Brain Age Index Methodology

Brain age is a way of asking a simple question: at this moment in your life, do your lifestyle, vascular health, sensory functions, and social environment look more like the average person your chronological age, or more like someone older or younger? The answer is not a diagnosis. It is a structured way to look at the modifiable forces that shape how brains age, decade by decade.

The Brain Age Index does not scan your brain. The most rigorous brain age estimators in research use deep learning models trained on tens of thousands of structural MRI scans, predicting chronological age from grey matter morphology and reporting the gap between predicted and actual age. That methodology requires a scanner, a radiologist, and a validated machine-learning pipeline. Our tool computes something different and complementary: a risk-factor-weighted estimate of the modifiable brain-age gap, derived from twenty-five self-reported inputs organized into six brain-health domains, calibrated to the modifiable factors that the world's leading dementia research consortium has identified as causally linked to cognitive aging.

The six domains are Cardiometabolic (blood pressure, LDL cholesterol, blood sugar, central adiposity), Lifestyle & Movement (physical activity, diet, smoking, alcohol), Sleep & Recovery (duration, quality, breathing-related disruption), Cognitive Reserve (education, occupational complexity, mental stimulation, multilingualism), Mind & Mood (mood, chronic stress, sense of purpose), and Sensory & Protective (hearing, vision, social connection, head injury, air pollution). Each domain is scored 0–100 and displayed on a radar, alongside an overall estimate, a percentile relative to age peers, a count of recoverable years, and a brain-health archetype.

The result is an estimate, not a measurement. We surface that distinction in every output. The tool is designed for personal reflection on cognitive health behaviors and is never a substitute for evaluation by a licensed healthcare professional.

More than 57 million people live with dementia worldwide, a number projected to reach 152 million by 2050 absent meaningful intervention. The 2024 Lancet Commission on dementia prevention, intervention, and care — chaired by Gill Livingston of University College London with 27 international experts — concluded that approximately 45% of global dementia cases could be prevented or delayed by addressing fourteen modifiable risk factors across the life course. This is one of the most consequential findings in modern public health: roughly half of dementia is, in principle, avoidable.

The Commission's framework rests on twenty-five years of cohort and longitudinal evidence drawn from studies including the Whitehall II Study, the Cardiovascular Health Study, the Three-City Study, the Rotterdam Study, the Framingham Heart Study, and the Honolulu–Asia Aging Study — collectively following hundreds of thousands of adults across decades. The 2017, 2020, and 2024 iterations of the Commission report represent successive refinements as evidence has matured. The 2024 update added two new factors that 2020 lacked sufficient evidence for: vision loss and elevated low-density lipoprotein (LDL) cholesterol.

The Commission's headline number — 45% — is a population attributable fraction (PAF), meaning the proportional reduction in dementia cases that would occur if a specific risk factor were eliminated from a population, assuming causality. PAFs are population-level, not individual-level: your personal risk depends on how the factors apply to you. But the population number frames the public health stakes, and the per-factor breakdown frames the personal action surface.

Our tool implements the modifiable risk factor framework articulated in the 2024 Lancet Commission report (Livingston et al., 2024), and converts it into years using the brain-age-gap (BAG) neuroimaging literature. Version 2.0 captures all of the Commission's self-reportable factors — including the two added in 2024, elevated LDL cholesterol and untreated vision loss — and extends them with additional evidence-based items: central adiposity (waist-to-hip ratio outperformed BMI as a BAG predictor in causal-modeling work), breathing-related sleep disruption, occupational complexity, multilingualism, and sense of purpose.

Each factor carries a relative weight informed by the Commission's population attributable fractions (PAFs). Less education carries roughly a 5% PAF; hearing loss roughly 7%; high LDL roughly 7%; untreated vision loss roughly 2%; with hypertension, diabetes, obesity, smoking, depression, physical inactivity, alcohol, social isolation, traumatic brain injury, and air pollution making up the remainder of the ~45% total. We translate these population-level associations into an individual year-estimate using effect sizes from the brain-age-gap literature, scaled conservatively (see below).

