Cumulative stress hyperglycemia ratio exposure and dynamic trajectories reveal prognostic determinants of acute hyperlipidemic pancreatitis: an 5-year cohort study

The stress hyperglycemia ratio (SHR) and glycemic variability (GV) both reflect acute glycemic fluctuations, with established roles in cardiovascular and stroke diseases. However, their prognostic value in acute hyperlipidemic pancreatitis (HTG-AP) remains underexplored. Traditional assessments based on single measurements fail to capture the dynamic evolution. Therefore, this study aims to evaluate the early predictive value of Cumulative stress hyperglycemia ratio (CumSHR) and trajectory combined with GV in patients with hyperlipidemic pancreatitis. This study collected data from 959 patients with HTG-AP. SHR and GV were calculated using standardized formulas; CumSHR was derived from the area under the curve (AUC) based on SHR data from the 7 days prior to admission. Multivariable analyses and RCS analysis assessed whether CumSHR predicted severe hyperlipidemic acute pancreatitis (HTG-SAP). Subgroup stratification was performed based on clinically distinct glucose metabolic states, while Latent class growth mixture model (LCGMM) identified dynamic SHR trajectory sub-phenotypes. Kaplan–Meier (K-M) curves compared survival rates across different risk trajectory groups. Finally, machine learning models predicted HTG-SAP risk, and SHapley Additive exPlanations (SHAP) identified key predictors. In the Jiangxi cohort, 162 cases (16.9%) of HTG-AP developed HTG-SAP. RCS analysis demonstrated a U-shaped association between CumSHR and HTG-SAP and persistent organ failure (POF) (P < 0.001). The LCGMM identified three dynamic trajectories: the sustained high-value group (SHG-T3) exhibited the highest risk of HTG-SAP (55.6%, compared with 10.6% in the low-value gradually decreasing group LDG-T1). Subsequently, stratified into normal glucose regulation (NGR), pre-diabetes Mellitus (Pre-DM), and diabetes mellitus (DM), high CumSHR + high GV showed increased risk compared to low CumSHR + low LGV (P < 0.05). Among machine learning models predicting HTG-SAP risk, the Naive-Bayes model demonstrated the highest predictive accuracy. SHAP analysis identified CumSHR as one of the most important predictors. This study demonstrates that CumSHR exhibits significant association with early assessment of severe conditions in patients with HTG-AP, with its trajectory effectively capturing dynamic changes to compensate for the limitations of static prediction.