Development and validation of a nomogram for predicting prolonged intensive care unit stay in patients with acute myocardial infarction and analysis of associated hospitalization costs

Background Prolonged intensive care unit (ICU) stay after acute myocardial infarction (AMI) is associated with increased complications, mortality, and resource use. However, early prediction tools for this population remain lacking. This study aimed to develop and validate a nomogram for predicting prolonged ICU stay in AMI patients and analyze cost differentials. Methods This study included 7,571 AMI patients admitted to a tertiary-hospital ICU in China (2019–2025). Prolonged ICU stay was defined as ≥4 days (75th percentile). The cohort was split 8:2 into training and test sets. From 48 candidate variables available within 24 h, Spearman correlation analysis and least absolute shrinkage and selection operator (LASSO) regression selected features. A logistic regression-based nomogram was constructed. Discrimination, calibration, decision curve analysis (DCA), and bootstrap validation (1,000 resamples) were evaluated. Costs were compared between groups and across risk quartiles. Results Of 7,571 patients, 2,145 (28.3%) experienced prolonged ICU stay. Nine predictors were retained: pH, lactate dehydrogenase (LDH), blood urea nitrogen (BUN), albumin (ALB), anterior myocardial infarction (Anterior MI), age, partial pressure of carbon dioxide (PaCO 2 ), serum chloride (Cl), and mean corpuscular hemoglobin concentration (MCHC). The area under the receiver operating characteristic curve (AUC) was 0.714 (95% confidence interval [CI] 0.700–0.728) and 0.709 (95% CI: 0.679–0.739) in the training and test sets, respectively (DeLong P > 0.05). Calibration was satisfactory (Brier score 0.178; Hosmer-Lemeshow P > 0.05 in both sets), and DCA confirmed sustained net benefit at 15%–80% threshold probabilities. Bootstrap-corrected C-index was 0.712 (optimism 0.002). At the optimal cutoff, sensitivity, specificity, positive and negative predictive values were 66.8%, 65.3%, 43.2% and 83.3% in the training set, and 64.8%, 68.3%, 44.7% and 83.1% in the test set. Discrimination was robust to the outcome definition (test-set AUC 0.715 under a 90th-percentile threshold). Conclusion The nomogram demonstrated moderate discrimination and good calibration. Its operating characteristics do not support use as a standalone tool for individual admission or discharge decisions, but it may aid group-level risk stratification and resource planning, in which predicted risk strata corresponded to graded cost increases. Multicentre external validation is a prerequisite for clinical application.