[Construction and validation of an in-hospital mortality risk prediction model for patients receiving VA-ECMO: a retrospective multi-center case-control study].

Abstract

To investigate the risk factors of in-hospital mortality and establish a risk prediction model for patients receiving venoarterial extracorporeal membrane oxygenation (VA-ECMO). We retrospectively collected the data of 302 patients receiving VA-ECMO in ICU of 3 hospitals in Guangdong Province between January, 2015 and January, 2022 using a convenience sampling method. The patients were divided into a derivation cohort (201 cases) and a validation cohort (101 cases). Univariate and multivariate logistic regression analyses were used to analyze the risk factors for in-hospital death of these patients, based on which a risk prediction model was established in the form of a nomogram. The receiver operator characteristic (ROC) curve, calibration curve and clinical decision curve were used to evaluate the discrimination ability, calibration and clinical validity of this model. The in-hospital mortality risk prediction model was established based the risk factors including hypertension (OR=3.694, 95% CI: 1.582-8.621), continuous renal replacement therapy (OR=9.661, 95%CI: 4.103-22.745), elevated Na2 + level (OR=1.048, 95% CI: 1.003-1.095) and increased hemoglobin level (OR=0.987, 95% CI: 0.977-0.998). In the derivation cohort, the area under the ROC curve (AUC) of this model was 0.829 (95% CI: 0.770-0.889), greater than those of the 4 single factors (all AUC < 0.800), APACHE II Score (AUC=0.777, 95% CI: 0.714-0.840) and the SOFA Score (AUC=0.721, 95% CI: 0.647-0.796). The results of internal validation showed that the AUC of the model was 0.774 (95% CI: 0.679-0.869), and the goodness of fit test showed a good fitting of this model (χ2=4.629, P>0.05). The risk prediction model for in-hospital mortality of patients on VA-ECMO has good differentiation, calibration and clinical effectiveness and outperforms the commonly used disease severity scoring system, and thus can be used for assessing disease severity and prognostic risk level in critically ill patients.