AI-derived myosteatosis on cardiac CT is associated with incident COPD risk: an AI-CVD study

Background Myosteatosis, the pathological fat infiltration into skeletal muscle, has emerged as an important marker of metabolic dysfunction and cardiovascular risk. However, its potential impact on lung health and development of chronic obstructive pulmonary disease (COPD) is not well established. Patients with COPD often exhibit muscle atrophy and fatty infiltration, which correlates with worse outcomes, but it is unknown if muscle fat infiltration in asymptomatic individuals predicts future COPD. Objectives We aimed to determine whether myosteatosis quantified from cardiac CT scans using artificial intelligence (AI) is associated with risk of incident COPD in an initially asymptomatic, community-based cohort. The AI-CVD initiative aims to rapidly and accurately extract clinically relevant opportunistic screening data from chest CT scans, thereby improving early detection and risk assessment. Methods We performed a retrospective cohort analysis in participants from Exam 1 of the Multi-Ethnic Study of Atherosclerosis (MESA), a cohort of men and women aged 45–84 free of CVD at baseline. Myosteatosis was quantified using baseline non-contrast cardiac CT scans using a validated AI algorithm (AI-CVD) to segment muscle and compute muscle fat infiltration (e.g., low muscle attenuation in Hounsfield units (HU) as a proxy for intramuscular fat). COPD was diagnosed using spirometry. Proportional hazards models were used to examine the association between myosteatosis and incident COPD. Myosteatosis was analyzed in quartiles using the highest quartile as reference and continuously using restricted cubic splines. Models were adjusted for available confounders including age, sex, pack-years of smoking, emphysema (low attenuation area in CT), body mass index, inflammatory markers, diabetes, and socioeconomic status. Results Over 15 years follow-up, 283 cases of COPD were identified. The cumulative incidence of COPD for individuals within the lowest quartile of muscle mean HU, representing myosteatosis, was significantly higher than other quartiles (p<0.0001). In minimally adjusted models (for age, pack years of smoking, emphysema), the HR for the lowest vs. highest quartile of muscle mean HU was 1.87 (1.31-2.67). In fully adjusted models, the HR for incident COPD remained significant (1.32 (1.09, 1.62)). Conclusions In this large asymptomatic cohort, AI-derived muscle fat infiltration on routine cardiac CT scans was an independent predictor of future COPD risk. This suggests adverse muscle composition, particularly those available in the field of view of cardiac CT such as the pectoralis, intercostal, and paraspinal muscles, precedes lung function decline and COPD onset. Myosteatosis measured opportunistically on CAC scans could serve as an early indicator for elevated COPD risk, potentially enabling earlier preventive interventions.