Machine learning prediction of in-hospital mortality and external validation in patients with cardiogenic shock: the RESCUE score.

Summary

Using four ML algorithms for feature selection and logistic regression for model building, the RESCUE score identified seven predictors and achieved AUC 0.86 (internal) and 0.80 (external) for in-hospital mortality in cardiogenic shock. The model generalized across etiologies and was validated in an independent 750-patient cohort.

Key Findings

• Seven predictors were selected: age, vasoactive inotropic score, LVEF, lactate, in-hospital cardiac arrest, need for CRRT, and mechanical ventilation.
• Model performance: AUC 0.86 (internal with 10-fold CV) and 0.80 (external validation in 750 patients).
• Applicable across all-cause cardiogenic shock, supporting generalizability.

Clinical Implications

RESCUE can inform MCS escalation, ICU triage, and timing of advanced therapies by quantifying mortality risk at presentation. Integration into care pathways and EHRs could standardize CS risk assessment.

Limitations

• Observational registry design with potential residual confounding and selection bias.
• Calibration and transportability to different health systems and care pathways were not fully explored.

Future Directions

Prospective, multi-regional impact studies assessing clinical decision support integration, calibration drift monitoring, and whether RESCUE-guided management improves outcomes.