To establish and validate a nomogram model for predicting in-hospital mortality in patients with sepsis-induced cardiomyopathy (SICM).

Based on the open-source database of medical information mart for intensive care-Ⅲ (MIMIC-Ⅲ), 848 patients meeting the diagnostic criteria of SICM were included. All patients were randomly divided into a training set (605 cases) and a validation set (273 cases) according to a ratio of 7 ∶ 3. The data of basic information, vital signs, comorbidities, laboratory tests, treatment measures and sequential organ failure assessment (SOFA) score were extracted. Predictive variables were selected by least absolute shrinkage and selection operator (LASSO) regression in the training set and a predictive model for in-hospital mortality of SICM patients was derived using multivariate logistic regression. Receiver operating characteristic (ROC) curves were used to evaluate the predictive ability of the nomogram model, calibration curves to evaluate its accuracy, and decision curve analysis (DCA) to evaluate its clinical practicability.

Multivariate logistic regression was conducted according to the variables screened by LASSO regression, which showed that the age [odds ratio (OR) = 1.024, 95% confidence interval (CI) (1.004, 1.046), P = 0.023], red blood cell distribution width (RDW) [OR = 1.164, 95%CI (1.001, 1.349), P = 0.046], respiratory rate (RR) [OR = 1.048, 95%CI (1.007, 1.092), P = 0.023], continuous renal replacement therapy (CRRT) [OR = 4.472, 95%CI (1.213, 17.612), P = 0.026] and SOFA score [OR = 1.147, 95%CI (1.043, 1.262), P = 0.005] were independent risk factors affecting the in-hospital mortality in SICM patients. ROC curves showed that the area under the curve (AUC) of the nomogram was 0.745 [95%CI (0.684, 0.806), P < 0.001] in the training set and 0.739 [95%CI (0.654, 0.824), P < 0.001] in the validation set, revealing that the model had moderate discriminative ability in both training and validation sets. The calibration curves of the nomogram basically fitted the ideal curve in the training and validation sets, which indicated that the actual value and predicted value of the model matched well. DCA showed that the threshold ranged from 0.1 to 0.9 in the validation set and from 0.1 to 0.8 in the training set, so the nomogram model had a high net benefit.

This nomogram prediction model based on LASSO regression and multivariate logistic regression has moderate discriminative and calibrating ability to predict in-hospital mortality of patients with SICM, and its clinical application can improve the prognosis.