An Evaluation of the Effect of App-Based Exercise Prescription Using RL on Satisfaction and Exercise Intensity

Summary

The PERFORM-RL study (Personalised Exercise Prescription for Remote Fitness Using Reinforcement Learning) was a 12-week randomised crossover trial designed to evaluate the effectiveness of an app-based exercise prescription system powered by reinforcement learning (RL). The study aimed to investigate whether exercise sessions tailored by RL would lead to greater user satisfaction and higher exercise intensity compared to generic, non-personalised exercise sessions.

The trial enrolled 62 participants (27 males, 42 females; mean age 42 years) who were randomly assigned to alternate between two conditions: an RL-driven intervention, which personalised exercise sessions based on user preferences and feedback, and a control condition with non-tailored, generic exercise sessions. Participants were instructed to complete three exercise sessions per week using the i80 BPM app, which offered a variety of video-guided exercises. The RL model customised these sessions based on user feedback, including satisfaction and perceived intensity, with the goal of optimising future sessions.

The primary outcome was user satisfaction, measured via the Physical Activity Enjoyment Scale (PACES-8) after each session. Secondary outcomes included exercise intensity, as assessed by the Borg Rating of Perceived Exertion (RPE) scale, and heart rate data collected through a Samsung Galaxy Fit 2 smartwatch.

The trial was conducted in Dublin, Ireland, and approved by the UCD Human Research Ethics Committee (LS-21-34-Tragos-Lawlor). Participants provided informed consent and were blinded to their group allocation. The trial was not registered prospectively, but steps are being taken for retrospective registration.

Conditions

• Exercise
• Mobile Applications

Interventions

DEVICE

Reinforcement Learning (RL) Personalised Exercise Prescription via i80 BPM App

This intervention involved the use of a smartphone app, i80 BPM, which delivered personalised exercise prescriptions using a reinforcement learning (RL) model. The RL algorithm tailored the exercise sessions by adapting variables such as intensity, duration, and exercise type based on individual user preferences, real-time feedback, and performance data. This dynamic personalisation was designed to enhance user satisfaction and engagement over the 12-week study period. Participants completed three exercise sessions per week.

DEVICE

Generic Non-Personalised Exercise Prescription via i80 BPM App

This intervention used the same i80 BPM smartphone app to deliver generic, pre-designed exercise sessions that did not adapt based on user preferences or feedback. The exercise sessions were standardised for all participants, with no customisation. The control arm served as a comparator to evaluate the impact of personalised, RL-driven exercise prescriptions. Participants completed three exercise sessions per week.

Sponsors & Collaborators

• University College Dublin

  lead OTHER

Study Design

Allocation

RANDOMIZED

Purpose

TREATMENT

Masking

DOUBLE

Model

CROSSOVER

Eligibility

Min Age

18 Years

Max Age

65 Years

Sex

ALL

Healthy Volunteers

Yes

Timeline & Regulatory

Start

2022-09-01

Primary Completion

2022-11-16

Completion

2022-11-30

Countries

• Ireland

Study Locations