Clinical subtyping of severe acute pancreatitis reveals heterogeneous associations with early management strategies: a multicenter retrospective cohort study

Background: The clinical prognosis of severe acute pancreatitis (SAP) varies significantly, and traditional scoring systems fail to reveal its intrinsic heterogeneity. Methods: This multicenter retrospective cohort study enrolled SAP patients from a Chinese Jiangxi cohort (n = 3,022) and the US MIMIC-IV cohort (n = 840). Consensus K-means clustering was applied to the derivation cohort to identify clinical subphenotypes, which were externally validated in the independent cohort. An XGBoost classifier was constructed based on selected variables and its performance was evaluated. Multivariable Cox regression models were used to analyze the association of each subphenotype with mortality and to assess heterogeneous associations with three key early interventions: early fluid intake, low molecular weight heparin (LMWH) anticoagulation, and early enteral nutrition. Results: Cluster analysis consistently identified three stable and reproducible SAP clinical subphenotypes: K1 (Stable type, n = 1,446), characterized by younger age, stable hemodynamics and renal function, and good nutritional status; K2 (Renal impairment-dominant type, n = 317), defined by severe renal insufficiency, hypoalbuminemia, and lower mean arterial pressure; and K3 (Elderly/Hepatic impairment type, n = 1,259), featuring the oldest age and significantly elevated liver injury markers. The K2 subphenotype had the highest in-hospital mortality (44.8%), significantly greater than that of K1 (8.3%) and K3 (14.9%) (p < 0.001). Multivariable Cox regression showed that, using K1 as reference, the K2 subphenotype was independently associated with an increased risk of 28-day mortality (adjusted hazard ratio [aHR] = 6.70, 95% CI 4.94-9.08). The constructed XGBoost subphenotype classifier demonstrated excellent discriminative performance in the training, test, and external validation sets (macro-average AUCs: 0.997, 0.994, and 0.958, respectively). Analysis of heterogeneous associations revealed that fluid overload (total intake ≥ 2.5 L within the first days) was significantly associated with increased mortality specifically in the K3 subphenotype. LMWH therapy showed a clear survival benefit only in the K2 subphenotype (aHR = 0.66, 95% CI 0.45-0.96, p = 0.031). Conclusion: This study identified three SAP subphenotypes with significantly heterogeneous clinical features and prognoses based on early routine indicators and developed a high‑performance classifier for bedside use. The distinct associations with fluid administration and anticoagulation across subphenotypes suggest that phenotype‑based strategies may optimize the clinical management of SAP, although these findings are hypothesis‑generating and require prospective validation.