Development and internal validation of a machine learning-based prediction model for pulmonary hypertension in COPD

Area of Science:

• Pulmonary Medicine
• Cardiology
• Medical Informatics

Background:

• Chronic obstructive pulmonary disease (COPD) frequently co-occurs with pulmonary hypertension (PH).
• PH significantly worsens COPD prognosis.
• Current PH diagnostic tools for COPD patients are often invasive or lack sensitivity for early detection.

Purpose of the Study:

• To develop and validate a machine-learning model for predicting PH in COPD patients.
• To identify key noninvasive predictors of PH in COPD.
• To facilitate earlier and more accurate PH diagnosis in COPD.

Main Methods:

• Retrospective analysis of 523 hospitalized COPD patients.
• Utilized 18 noninvasive clinical and physiological variables.
• Trained and compared eight machine-learning algorithms, including CatBoost, using metrics like AUC, accuracy, sensitivity, and specificity.

Main Results:

• The CatBoost model achieved high discrimination with an AUC of 0.848, accuracy of 0.830, and specificity of 0.866.
• Key predictors identified by SHAP analysis included right ventricular diameter, pulmonary artery diameter, PaCO2, right atrial transverse diameter, and age.
• Model interpretability was assessed using Shapley additive explanations (SHAP).

Conclusions:

• A CatBoost-based model using noninvasive variables can accurately estimate PH risk in COPD patients.
• This model offers transparent feature-level explanations, aiding clinical decision-making.
• The findings support earlier PH detection and risk-stratified management in COPD.