Machine Learning Multicenter Risk Model to Predict Right Ventricular Failure After Mechanical Circulatory Support:...

Area of Science:

• Cardiology
• Medical Devices
• Heart Failure Research

Background:

• Existing models for predicting right ventricular failure (RVF) after left ventricular assist device (LVAD) implantation have limitations, including lack of external validation and marginal predictive power.
• Intraoperative characteristics have not been adequately incorporated into current RVF prediction models.

Purpose of the Study:

• To derive and validate a robust risk prediction model for RVF following LVAD implantation.
• To develop a personalized risk assessment tool for LVAD candidates to quantify their risk of RVF.

Main Methods:

• A hybrid prospective-retrospective multicenter cohort study involving derivation (5 institutions) and external validation (1 institution) cohorts.
• Utilized bootstrap imputation and adaptive least absolute shrinkage and selection operator (LASSO) for variable selection to create the STOP-RVF predictive model.
• Defined RVF as the need for RV assist device or inotropes for >14 days; compared STOP-RVF performance against existing scores.

Main Results:

• The STOP-RVF calculator achieved good predictive accuracy in both derivation (C statistic, 0.75) and validation (C statistic, 0.73) cohorts.
• Identified preimplant variables including cardiomyopathy type, mechanical circulatory support use, hemodynamic parameters, and laboratory values as predictors of RVF.
• STOP-RVF demonstrated superior performance compared to previously published risk scores (Kormos et al. and Drakos et al.).

Conclusions:

• The STOP-RVF calculator, derived and validated using routine clinical data, serves as an effective tool for personalized RVF risk prediction in LVAD candidates.
• Accurate RVF prediction using STOP-RVF is associated with improved cumulative survival rates.
• This validated tool can guide clinical decision-making and improve outcomes for patients undergoing LVAD support.