A predictive atlas of disease onset from retinal fundus photographs

Abstract Early detection of high-risk individuals is crucial for healthcare systems to cope with changing demographics and an ever-increasing patient population. Images of the retinal fundus are a non-invasive, low-cost examination routinely collected and potentially scalable beyond ophthalmology. Prior work demonstrated the potential of retinal images for risk assessment for common cardiometabolic diseases, but it remains unclear whether this potential extends to a broader range of human diseases. Here, we extended a retinal foundation model (RETFound) to systematically explore the predictive potential of retinal images as a low-cost screening strategy for disease onset across >750 incident diseases in >60,000 individuals. For more than a third (n=308) of the diseases, we demonstrated improved discriminative performance compared to readily available patient characteristics. This included 281 diseases outside of ophthalmology, such as type 2 diabetes (Delta C-Index: UK Biobank +0.073 (0.068, 0.079)) or chronic obstructive pulmonary disease (Delta C-Index: UK Biobank +0.047 (0.039, 0.054)), showcasing the potential of retinal images to complement screening strategies more widely. Moreover, we externally validated these findings in 7,248 individuals from the EPIC-Norfolk Eye Study. Notably, retinal information did not improve the prediction for the onset of cardiovascular diseases compared to established primary prevention scores, demonstrating the need for rigorous benchmarking and disease-agnostic efforts to design cost-efficient screening strategies to improve population health. We demonstrated that predictive improvements were attributable to retinal vascularisation patterns and less obvious features, such as eye colour or lens morphology, by extracting image attributions from risk models and performing genome-wide association studies, respectively. Genetic findings further highlighted commonalities between eye-derived risk estimates and complex disorders, including novel loci, such as IMAP1 , for iron homeostasis. In conclusion, we present the first comprehensive evaluation of predictive information derived from retinal fundus photographs, illustrating the potential and limitations of easily accessible and low-cost retinal images for risk assessment across common and rare diseases. Research in context Evidence before this study Before undertaking this study, we reviewed the literature on the predictive utility of medical imaging for disease onset, focusing particularly on retinal fundus photographs. We conducted searches in databases including PubMed and Google Scholar, spanning from the inception of these databases to January 1, 2023. Our search terms included “retinal fundus photography”, “disease prediction”, “machine learning”, “deep learning”, and “healthcare AI”, without language restrictions. Prior research has shown the promise of retinal images in diagnosing and predicting a range of conditions, notably within ophthalmology and specific systemic diseases such as diabetes and cardiovascular diseases. However, a comprehensive evaluation of retinal images’ predictive potential across a broad spectrum of diseases, particularly those without known associations to retinal changes, was lacking. Studies identified varied in quality, with many focusing on single diseases or small datasets, indicating a potential risk of bias and overfitting. Added value of this study Our study extends the application of retinal fundus photographs from ophthalmological and systemic diseases to more than 750 incident diseases, leveraging a foundation model combined with a deep multi-task neural network. This represents the first systematic exploration of the predictive potential of retinal images across the human phenome, significantly expanding the scope of diseases for which these images could serve as a low-cost screening strategy. Moreover, we rigorously compare the predictive value of retinal images against established primary prevention scores for cardiovascular diseases, showing both the strengths and limitations of this approach. This dual focus provides a nuanced understanding of where retinal imaging can complement existing screening strategies and where it may not offer additional predictive value. Implications of all the available evidence The evidence from our study, combined with existing research, suggests that retinal fundus photographs hold promise for predicting disease onset across a wide range of conditions, far beyond their current use. However, our work also emphasizes the importance of contextualizing these findings within the broader landscape of available prediction tools and established primary prevention. The implications for practice include the potential integration of retinal imaging into broader screening programs, particularly for diseases where predictive gains over existing methods are demonstrated. For policy, our findings advocate for further investment in AI and machine learning research in healthcare, particularly in methods that improve upon or complement existing prediction models. Future research should focus on refining these predictive models, exploring the integration of retinal imaging with other biomarkers, and conducting prospective studies to validate the clinical utility of these approaches in diverse populations.