Machine Learning-Based Score Predicts Diagnostic Yield of Endomyocardial Biopsy
Overview
A machine learning model using non-invasive clinical data was developed and validated to predict the diagnostic yield of endomyocardial biopsy (EMB) in heart failure patients. The model demonstrated excellent discrimination with AUCs above 0.8 in both internal and external cohorts, potentially guiding clinical decision-making for EMB.
Background
Endomyocardial biopsy remains the gold standard for diagnosing various cardiomyopathies despite its relatively low diagnostic yield. The procedure is invasive and carries risks, making patient selection critical. Non-invasive predictors that can accurately estimate the likelihood of a diagnostic EMB could optimize clinical use. Machine learning approaches offer promise in integrating multiple clinical variables to improve prediction accuracy.
Data Highlights
Parameter
Value
Study population size
775 patients (development cohort)
External validation cohort size
171 patients
Male patients
72.1%
NYHA class III–IV
~50%
Diagnostic yield of EMB
19.9%
Most common diagnosis on EMB
Amyloidosis (50% of diagnostic cases)
Model AUC (cross-validation)
0.92 (95% CI 0.89–0.96)
Model AUC (testing set)
0.91 (95% CI 0.86–0.98)
Model AUC (external validation)
0.82 (95% CI 0.76–0.89)
Key Findings
A random forest algorithm was selected for score development due to superior discriminative performance.
Right ventricular late gadolinium enhancement (LGE) on cardiac MRI was the strongest predictor of diagnostic EMB yield.
Other important predictors included left ventricular and atrial LGE, NTproBNP levels, and renal function.
The predictive score ranges from 0 to 100, integrating key non-invasive clinical variables.
The model maintained robust performance across internal cross-validation, testing, and external validation cohorts.
EMB diagnostic yield was low overall (19.9%), highlighting the need for better patient selection tools.
Clinical Implications
This machine learning-based scoring system can assist clinicians in identifying heart failure patients most likely to benefit from EMB, potentially reducing unnecessary invasive procedures. Incorporating cardiac MRI findings and biomarkers into decision-making may improve diagnostic efficiency. The tool supports personalized risk stratification and could be integrated into clinical workflows to optimize EMB utilization.
Conclusion
The validated machine learning score offers a promising non-invasive approach to predict the diagnostic efficacy of EMB in heart failure patients. Its use may enhance clinical decision-making and improve patient selection for biopsy.
References
Arbelo et al. 2023 ESC Guidelines for the management of cardiomyopathies
Cooper et al. 2007 Scientific statement on the role of endomyocardial biopsy
McDonagh et al. 2021 ESC Guidelines for diagnosis and treatment of heart failure
Sinagra et al. 2021 Standardizing the role of endomyocardial biopsy
Porcari et al. 2023 Endomyocardial biopsy in clinical context
by Christian Basile, Christian L. Polte, Piero Gentile, Entela Bollano, Araz Rawshani, Anders Oldfors, Charlotta Ljungman, Sven-Erik Bartfay, Pia Dahlberg, Clara Hjalmarsson, Marie Björkenstam, Elena Gualini, Antonio Cannatá, Patrizia Pedrotti, Andrea Garascia, Gianluigi Savarese, Aldo Pietro Maggioni, Kristjan Karason, Emanuele Bobbio
