Predicting exercise pulmonary hypertension: the right-net machine learning model a pilot study

Abstract

Background: Exercise-transthoracic Doppler echocardiography (Ex-TTE) determination of mean pulmonary arterial pressure (mPAP)/cardiac output (CO) slope may offer key diagnostic and prognostic information in cardiorespiratory diseases. However, its applicability and reliability in routine clinical practice remain to be established. Herein, the aim of the present study was to apply a machine learning (ML) model to predict abnormal exercise TTE-derived mPAP/CO slope (>3 mmHg/L·min) in individuals at risk of pulmonary hypertension (PH), based only on clinical and resting TTE parameters.

Methods: The study population (221 healthy adults and 196 patients with connective tissue disease) was grouped according to mPAP/CO slope ≤3 vs. >3 mmHg/L·min (n = 222 and n = 195, respectively). Three different ML models (Elastic Net-Regularized Generalized Linear Model, Classification and Regression Tree, LogitBoost) were trained on resting clinical and TTE parameters to predict mPAP/CO slope >3 mmHg/L·min. Data were split into training/test sets to evaluate performance. The model with the highest area under the curve (AUC) on the test set was selected.

Results: The Elastic Net model achieved the best performance (AUC=0.92). Lower tricuspid annular plane systolic excursion/systolic PAP ratio, female sex, and smaller left ventricular outflow tract diameter were the key features predicting TTE-derived mPAP/CO slope >3 mmHg/L·min.

Conclusions: An ML algorithm using resting clinical and TTE parameters can effectively predict exercise TTE-derived mPAP/CO slope >3 mmHg/L·min, supporting its use as a noninvasive tool to identify individuals at risk of exercise PH.

Keywords: Exercise Doppler echocardiography; Exercise pulmonary hypertension; Machine learning.