Development and validation of a prediction model for in-hospital mortality in patients with sepsis

Abstract

Abstract Objective The aim of this study is to develop and validate a multivariate prediction model for mortality risks at 28, 42, and 56 days in patients with sepsis in the intensive care units (ICUs) by utilizing locally sourced datasets, eschewing reliance on open-source clinical databases in developing nations. Methods A retrospective cohort study was conducted on 2389 sepsis patients admitted to ICUs across two campuses of a tertiary hospital from January 1, 2020, to June 30, 2022. An independently developed clinical decision support system captured electronic data. Enrolled patients were randomly divided into a training set (n = 1673) and a validation set (n = 716) in a 7:3 ratio. Variables identified through Least Absolute Shrinkage and Selection Operator (LASSO) regression analysis were integrated into a multivariate Cox proportional hazards regression model to construct a nomogram. Model accuracy was assessed using the area under the receiver operating characteristic curve (AUROC). Nomogram performance was evaluated for discrimination, calibration, and clinical utility in both sets. Results The risk score was developed based on 9 independent predictive factors from an original pool of 32 potential predictors. Notably, the prognostic nomogram revealed the minimum APACHE II score's paramount influence on prognosis, followed by days of mechanical ventilation, number of vasopressors, maximum and minimum SOFA scores, infection sources, gram-positive or gram-negative bacteria, and malignancy. A publicly accessible online calculator implementing this nomogram is available at (https://tingyutongji.shinyapps.io/Nomogram/). The nomogram demonstrated superior discriminative ability, with AUROC values of 0.882 (95%CI, 0.855–0.909) and 0.851 (95%CI, 0.804–0.899) at 4 weeks; 0.836 (95%CI, 0.798–0.874) and 0.820 (95%CI, 0.761–0.878) at 6 weeks; and finally, at week 8, it achieved AUROC values of 0.843 (95%CI, 0.800-0.887) and 0.794 (95%CI, 0.720–0.867) in both training and validation sets. Furthermore, both sets exhibited strong discrimination and calibration, supported by C-indexes of 0.872 and 0.839, respectively, confirmed through decision curve analysis, highlighting the significant net clinical benefit provided by the developed nomogram. Conclusion A risk assessment model and web-based calculator have been devised to predict in-hospital mortality among ICU sepsis patients. Targeting factors identified as relevant in the model could potentially enhance survival rates for critically ill patients during their hospital stay.