← Back to Home

How Much Health Information Does Gait Hold? AI Study Links Movement to Measures Across 18 Body Systems

A study of 3,328 adults found that movement patterns across the body's joints can add information about health differences that age and body fat data do not explain. This suggests movement analysis could become a new tool for clinical research, but prospective validation is still needed before it can be used to screen for or diagnose disease.

By SURL BioNews

Watching someone walk may reveal more than balance and muscle strength. Steps, trunk movement and joint coordination may also reflect differences in health, including metabolism and sleep. A study published online in Nature Medicine on November 6 used artificial intelligence to try to read these details: among 3,328 adults, movement data had predictive ability across multiple types of health measures, offering a research direction for low-burden health assessment.

The study used data from the Human Phenotype Project cohort, with a single depth camera recording three-dimensional skeletal movement as participants performed five movement tasks. According to information released by Newton Tech, the team analyzed 17,589 movement sequences, captured and processed using the company's Movement Lab system. The analysis examined skeletal dynamics during standardized tasks; it does not mean that a casually recorded everyday video can provide a health assessment.

Traditional gait analysis often reduces movement to measures such as cadence and stride length. This study's foundation model instead converted patterns of whole-body joints moving together over time into a numerical representation. The correlations between the model's predictions and measured age, body mass index (BMI) and visceral fat area were 0.69, 0.91 and 0.83, respectively, outperforming the traditional gait features compared in the study. These figures describe the degree of correlation between predicted and measured values and cannot be interpreted as disease diagnostic accuracy.

Of the 3,210 health phenotypes tested, the model showed statistically significant predictive ability for 1,887. More importantly, after establishing a baseline model using age, BMI, visceral fat and height, adding movement representations still improved predictions across all 18 body system categories in men and 17 of them in women. This indicates that movement provided information beyond body composition, but does not mean that every measure has reached a level of precision suitable for clinical use.

This information was not distributed evenly throughout the body. By removing information from different body regions, the researchers found that leg dynamics were more important for predictions related to metabolism and frailty, while trunk dynamics carried clues about sleep and lifestyle. This helps investigate what the model was detecting, but associations alone still cannot establish how a particular health condition causes specific changes in movement.

For clinical trials, a potential use is to retain more detail in functional assessments: examining changes in joint coordination and movement patterns alongside the time taken to complete a task. Newton Tech suggested that this type of analysis could potentially support participant stratification, exploratory endpoints and assessment of treatment responses. However, predicting existing health differences and reliably tracking drug effects are two questions that require separate validation.

The study's relationship with the equipment supplier also needs to be considered: the paper disclosed that three authors were employed by Newton VR Ltd., which developed the recording system, and two other authors served as paid consultants to Pheno.AI. The company's press release also explicitly stated that screening and diagnostic applications remain at the research stage and require validation through prospective studies and in more diverse populations. The findings so far identify a biological signal that could potentially complement health assessment; whether it can improve medical decisions remains a question for future research.

References

  1. Medical Xpress / Newton Tech
  2. PubMed / Nature Medicine
  3. Newton Tech press release via BrightSurf