Endoscopic intervention for patients of renal insufficiency with upper gastrointestinal bleeding: efficacy evaluation and predictive model construction.

OBJECTIVE

To explore disease severity thresholds beyond which endoscopic intervention fails to reduce mortality in the context of renal insufficiency and upper gastrointestinal bleeding (UGIB), and to develop predictive models for mortality.

METHODS

This study retrospectively collected data from the Medical Information Mart for Intensive Care (MIMIC-IV). Disease severity thresholds were explored using restricted cubic splines. Patients in our hospital (n = 37) were included as an external validation cohort. Logistic regression was used to identify independent factors for mortality. Predictive models were developed using Logistic regression, LASSO, XGBoost, and random forest algorithms. Models were evaluated in the hospital cohort and the National Inpatient Sample (NIS, n = 18248) database.

RESULTS

In the context of renal insufficiency and UGIB, compared with the non-endoscopic group (n = 372), patients receiving endoscopic intervention (n = 1177) had lower in-hospital mortality (12.74% vs. 21.77%, p < 0.001). However, no significant difference was observed when creatinine exceeding 354 µmol/L (p = 0.075). In the hospital cohort, no improvement on mortality, advanced respiratory or circulatory support was observed in patients with creatinine over 354 µmol/L (p = 0.335). For patients receiving endoscopic intervention, hemorrhagic stroke, disseminated intravascular coagulation, acute respiratory failure, hemodialysis, non-ST-elevation myocardial infarction, hypovolemic shock, acute kidney injury, and age emerged as independent risk factors for mortality. The Logistic predictive model was developed, achieving an area under the receiver operating characteristic curve (AUROC) of 0.851, 0.809, and 1.000 in the test set, the NIS cohort, and the hospital cohort, respectively.

CONCLUSIONS

In the context of renal insufficiency and UGIB, endoscopic intervention failed to improve mortality when creatinine > 354 µmol/L, and comprehensive evaluations should be made in such cases. Independent risk factors for mortality in patients receiving endoscopic intervention were revealed, and the Logistic model stratified risk of mortality effectively.