Zhonghua Yi Xue Za Zhi. 2026 Jul 21;106(26):2730-2736. doi: 10.3760/cma.j.cn112137-20260109-00102.
ABSTRACT
Objective: To develop a predictive model for discharge survival in patients with cardiac arrest treated with extracorporeal cardiopulmonary resuscitation (ECPR). Methods: The clinical data of patients with cardiac arrest who received ECPR at Beijing Anzhen Hospital, Capital Medical University from January 2017 to December 2024 were retrospectively analyzed. Three models were constructed using predefined mandatory variables and candidate variables. Model 1 only included pre-specified mandatory variables, Model 2 was a simplified clinical model constructed via variable selection using the least absolute shrinkage and selection operator (LASSO) regression, and Model 3 included both pre-specified mandatory variables and candidate variables with significant contribution, screened by LASSO regression. A multivariable logistic regression model was employed to identify predictors of in-hospital survival, and a prognostic nomogram was developed. Model efficacy was assessed by the area under the receiver operating characteristic curve (AUC), calibration plot, decision curve analysis, optimism-corrected bias, and Brier score. Results: A total of 149 patients were included, with a hospital survival rate of 31.5% (47/149). Multivariable logistic regression analysis revealed that pre-extracorporeal membrane oxygenation (ECMO) lactate was an independent risk factor for in-hospital survival across all three models (all P<0.05). In Model 2, 24-hour lactate clearance was a protective factor for in-hospital survival (OR=2.676, 95%CI: 1.515-6.264, P=0.007), whereas renal replacement therapy was a risk factor (OR=0.277, 95%CI: 0.107-0.664, P=0.006). In Model 3, both 24-hour lactate clearance (OR=2.691, 95%CI:1.426-6.991, P=0.015) and renal replacement therapy (OR=0.308, 95%CI:0.114-0.775, P=0.015) showed statistically significant differences.The AUC of Model 1, Model 2, and Model 3 were 0.716 (95%CI: 0.630-0.802), 0.821 (95%CI: 0.751-0.891), and 0.827 (95%CI: 0.757-0.898), respectively. The DeLong test demonstrated that the discriminative abilities of both Model 2 and Model 3 were superior to that of Model 1 (both P<0.05), whereas no statistically significant difference was observed between Model 2 and Model 3 (P=0.453). The optimism values of the three models were 0.028, 0.020, and 0.047, respectively, and the Brier scores were 0.189, 0.156, and 0.153, respectively, indicating a lower risk of overfitting in Model 2. Calibration curves and decision curve analysis demonstrated favorable consistency and net benefit of Model 2. Conclusion: Current model which integrates pre-ECMO lactate, 24-hour lactate clearance rate, renal replacement therapy, age, achieves favorable predictive performance and stability, and can be applied to predict hospital discharge survival in cardiac arrest patients undergoing ECPR.
PMID:42477943 | DOI:10.3760/cma.j.cn112137-20260109-00102