcm9082603

Abstract

Clinical risk-scoring systems are important for identifying patients with upper gastrointestinal bleeding (UGIB) who are at a high risk of hemodynamic instability. We developed an algorithm that predicts adverse events in patients with initially stable non-variceal UGIB using machine learning (ML). Using prospective observational registry, 1439 out of 3363 consecutive patients were enrolled. Primary outcomes included adverse events such as mortality, hypotension, and rebleeding within 7 days. Four machine learning algorithms, namely, logistic regression with regularization (LR), random forest classifier (RF), gradient boosting classifier (GB), and voting classifier (VC), were compared with the Glasgow–Blatchford score (GBS) and Rockall scores. The RF model showed the highest accuracies and significant improvement over conventional methods for predicting mortality (area under the curve: RF 0.917 vs. GBS 0.710), but the performance of the VC model was best in hypotension (VC 0.757 vs. GBS 0.668) and rebleeding within 7 days (VC 0.733 vs. GBS 0.694). Clinically significant variables including blood urea nitrogen, albumin, hemoglobin, platelet, prothrombin time, age, and lactate were identified by the global feature importance analysis. These results suggest that ML models will be useful early predictive tools for identifying high-risk patients with initially stable non-variceal UGIB admitted at an emergency department.

Highlights

The morbidity and mortality rates of upper gastrointestinal bleeding (UGIB) have decreased recently, this condition remains a burden to public health, with a mortality rate of 6–12% and hospital cost of more than $2.5 billion yearly in the United States [1,2]
Several scoring systems such as the Glasgow–Blatchford score (GBS) and the Rockall score [4,5] have been developed for assessing patients with UGIB; they have limitations in detecting high-risk patients with UGIB who require endoscopy, embolization, or surgical treatment and who have higher risk of developing hemodynamic instability [6,7,8,9]
Patients who were already in a hypotensive state, defined by systolic blood pressure (SBP) of

Summary

Introduction

The morbidity and mortality rates of upper gastrointestinal bleeding (UGIB) have decreased recently, this condition remains a burden to public health, with a mortality rate of 6–12% and hospital cost of more than $2.5 billion yearly in the United States [1,2]. Current guidelines recommend early risk stratification of patients with non-variceal UGIB to identify high- or low-risk patients in order to help in decision-making, including timing of endoscopy, disposition (admission vs outpatient), and level of care (general ward vs intensive care unit) [3] Several scoring systems such as the Glasgow–Blatchford score (GBS) and the Rockall score [4,5] have been developed for assessing patients with UGIB; they have limitations in detecting high-risk patients with UGIB who require endoscopy, embolization, or surgical treatment and who have higher risk of developing hemodynamic instability [6,7,8,9]. Studies of ML models in UGIB are limited by relatively small sample sizes (all but two studies had

Objectives

Results

Conclusion