Abstract Background Optical Coherence Tomography (OCT) enables high-resolution, non-invasive imaging of the retina and has shown promise as a highly accessible tool for accurate assessment of cardiovascular risk. Causal relationships between retinal OCT phenotypes and cardiovascular disease (CVD) has not been previously assessed. Purpose This study leveraged large-scale OCT images, genomics, and health data from the UK Biobank to provide a comprehensive evaluation of the causal relationships between OCT phenotypes and major CVD and risk factors. Methods We developed a three-stage framework: 1) Retinal layer segmentation: we applied the state-of-the-art nn-UNet model to segment ten retinal layers in 42,350 UK Biobank participants, validated on 50 ground-truth cases (Figure 1). 2) Genome Wide Association Study (GWAS) on retinal features: we extracted quantitative imaging features from layer-wise retinal thickness maps using an unsupervised autoencoder-based phenotyping algorithm (64 latent vectors). GWAS established the genetic basis of these features. 3) Mendelian randomization (MR): we explored causal relationships between cardiovascular traits and OCT-derived features using Inverse-variance weighted two-sample MR (IVW-MR). Results Of 20 cardiovascular traits analysed, 5 showed causal associations with retinal traits. GWAS of the learned imaging features identified 93 statistically significant loci, including variants previously linked to retinal thickness, ophthalmic disorders, and CVD. Five cardiovascular traits were causally linked to the autoencoder’s 64-dimensional embeddings (Table 1):• Coronary artery disease was associated with z6 (β = 0.021, p = 0.0337), z7 (β = -0.047, p = 0.0014), z11 (β = 0.032, p = 0.0486), z22 (β = 0.043, p = 0.0007), z27 (β = -0.043, p = 0.0084), z28 (β = -0.033, p = 0.0032) and z48 (β = 0.041, p = 0.0071)• Body fat percentage showed associations with z8 (β = 0.221, p = 0.0027), z21 (β = 0.139, p = 0.0363), z23 (β = -0.178, p = 0.0161), z45 (β = -0.274, p = 0.0043) and z50 (β = -0.194, p = 0.0108)• Stroke was significantly associated with z20 (β = 0.064, p = 0.0124) and z50 (β = -0.086, p = 0.0071)• Triglycerides were associated with z8 (β = 0.059, p = 0.0124), z10 (β = 0.051, p = 0.0269), z20 (β = -0.060, p = 0.0219), z30 (β = -0.054, p = 0.0229), z46 (β = -0.063, p = 0.0159), z49 (β = 0.060, p = 0.0424), z57 (β = -0.060, p = 0.0081), z60 (β = -0.062, p = 0.0127)• Myocardial infarction (MI) was associated with z17 (β = 0.054, p = 0.0011), z20 (β = -0.083, 5.09 × 10−5 ), z31 (β = 0.051, p = 0.0001) and z33 (β = 0.055, p = 0.0003) Conclusions Our study identified, for the first time, causal links between CVD, risk factors, and the thickness of ten OCT-derived retinal layers using MR. Notably, MI showed the strongest causal association with the z20 retinal feature. These findings suggest the potential of OCT imaging combined with genetic analysis for cardiovascular disease screening and risk assessment.Figure 1
Maldonado-Garcia et al. (Sat,) studied this question.