Development and internal validation of a nomogram for predicting in-hospital cardiac arrest among adult inpatients

Background In-hospital cardiac arrest (IHCA) is associated with substantial mortality. We aimed to develop and internally validate a nomogram to predict IHCA after the 24-hour landmark among adult inpatients. Methods This single-center retrospective case-control study included 216 IHCA cases and 649 randomly selected non-IHCA controls admitted from January 2020 to December 2024. All were alive and hospitalized at the 24-hour landmark; predictors were obtained during the first 24 h, and the outcome was subsequent IHCA before discharge. An 80:20 split yielded a training cohort ( n = 692; 173 cases) and an internal validation cohort ( n = 173; 43 cases). Missing data were handled using 20 imputed datasets, with imputation models fitted in the training cohort and applied to the validation cohort. We used least absolute shrinkage and selection operator (LASSO) logistic regression to screen 46 candidate predictors, followed by backward stepwise logistic regression. We assessed discrimination, calibration, Brier score, decision curve analysis, and bootstrap validation. We also conducted an exploratory comparison with the National Early Warning Score (NEWS), calculated from admission observations. Results The final model included hemoglobin, lactate, log-transformed N-terminal pro-B-type natriuretic peptide (NT-proBNP), fasting blood glucose, serum potassium, and creatine kinase-MB (CK-MB). The area under the receiver operating characteristic curve (AUC) was 0.801 (95% confidence interval [CI], 0.764–0.838) in training and 0.761 (95% CI, 0.679–0.843) in internal validation. Bootstrap internal validation yielded a mean optimism-corrected AUC of 0.761 and a mean optimism-corrected calibration slope of 0.950. The validation Brier score was 0.157 (95% CI, 0.123–0.192). After excluding 41 non-IHCA in-hospital deaths, the internal validation AUC was 0.782 (95% CI, 0.699–0.865). NEWS had an AUC of 0.638 (95% CI, 0.539–0.738); the difference between the model and NEWS was not statistically significant (paired DeLong test, P = 0.076). Conclusions The nomogram showed moderate discrimination. Given the case-control sampling design, absolute-risk calibration and clinical utility require cautious interpretation. External validation and prospective evaluation are needed before clinical application. Clinical trial number Not applicable.