Early risk stratification in contemporary acute myocardial infarction–associated cardiogenic shock: development and temporal validation of an admission-based mortality model
K Mihali, B Markus, L Harbaum, B Schieffer, J KreutzAbstract
Background
Acute myocardial infarction complicated by cardiogenic shock (AMI-CS) remains associated with high in-hospital mortality despite contemporary revascularization and increasing use of temporary mechanical circulatory support (tMCS). Established scores (CardShock, IABP-SHOCK II, GRACE, ENCOURAGE) were derived in heterogeneous or pre–device-era populations and may not reflect contemporary AMI-CS cohorts.
Purpose
To develop and temporally validate an admission-based mortality model and compare its performance with established scores.
Methods
In this single-center study at the Department of Cardiology, 479 consecutive AMI-CS patients were analyzed. The primary endpoint was all-cause in-hospital mortality. Admission variables associated with mortality were entered into multivariable logistic regression to derive a parsimonious prediction model. Temporal validation used a derivation cohort (2017–2022; 2/3) and a validation cohort (2023–2025; 1/3). Discrimination was assessed using AUC with 95% CI; calibration was assessed using slope and calibration-in-the-large; and comparative performance was assessed using DeLong testing. Subgroup analyses were performed in the no-tMCS, Impella, VA-ECMO, and ECMELLA cohorts.
Results
In-hospital mortality was 42.6% (204/479). The final admission-based model included age, left ventricular ejection fraction, prior cardiopulmonary resuscitation, glomerular filtration rate, vasoactive-inotropic score, and ST-segment elevation on admission ECG. In the derivation cohort (n=338), discrimination was good (AUC 0.81; 95% CI 0.76–0.86). Performance remained stable in the validation cohort (n=141) (AUC 0.81; slope 0.93; calibration-in-the-large −0.04). In the overall cohort, the model achieved an AUC of 0.81 (95% CI 0.769–0.851), exceeding CardShock (AUC 0.74; ΔAUC 0.08; p=0.0015), IABP-SHOCK II (AUC 0.70; ΔAUC 0.13; p<0.001), and GRACE (AUC 0.69; ΔAUC 0.14; p<0.001). Established scores showed heterogeneous and attenuated discrimination, particularly in VA-ECMO– and ECMELLA-treated patients. Across treatment strata, the admission-based model maintained consistent accuracy (AUC range 0.79–0.85).
Conclusions
In a contemporary AMI-CS cohort, an admission-based model demonstrated stable temporal validation and consistent discrimination across mechanical support strategies. Established scores showed variable performance, particularly in device-escalated patients. These findings highlight the need for next-generation risk models that incorporate early integrative predictors and explicitly account for the growing heterogeneity of AMI-CS in the era of device-supported care.Figure 1 Abstract