Interim metabolic fitness

A placeholder for the terrain's metabolic-fitness coordinate, built from routine NHANES measures so it can be checked against mortality now. The framing follows Brian Fertig and Jack Tuszynski; this page tests that framing against mortality and does not validate it.

Definition

Capacity uses VO₂max estimated from the NHANES submaximal treadmill test, which was run only in the early survey cycles and only in adolescents and younger adults. It is the closest routine proxy for whole-body metabolic efficiency. It is not P/O.

Each component against mortality

Hazard ratio per standard deviation, adjusted for age and sex, with its confidence interval. Left of the line means higher values go with lower mortality. A per-SD ratio is a straight-line summary and says nothing about shape; the curves below show shape.

The shape of each component's curve

Log hazard ratio across each marker's range, from the NHANES restricted cubic spline (age and sex adjusted), on one shared vertical scale. Higher is worse. The shaded band is the confidence interval and the dashed line marks the lowest risk. Beneath each curve, the strip shows where adults sit in a recent survey cycle: the thin line spans the middle ninety per cent, the thick bar the middle half, the tick the median. Components with no NHANES curve in this release are left empty, not filled in.

In published cohorts, bicarbonate's curve is U-shaped and partly explained by kidney function, and the albumin-adjusted anion gap rises roughly in a straight line.

The composite

Discrimination on held-out data

Age alone already separates well who dies and who does not, so a raw C-index flatters any score. What counts is the gain over age and sex, and over a conventional risk set, measured within age bands on people the model never saw.

Not yet reported: a comparison with PhenoAge, calibration by decile of predicted risk, and a decision curve. Each needs a rerun of the analysis.

A hypothesis this page does not yet test

Hyperinsulinaemia versus hyperglycaemia

Does high insulin cost lives even when glucose is held normal? Adults in the NHANES morning fasting subsample with linked mortality, pooled across survey cycles, insulin users excluded because a measured insulin would then be the drug. One cycle is absent: its body-measures file could not be fetched, so its participants had no BMI.

Insulin at fixed glycaemia

The hazard ratio per standard deviation of log fasting insulin, holding HbA1c and fasting glucose fixed with splines, at three levels of adjustment. Insulin is standardised within each survey cycle because the assay changed between cycles.

The four cells of the metabolic terrain

Position on the terrain is read from fasting values. The valley is normal glucose with low insulin. The mountainside is normal glucose held there by high insulin. Retreat is raised glucose with low insulin, the state the model reads as insulin falling back as a protective withdrawal. Overload is raised glucose with high insulin. High insulin means the top third within a survey cycle; raised glucose means prediabetes or diabetes by HbA1c, fasting glucose, diagnosis or glucose-lowering pills.

At the same glucose state, lower insulin is safer: retreat carries less risk than overload, which fits the hypothesis. Holding raised glucose down with high insulin, the mountainside, is not worse than letting insulin fall. Among people with diabetes the picture turns: there low insulin is as deadly as high, and reads as a failing pancreas rather than a retreat.

Recovery adds information beyond position

In people without diabetes, how far glucose still stands above its fasting value two hours after a glucose load adds information beyond the fasting values alone: the stability of the system, not only its position.

Robustness

Sensitivity analyses on the key estimate; each survey cycle dropped in turn, which also bounds what the absent cycle could change; and a survey bootstrap that resamples sampling units within strata, the usual design-based check.

The efficiency slot: two rival models

The open problem asks whether falling mitochondrial efficiency is a readout of allostatic load or the capacity term that sets how close a person is to the fold. Written as models of one marker held in a valley: in the readout model, efficiency tracks load but the valley's stiffness is set elsewhere; in the capacity model, the pull back toward the setpoint is paid for in energy, so it weakens as efficiency falls. Both reproduce what is already seen, so cross-sectional correlations cannot separate them. They differ in dynamics.

The smallest study that separates them

Under the capacity model, people with lower efficiency recover more slowly at the same load; under the readout model they do not. The discriminating statistic is the partial correlation of recovery time with efficiency at fixed load. The table gives the correlation the capacity model implies and the number of people for eighty per cent power, across assumptions: how much of the stiffness is paid in energy, how much efficiency varies between people at the same load, and how reliably efficiency is measured. One point was checked by simulation. Such a study needs a direct efficiency measure (phosphocreatine recovery by magnetic resonance spectroscopy, or respirometry on a biopsy), a challenge test with a recovery time per person, and a load score from routine blood tests.

What the release already says

The capacity model predicts the slow pull weakens with age, as efficiency falls. In the stiffness tables a few measures stiffen with age and the rest resolve no trend; the challenge recovery in MIDUS shows no trend across age bands. This is weak evidence against the capacity model at the population level, because the cohort pull is a lower bound and survivors are selected; it does not settle the question.