Estimating individual treatment effects on COPD exacerbations by causal machine learning on randomised controlled...

Area of Science:

• Pulmonary Medicine
• Data Science
• Pharmacogenomics

Background:

• Estimating individual treatment effects (ITE) is crucial for personalized medicine.
• Identifying patient response before treatment can optimize therapeutic strategies.

Purpose of the Study:

• Develop machine learning (ML) models to estimate ITE for interventions.
• Apply ML models to predict ITE of COPD treatments on exacerbation rates.

Main Methods:

• Utilized data from the SUMMIT and IMPACT randomized controlled trials for COPD patients.
• Employed Causal Forest, a machine learning model, for causal inference.
• Developed and applied a Q-score metric to assess causal inference model performance.

Main Results:

• Causal Forest models showed good performance in predicting ITE in both trials (Q-scores 0.61 and 0.21).
• Patients with the strongest predicted ITE demonstrated significantly larger reductions in exacerbation rates.
• Poor lung function and blood eosinophil counts were key predictors of ITE.

Conclusions:

• ML models for causal inference can effectively identify individual responses to COPD treatments.
• These models have the potential to become valuable clinical tools for personalized treatment decisions in COPD management.