By Deborah Borfitz
July 27, 2021 | Researchers with Singapore-based longevity startup Gero have used wearable sensor data and artificial intelligence to develop digital biomarkers of longevity, stress, and resilience. The novel GeroSense biological age acceleration model is being positioned as a practical, high-quality alternative to blood tests or genomic profiling for large-scale population health studies, according to Tim Pyrkov, Ph.D., head of mHealth R&D at Gero.
In a study recently published in Aging (DOI: 10.18632/aging.202816), GeroSense was found to be better than daily step counting at predicting life expectancy and just as good as blood-based testing when it came to forecasting the effects of lifestyle choices and prevalence of future disease. The model was trained and validated on steps-per-minute recordings collected by consumer wearable devices among 97,320 participants in the UK Biobank and 6,510 in the National Health and Nutrition Examination Survey of the U.S. Centers for Disease Control and Prevention, along with another 2,599 samples from longitudinal data obtained for smartphone and smartwatch users.
The number of steps walked per day has been associated with chronic diseases, future health, and risk of death, and the basis of the widespread recommendation to walk at least 10,000 steps a day, says Pyrkov, the paper’s first author. But step counts “do not tell you much about the actual health status of the fitness tracker users [or] if walking around more is better for you.” Step count is also an unreliable predictor of life expectancy across countries.
Among people working in various professional occupations, GeroSense did a better job of scoring life expectancy than tracking the “ebb and flow of the average physical activity levels” due to both social factors and work schedules influenced the predictions. This feature of the model might be useful in applications involving health risk assessment, including life insurance, says Pyrkov. Theoretically, GeroSense could also provide a model of biological age that is geography-agnostic.
Biological aging is more reflective of stress than one’s age based on birthdate, continues Pyrkov. As revealed by measurements taken continuously with wearable sensors, lifestyle as well as chronological age drive resilience (e.g., recovery time from acute infections, lack of sleep, or vigorous physical activity).
GeroSense was trained to predict the prevalence of at least one major age-related chronic disease, but factors actual age into its scoring methodology, he notes. Its output is a single dynamic variable that gets estimated once every seven days. Prediction of years of life gained or lost due to healthy or unhealthy lifestyle and nutrition choices is based on detected patterns associated with changes in life expectancy.
By averaging over longer motion tracks, GeroSense succeeded in improving the signal-to-noise ratio of existing activity-based trackers, Pyrkov says. Notably, after just a few months, a wristband signal could detect the effects of chronic diseases and smoking and predict the severity of seasonal infectious diseases (including COVID-19) at the accuracy level of blood-based tools.
GeroSense also appears to have outperformed measures of the average physical activity level, which dropped sharply when the pandemic struck, in predicting risk of infection, he adds. The researchers hypothesize that the effects of lockdown on morbidity risk may be smaller than what fitness apps alone would indicate.
The research team is hopeful GeroSense, available as a web-based protocol for integration in any mobile health and fitness application, will “democratize” the use of aging biomarkers in population health studies, says Pyrkov. The popularity of such apps grew by 46% during the COVID-19 lockdown.
It is currently impractical to conduct clinical trials of interventions for improving human lifespan because of the long wait for outcomes (inevitable death) and the exorbitant cost, he says. The best that can be done is see if a therapy changes aging risk based on objective biological signals.
For clinical trial purposes, the digital biomarkers answer the need for reliable and immediate feedback on health status changes in response to experimental treatments, Pyrkov says. Activity tracking used in clinical studies have been hampered by issues related to inaccuracies of sensor data, missing data, outliers, varying measurements between devices of different manufacturers, and seasonal variation of physical activity. The neural network architecture of GeroSense is specifically designed to resolve the missing data and transferability problems.
Gero is already partnering with two other longevity-focused companies—digital health platform AgelessRx, which is conducting longevity clinical trials, and Humanity Inc., which has added GeroSense to other clinical and genetic biomarkers in its consumer app for monitoring aging rate. The benefits of GeroSense here are in personalized self-monitoring over different periods of time, says Pyrkov, so users can learn which lifestyle choices are likely to gain them more years of healthy life expectancy.
Pyrkov imagines “smartphones in our pockets may become the center of longevity fitness for each person individually.” Greater adoption of GeroSense will also provide the company with additional longitudinal activity data to enhance the tool’s predictive capabilities. Collaborations with academic and industry researchers are welcomed and, ultimately, could include anti-aging drug discovery partnerships.