Artificial intelligence machine learning based evaluation of elevated left ventricular end-diastolic pressure: a Cleveland Clinic cohort study

Background: Left ventricular end-diastolic pressure (LVEDP) is a key indicator of cardiac health. The gold-standard method of measuring LVEDP is invasive intra-cardiac catheterization. Echocardiography is used for non-invasive estimation of left ventricular (LV) filling pressures; however, correlation with invasive LVEDP is variable. We sought to use machine learning (ML) algorithms to predict elevated LVEDP (>20 mmHg) using clinical, echocardiographic, and biomarker parameters.

Methods: We identified a cohort of 460 consecutive patients from the Cleveland Clinic, without atrial fibrillation or significant mitral valve disease who underwent transthoracic echocardiography within 24 hours of elective heart catheterization between January 2008 and October 2010. We included patients' clinical (e.g., heart rate), echocardiographic (e.g., E/e'), and biomarker [e.g., N-terminal brain natriuretic peptide (NT-proBNP)] profiles. We fit logistic regression (LR), random forest (RF), gradient boosting (GB), support vector machine (SVM), and K-nearest neighbors (KNN) algorithms in a 20-iteration train-validate-test workflow and measured performance using average area under the receiver operating characteristic curve (AUROC). We also predicted elevated tau (>45 ms), the gold-standard parameter for LV diastolic dysfunction, and performed multi-class classification of the patients' cardiac conditions. For each outcome, LR weights were used to identify clinically relevant variables.

Results: ML algorithms predicted elevated LVEDP (>20 mmHg) with good performance [AUROC =0.761, 95% confidence interval (CI): 0.725-0.796]. ML models showed excellent performance predicting elevated tau (>45 ms) (AUROC =0.832, 95% CI: 0.700-0.964) and classifying cardiac conditions (AUROC =0.757-0.975). We identified several clinical variables [e.g., diastolic blood pressure, body mass index (BMI), heart rate, left atrial volume, mitral valve deceleration time, and NT-proBNP] relevant for LVEDP prediction.

Conclusions: Our study shows ML approaches can robustly predict elevated LVEDP and tau. ML may assist in the clinical interpretation of echocardiographic data.