# A predictive model for upper gastrointestinal bleeding in patients with acute myocardial infarction complicated by cardiogenic shock during hospitalization

## TL;DR

This study identifies risk factors for upper gastrointestinal bleeding in patients with heart attacks and shock, and creates a model to predict it.

## Contribution

A novel predictive model for UGIB in AMICS patients based on baseline risk factors is developed and validated.

## Key findings

- UGIB occurred in 22.9% of patients with AMICS undergoing emergency PCI.
- The predictive model achieved an AUC of 0.768 with high sensitivity and moderate specificity.
- Bleeding group patients had higher mortality and longer hospital stays.

## Abstract

To explore the current status and characteristics of upper gastrointestinal bleeding (UGIB) in patients with acute myocardial infarction complicated by cardiogenic shock (AMICS) following emergency percutaneous coronary intervention (PCI), and to develop and validate a predictive model based on baseline risk factors at the time of admission.

We selected patients diagnosed with AMICS who underwent emergency PCI. Patients were categorized into the non-bleeding group and the bleeding group based on the occurrence of UGIB during hospitalization. Logistic regression analysis was employed to construct a predictive model for UGIB based on baseline risk factors at admission.

A total of 253 patients were included in the study, of whom 58 experienced UGIB, resulting in an incidence rate of 22.9%. Univariate analysis indicated that the levels of uric acid, lactate, and alanine aminotransferase (ALT) were higher in the bleeding group compared to the non-bleeding group. Conversely, the estimated glomerular filtration rate (eGFR), left ventricular ejection fraction (LVEF), and albumin were lower in the bleeding group. Additionally, the bleeding group had a higher stage of American Society for Cardiovascular Angiography and Interventions-Cardiogenic Shock(SCAI-CS). The least absolute shrinkage and selection operator (LASSO) regression identified 5 non-zero coefficient variables, and the variance inflation factor (VIF) test excluded the collinearity relationships among these 5 variables. Continuous variables were converted into categorical variables and assigned as follows: albumin was categorized based on whether below 35 g/L, indicating hypoproteinemia; eGFR was categorized based on whether below 45 ml/(min·1.73 m2), indicating moderate to severe renal function decline;The receiver operating characteristic curve (ROC) curve was employed to identify the node with the highest Youden index, which determines the cut-off value: For lactate, the corresponding value is 6.95 mmol/L, rounded to 7 mmol/L. For LVEF, the corresponding value is 35.5%, rounded to 36%.Multivariate logistic regression analysis identified 3 risk factors for UGIB following AMICS emergency PCI: baseline SCAI-CS stage D + E, baseline eGFR < 45 ml/(min·1.73 m2), and baseline LVEF < 36% (P < 0.05). The predictive model based on multivariate logistic regression results, demonstrated good fit according to the Hosmer-Lemeshow test (χ² = 6.968, P = 0.324). The model achieved an AUC of 0.768 [95% CI (0.700, 0.837)] in predicting UGIB in patients undergoing emergency PCI for AMICS, with a sensitivity of 87.9% [95% CI (0.700, 0.837)] and a specificity of 52.3% [95% CI (27.8%, 61.2%)]. The DeLong test indicated that the AUC of the predictive model was superior to that of any individual indicator (P < 0.05), and the DCA curve confirmed the model's clinical utility. Recent Prognosis: The mortality rate during hospitalization in the bleeding group was significantly higher than that in the non-bleeding group (63.8% vs. 29.7%, P < 0.05). Additionally, the duration of hospital stays for surviving patients in the bleeding group was significantly longer than that in the non-bleeding group [19 (14,37) vs. 12 (9,16), (P < 0.001)].

High baseline SCAI-CS stage, poor kidney function, and low LVEF are independent risk factors for UGIB during hospitalization. The constructed predictive model demonstrates high predictive efficacy.