Gait Model Predicts Over 1,800 Health Measures

A self-supervised model trained on 3D skeletal motion data from 3,328 adults links walking patterns to a wide range of body-system health markers.

Updated on Nov. 9, 2026 in Diseases — General

Isometric editorial illustration of a jointed chain of matte blocks arcing across a cream field, in oxblood, teal, and slate, evoking measured body motion.
Researchers at an international team reported that a self-supervised model trained on walking data from 3,328 adults predicted 1,887 of 3,210 health measures. AI Illustration. Upload story photo >

Live Poll

Would you be comfortable with a camera analyzing your walk to screen for health conditions?

Researchers developed a gait foundation model using three-dimensional skeletal motion data from 3,328 deeply phenotyped adults recorded with a depth camera. The embeddings significantly predicted 1,887 of 3,210 phenotypic targets.

Why it matters

Gait has long been proposed as a broad indicator of health, yet most instrumented gait research has addressed narrow technical or condition-specific questions. A 2020 systematic review found that over 99% of 2,700 such studies did so, leaving gait's wider diagnostic potential largely unexplored.

The evidence comes from predictive modeling on the Human Phenotype Project cohort of 3,328 adults, with embeddings significantly predicting 1,887 of 3,210 targets across 18 body systems in men and 17 of 18 in women. The extraction does not report how these gains compare to baseline clinical models, nor whether they hold in populations outside this cohort.

The players

Human Phenotype Project

A deeply phenotyped adult cohort study that supplied the participant data used to train and validate the gait model.

The details

The model adapts masked autoencoding, a self-supervised learning method, to three-dimensional spatiotemporal skeletal motion data captured during five motor tasks. The embeddings improved prediction of clinical diagnoses and medication use in addition to the broader body-system targets.

Timeline

  1. 2009: A white paper established gait as a sixth vital sign.

  2. 2020: A systematic review of instrumented gait studies was published.

  3. 2022: The masked autoencoder method was established at CVPR.

  4. 2025: The Human Phenotype Project cohort article appeared in Nature Medicine.

  5. November 9, 2026: The gait foundation model study was published in Nature Medicine.

Big Picture

The 2009 white paper proposed gait as a sixth vital sign, but the field's instrumented studies stayed narrow, as a 2020 review of 2,700 studies showed. This gait foundation model extends that proposal by testing gait as a broad health signal across body systems.

If gait measures become clinically usable, they could offer a low-cost, non-invasive way to screen for conditions across body systems. The extraction does not indicate when such tools would reach routine clinical use.

The takeaway

Walking patterns may reveal far more about overall health than previously assumed, and broader gait-based screening could emerge as research matures. Readers should watch for validation studies beyond the original cohort.

Further reading

Explore more in Diseases — General, where gait-based health research connects to broader disease detection.

Source note: This article includes information reported by Nature.

Live Poll

Would you be comfortable with a camera analyzing your walk to screen for health conditions?