Brains Aged Faster In Every Condition Tested Except Autism And ADHD, Where The Effect Was Zero

Show a machine 45,900 healthy brains and it learns to read age off gray matter the way a forester reads rings off a stump. Feed it a scan it has never seen and it returns a number, usually within a couple of years of the birth certificate. Sometimes, though, it misses. And the size of that miss, whether the brain looks older than the person carrying it, is what a team led by Chuang Liang and Shile Qi at Nanjing University of Aeronautics and Astronautics has just measured across nine different brain conditions at once.

Their results, out today in PLOS Medicine, sort those conditions into a ranking nobody had assembled inside a single framework before. Dementia sits at the top, addiction close behind, psychiatric illness after that, and developmental conditions at the bottom on precisely zero.

The metric goes by predicted age difference, or PAD. Subtract chronological age from the model’s guess; positive means the tissue looks older than it ought to. Researchers have poked at PAD for well over a decade, and it has always suffered from a particular weakness, which is that a single number for an entire brain tells you something has gone wrong without telling you where or why. Useful for headlines about smoking and drinking. Less useful in a clinic.

Liang’s team went at that problem with volume. Gray matter from 1,016 brain regions, gradient-boosted decision trees, healthy training sets drawn from the Human Connectome Project, the Brain Genomics Superstruct Project and UK Biobank, then applied to 2,698 patients pulled from ADNI, ABIDE, ADHD-200 and hospital cohorts in Beijing, Chengdu, Zhejiang, Xinxiang and Albuquerque.

The Ranking Nobody Expected to Bottom Out

Alzheimer’s produced the biggest gap, a Cohen’s d of 0.97, which in a field used to effect sizes around 0.2 is enormous. People with both alcohol and tobacco use disorder came next at 0.84, tobacco alone at 0.72, alcohol alone 0.62. Schizophrenia landed at 0.53, bipolar disorder 0.46, mild cognitive impairment 0.45, depression a modest 0.28. Then autism: 0.06, and not statistically distinguishable from controls. ADHD: 0.01. Nothing at all.

That null result may be the more interesting half. Developmental conditions announce themselves in childhood, when the brain is still myelinating and pruning and generally rebuilding itself, and the team suggests that plasticity might simply buffer the structural damage that an older cortex would carry permanently. Whether that is really what’s happening, we don’t yet know. But if accentuated aging reflects genuine tissue pathology rather than a statistical shadow, this is roughly the pattern you would predict.

The ordering also refused to break. Swap the training set, shrink the atlas from 1,016 regions down to 216, replace the algorithm with support vector regression or a neural network or a random forest, and the ranking held; Spearman correlations between the original sequence and each alternative ran from 0.83 to 1.00. Whatever this signal is, it isn’t an artifact of one modeling choice.

One Region Turns Up Everywhere

The team then borrowed a tool from cooperative game theory — Shapley values — to work out which of those 1,016 regions were actually driving each disorder’s excess years. The maps came out different, and that’s the point. Psychiatric conditions lit up the frontotemporal network. Addiction picked out the default mode and salience networks, along with the putamen and thalamus, deep structures that handle habit and reward. Which is about what you’d expect from conditions defined by compulsive wanting. Dementia claimed frontal and occipital territory plus the fusiform gyrus.

Genes mapped onto those patterns from the Allen Human Brain Atlas fell into correspondingly different buckets: translation and membrane transport in psychiatric illness, energy metabolism and synaptic signaling in addiction, protein localization and DNA damage repair in dementia. Those last two fit neatly with what’s already known about aging neurons accumulating unrepaired damage.

But one structure showed up in every single group. The prefrontal cortex. Schizophrenia, bipolar, depression, alcohol, tobacco, Alzheimer’s, mild cognitive impairment, all of them. It’s the region that ages fastest in ordinary life and, apparently, the region that suffers first whatever else goes wrong. Psychiatric taxonomy insists these are separate conditions with separate causes; the aging signal hints they might share a final common pathway, or at least a shared vulnerability, in the same few centimetres of frontal tissue. As the authors put it in a statement accompanying the paper, “Different neurological disorders appear to leave different signatures on the brain aging clock, which may help researchers better understand the neural and biological pathways involved in these conditions.”

Some caution is warranted. These are snapshots, cross-sectional scans pressed into describing a process that unfolds across decades. Psychiatric illness and substance use overlap enormously in real populations, and that comorbidity was not modeled, meaning some of the schizophrenia signal could conceivably be cigarettes. The gene expression came from six donated brains, none of whom had any of the conditions in question. And the whole thing is correlational.

There is also a negative result buried in the supplementary material. When the same pipeline was fed functional MRI measures instead of structural ones, the disorder differences vanished almost entirely. Brain aging, if that’s what this is, shows up in tissue volume long before it shows up in activity.

What might come of it is a test that doesn’t exist yet. The divergence analysis found Alzheimer’s patients starting to pull away from controls at around 57 years old, and PAD tracked cognitive scores in both dementia groups (though nowhere else). Somewhere in there sits the germ of a tool for flagging which people with mild cognitive impairment are heading toward Alzheimer’s and which are not, a question clinicians currently answer by waiting. The authors are careful to call this future work rather than an achievement.

Still. A model that has never met you, looking only at the folds and volumes of your cortex, can apparently tell whether your brain is keeping time. Quite a lot may end up riding on what gets done with the answer.

Frequently Asked Questions

What does it actually mean if your brain age is older than your real age?

If your brain age is older than your real age, it means a machine learning model, looking at the volume of gray matter across your brain, estimated you to be older than you are. Researchers call this gap the predicted age difference, and a positive value indicates accentuated aging. It is a population-level research measure rather than a clinical test, and it does not by itself diagnose any condition.

Why did autism and ADHD show no increased brain aging when other conditions did?

Autism and ADHD showed no increased brain aging in this study, with effect sizes of 0.06 and 0.01 that were statistically indistinguishable from healthy controls. The researchers suggest this may be because developmental conditions emerge in childhood, when the brain is still highly plastic and capable of reorganization and repair that could buffer structural damage. Older brains have less of that capacity, so pathology may leave a more permanent mark.

Is it true that addiction ages the brain almost as much as dementia?

It is true that addiction aged the brain almost as much as dementia in this analysis. Combined alcohol and tobacco use disorder produced an effect size of 0.84, close behind Alzheimer’s disease at 0.97 and ahead of every psychiatric condition measured. Both alcohol and tobacco are known to damage neurons directly, which may explain why their aging signature is so pronounced.

Why does the prefrontal cortex matter so much in these findings?

The prefrontal cortex matters because it was the one brain region associated with accentuated aging in every disorder group studied, from schizophrenia and depression to addiction and Alzheimer’s. Each condition otherwise had its own distinct spatial pattern. That shared involvement suggests these conditions may not be entirely independent at the level of brain structure, though the study cannot establish cause.

Could this be used to predict who will develop Alzheimer’s disease?

This could potentially be used to predict Alzheimer’s risk, but no such test exists yet. The researchers found that brain aging correlated with cognitive scores in dementia and mild cognitive impairment, and that Alzheimer’s patients began diverging from controls around age 57. They explicitly frame clinical prediction as a question for future studies rather than something their results already deliver.