The Commission's methodology is itself transparent: each factor's PAF is derived from a meta-analysis of the cohort evidence and a separate estimate of the factor's prevalence in the population. The Commission does not claim that eliminating these factors would prevent 45% of dementia in any individual; it claims that doing so at the population level would reduce the population burden by approximately that amount. We carry that distinction forward into every result: this tool estimates the modifiable portion of brain aging, not a clinical brain age.

The tool produces a full profile: an estimated brain-age gap (with a confidence band), six domain scores, an overall brain-health score, a percentile relative to age peers, a count of recoverable years, and a brain-health archetype. The algorithm proceeds as follows.

Step one: each answer maps to two values. Every response carries a signed year-contribution (some factors protect and subtract years, others accelerate and add them) and a 0–1 sub-score for its domain (1 = healthiest). The year-contributions are anchored to effect sizes in the brain-age-gap literature — for example, suboptimal sleep is associated with one to three years of additional MRI-derived brain aging across large cohorts.

Step two: age-conditional salience. Each factor's year-contribution is scaled by a salience multiplier that reflects the Lancet Commission's life-course framing. Midlife-salient factors (blood pressure, LDL, obesity) carry peak weight from roughly 45 to 65; late-life-salient factors (vision, social isolation) ramp in weight after 60. This is a refinement over a flat additive model: the same factor matters differently at different ages.

Step three: conservative aggregation. Because lifestyle and cardiometabolic factors explain no more than about 21% of the variance in the measured brain-age gap (Rotterdam Study), the summed year-contributions are scaled by a global conservatism factor and clamped to a defensible range (roughly −6 to +10 years). This keeps the headline estimate honest rather than overclaiming large swings from self-report alone.

Step four: domains, archetype, and recoverable years. Each domain score is the mean of its sub-scores, mapped to 0–100; the overall score is the mean of the six domains. The recoverable years figure sums the brain-aging years currently being added by factors you can change — the upper bound of what optimizing them could plausibly recover. Finally, an archetype is assigned from the interaction of cognitive reserve and modifiable risk load, with care-aware routing when mood and sleep are both strained.

Confidence band. The output is presented with an explicit uncertainty range of plus or minus three years. This range is not derived from a closed-form variance estimate — it would be misleading to imply that level of statistical precision from self-report data — but it reflects both the typical margin of error in cohort-based brain-age estimators (mean absolute error ~2.4–2.5 years) and the limited share of variance that self-reported factors explain. We display the band rather than a point estimate to discourage over-interpretation of small differences.

Each variable below maps directly to a 2024 Lancet Commission factor. The "weight" column reflects the Commission's published population attributable fraction for that factor at the global level. Where source instruments are well-known validated measures, we have shortened them while preserving construct validity.

PAF values are the 2024 Lancet Commission global population-attributable fractions where published. Factors marked — are evidence-based brain-aging contributors drawn from the brain-age-gap literature that sit outside the Commission's core fourteen and therefore have no Commission PAF; they are weighted from their effect sizes in that literature, scaled conservatively.

end-table-note

Version 2.0 captures all of the Commission's self-reportable factors, including elevated LDL cholesterol and untreated vision loss (both added in 2024). For factors that clinically require instrumentation — audiometric hearing loss, sensor-measured air pollution — we use validated subjective substitutes rather than excluding them. We additionally include several items beyond the Commission's core fourteen where the brain-aging evidence is strong: central adiposity, breathing-related sleep disruption, occupational complexity, multilingualism, and sense of purpose.

The PAF weights are drawn from the 2024 Commission's meta-analytic synthesis, which itself aggregates effect sizes from cohort studies conducted predominantly in high-income settings: the United States, the United Kingdom, the European Union, Australia, and parts of East Asia. This is a meaningful limitation for users outside those populations. Independent replication studies in low- and middle-income countries have found higher total PAFs — 56% in Latin America, 51% in Ethiopia, 41% in India, 40% in China — reflecting higher baseline prevalence of risk factors rather than different underlying biology.

We have applied the global weights without ethnic-specific adjustment. The within-population evidence for ethnic-specific weights is too thin to justify recalibration, though future versions may incorporate cohort-specific calibrations as more diverse longitudinal data becomes available, particularly from the UK Biobank, the Health and Retirement Study, and the SAGE cohort.

