Methodology

1. The model

Directional markers use the z-score at your chronological age and sex, multiplied by a configured years-per-SD coefficient to shift that age. Band/floor markers instead use deviation from configured ranges. The result combines their weighted mean with activity and sleep, limited to ±3 years. Final age is clamped to ±15 years around chronological age and to 18–80. This is not an age-search or best-fit procedure, nor PhenoAge.

matchedAge_i = clamp(age + z_i · dir_i · yearsPerSD_i,  age ± 15,  [18, 80])
      z_i = (value_i − μ_i(age, sex)) / σ_i(age, sex)

optimalBand   deviation = max(0, lo − value) + max(0, value − hi)
              matchedAge_i = clamp(age + (deviation / σ_i) · yearsPerSD_i,  age ± 15,  [18, 80])

optimalFloor  matchedAge_i = clamp(age + max(0, threshold − value) / σ_i · yearsPerSD_i,  age ± 15,  [18, 80])

factorAge_j = weightedMean(matchedAge_i in factor j)
weight_j = Σ w_i in factor j
BA_fit = Σ(weight_j · factorAge_j) / Σ weight_j
BA     = clamp(clamp(BA_fit + lifestyle,  age ± 15),  [18, 80])
lifestyle ∈ [−3, +3]
scenario = clamp(clamp(Σ(weight_j · min(factorAge_j, age)) / Σ weight_j − 3,  age ± 15),  [18, 80])
delta  = round1(BA − age)          Σ factor.impact === delta

2. Marker directions

These directions describe the model's arithmetic, not universal health rules. Sample averages are not health targets. BMI, waist-to-height ratio and blood pressure use configured bands, and vitamin D uses a floor. Very low glucose or very high HDL, for example, must not be assumed healthy because they lower this model's estimate.

higherWorse

Higher increases the model age relative to the sample mean for your age and sex; lower is not necessarily medically safer.

higherBetter

Higher decreases the model age using the reversed z-score; this does not imply unlimited health benefit.

optimalBand

Configured band: either-side deviation increases the model age; within the band is neutral.

optimalFloor

Configured floor: below-threshold values increase the model age; at or above is neutral, not necessarily safe.

3. Markers, weights and references

Weights and years per SD are heuristic product coefficients, not fitted or externally validated clinical calibrations. Weights are normalised across supplied markers. Years per SD converts one standard deviation into a model age shift. Listed sources concern clinical ranges or marker associations; they do not establish these numerical coefficients. Configured references are not individual treatment targets.

4. Attribution

Each factor card shows how many years that marker contributed. Contributions are distributed with largest-remainder rounding, so they always add up exactly to your total difference — no hidden residual.

5. Estimate vs hypothetical scenario

The hypothetical scenario takes the lower of each grouped factor age and your chronological age, preserving negative shifts and neutralising positive ones. Blood pressure and lipids are each grouped before this step. It applies the best lifestyle adjustment (−3 years) and the same age limits. The difference is a model scenario gap, not evidence of years recoverable through treatment or lifestyle changes.

6. Data sources

Measured-marker means, standard deviations and percentiles use NHANES 2017–2018 records with completed interview/examination and recorded ages 18–80, in six bands by sex. Recorded age 80 includes everyone aged 80 or older. The processor uses no survey weights, trims pooled marker values at the 0.1st and 99.9th percentiles, and omits groups with fewer than 30 observations. These are sample statistics, not representative US estimates or healthy targets. Means and SDs are interpolated between band midpoints, with endpoint values outside 24–75.

Age-band midpoints: 18-29 → 24 · 30-39 → 34.5 · 40-49 → 44.5 · 50-59 → 54.5 · 60-69 → 64.5 · 70-80 → 75

Measured-marker statistics in the static JSON can be reproduced from the local parquet cache with scripts/process_nhanes.py --audit, without replacing it. Fresh CDC downloads have not been independently compared with that cache. Fasting eligibility is not explicitly filtered; tiny positive values in cached weights and diastolic readings need review. Pregnancy is not excluded. Activity and sleep references are legacy model inputs with publication-level provenance still unverified, not statistics computed by this processor.

• NHANES 2017–2018, US Centers for Disease Control and Prevention
• WHO — Body mass index classification
• Whelton et al. — 2017 ACC/AHA Hypertension Clinical Practice Guideline
• Grundy et al. — 2018 ACC/AHA Cholesterol Clinical Practice Guideline
• Ashwell et al. — Waist-to-height ratio meta-analysis, Obesity Reviews 2012 · NICE CG189
• Holick et al. — Endocrine Society Vitamin D Guideline, 2011
• Demay et al. — Vitamin D for Prevention of Disease, 2024

7. Limitations

No external validation of this tool against mortality, clinical outcomes or epigenetic age has been established. Heuristic coefficients, unweighted US sample references, correlated markers, self-report and age limits affect the estimate. Coverage counts optional grouped factors, not an uncertainty interval. The historical vitamin D threshold comes from 2011; the 2024 Endocrine Society update states that optimal levels for disease prevention remain uncertain. Do not change medication, supplements or sun exposure to lower this score; discuss abnormal measurements with a healthcare professional.