Early Proteolytic Biomarkers in Acute Pancreatitis: Associations with Clinical, Biochemical, and Imaging Severity in a Retrospective Observational Study

Background: Early identification of patients at increased risk of severe acute pancreatitis (AP) remains challenging because clinically relevant disease heterogeneity may not be fully apparent during the initial evaluation. Premature activation of pancreatic digestive zymogens within acinar cells is a central event in the pathophysiology of AP, and biomarkers directly related to this proteolytic process may provide information that is biologically distinct from nonspecific inflammatory markers. This study investigated the prognostic relevance of early serum and urinary trypsinogen-2 and trypsin activation peptide (TAP) in patients with AP, using complementary statistical, Bayesian, and machine-learning approaches. Methods: This retrospective observational cohort included 73 adults with AP who underwent serum and urinary proteolytic biomarker assessment within the first 24 h after admission. Imaging severity was evaluated using the modified Computed Tomography Severity Index (mCTSI) and was available for 66 patients. Associations between biomarkers and clinical or imaging variables were assessed using Spearman correlation and Bayesian inference. Receiver operating characteristic (ROC) analysis evaluated urinary trypsinogen-2 for mCTSI ≥ 4 and mCTSI ≥ 8. A multiple mediation model assessed indirect pathways linking C-reactive protein (CRP) with mCTSI. Unsupervised K-means clustering was used to identify proteolytic profiles, while Elastic Net and Random Forest models were developed to predict higher imaging severity (mCTSI ≥ 6) and log-transformed CRP. No imputation was performed; each analysis used the available valid observations. Results: Urinary trypsinogen-2 concentrations increased significantly across mild, moderate, and severe imaging categories (medians 5149.8, 11,952.8, and 27,008.6 pg/mL, respectively; Kruskal–Wallis p = 0.0073). For mCTSI ≥ 8, urinary trypsinogen-2 yielded an AUC of 0.791 (95% CI, 0.638–0.944), with 77.8% sensitivity and 75.4% specificity at a threshold > 14,707 pg/mL, and a negative predictive value of 95.6%. For mCTSI ≥ 4, the AUC was 0.683 (95% CI, 0.550–0.817). Urinary trypsinogen-2 correlated positively with mCTSI (rho = 0.412, p < 0.001), serum trypsinogen-2 (rho = 0.512, p < 0.001), and urinary TAP (rho = 0.447, p < 0.001), but not with CRP, amylase, or lipase. Bayesian analysis provided strong evidence for the association between urinary trypsinogen-2 and mCTSI (BF10 = 42.37). In the mediation model, the indirect pathway through urinary trypsinogen-2 was statistically significant (β = 0.14, p = 0.009). K-means clustering identified three proteolytic profiles that differed significantly in age, CRP, and BISAP score (p = 0.003, p < 0.001, and p < 0.001, respectively). For mCTSI ≥ 6, Random Forest achieved an AUC of 0.86 and Elastic Net an AUC of 0.81, with urinary trypsinogen-2 the most important predictor in both models. In CRP regression, R2 was 0.53 for Random Forest and 0.41 for Elastic Net. Conclusions: Early urinary trypsinogen-2 was consistently associated with imaging severity and represented the strongest predictor across multiple analytical frameworks, including frequentist correlation, Bayesian inference, mediation analysis, clustering, and machine learning. These findings support its potential role as a component of multimodal early risk stratification in AP, providing information complementary to conventional clinical and biochemical assessment. However, the identified ROC thresholds and machine-learning models remain exploratory and require prospective, multicenter external validation before clinical implementation.