Association of red blood cell distribution width combined with alanine aminotransferase with gastrointestinal hemorrhage in patients with acute pancreatitis: a retrospective MIMIC-IV study

Background Readmission is common in acute pancreatitis (AP), with a rate of about 20%. Acute kidney injury (AKI), shock, and gastrointestinal (GI) hemorrhage are serious complications related to adverse outcomes in AP. This study aimed to identify the complication most strongly associated with readmission in AP for further risk factor analysis. Methods Data were downloaded from the MIMIC-IV database. Chi-square test and the Phi coefficient were used to explore which complication was most strongly associated with readmission in AP. Then, patients were divided into with/without GI hemorrhage groups to observe the baseline characteristics across the two groups. LASSO regression and exploratory machine-learning algorithms involving XGBoost, random forest, and K-nearest neighbor were employed for feature selection. Moreover, restricted cubic spline (RCS) and logistic regression analyses were used to evaluate the associations of the variables with GI hemorrhage. Results There were 2,004 participants with AP enrolled with 445 readmission records. Among the three complications examined, GI hemorrhage showed the strongest association with readmission. Twelve parameters related to GI hemorrhage were enrolled for feature selection, with three consistently prioritized variables: age, alanine aminotransferase (ALT), and red blood cell distribution width (RDW). GI hemorrhage had a negative relationship with ALT but was positively linked to RDW (all P 0.05). When combining RDW and ALT, the multivariable regression results showed that AP patients in the low-ALT + high-RDW group exhibited the highest odds ratio (OR) of GI hemorrhage (OR = 2.315). Conclusion This study highlights that GI hemorrhage showed the strongest association with readmission among the complications examined in AP. Moreover, RDW and ALT were independently associated with GI hemorrhage risk. The low-ALT + high-RDW pattern may help identify patients at higher risk, but these findings are exploratory and require external validation before clinical application.