Several limitations are important enough to surface explicitly. The tool relies on self-report for every input. Self-report is subject to recall bias, social desirability bias, and the known tendency of healthier and more health-conscious users to disproportionately complete tools like this. The result is a snapshot of a single point in time, not a trajectory. Brain aging trajectories are by definition longitudinal; a single result tells you very little compared to two results six months apart.

The tool does not measure genetic risk factors such as APOE4 status, which has a substantial effect on dementia risk independent of modifiable factors. A 2025 replication of the Lancet model in the Wisconsin Longitudinal Study (5,526 participants, 70 years of data) found that risk factor associations diverged meaningfully by APOE4 carrier status — mid-life diabetes was the primary risk factor among carriers, while mid-life hearing loss was the strongest signal among non-carriers. This kind of gene-environment interaction is invisible to our tool.

The overall score also collapses twenty-five distinct inputs into a single number. This is a compression: a person with mild deficits across many factors may score similarly to a person with severe deficits in two or three. This is precisely why version 2.0 reports a six-domain radar, a ranked list of your biggest modifiable levers, and a count of recoverable years alongside the headline figure — the domain profile and the per-factor view are the more actionable outputs. Treat the single number as a conversation starter and the domain breakdown as the working surface.

Finally: a single result is not a diagnosis, a forecast, or a verdict. It is a structured way to think about cognitive health behaviors at a particular moment, in light of what the published research currently establishes as modifiable. If you have concerns about your cognitive health, please consult a qualified healthcare professional.

The analytical modeling and results-analysis logic of this tool is independently reviewed by a domain expert in statistical modeling and machine learning. The reviewer validates that tool outputs faithfully implement the cited peer-reviewed methodology, tests edge cases at the boundaries of the input space, and confirms that results match the underlying mathematics. See our About page for reviewer credentials.

The reviewer's role is methodological, not editorial. Review covers the analytical model and how it converts inputs to outputs, not the framing of the prose surrounding it. The framing, including this methodology page, is the responsibility of the author.

• v2.0 (May 2026) — Major rework. Reorganized into six brain-health domains scored 0–100 with a radar profile; added a brain-age-gap year-conversion calibrated to the neuroimaging literature; introduced age-conditional salience weighting, a global conservatism factor, recoverable-years estimation, six brain-health archetypes with care-aware routing, and percentile positioning. Expanded from twelve to twenty-five inputs, adding LDL cholesterol, untreated vision loss, central adiposity, breathing-related sleep disruption, occupational complexity, multilingualism, and sense of purpose.
• v1.0 (May 2, 2026) — Initial public release. Implemented twelve self-reportable factors from the 2024 Lancet Commission framework as a single additive estimate.

• Livingston, G., et al. (2024). Dementia prevention, intervention, and care: 2024 report of the Lancet standing Commission. The Lancet, 404(10452), 572–628.
• Livingston, G., et al. (2020). Dementia prevention, intervention, and care: 2020 report of the Lancet Commission. The Lancet, 396(10248), 413–446.
• Stephan, B. C. M., et al. (2024). Population attributable fractions of modifiable risk factors for dementia: a systematic review and meta-analysis. The Lancet Healthy Longevity, 5(6), e406–e421.
• Williams, V. J., et al. (2025). Life Course Modifiable Risk Factors of Dementia: Replicating the 2024 Lancet Commission Model in a Single Longitudinal Cohort. Wisconsin Longitudinal Study analysis.
• Cole, J. H., & Franke, K. (2017). Predicting age using neuroimaging: innovative brain ageing biomarkers. Trends in Neurosciences, 40(12), 681–690.
• World Health Organization (2020). WHO guidelines on physical activity and sedentary behaviour. Geneva: WHO.

The constructs measured by this tool, defined in the LifeByLogic glossary:

Browse the full glossary →

• Take the Brain Age Index assessment
• Sleep-Cognition Optimizer methodology
• About the author — Abiot Y. Derbie, PhD
• LifeByLogic editorial policy
• Educational decision support disclaimer