A 58-year-old executive sat in my office last year convinced something had shifted. His annual physical was clean. His MRI read normal. His cognitive screen landed in the average range for his age. Every test I could order told him he was fine, and none of them measured the thing he actually wanted to know: how healthy is my brain, really, and which direction is it heading? We have exquisite tools for finding tumors and strokes. We have almost nothing that gives a person a single, honest readout of overall brain health. A group at Mass General Brigham may have just built one, and they built it out of a night's sleep.
What the NEJM AI Study Found
In February 2026, Brandon Westover, Wolfgang Ganglberger, and their colleagues published a study in NEJM AI that I have not stopped thinking about. They took 36,000 overnight sleep recordings from 27,000 people across six separate research cohorts and fed the raw EEG signal into a deep neural network. They did not hand the model the features sleep scientists usually rely on, things like spindle density or REM fraction. They let it learn directly from the brain's electrical activity and discover its own markers. The network compressed each night into a 1,024-dimensional representation of brain health, then distilled that down to a single score.
The score did real work. It predicted cognitive test performance better than age and demographics alone, with correlations climbing into the moderate range. It separated people with neurological and psychiatric disease from healthy controls with accuracy well above chance. Most striking to me, in models adjusted for age, every one-standard-deviation increase in the brain health score corresponded to a 31% to 35% lower risk of dying during follow-up. That is a hazard ratio between 0.65 and 0.69, with a P value below .0001. One night of sleep, read by a machine, carrying a signal about mortality that conventional sleep metrics missed entirely.
A companion analysis from the same group, published in JAMA Network Open in March, pulled a sleep-derived brain age index from more than 7,000 people and linked it to dementia risk years before diagnosis. These two papers express a similar principle: The nature of one's sleep has dramatic implications for their brain health and overall longevity as well.
Why Sleep Is the Right Place to Read the Brain
It is not an accident that the signal lives in sleep. When you are awake, your brain is busy responding to the world, and the EEG is dominated by whatever you happen to be doing. In sleep, the brain runs its own internal maintenance, and the electrical patterns it produces reflect the integrity of the underlying machinery.
During deep sleep, the glymphatic system flushes metabolic waste, including the amyloid and tau proteins tied to neurodegeneration. Slow waves coordinate across the cortex to consolidate the day's memories. The thalamus and cortex hand signals back and forth in rhythms that depend on circuits being intact. When those circuits start to fail, the rhythms change in ways too subtle for a human reader to catch but well within reach of a model trained on tens of thousands of nights. The brain is, in effect, running a nightly self-test, and sleep EEG is the printout.
A Number You Can Actually Move
Here is the part that matters most for anyone thinking about their own brain over the next twenty years. A brain health score is not a verdict. It is a measurement, and measurements can be changed.
I have written before that your brain has a biological age that can run years ahead of or behind your chronological age. A sleep-derived score is the same idea made concrete and trackable. The things that move it are the things we already know shape brain trajectory: how much deep sleep you actually get, your vascular health, your metabolic state, your exposure to chronic inflammation. The difference is that instead of guessing whether your interventions are working, you would have a number to watch. This is the logic behind precision brain medicine, where a single test never tells the whole story and the goal is to combine streams of data into a picture of where a brain is headed and what will change its course. It is also the same direction as the AI systems learning to detect cognitive change from your own baseline rather than a population average.
What a Score Does Not Do Yet
It's going to take time before this work becomes available to everyone. This is not a test you can order at your next visit. The model was built and validated on research cohorts, the cognitive correlations are moderate rather than decisive, and turning a population-level tool into something that gives one specific person a reliable individual readout is a real scientific step that has not been finished. Nobody should walk away thinking a smartwatch reading of their sleep can score their brain tonight.
What the work does establish is direction. Integrative brain biomarkers, the kind that summarize the health of the whole organ rather than hunting for one disease, are arriving. Sleep is turning out to be one of the richest places to find them, and AI is what makes the signal legible.
What This Means for You
The immediate, practical takeaway does not require waiting for the technology to mature. Sleep is both an input to brain health and a readout of it, which means protecting your deep sleep is one of the highest-yield things you can do for your brain, and paying attention to how you sleep gives you an early window into how your brain is doing. If your sleep has degraded and your thinking feels different, that pairing deserves a real workup, not reassurance that you are fine for your age.
That kind of measurement is exactly what our Intensive Brain Health Program is built around: not a single snapshot, but multiple streams of data assembled into a picture of where your brain stands and where it is going. The sleep-EEG score is not in the clinic yet, but the principle behind it already is. In the Neuroeconomy, your cognition is the asset that everything else depends on, and you cannot optimize what you have never measured. We are entering an era where the brain finally becomes quantifiable, and a single night of sleep may turn out to be one of the most honest measurements we have.