Derivation and validation of a machine learning-driven score to predict the diagnostic yield of endomyocardial biopsy.

Despite its low diagnostic yield, endomyocardial biopsy (EMB) remains the gold standard for establishing a definitive diagnosis in many cardiomyopathies. We developed and validated a machine-learning-based score to predict the likelihood of diagnostic EMB using non-invasive data. We retrospectively analyzed 775 heart failure patients who underwent EMB. A random forest algorithm was selected for score development based on superior discriminative performance. The model was externally validated in an independent cohort (n = 171). The study population was predominantly male (72.1%), with half of the patients in NYHA class III-IV. EMB yielded a definitive diagnosis in 19.9% of cases, most commonly amyloidosis (50%). A predictive score (0-100 range) was derived from key non-invasive predictors. Right ventricular late gadolinium enhancement (LGE) on cardiac magnetic resonance emerged as the strongest predictor, followed by left ventricular and atrial LGE, NTproBNP levels, and renal function. The model demonstrated excellent discrimination, with an area under the curve of 0.92 (95% CI = 0.89-0.96) in cross-validation and 0.91 (95% CI = 0.86-0.98) in the testing set, with consistent performance on external validation (AUC 0.82, 95% CI = 0.76-0.89). This machine-learning-based score may provide a non-invasive tool to support EMB decision-making in clinical practice.