Introduction
Regular physical activity supports healthy aging, yet guidelines commonly follow a one-size-fits-all approach and pay limited attention to non‑cardiovascular fitness components such as balance, flexibility, and strength. This study introduces a novel home‑based platform that uses a smartphone’s sensors to remotely assess motor fitness and deliver AI‑generated, personalized exercise programs tailored to older adults’ unique needs. The goal is to determine whether remotely delivered, multicomponent exercise improves balance, flexibility, and strength more effectively than generic WHO guidelines or no intervention.
Methods
Design: A randomized controlled trial with three arms: (1) experimental group receiving personalized, AI‑driven exercises, (2) active-control group following World Health Organization (WHO) general exercise guidelines, and (3) control group with no intervention. The study took place in community settings in Israel between 2020 and 2023.
Participants: 239 community-dwelling adults aged 65+ completed the program (mean age about 73 years; 155 women). Eligibility included independent living, ability to walk unaided, Hebrew fluency, and basic smartphone competence. Exclusions included cognitive impairment, recent hospitalization for cardiac or neurological conditions, and high fall risk.
Assessments: Baseline, 4 weeks, 8 weeks, and 12 weeks included tests of static and dynamic balance, upper body flexibility, and upper and lower body strength. Balance was measured through multiple static and dynamic tests, while flexibility and strength were assessed via targeted arm and leg movements and sit-to-stand tasks. Digital markers were captured using a smartphone accelerometer and gyroscope, then transformed into standardized z‑scores for comparison.
Interventions: The experimental group received an 8‑week, five‑times‑per‑week, home‑based program that combined balance, flexibility, and strength exercises. The program used machine learning to tailor exercise difficulty (A or B) for each domain, based on the baseline motor fitness profile and ongoing performance. The WHO group followed recommended aerobic and multicomponent activities, while the control group maintained usual routines and received no intervention during the trial.
Outcomes and analysis: Primary outcomes included changes in balance, flexibility, and strength from baseline to 8 weeks. A mixed repeated measures ANOVA tested group-by-time interactions. Secondary analyses explored adherence and maintenance of effects at the 12‑week follow‑up. Power calculations anticipated 300 participants, but the final analyses included data from 189–231 participants across various tests to reflect attrition and data completeness.
Intervention Details
The personalized program was delivered via a study app on participants’ smartphones, with audiovisual instructions and remote data streaming to a central database. Exercises spanned three categories: (1) balance and lower body, (2) upper body flexibility, and (3) upper body strength. Two difficulty levels (A/B) were used to match the participant’s evolving fitness profile. Adherence was tracked by the app; post‑hoc analyses used a 1.5 sessions/week threshold to define adherers, a level at which meaningful improvements were observed in several fitness domains.
Key Findings
Compared with the active-control and control groups, the personalized exercise group showed greater improvements in several motor fitness components. Notably, dynamic balance (as measured by tandem walk tests) and upper‑body strength and flexibility improved more in adherers who followed the tailored program. Some balance gains emerged as early as 4 weeks, with continued improvement by 8 weeks. In contrast, static balance measures showed mixed results, suggesting that the platform’s precision is particularly strong for dynamic, functional balance tasks that translate to fall prevention in real-world settings.
Flexibility gains were evident mainly in arm movements, with modest changes in torso rotation. Arm strength and certain arm movements improved significantly among adherers, while lower body strength improvements were less pronounced, potentially due to testing methods and ceiling effects in sit-to-stand measures.
Discussion and Implications
The study demonstrates that a smartphone‑based platform can perform remote, detailed assessments of balance, flexibility, and strength and translate these data into personalized exercise programs for older adults. The approach offers several potential benefits: scalable delivery of multicomponent exercise, reduced need for in‑person supervision, and continual adaptation of training based on real‑time performance. While the observed effect sizes varied and some measures showed small changes, the intervention notably enhanced dynamic balance and upper body function—key components in reducing falls and maintaining independence.
Limitations include adherence variability, reliance on trained testers for initial testing in the study (with future work aiming for user‑led testing), potential sampling bias toward higher socioeconomic groups, and a limited set of exercise difficulty levels. Nevertheless, the technology shows promise for broader adoption, including populations with mobility or cognitive challenges, once customized interfaces and safety considerations are optimized.
Conclusion
In this randomized trial, a mobile, AI‑driven platform that remotely assesses motor fitness and delivers personalized, multicomponent exercise programs improved several key fitness domains in older adults more effectively than generic WHO guidelines or no intervention. The ability to deliver tailored exercise remotely through smartphones could transform health promotion for aging populations, enabling proactive maintenance of mobility, resilience, and independence in everyday life.
Future Directions
Future work should expand to diverse populations, enhance user‑friendliness for older adults with limited technology experience, and incorporate additional fitness components and longer follow‑ups. Integrating genetic, clinical, and behavioral data could further refine personalization, while larger trials would help establish robust clinical guidelines for smartphone‑driven, AI‑assisted physical activity in aging populations.
Keywords
older adults, smartphone sensors, AI‑generated exercise, remote assessment, balance, flexibility, strength, randomized controlled trial, multicomponent exercise
