医学
肝硬化
上消化道出血
内科学
胃肠病学
肝性脑病
食管静脉曲张
胃肠道出血
肝细胞癌
逻辑回归
终末期肝病模型
门脉高压
内窥镜检查
肝移植
移植
作者
Halim Bou Daher,Mathew J. Gregoski,Don C. Rockey
标识
DOI:10.1177/10815589251357622
摘要
Upper gastrointestinal bleeding (UGIB) is associated with poor outcomes in patients with cirrhosis. Here, we developed a new-and examined previous UGIB predictor models-in cirrhotic patients with UGIB. We analyzed consecutive patients with cirrhosis and UGIB admitted to our center from 2011 to 2018. Predictors of mortality during index admissions were identified using logistic regression and existing scoring system were compared using Area Under the Receiver Operating Characteristic Curve (AUROCs). In addition, classification and regression trees (CART) analyses were conducted with v-fold cross-validation. Three hundred thirty-three patients with cirrhosis were admitted with UGIB; 294 (88%) survived and 39 (12%) expired. Those who expired were more likely to have Child-Pugh C cirrhosis (67% vs 32%, p < 0.001), hypotension, hepatic encephalopathy, and hepatocellular carcinoma. Endoscopic sources of bleeding were similar in both groups, with esophageal varices being the most common culprit lesion. Regression analysis yielded a model including systolic blood pressure, model of end-stage liver disease-sodium (MELD-Na), and alanine aminotransferase on admission as having the best mortality predictive capability (AUROC, 0.83). MELD-Na, MELD 3.0, MELD, Lyles-Rockey, AIMS65, Rockall, and Glasgow-Blatchford scores were all significantly higher in patients who expired vs survivors; of these scoring systems, MELD-Na and MELD were the best predictors of death (AUROCs = 0.81 and 0.80), respectively. In addition, CART identified MELD-Na as the strongest predictor of mortality. The MELD score appears to be an accurate predictor of mortality in patients with cirrhosis and UGIB; since the MELD score is well-established and widely used in patients with cirrhosis, we suggest that it be the primary tool utilized to predict mortality in practice.
科研通智能强力驱动
Strongly Powered by AbleSci AI