Blood-based aging clocks: prediction target outweighs deep-learning model capacity
In an abstract on blood-chemistry aging clocks, higher model capacity barely improved out-of-sample age prediction and lowered test–retest reliability; the prediction target (chronological age versus mortality-derived phenotypic age) drove most of the mortality-related validity. [1]
Automatic summary · not editorially reviewed
The key findings
Six blood-chemistry aging clocks crossed three model capacities with two prediction targets, trained on NHANES III and externally validated in Continuous NHANES 1999–2010 linked to all-cause mortality. [1]
Higher capacity predicted chronological age slightly better in internal cross-validation but the advantage nearly vanished out of sample, and higher capacity lowered test–retest reliability. [1]
The prediction target, not model capacity, governed the mortality association of the age gap: switching target multiplied the hazard ratio by 1.29×, while moving from linear to deep multiplied it by 0.93×. [1]
The source at a glance
- Study type / format
- Abstract of a modeling study building and validating blood-based aging clocks
- Population / context
- Participants in NHANES III (training, n=9,200) and Continuous NHANES 1999–2010 (external validation, n=16,400) with National Death Index–linked all-cause mortality.
- Summary based on
- Study abstract
- Publication status
- Preprint · peer-review status not confirmed
- Original publication
- July 22, 2026
NHANES blood-chemistry clocks with mortality follow-up
The abstract describes six aging clocks built on routine blood chemistry in a fully crossed 2 × 3 design: three model capacities (elastic net, gradient-boosted trees, a deep multilayer perceptron) crossed with two prediction targets (chronological age versus a mortality-derived phenotypic age). NHANES III served as training data (n=9,200) and Continuous NHANES 1999–2010 as external validation (n=16,400), linked to National Death Index all-cause mortality with a median follow-up of 20.6 years and 4,368 validation deaths.
Supporting source [1]
Capacity helped in-sample, faded out-of-sample
Higher-capacity clocks predicted chronological age slightly better in internal cross-validation (deep multilayer perceptron mean absolute error 5.06 years versus elastic net 5.43 years). The advantage nearly vanished out of sample (both about 5.48 years), and higher capacity lowered test–retest reliability (intraclass correlation 0.970 versus 0.985).
Supporting source [1]
Target drove mortality validity of the age gap
Biomarker validity was defined as the mortality association of the age gap (age acceleration) beyond chronological age. Adjusted for chronological age and sex, the age-gap hazard ratio per standard deviation was 1.14–1.23 for chronological-age clocks and 1.48–1.59 for phenotypic clocks. Switching target multiplied the hazard ratio by 1.29× (95% CI 1.26–1.33), whereas moving from linear to deep multiplied it by 0.93×, i.e. no improvement. A two-way variance decomposition attributed 95% of between-clock variance in age-gap log-hazard-ratio to the target and 5% to capacity; the dissociation was stable across internal and external validation and within sex strata.
Supporting source [1]
How to interpret this
This is based on the abstract only; the full text was not reviewed. Design details, model specifications and any further analyses beyond those reported in the abstract are not assessed here. The findings concern blood-chemistry clocks within this described regime and are associations, not evidence of causation or of human lifespan benefit.
Based on the study abstract; the full paper, tables and supplementary data were not reviewed.
Source & supporting passages
Feng J. · europepmc.org · July 22, 2026