Simple physical function tests may identify adults at mobility risk

By Dr. Liji Thomas, MDReviewed by Lauren HardakerMay 15 2026

A recent study published in the journal JMIR Aging reported the development of a predictive model for home-based self-assessment to identify early mobility limitations (EMLs).

Aging populations face rising mobility-related health burdens

The steadily aging global population poses a public health challenge due to the high prevalence of age-related medical conditions, including the onset of limited mobility. Declining mobility makes activities of daily living (ADL) and social participation difficult. It also predisposes to the loss of independence, reduced quality of life, and soaring healthcare costs. In addition, it often precedes the onset of loss-of-function.

Early mobility decline links frailty and sarcopenia

EMLs occur at the intersection of sarcopenia (the generalized loss of muscle mass) and physical frailty. The ability to identify EMLs and provide timely interventions could help preserve mobility throughout old age. Mobility decline can be addressed by improving nutritional intake and encouraging physical activity within the EML window of opportunity.

Taken together, these factors indicate the importance of mobility limitation as an interventional research target. The first step is to identify EML, which, by definition, is often subtle and may not yet be clearly apparent on routine clinical assessment.

Existing predictive models for EML are based on physical performance data from older adults, but these measures are relatively insensitive in younger adults. One key to improving predictive performance could be incorporating muscle power decline, which sets in earlier and faster than loss of muscle strength and could possibly have greater predictive value for EMLs.

Integrating specific dietary factors could also sharpen the association between physical performance, specifically muscle power, and EMLs in mid-life and beyond.

Simple home-based mobility risk measures 

The current study used a set of modifiable lifestyle factors and a few routinely obtainable parameters.

The study was conducted on 1,344 community-dwelling adults aged 45 years or more, forming part of the Guangzhou Nutrition and Health Study. The median age at baseline was 62 years, with 70 % women.

All participants were healthy, nonfrail adults reporting no limitations in activities of daily living (ADL) at baseline.

At six years, participants were assessed again for physical performance using two parameters: walking speed and handgrip strength. Reported difficulty in walking or climbing stairs was also considered indicative of EML.

Using machine learning, a set of six models was trained and then tested for predictive performance of EMLs.

Age, BMI, and diet strongly predicted mobility decline

The analysis showed that among the total number of participants, 206 (15.3 %) developed EMLs after a median follow-up of 6.67 years. Participants with EMLs were older, with a higher body mass index (BMI), a lower Mediterranean diet score, and poorer sit-to-stand (STS) test scores. They also had lower estimated muscle power based on the baseline STS.

In the final models, six predictive factors were scored: age, sex, BMI, alternate Mediterranean diet score, STS power, and dietary calcium intake. Four of the six models showed modest to acceptable discriminative ability for EML identification.

There was not much to choose between the models in terms of performance, but logistic regression and LASSO models were easier to interpret and less likely to overfit.

Further analysis showed that the most important predictors of EMLs were older age, lower adherence to the Mediterranean diet, lower (baseline STS-based) estimated muscle power, and higher baseline BMI.

Modest model performance may still support prevention

A larger study recently showed good predictive ability of similar models for frailty among older adults in Canada, which aligns with the results of the current study. 

The authors also noted that predicting subtle mobility decline in relatively healthy adults is inherently difficult, which may partly explain the models’ modest predictive accuracy.

Limitations of the models

However, clinical utility was observed only for the LASSO and logistic regression models trained on unbalanced datasets, with fewer EML cases, and only at relatively low intervention thresholds of 0 %-35 %. Attempts to improve performance using balanced datasets with bootstrapping and resampling techniques did not provide additional net benefit.

The researchers also found that the models generally produced overconfident predictions, highlighting the need for further calibration before broader use. Moreover, most machine learning approaches did not meaningfully improve the ability to distinguish between individuals at higher or lower EML risk compared with standard logistic regression, except for a modest gain with the random forest model.

The authors note that these findings should also be interpreted in light of the relatively low number of observed EML cases within the cohort, which likely reduced overall predictive accuracy.

STS-based muscle power estimation

The study cohort appeared to have optimal physical function as estimated by the STS data, irrespective of EML risk.

In addition to established risk factors such as advanced age and higher BMI, the study sheds new light on the role of STS-based estimated muscle power and overall diet quality in predicting EML. Notably, this study found STS-based muscle power to be more significant than the number of STS repetitions, agreeing with earlier observations of its clinical relevance to physical function and sarcopenia in the elderly.

Higher alternate Mediterranean diet scores were associated with lower EML risk, perhaps reflecting known links between these diets, skeletal muscle mass, mobility, and less frailty.

External validation is still needed before wider use

Despite selecting a healthy community-dwelling cohort and using simple criteria, the study has some limitations. Lacking a standardized EML measure, the researchers used both patient-reported and physical performance to identify this outcome.

A low positive predictive value (PPV) was another limitation, but should be interpreted in light of the low prevalence of EML, which may in part be due to the loss to follow-up of over 50 % of participants. The PPV rose markedly when the prevalence in China was incorporated, suggesting the model’s greater utility in the appropriate setting.

No external validation of the models was conducted, and the dataset was unbalanced, which could bias the training performance.

Simple home tools may detect mobility decline earlier

The authors present this as among the first studies to develop a predictive model for EMLs in Chinese adults aged 45 years or more, based on easily obtained home-based demographic, nutritional, and physical performance measures.

The authors comment, “Given the models’ intended use for early identification of individuals who may benefit from low-risk lifestyle interventions, their limited discrimination remains potentially valuable.”

In particular, they emphasize the home-based collection of data on modifiable risk factors, including diet quality, micronutrient content, and physical function, requiring only a chair, a stopwatch, and a short diet questionnaire. However, the authors also stress that the models still require external validation and further calibration before routine implementation in real-world settings.

Once validated, these findings could help identify individuals with an estimated EML risk of 10 %-20 % who might benefit from targeted low-risk lifestyle interventions.

Download your PDF copy by clicking here.