Incremental value of routine admission variables added to clinical scores for early prediction of ICU admission and systemic complications in acute pancreatitis

We assessed whether routine admission variables add predictive information to Acute Physiology and Chronic Health Evaluation II (APACHE II) for intensive care unit (ICU) admission and systemic complications in acute pancreatitis. This retrospective cohort included 264 patients. Endpoint-specific complete-case logistic models were assessed using discrimination, calibration, reclassification, and decision curves. We compared included and excluded patients and performed 1,000 bootstrap resamples with fixed predictor sets. The ICU analysis included 160 patients with 34 events; excluded patients had a higher ICU admission rate (39.4% versus 21.3%). Of 104 patients without an exact numeric C-reactive protein (CRP) value, 89 had blank records and 15 had assay-limit reports. The systemic-complication analysis included 226 patients with 89 events. Apparent areas under the receiver operating characteristic curve (AUCs) for the Combined model versus APACHE II were 0.947 versus 0.928 for ICU admission and 0.968 versus 0.962 for systemic complications; neither difference was statistically significant ( P = 0.0958 and P = 0.1536). Optimism-corrected Combined AUCs were 0.937 and 0.962, respectively. Apparent category-free net reclassification improvement was 0.877 and 0.674, and integrated discrimination improvement was 0.090 and 0.027, respectively. CRP selection stability was limited. Routine variables showed preliminary incremental information without a significant AUC improvement over APACHE II. Selection bias, limited events, and endpoint heterogeneity constrain interpretation. The exploratory Combined models require independent validation before clinical use.