A gait foundation model for prediction of multisystem health phenotypes from 3D skeletal motion

Gait is increasingly recognized as a vital sign, yet current approaches treat it as a symptom of specific pathologies rather than a systemic biomarker.

Here we developed a gait foundation model for three-dimensional skeletal motion, which was recorded via a single-depth camera during 5 motor tasks, from 3,328 deeply phenotyped adults. Our purpose was to capture the full pattern of a person's movement and use it to predict health phenotypes such as grip strength and sleep quality across 18 body systems.

Gait is conventionally described by engineered features (for example, calculated measurements such as cadence and stride length), which may fail to capture the fine detail of how a person moves. Our model instead translates movement into learned embeddings: sets of numbers that describe how all body joints move together over time.

Learned embeddings outperformed engineered features for prediction of age (Pearson's r = 0.69), body mass index (r = 0.91) and visceral adipose tissue area (r = 0.83), and embeddings significantly predicted 1,887 of 3,210 tested phenotypes.

Relative to a baseline model using age, body mass index, visceral adipose tissue and height alone, gait embeddings improved prediction in all 18 systems in males and 17 of 18 in females, as well as prediction of clinical diagnoses and medication use.

Anatomical ablation analyses revealed that leg dynamics dominated metabolic and frailty predictions, and torso dynamics encoded sleep and lifestyle phenotypes. These findings establish gait as an independent, multisystem biosignal, motivating integration of gait analysis into routine clinical assessment as a scalable, passive vital sign.