Short-term mortality risk in the patients with COVID-19 complicated by sepsis based on LASSO-Logistic regression and nomogram models

Liang Menglin; Li Chen; He Xinhua

doi:10.3969/j.issn.1002-1949.2025.05.010

Abstract：Objective To construct a prediction model for the 28-day in-hospital mortality risk of the patients with the coronavirus disease 2019(COVID-19)complicated by sepsis based on the LASSO-Logistic regression and nomogram models.Methods A total of 126 patients with COVID-19 complicated by sepsis treated in Beijing Chaoyang Hospital Affiliated to Capital Medical University from May 2023 to July 2023 were selected.Based on the survival outcomes of patients 28 days after admission,they were divided into a death group(32 cases)and a survival group(94 cases).The demographic characteristics,past medical histories,laboratory test results(including blood routine,coagulation function indexes,immunological indexes,etc.)and sequential organ failure assessment(SOFA)score of the two groups were collected.Important predictive factors were screened through univariate and LASSO-Logistic regression.Subsequently,tools such as the nomogram model,calibration curve,Akaike information criterion(AIC),Bayesian information criterion(BIC),the area under curve(AUC)and Brier score were used to evaluate and validate the constructed 28-day mortality risk prediction model.Results In univariate Logistic regression variables including sex,age,hypertension,myocardial infarction,chronic heart failure,neutrophil percentage,platelet count,activated partial thromboplastin time,serum urea nitrogen,SOFA score,heart rate,respiratory rate,CD3+,CD3+％,CD4+,CD4+％,CD8+,CD8+％,CD16+/CD56+and BCD19+showed statistical significance(P＜0.1).The LASSO-Logistic model with variable screening and cross-validation identified λ=0.09511 as the optimal parameter,age,myocardial infarction,chronic heart failure,neutrophil percentage,platelet count,activated partial thromboplastin time,respiratory rate and CD3+were included as variables for model construction.The multivariate Logistic prediction model had AIC 118.478,BIC 178.040,AUC 0.913(95％CI 0.865～0.961),Brier 10.1(6.9,13.4).The LASSO-Logistic prediction model had AIC 103.237,BIC 128.764,AUC 0.878(95％CI 0.859～0.983),Brier 11.4(7.9,14.8).Conclusions The LASSO-Logistic regression model with fewer variables demonstrates better discriminative ability and higher calibration compared to the multivariate Logistic model.It effectively predicts the 28-day in-hospital mortality risk for the patients with COVID-19 complicated by sepsis.