Individual response to exercise training — a statistical perspective

methodological paper; no new sample

N/A — a statistical argument about how subject-by-training interaction must be estimated.

J Appl Physiol 118(12):1450-9. A methods paper, so grade D on design while being high-value on content: it is the argument Renwick 2024 later operationalised. Its central move is to treat "individual response" as a subject-by-training INTERACTION term that has to be separated from two confounding variance sources before it can be claimed to exist.

Two structural obstacles to the trainability claim. FIRST, measurement error limits the accuracy of any individual response estimate, and — the key asymmetry — "a low signal-to-noise ratio may not be compensated by increasing sample size (1 case)". You cannot power your way to an individual prediction. SECOND, and less appreciated: WITHIN-subject variation in training efficacy (the same person responding differently to the same training on different occasions) "may not be disclosed by comparison to a control group but calls for repeated interventions." So even the control-arm designs this database currently treats as the gold standard are NOT sufficient to establish stable trainability — you need the same person trained twice. The authors conclude that conclusive inference on subject-by-training interaction is beset with practical difficulties and that pragmatic alternatives are warranted.

Upstream of heritability entirely. If the trait "trainability" cannot be cleanly measured in an individual, a heritability estimate for it inherits that measurement problem — which is the technical reason HERITAGE's 47% is an upper bound rather than an estimate.