MRI Brain-Age Gaps Across Seven Disorders: A Shared Research Signal

On October 7, 2026, researchers affiliated with Jena University Hospital in Germany published “Accelerated brain aging as a transdiagnostic biomarker: A lifespan MRI brain age study across seven disorders” online in Psychological Medicine. The study tested an MRI-based age estimate across seven diagnoses using a model trained on healthy participants.

What did the MRI study find across seven disorders?

The study found higher control-adjusted brain-age deviations in six of seven patient groups, suggesting that an MRI-derived age measure may capture structural patterns shared across different diagnoses. The authors present brain-PAD as a candidate transdiagnostic or complementary research biomarker. Here, transdiagnostic means that the same measure is examined across diagnostic categories rather than being tied to one condition. Its relevance to healthy-aging research is therefore a shared way to study brain structure, not a demonstrated intervention for extending healthspan or longevity.

Was this study conducted in humans?

This study was conducted in humans using structural MRI, not in vitro or in animal models.

How was the reference model built?

According to the paper, the researchers trained their brain-age model on structural MRI data from 25,425 healthy participants aged 2 to 95. This provided the reference population against which the model estimated age from brain scans. The clinical analyses then compared 1,737 patients across seven disorders with 1,793 age-matched controls.

The study also assessed the model in independent Dallas Lifespan Brain Study cohorts. These included 270 cross-sectional participants and 188 participants with longitudinal scans. Cross-sectional assessment examines participants at a particular point in time, while longitudinal assessment uses scans collected over time.

These components served different purposes within the research. The healthy reference data supported model training. The independent cohorts provided another setting for assessing its predictions. The patient and control comparisons addressed whether the resulting brain-age measure differed across diagnostic groups.

For readers evaluating the paper, separating those purposes is more useful than treating every participant as part of one undifferentiated sample. A model’s training population answers a different question from its clinical comparison groups. The study’s design therefore needs to be read alongside its headline findings.

What does a brain-age gap represent?

The paper’s brain-age measure compares an MRI-based age prediction with chronological age. A positive gap means that a scan appears older than expected relative to the model’s reference. The central measurement is a relationship between a model prediction and a person’s actual age.

There is an additional comparison in the disorder analyses. The reported Δbrain-PAD values describe deviations relative to age-matched controls. The patient-group results therefore concern how brain-age estimates differed from those in the comparison group, rather than simply reporting the predicted age of each participant.

Keeping those comparisons distinct makes the results easier to follow. First comes the age estimate generated from the MRI. Next comes its difference from chronological age. In the clinical analysis, the researchers compare that deviation with the corresponding measure in age-matched controls.

A useful reading framework is to identify both the measurement and the claim attached to it. AgeAmaze’s guide to different levels of aging measurement provides a broader way to organize questions about biomarkers. For this paper, the relevant starting point is an MRI-derived estimate referenced to age.

How large were the reported group differences?

The authors reported the following control-adjusted brain-age deviations in their sample:

• Schizophrenia: +7.64 years.
• Frontotemporal dementia: +7.61 years.
• Alzheimer’s disease: +3.24 years.
• Chronic substance use disorder: +2.91 years.
• Major depressive disorder: +2.76 years.
• Bipolar disorder: +2.08 years.

These six groups had significantly higher deviations than their age-matched controls. The researchers found no significant increase for ADHD, the seventh diagnosis examined.

The shared direction of the findings is central to the transdiagnostic idea. The paper asks whether one measure can describe a structural difference across several diagnoses. It does not require the reported differences to be identical for that research question to be useful.

Reading the results as a set also preserves the study’s scope. The finding concerns the pattern across the sampled diagnostic groups. The numerical differences describe how that pattern appeared in this analysis.

How accurate were the age predictions?

The authors reported mean absolute errors of approximately 5 to 7 years across validation and independent cohorts. This means that predicted brain age was typically several years away from chronological age. Mean absolute error summarizes the size of those prediction differences without letting positive and negative errors cancel each other out.

