Introduction: Rethinking Exercise for Healthy Aging
Regular physical activity remains a cornerstone of healthy aging, yet traditional guidelines often adopt a one-size-fits-all approach that may neglect important fitness components beyond cardiovascular endurance. Balance, flexibility, and strength are critical for maintaining independence and reducing fall risk in older adults, but there is limited guidance for accurately assessing these domains outside laboratory settings. This study introduces a smartphone-based platform that uses mobile sensors to remotely assess motor fitness and deliver AI-generated, personalized exercise programs tailored to each individual’s needs.
By integrating accessible technology with machine learning, the platform aims to extend personalized medicine into routine home-based exercise. The remote assessment relies on a suite of movement tests captured by a smartphone’s accelerometer and gyroscope, enabling assessment of static and dynamic balance, upper body flexibility, and upper- and lower-body strength. The resulting data feed a machine learning model that crafts a customized exercise plan delivered as video on the user’s phone, allowing older adults to engage with a high-quality, individualized program without visiting a clinic or gym.
Methods: A Randomized Controlled Framework
The trial randomized healthy community-dwelling adults aged 65 and older into three groups for an 8-week intervention: (1) personalized exercise (experimental), (2) general WHO-guideline exercise (active control), and (3) no-intervention control. Primary outcomes focused on motor fitness components: balance (static and dynamic), flexibility (torso rotation and arm movements), and strength (arm and sit-to-stand tests). Assessments occurred at baseline, 4 weeks, 8 weeks, and a 12-week follow-up.
Participants were recruited in Israel from diverse settings to reflect urban and rural living. Inclusion criteria emphasized independence in daily activities and smartphone competency, while exclusions included cognitive impairment or recent cardiovascular or neurological events. The randomization used a block design with stratification by age group and gender to minimize imbalance across arms.
The experimental arm received an AI-personalized program generated from each participant’s digital markers derived from smartphone tests. The program emphasized three target domains: balance and lower body, upper body flexibility, and upper body strength, with two difficulty levels (A and B). Participants were advised to perform the tailored program five times per week for eight weeks. The active control group followed WHO recommendations (150–300 minutes of aerobic activity per week plus balance and strength components), while the control group continued usual activities with no structured program.
Results: Efficacy of Personalization in Real-World Use
Analyses revealed a significant Group × Time interaction for dynamic balance, with the personalized-adherer subgroup showing meaningful improvements from baseline to the end of the intervention. Notably, improvements in arm flexion and extension, as well as overall arm strength, were more pronounced in adherers to the personalized program than in the active-control or control groups. Balance measures also favored adherers, particularly in tandem balance tasks and the composite balance score when compared with controls.
Flexibility showed notable gains in arm movements, especially on the right side, among those following the personalized regimen. Torso rotation did not demonstrate a detectable change, which may reflect measurement limitations with smartphone sensors rather than a true lack of physiological adaptation. In contrast, lower body strength (sit-to-stand) did not show a robust group difference, although some positive trends were observed in the personalized group.
Importantly, improvements tended to diminish during a 4-week follow-up without continued exercise, underscoring the need for ongoing engagement to sustain gains. The study also found that baseline habitual physical activity predicted adherence: more active individuals tended to remain engaged with the program, while those with higher sedentary time exhibited lower adherence after enrollment.
Discussion: Implications for Remote, AI-Driven Elder Care
The smartphone platform demonstrated that a multicomponent, remotely delivered exercise program—driven by objective remote assessments—can outperform generic WHO-guideline advice and no intervention in several motor fitness domains. The approach offers several practical advantages: it removes the need for laboratory testing or in-person supervision, provides continuous monitoring through digital markers, and enables ongoing personalization as a user’s fitness profile evolves.
While the findings are promising, several limitations are acknowledged. Adherence varied, and some sensors’ precision may differ across smartphone models. The sample largely consisted of cognitively intact, high-functioning older adults, which may limit generalizability to those with frailty, cognitive impairment, or multiple comorbidities. Future work should expand to broader populations and refine the AI algorithms to incorporate additional health and behavioral data, potentially enhancing precision further.
Conclusion: A Pathway to Scalable, Personal, Home-Based Exercise
This randomized trial supports the feasibility and effectiveness of a smartphone-based, AI-generated personalized exercise program for older adults. By remotely assessing motor fitness and tailoring exercises to the individual, the platform holds promise for scalable health promotion, improved intrinsic capacity, and reduced risk of functional decline in aging populations. Ongoing engagement and broader inclusion will be key to translating these gains into real-world, long-term health outcomes.
