ZORV Vitality Age — Methodology

2. The seven domains

Domains are combined with default weights that sum to 100%. Order below matches the engine’s aggregation order.

3. Gompertz conversion

Each composed hazard ratio for a domain (or sub-metric path) is mapped to an effective age offset in years using a Gompertz-style baseline hazard for mortality scaling:

offset_years = ln(HR) / ln(b)

Here b is the age-stratified annual mortality multiplier for a reference band (approximate Gompertz slope in log-hazard per year). ZVA uses:

Chronological age selects the band (values outside 20–80 clamp to the nearest band).

Examples (40–59 band, b = 1.10)

• HR = 1.09 (e.g. higher resting HR per 10 bpm): offset ≈ ln(1.09)/ln(1.10) ≈ +0.90 y.
• HR = 0.87 (e.g. protective cardiorespiratory fitness per MET): offset ≈ ln(0.87)/ln(1.10) ≈ −1.47 y.
• HR = 4.0 (e.g. impaired heart-rate recovery): offset ≈ ln(4)/ln(1.10) ≈ +14.5 y (large HRs map to large offsets by design).

4. Bayesian aggregation

ZVA combines domain-level offsets into one aggregate vitality offset and an uncertainty range. The implementation is a pragmatic MVP: it is not a full hierarchical Bayesian posterior with MCMC; it is documented as Bayesian-style because it pools uncertainty from domains with explicit priors.

How domains are combined

1. Effective weights: each domain’s configured weight is multiplied by that domain’s input confidence (0–1). If the sum of effective weights is positive, weights are renormalized to sum to 1. If all confidences are negligible, the model falls back to uniform weights over domains that have any inputs (or over all domains as a last resort).
2. Aggregate offset: the weighted sum of per-domain offset_years values.
3. Per-domain uncertainty: a stochastic term combines low confidence with larger offset magnitude, plus a base uncertainty per domain (years) so sparse domains do not collapse uncertainty.
4. Pool: independent combination yields raw combined spread = sqrt(sum (w_i s_i)²).
5. Correlation inflation: because domains overlap physiologically, a factor in [1.0, 1.4] inflates the pooled spread based on average inter-domain correlation among active domains (fixed correlation prior matrix in code).
6. 95% interval on offset: the aggregate offset ± 1.96 × inflated combined spread. Vitality age = chronological age + offset; the reported band applies to that offset (and thus to vitality age).

What the interval means

This is best read as a parametric uncertainty interval around the aggregate offset under the MVP independence/correlation-inflation model. It reflects input completeness and domain-level spread, not a full posterior over latent biological age. As longitudinal data grows, the model may graduate to richer covariance or explicit Bayes inference (see in-code “Phase 3” notes).

5. Periodization gate

Structured training intentionally changes acute markers (e.g. HRV suppression in build/overload; reduced strength stimulus in taper/race). ZVA applies gates so population-norm comparisons do not blindly penalize athletes in appropriate phases.

Autonomic (build / overload)

When training phase is build or overload, HRV is evaluated against an expected suppression versus baseline (and population z-scores). If the gated path yields a more favorable (lower) unfavorable offset than the naive population comparison, a periodization gate flag is set and the gated offset is used. Optional baseline rMSSD improves detection of excess suppression beyond expected.

Musculoskeletal (taper / race)

When phase is taper or race, strength and movement-diversity hazard ratios are forced to neutral (HR = 1) so short-term reductions in lifting volume or sport variety do not register as long-term risk. A gate flag records that this neutralization applied.

6. Data tiers

Your snapshot includes a data tier label:

• Complete: every domain reports at least two tracked inputs available (inputs_available ≥ 2 per domain).
• Enhanced: otherwise, if you have metabolic lab signals on the engine’s metabolic LOINC path or body composition (lean mass or body fat) plus at least one lab value.
• Core: wearable-forward estimate when enhanced/complete criteria are not met; domains may be sparse.

7. Limitations & caveats

• Population hazard ratios on individuals: published HRs are group-level associations. They are used as transparent priors, not individualized causal effects.
• Wearable accuracy: device error, wrist vs chest, algorithm changes, and missing nights affect inputs; confidence weighting and uncertainty partially reflect this but cannot remove all measurement bias.
• Not a medical diagnostic: ZVA is a wellness and training-context estimate. It does not diagnose disease or replace clinical judgment or screening guidelines.
• Additive / independence limits: domain-level independence and fixed correlation inflation approximate overlapping biology; interaction effects (e.g. sleep × training load) are not fully captured.
• Living model: priors, weights, and aggregation may change as methodology improves. Check the product version string when comparing snapshots over time.