# Analysis of factors associated with 1-year rebleeding in patients with acute upper gastrointestinal bleeding and a Glasgow- Blatchford Score ≥ 6 based on serological indicators

## TL;DR

This study develops a new model to predict rebleeding in patients with upper gastrointestinal bleeding, outperforming existing scoring systems.

## Contribution

A novel risk prediction model using serological indicators and clinical factors that outperforms GBS and AIMS65 in predicting 1-year rebleeding.

## Key findings

- The new model achieved an AUC of 0.938 in training and 0.940 in validation, surpassing GBS and AIMS65 scores.
- The model includes factors like heart rate, hemoglobin, and comorbid liver disease for rebleeding prediction.
- Decision and calibration curves confirmed the model's superior predictive accuracy and clinical benefit.

## Abstract

Analysis of serological indicators for recurrent bleeding within 1 year in patients with upper gastrointestinal bleeding and a Glasgow-Blatchford Score (GBS) ≥ 6, identification of independent risk factors, and development of a risk prediction model for recurrent bleeding within 1 year.

This study enrolled 575 patients with acute upper gastrointestinal bleeding and a GBS score ≥ 6. Feature selection was performed using Lasso regression to identify statistically significant variables. The cohort was then randomly split into a training set (n = 400) and a validation set (n = 175) at a 7:3 ratio. A prediction model was developed through logistic regression analysis on the training set. Additionally, Random Forest analysis was applied, and its outcomes were visualized. The predictive performance of the newly developed model was assessed by means of a receiver operating characteristic curve, a line graph, and calibration and decision curves. Furthermore, the model was compared against the AIMS65 and GBS scoring systems to evaluate its comparative predictive value.

A new model was developed based on heart rate, blood pressure, hemoglobin, history of peptic ulcer, comorbid liver disease, albumin, platelet count, and CRP. The model achieved an AUC of 0.938 (95% CI: 0.915–0.961) in the training cohort, substantially higher than the AUCs for the GBS (0.590) and AIMS65 (0.562) scores. In the validation cohort, the model maintained an AUC of 0.940, compared to 0.612 for GBS and 0.637 for AIMS65. Superior performance was further evidenced by decision and calibration curve analyses, which indicated advantages in predictive accuracy, calibration, and net clinical benefit over the two conventional scores.

The model exhibits robust performance in the prediction of 1-year rebleeding among patients scoring GBS ≥ 6, underscoring its potential clinical applicability.