Stress-Induced Hyperglycemia as an Independent Predictor of Infectious Pancreatic Necrosis in Acute Pancreatitis: A Machine Learning-Driven Prognostic Model

Abstract

Objective: To investigate the impact of stress-induced hyperglycemia (SHG) at admission on clinical outcomes in acute pancreatitis (AP) by collecting and analyzing relevant clinical data. Methods: This study enrolled AP patients diagnosed at Shanxi Bethune Hospital from January 1, 2017, to December 31, 2022. Clinical data and 24-h laboratory indicators were retrospectively collected. We employed propensity score matching (PSM) to compare the impact of SHG on AP clinical outcomes before and after matching. A temporal split allocated patients into training/validation cohorts for developing and validating a clinical prediction model for infected pancreatic necrosis (IPN). Results: This study included 1343 acute pancreatitis patients, with 348 having SHG at admission. Before PSM, SHG patients showed significantly longer hospital stays (13.8 vs 12.28 days, p<0.001), higher ICU admission rates (6% vs 2%, p<0.001), and increased infected pancreatic necrosis (IPN) (15% vs.6%). After using PSM to control for confounding factors, SHG patients maintained longer hospitalizations (13.61 vs 12.28 days, p=0.004), higher ICU admissions (6% vs 2%, p=0.005), and IPN rates (15% vs 6%, p<0.001). These results confirm SHG as an independent poor prognostic factor for AP rather than a reflection of baseline differences. In the training cohort, seven independent IPN predictors were identified: hyperlipidemia, SHG, modified CT severity index (MCTSI), systemic inflammatory response syndrome (SIRS), Prothrombin Time Activity (PT%), LDL-C, and peripancreatic effusion. The clinical prediction model demonstrated good performance in the validation cohort, with an area under the receiver operating characteristic curve (AUC) of 0.891. Conclusion: PSM confirmed that SHG adversely impacts clinical outcomes in acute pancreatitis. The prediction model incorporating seven variables—hyperlipidemia, SHG, MCTSI, SIRS, PT%, LDL-C, and peripancreatic effusion—demonstrated favorable predictive performance and clinical utility for infected pancreatic necrosis (IPN) in acute pancreatitis patients. Meanwhile, we developed a web-based calculator to enhance its clinical utility.