Construction of a risk prediction model for non-variceal upper gastrointestinal bleeding and rebleeding based on multi-dimensional indicators and its application research.

OBJECTIVE

To construct a predictive model for the risk of rebleeding in non-variceal upper gastrointestinal bleeding (NVUGIB) based on multidimensional indicators to provide an assessment tool for early screening of rebleeding in NVUGIB.

METHODS

Retrospective analysis of the 3-month follow-up data of 85 patients with NVUGIB diagnosed at the Fifth Hospital of Wuhan from January 2019 to December 2021 who were discharged from the hospital after medical treatment. Patients were divided into a rebleeding group (n=45) and a non-rebleeding group (n=95) based on whether they rebleed during follow-up. The demographic characteristics, clinical characteristics and biochemical indicators of the two groups were compared. A multivariate logistic regression was used to analyze the influencing factors of NVUGIB rebleeding. A nomograph model was built using the screening results. The area under the working characteristic curve of the subject (AUC) was used to analyze the model differentiation, evaluate the model specificity and sensitivity, and verify the prediction performance of the model with the validation set.

RESULTS

There were significant differences in age, hematemesis, red blood cell count (RBC), platelet (PLT), albumin (Alb), prothrombin time (PT), TT, fibrinogen (Fib), plasma D-dimer (D-D), and blood lactate (LAC) levels between the two groups (all <0.05). Logistic regression analysis shows that, age ≥75, hematemesis more than 5 times, PLT≤100*10/L, D-D>0.5 mg/L were associated with greater risk of rebleeding. The nomogram model was constructed based on the above four indicators. The AUC of the training set (n=98) for predicting the risk of NVUGIB rebleeding was 0.887 (95% CI: 0.812-0.962), the specificity was 0.882, and the sensitivity was 0.833. The AUC of the validation set (n=42) was 0.881 (95% CI: 0.777-0.986), the specificity was 0.815, and the sensitivity was 0.867. After 500 times of sampling by bootstrap method, the mean absolute error of the calibration curve of the validation set model was 0.031, indicating that the calibration curve and the ideal curve fit well, and the predicted value of the model was in good agreement with the actual value.

CONCLUSION

Age ≥75, hematemesis >5 times, lower PLT, and higher D-D levels rise the risk of rebleeding in NVUGIB patients and have some reference value in clinical diagnosis and disease assessment.