Using AI to Assess Eye Scans, Scientists Find “Retinal Age Gap” Predicts Mortality Risk

Using deep learning to predict “retinal age” from images of the internal surface of the back of the eye, an international team of scientists has found that the difference between the biological age of an individual’s retina and that person’s real, chronological age, is linked to their risk of death. This ‘retinal age gap’ could be used as a screening tool, the investigators suggest.

Reporting on development of their deep learning model and research results in the British Journal of Ophthalmology, first author Zhuoting Zhu, PhD, at Guangdong Academy of Medical Sciences, together with colleagues at the Centre for Eye Research Australia, Sun Yat-Sen University, and colleagues in China, Australia, and Germany, concluded that in combination with previous research, their study results add weight to the hypothesis that “… the retina plays an important role in the aging process and is sensitive to the cumulative damages of aging which increase the mortality risk.”

The team’s published paper is titled “Retinal age gap as a predictive biomarker for mortality risk,” in which they concluded, “To the best of our knowledge, this is the first study that has proposed retinal age gap as a biomarker of aging … Our findings have demonstrated that retinal age gap might be a potential biomarker of aging that can predict mortality risk.”

Estimates suggest that the global population aged 60 years and over will reach 2.1 billion in 2050, the authors noted. “Aging populations place tremendous pressure on healthcare systems.”  But while the risks of illness and death increase with age, these risks vary considerably between different people of the same age, implying that ‘biological aging’ is unique to the individual and may be a better indicator of current and future health. As the authors pointed out, “Chronological age is a major risk factor for frailty, age-related morbidity and mortality. However, there is great variability in health outcomes among individuals with the same chronological age, implying that the rate of aging at an individual level is heterogeneous. Biological age rather than chronological age can better represent health status and the aging process.”

Several tissue, cell, chemical, and imaging-based indicators have been developed to pick up biological aging that is out of step with chronological aging. But these techniques are fraught with ethical/privacy issues as well as often being invasive, expensive, and time consuming, the researchers noted.

A growing body of evidence suggests that the network of small vessels (microvasculature) in the retina might be a reliable indicator of the overall health of the body’s circulatory system and the brain. “The retina is considered as a window to the whole body, which shares similar embryological origins, physiological features and anatomical structures with vital organs such as the heart, the brain and the kidney,” the team noted. “A growing number of studies have suggested that the retinal microvasculature could reliably reflect the systemic circulation in vivo and the retinal neural tissue shared common pathological alterations of neurodegenerative diseases with the brain.”

With recent studies having also demonstrated the use of deep learning (DL) models to predict age using clinical images, the team considered the potential to predict biological age by applying DL to retinal images. Deep learning is a type of machine learning and artificial intelligence (AI) that imitates the way people acquire certain types of knowledge. But unlike classic machine learning algorithms that are linear, deep learning algorithms are stacked in a hierarchy of increasing complexity.

For their reported study, the researchers developed a deep learning model to see if it might accurately predict a person’s retinal age from images of the back internal surface of the eye —the fundus—and to see whether any difference between this and a person’s real age— retinal age gap—might be linked to a heightened risk of death. “We therefore developed a DL model that can predict age from fundus images, known as retinal age,” they wrote. “Using a large population-based sample of middle-aged and elderly adults, we investigated the association between retinal age gap, defined as the difference between retinal age and chronological age, and mortality.”

The researchers drew on 80,169 fundus images taken from 46,969 adults aged 40–69 years, who were were part of the U.K. Biobank, a large, population-based study of more than half a million middle aged and older U.K. residents. The team used 19,200 fundus images from the right eyes of 11,052 participants in relatively good health at the initial Biobank health check to validate the accuracy of the deep learning model for retinal age prediction. The results indicated a strong association between predicted retinal age and real age, with an overall accuracy to within 3.5 years.

The retinal age gap was then assessed in the remaining 35,917 participants during an average monitoring period of 11 years. During this time, 1871(5%) participants died. Of these, 321(17%) died of cardiovascular disease, 1018 (54.5%) died of cancer, and for 532 (28.5%) death was due to other causes, including dementia. The proportions of ‘fast agers’ —those whose retinas looked older than their real age—with retinal age gaps of more than 3, 5, and 10 years were, respectively, 51%, 28%, and 4.5%.

Large retinal age gaps in years were significantly associated with 49–67% higher risks of death, other than from cardiovascular disease or cancer. And after adjusting for potential confounding factors, each 1 year increase in the retinal age gap was associated with a two percent increase in the risk of death from any cause and a three percent increase in the risk of death from a specific cause other than cardiovascular disease and cancer, after accounting for potentially influential factors, such as high blood pressure, weight (BMI), lifestyle, and ethnicity. The same process applied to the left eyes produced similar results.

The reported work is an observational study, and as such, can’t establish cause, and the researchers also acknowledged that the retinal images were captured at one moment in time, and that the participants may not be representative of the U.K. population as a whole. Nevertheless, they wrote, “Our novel findings have determined that the retinal age gap is an independent predictor of increased mortality risk, especially of non-[cardiovascular disease]/ non-cancer mortality.”

The retina thus offers a unique, accessible ‘window’ to evaluate underlying pathological processes of systemic vascular and neurological diseases that are associated with increased risks of mortality, the team suggested. This hypothesis is supported by previous studies, which have suggested that retinal imaging contains information about cardiovascular risk factors, chronic kidney diseases, and systemic biomarkers. “… the fast, non-invasive, and cost-effective nature of fundus imaging enables it to be an accessible screening tool to identify individuals at an increased risk of mortality,” they concluded.