Two people can be the same age and have very different-looking brains on an MRI. A study of 61,437 UK Biobank participants asks whether that difference carries information about health beyond the brain. People whose scans looked older than expected for their age had higher rates of death during follow-up and were more likely to develop several diseases, including conditions outside the nervous system.
The peer-reviewed study, published November 3 in npj Aging, adds weight to the idea that brain structure can reflect aspects of whole-body health. The researchers studied patterns across a population. They did not establish that an older-looking brain causes illness or show that a scan can predict an individual’s lifespan.
Evidence level Human Observational Research. The researchers compared MRI estimates with health records and other measurements in an existing cohort.
How an MRI becomes an age estimate
The team started with features that can be measured on brain MRI, including the volume and thickness of different regions and measures from diffusion imaging, which captures aspects of how water moves through tissue. These measurements change with age, but the changes are uneven across the brain. An age model combines the features so that it can learn which patterns tend to occur at different ages.
The researchers used a statistical method called LASSO regression. It limits how much the model can depend on any one measurement and can drop features that contribute little to its predictions. The team reserved about a fifth of the imaging sample to test age prediction. It also corrected for a common bias in age models, which tend to guess too high for younger people and too low for older people. The team applied the model to repeat scans from UK Biobank as a further check.
The difference between the adjusted estimate and a participant’s age at the scan is what the paper calls brain-age acceleration. If the model estimates 65 for a 60-year-old, the gap is five years. Despite the name, that calculation can be made from a single scan. It compares the scan with an age-related pattern and does not measure how fast that person’s brain changed from one year to the next.
Why the link extends beyond the brain
Participants with larger gaps had a higher rate of death from any cause during follow-up. Larger gaps also tracked the incidence of diseases across several organ systems. The authors’ models estimated shorter life expectancy after age 40 for participants in the highest quarter of brain-age gaps compared with those in the lowest quarter. That comparison describes groups and depends on the assumptions of the survival models.
A brain scan can capture signs of processes that also affect the rest of the body. Blood vessels are one place to look. When researchers compared genetic associations with patterns of gene activity in individual brain cells, some of the strongest links involved cells at or around blood vessels. These included endothelial cells, which line the vessels, and pericytes, which help support small vessels. The analysis also implicated astrocytes, support cells that interact with both neurons and the blood supply. That points to places to investigate experimentally.
The team also found associations with smoking, alcohol intake and diet quality. Statistical models suggested that the MRI gap could account for part of some relationships between these factors and health outcomes. Those calculations depend on assumptions about what comes first and which other influences have been accounted for. Illness can change both behaviour and the brain, and a shared influence can affect both a scan and later health. The study leaves open whether changing the MRI score would change someone’s risk.
What preventive medicine would need to know
A useful clinical test has to change a decision. Researchers would first need to know whether this MRI score improves risk estimates beyond information clinicians already have, such as age, blood pressure, smoking history and known disease. Then they would need to show that acting on the extra information helps patients. This study did not test care guided by the score.
The scans used to build and check the model came from the same research programme. UK Biobank volunteers are generally healthier than the wider population, and the study cohort was predominantly of European ancestry. Testing the model independently in more diverse groups and across scanner types would help establish how well the estimates travel.
Repeated scans could answer another practical question. If a person’s score changes, researchers need to separate a lasting change in brain structure from measurement noise and determine whether that change tracks outcomes patients care about. The size of this study makes the links worth following. Independent testing will determine whether they add anything to preventive care.
AI-generated editorial illustration of a brain with MRI-style slices.