OR WAIT null SECS
Using AI for COPD Diagnosis, Equity, Guideline Gaps: Alvar Agusti, MD, PhD
Roughly 70% of people with chronic obstructive pulmonary disease (COPD) worldwide remain undiagnosed, and spirometry, the test required to confirm the disease, is often unavailable or misinterpreted where it is performed.1 The GOLD Science Committee has identified underdiagnosis, misdiagnosis, and late diagnosis as the 3 key bottlenecks limiting timely COPD care.
Deep learning applied to low-dose CT scans obtained for lung cancer screening identified COPD with an area under the curve of 0.87, compared with 0.68 for traditional quantitative measures of emphysema in the same dataset.2 AI models mining electronic health records for smoking history, repeated infections, and symptom descriptions can similarly flag patients who warrant spirometry. Foundation models trained on broad, generic datasets do not consistently outperform disease-specific models on tasks like COPD diagnosis or risk prediction.
GOLD 2026 also introduced disease activity, disease stability, and clinical control as new conceptual frameworks for primary care, alongside tools such as the RADAR score, derived from the COPD Clinical Control Questionnaire, still in early validation.3 Non-representative training data is a separate, named risk in the GOLD Perspective's list of AI implementation challenges, since models built on skewed cohorts can reinforce existing disparities in COPD care.
Before recommending an AI intervention, developers must validate it in cohorts large and representative enough to support its use, a standard the GOLD Perspective applies to any new tool regardless of specialty. Guidelines remain essential in this AI-enabled future, since they represent human consensus built on the best available evidence, precisely the quality input data AI models need.
Alvar Agusti, MD, PhD, professor of medicine at the University of Barcelona and chair of the GOLD Board of Directors, led the writing of a new GOLD Perspective on AI in COPD. In the following interview, Agusti discusses which diagnostic bottleneck AI is closest to solving, how equity fits into AI development, and where clinical guidelines stand as AI tools mature.
HCPLive: Which of the 3 diagnostic bottlenecks, underdiagnosis, misdiagnosis, or late diagnosis, represents the biggest near-term opportunity for AI?
Alvar Agusti, MD, PhD: Underdiagnosis, misdiagnosis, and late diagnosis of COPD are key bottlenecks currently. For underdiagnosis, we need to use spirometry, but spirometry is not always interpreted correctly, and it's not always done. AI can help identify individuals who really might benefit from spirometry, and can help physicians interpret it correctly.
HCPLive: Why is it important to distinguish between generalist foundation models and specialty models for clinical tasks?
Alvar Agusti, MD, PhD: AI is here, no question. AI is in our life for many different things, and in my own opinion, AI can be of great help in many different aspects of our lives. However, AI can make mistakes, so it's very important, in relation to health and COPD management in particular, that doctors can somehow check the information provided by AI tools.
HCPLive: Why were disease activity, stability, and clinical control included in GOLD 2026, and how far is the RADAR score from routine use?
Alvar Agusti, MD, PhD: The GOLD 2026 report, for the first time, talks about disease activity, which refers to biological activity, and also disease stability and clinical control, which refer to the clinical manifestation of the disease. This is an area that needs more research, discussion, and consensus, but probably the RADAR score might be a very good tool to address this.
HCPLive: How do you build equity into AI tool development and validation from the outset?
Alvar Agusti, MD, PhD: As in any other aspect of research, medicine in general, before you recommend the use of any intervention or tool, you need to validate it in cohorts that are large enough and representative enough. This is what we are doing now, and I hope that in the very near future, we will have the data, the evidence, to support that.
HCPLive: What will the relationship between AI and clinical guidelines look like in 5 to 10 years?
Alvar Agusti, MD, PhD: I was in a meeting a year ago where someone said, now that we have AI, we do not need guidelines anymore. The end result of the discussion was just the opposite. Now that we have AI, guidelines are more important than ever before, because guidelines are the quality, evidence-proven input data for AI. We will still need guidelines that are the result of human consensus on available evidence to inform AI. But once we have this, AI can be a great tool to help physicians around the world treat their patients better.
HCPLive: Is there anything else you'd like to highlight about this GOLD Perspective paper?
Alvar Agusti, MD, PhD: We published this paper on AI in COPD a few months ago. It was the end result of a long series of debates and discussions among GOLD members. It's not the final word on this; it's a rapidly evolving field. But I would suggest that someone interested in AI and COPD read the paper, because it's, to my knowledge, the most up-to-date summary of how AI can help COPD management.
Editor's Note: This transcript has been edited for grammar and clarity using artificial intelligence tools. Alvar’s disclosures include GSK, Sanofi, AstraZeneca, and more.
References
Agusti A, Vila M, Faner R, et al. How artificial intelligence could improve the diagnosis and management of COPD: a perspective from GOLD. Am J Respir Crit Care Med. Published online 2026. doi:10.1093/ajrccm/aamag337
Tang LYW, Coxson HO, Lam S, Leipsic J, Tam RC, Sin DD. Towards large-scale case-finding: training and validation of residual networks for detection of chronic obstructive pulmonary disease using low-dose CT. Lancet Digit Health. 2020;2(5):e259-e267.
Soler-Cataluna JJ, Villagrasa M, Catalan P, et al. Risk validation of a new quantitative score for clinical control of chronic obstructive pulmonary disease: the RADAR score. Arch Bronconeumol. 2026;62(1):28-34.
