Machine learning models for predicting COPD exacerbations and long-term prognosis are not yet ready for routine clinical decisions, Jean Bourbeau, MD, MPH, told HCPLive at CHEST 2026 in Phoenix.
The GOLD 2026 report introduced its first chapter on AI and emerging technologies in COPD, alongside a treatment framework centered on disease activity and stability.¹,² In this third part of his interview, Bourbeau addressed where prognostic models stand and how digital tools may support the new framework.
“If you’re not giving all the right information, it may not necessarily give you the right answer at the end,” said Bourbeau, professor in the Division of Respiratory Medicine at McGill University and director of the COPD clinic and pulmonary rehabilitation at the McGill University Health Centre. “This may be one of the limitations we have now.”
Machine Learning for COPD Exacerbation and Prognosis Prediction
Bourbeau described AI-based prognosis as a large field in its own right, though few studies have applied machine learning to establish prognosis at different points along the disease trajectory. None of these models has yet proven superior to the clinical scores already used in practice, he said.
He attributed part of this gap to incomplete data. Machine learning can only transform the information it receives, and models built on partial clinical inputs are unlikely to deliver accurate predictions, according to Bourbeau.
Bourbeau expects performance to improve as the field identifies new predictive biomarkers. Candidate sources he named include CT imaging, blood, voice recognition, clinical features, and measurements from wearable devices. Until more research is completed, he said the technology is not ready for clinical use, and no current evidence shows these models outperform existing prognostic tools.
Digital Monitoring and Disease Activity in GOLD 2026
Bourbeau noted the concepts of disease activity and stability, newly introduced in GOLD 2026, are not new to medicine.¹ The framework originates in other chronic inflammatory diseases, including rheumatoid arthritis, multiple sclerosis, and ulcerative colitis.
He said the concept now applies to COPD as treatment becomes more effective, beginning with triple therapy for patients at high risk of exacerbation and extending to biologics targeting specific airway inflammatory mechanisms. Recognizing the right phenotype and endotype is central to this approach.
AI could help integrate data gathered at clinic visits, blood tests, and CT, with home monitoring adding another layer of precision, according to Bourbeau. Consumer devices, such as a ring capturing multiple physiological measures, make home data collection possible in ways unavailable 5 to 10 years ago. He described this integration as the near future, not current practice.
For now, Bourbeau said clinicians should put the disease activity concept into practice. The goals he outlined are reducing exacerbations with an aim of zero, maintaining or improving symptoms with the lowest possible treatment burden, and preventing structural damage and progressive lung function decline.
“It is as important to say when the patient is stable…[as] to recognize early on the patient is not stable anymore,” Bourbeau said. He sees AI-supported home monitoring as a potential way to raise a red flag when a patient loses stability.
Bourbeau said research will determine whether home-based measures add value to clinical assessment. How AI models integrate these data sources will be important to their eventual role in practice, he added.
Watch parts 1 and 2 of our interview with Bourbeau at CHEST here: AI and Emerging Technologies in COPD Care, With Jean Bourbeau, MD and AI Could Help Find Undiagnosed COPD on Lung Cancer Screening CT
Bourbeau has no reported disclosures.
References
Global strategy for prevention, diagnosis and management of COPD: 2026 report. Global Initiative for Chronic Obstructive Lung Disease. Published November 10, 2025. Accessed October 27, 2026. https://goldcopd.org/wp-content/uploads/2026/01/GOLD-REPORT-2026-v1.3-8Dec2025_WMV2.pdf