That accuracy measure belongs beside the group results because it describes a different part of the study. Prediction error asks how closely estimated age matches chronological age. The disorder analysis asks whether the brain-age deviations differ between patient and control groups.

The clearest reading keeps both results visible. The model produced age estimates with the reported error, and the clinical comparison identified differences between groups. Neither result replaces the other. An assessment of the research should account for the model’s performance as well as the pattern it identified.

What limits the interpretation of these findings?

The paper and related methodological work raise several distinct issues. Each affects a different step between producing an age estimate and establishing a useful health measure.

• Group findings do not establish individual diagnosis or prognosis. The reported differences are group comparisons rather than predictions of a particular person’s disease status or future health. A positive gap is not evidence that someone’s brain is literally aging by a specified number of extra years. Nor is the MRI estimate a direct reading of overall biological age.
• Average error is not a personal accuracy guarantee. The reported mean absolute error summarizes performance across participants. It should not be interpreted as a promise that every individual prediction falls within that distance of chronological age. This matters when a research result is presented as a precise personal score.
• Clinical outcomes were not directly assessed. The authors describe brain-PAD as a possible complementary research measure rather than a validated standalone diagnostic test. The study did not establish that obtaining a brain-age score improves clinical decisions or health outcomes. Demonstrating a group association and demonstrating clinical benefit are separate tasks.
• Longitudinal validation is not the same as tracking disease progression. The model was assessed in participants with repeated scans from the Dallas Lifespan Brain Study. That assessment should not be confused with longitudinal evidence showing how the reported differences develop within patient groups or relate to their subsequent clinical course.
• Clinical information was incomplete. The authors identify this as a study limitation. It constrains interpretation of the differences between groups. The reported metric is not a direct measure of symptoms or disease duration, and these associations do not explain why a disorder occurs.
• Age-related prediction uncertainty remains a methodological question. A separate 2026 bioRxiv preprint discusses dependence on chronological age alongside regression to the mean and interpretability. It proposes uncertainty-aware age intervals as an alternative approach. This is emerging methodological work rather than settled consensus about how every brain-age model should be interpreted.
• Health-dependent prediction bias needs attention. Another 2026 preprint examining several biological-age markers reports this concern for measures including brain MRI. It cautions against assuming that age gaps reliably assess individual health. That finding is contextual evidence from a separate analysis, not a result established for this specific model by the Psychological Medicine paper.

How should readers evaluate a brain-age claim?

Start with the question the research actually tested. Here, that question was whether an MRI-derived age deviation differed across diagnostic groups. A useful account of the study should identify the reference population and explain the comparison with age-matched controls. It should also report prediction error alongside the clinical findings.

Next, look for the evidence supporting any proposed use of the score. If a claim concerns future health or better care, ask which outcomes were measured and how participants were followed. AgeAmaze’s overview of questions that matter when validating biomarkers offers a framework for reading that next stage of evidence.

For personal decisions, the relevant conversation is with a qualified clinician about symptoms and the purpose of any proposed testing. The distinction between an aging measurement and a medical test helps keep that conversation focused on the decision the test is meant to inform. The practical question is what useful information a result would add.

What research should come next?

The next step is longitudinal research in patient groups that tests whether brain-age deviations track meaningful clinical outcomes and whether the measure adds useful information to clinical assessment.

How AgeAmaze helps

MRI-derived age differences become overstated when a research comparison is packaged as a personal brain-health verdict. AgeAmaze addresses inflated claims about biological age through readable explanations of biomarkers and emerging interventions, helping research-minded adults assess a reported association without confusing it with an established clinical benefit.

Read the research

Sources

1. Accelerated brain aging as a transdiagnostic biomarker: A lifespan MRI brain age study across seven disorders | Psychological Medicine | Cambridge Core
2. Quantifying Uncertainty in Brain Age Predictions via Conformal Prediction
3. Prediction bias in biological ageing markers | arXiv Science