External validation of ECPR prognostic models derived from pre-ECMO indicators in patients undergoing extracorporeal cardiopulmonary resuscitation.
To externally validate previously published prognostic models developed exclusively from pre-extracorporeal cardiopulmonary resuscitation (ECPR) variables in a contemporary ECPR cohort.
We conducted a bicenter retrospective external validation of four published pre-ECPR prognostic models (Lee, RESCUE-IHCA, CHIU-S1, and CHIU-S2) in adult patients treated with ECPR between January 2015 and December 2024. Model performance was evaluated for in-hospital survival and favorable neurological outcome (FNO; Cerebral Performance Category 1-2) in the overall cohort, in-hospital cardiac arrest (IHCA), and cardiac-origin cardiac arrest (Cardio_CA) subgroups. Discrimination (the area under the receiver operating characteristic curve, AUROC), calibration, overall model fit (Brier score), and decision curve analysis (DCA) were assessed. For point-based CHIU scores, validation focused on observed outcome rates across predefined risk strata.
Among 214 patients, 79.0% (169/214) had IHCA; survival to discharge was 45.8% and FNO occurred in 24.8%. Discrimination for survival was modest across models (overall ECPR AUROC 0.608-0.709; IHCA 0.586-0.672; Cardio_CA 0.591-0.689) but was higher for FNO (overall ECPR 0.709-0.764; IHCA 0.696-0.744; Cardio_CA 0.698-0.718). The Lee model showed poor calibration with slopes far below 1, whereas RESCUE-IHCA model underestimated survival but demonstrated better calibration (slopes close to 1), higher overall accuracy (lower Brier scores) and broader clinical utility (wider net-benefit ranges in DCA). CHIU models provided limited risk separation between adjacent strata.
In this external validation, pre-ECPR models showed modest performance, with better discrimination for neurological outcome than for survival. RESCUE-IHCA showed the most favorable overall performance. Future studies should develop and validate more robust, transportable tools.
We conducted a bicenter retrospective external validation of four published pre-ECPR prognostic models (Lee, RESCUE-IHCA, CHIU-S1, and CHIU-S2) in adult patients treated with ECPR between January 2015 and December 2024. Model performance was evaluated for in-hospital survival and favorable neurological outcome (FNO; Cerebral Performance Category 1-2) in the overall cohort, in-hospital cardiac arrest (IHCA), and cardiac-origin cardiac arrest (Cardio_CA) subgroups. Discrimination (the area under the receiver operating characteristic curve, AUROC), calibration, overall model fit (Brier score), and decision curve analysis (DCA) were assessed. For point-based CHIU scores, validation focused on observed outcome rates across predefined risk strata.
Among 214 patients, 79.0% (169/214) had IHCA; survival to discharge was 45.8% and FNO occurred in 24.8%. Discrimination for survival was modest across models (overall ECPR AUROC 0.608-0.709; IHCA 0.586-0.672; Cardio_CA 0.591-0.689) but was higher for FNO (overall ECPR 0.709-0.764; IHCA 0.696-0.744; Cardio_CA 0.698-0.718). The Lee model showed poor calibration with slopes far below 1, whereas RESCUE-IHCA model underestimated survival but demonstrated better calibration (slopes close to 1), higher overall accuracy (lower Brier scores) and broader clinical utility (wider net-benefit ranges in DCA). CHIU models provided limited risk separation between adjacent strata.
In this external validation, pre-ECPR models showed modest performance, with better discrimination for neurological outcome than for survival. RESCUE-IHCA showed the most favorable overall performance. Future studies should develop and validate more robust, transportable tools.
Authors
Li Li, Yan Yan, Liu Liu, Zhao Zhao, Zhao Zhao, Sun Sun, Zhao Zhao, Zhang Zhang, Liu Liu, Teng Teng, Wang Wang, Li Li, Bian Bian, Fu Fu, Ji Ji
View on Pubmed