Using random forest machine learning on data from a large, representative cohort of the general population improves...

Area of Science:

• Pulmonary Medicine and Respiratory Research
• Biostatistics and Machine Learning Applications
• Public Health and Epidemiology

Background:

• Spirometry, a key tool for diagnosing lung diseases like COPD, faces challenges leading to misdiagnosis.
• Existing clinical reference values for forced vital capacity (FVC) and forced expiratory volume in one second (FEV1) may lack precision for diverse populations.
• This study explores the utility of Random Forest (RF) algorithms to refine FVC and FEV1 prediction models using a large, representative US population dataset.

Discussion:

• The study compared RF model accuracy against traditional Danish clinical references and multiple linear regression (MLR) models.
• RF models demonstrated superior correlation coefficients for both FVC and FEV1 compared to existing references.
• Analysis of slope and intercept values indicates RF's potential for more accurate individual patient lung function estimation.

Key Insights:

• Random Forest models achieved significantly higher accuracy (FVC: 0.85, FEV1: 0.92) in predicting lung function compared to current clinical references (FVC: 0.66, FEV1: 0.69).
• RF models showed improved slope and intercept values, suggesting better precision in estimating individual FVC and FEV1.
• Machine learning, specifically RF, shows promise in enhancing the reliability of spirometry reference values.

Outlook:

• Machine learning models like RF hold significant potential for improving the prediction of estimated lung function in clinical practice.
• Further research is needed to optimize RF models, particularly to reduce the magnitude of intercept values for enhanced clinical utility.
• Refined lung function predictions can lead to more accurate diagnoses and better management of respiratory conditions.