A machine learning approach to triaging patients with chronic obstructive pulmonary disease

Area of Science:

• Pulmonary Medicine
• Artificial Intelligence in Healthcare
• Clinical Decision Support

Background:

• Chronic Obstructive Pulmonary Disease (COPD) patients face daily risks of acute exacerbations and loss of disease control.
• Effective on-demand decision support tools are needed to mitigate these risks.
• Early detection and triage of exacerbations are crucial for timely intervention.

Purpose of the Study:

• To present a machine learning-based strategy for the early detection and triage of COPD exacerbations.
• To develop and validate a supervised prediction algorithm trained on physician-validated patient cases.
• To assess the performance of the algorithm against individual pulmonologists.

Main Methods:

• A supervised prediction algorithm was trained using physician opinions on a comprehensive set of patient cases.
• The algorithm's accuracy was evaluated against a panel of physicians triaging identical cases.
• A validation set of 101 patient cases was used for assessment.

Main Results:

• The machine learning algorithm demonstrated superior accuracy and safety indicators compared to individual pulmonologists in identifying exacerbations and predicting triage consensus.
• The algorithm achieved top performance in sensitivity, specificity, and positive predictive value (PPV) for predicting the need for emergency care.
• The AI tool successfully identified exacerbations and guided triage decisions.

Conclusions:

• Machine learning-based decision support tools can effectively aid in the early detection and triage of COPD exacerbations.
• The developed algorithm surpasses individual clinician performance, offering a potentially safer and more accurate approach to managing COPD exacerbations.
• This technology has the potential to significantly improve the management of COPD patients by providing reliable on-demand decision support.