An interventional study of Uniform random message delivery and Reinforcement learning message delivery in Mobile Health, Physical Activity and Exercise, sponsored by University of California, Berkeley. Completed at 1 site in United States. Open to participants aged 18 Years to 65 Years, including healthy volunteers. Per ClinicalTrials.gov, last updated 2020-06-24.
Sponsored by University of California, Berkeley · Not applicable, Interventional, and Prevention
Background: Insufficient physical activity is one of the leading risk factors of death worldwide. Behavioral treatments delivered via smartphone apps, hold great promise for helping people engage in healthy behaviors including becoming more physically active. However, similar to 'face-to-face' treatments, effects typically do not seem to be sustained over longer periods of time.
Methods: the investigators developed a smartphone application that uses different types of motivational and feedback text-messaging to motivate individuals to increase physical activity. Here, participants are randomized to either receive messages by a uniform random distribution (n=50), or chosen by a reinforcement learning algorithm (n=50), which learns from daily participant data to personalize the frequency and type of motivation of messages.
Objectives: In the current study, the investigators examine this application in undergraduate and graduate students at the University of California, Berkeley. The investigators compare whether participants in the uniform random or adaptive group have higher increases in steps during the study. The investigators also examine the effect of the different types of messages on step counts. Further the investigators assess the influence of patient characteristics, such as socio-demographic, psychological questionnaire scores and baseline physical activity on the effect of the adaptive arm and effectiveness of the messages. Finally, the investigators assess participant qualitative feedback on the text-messaging program, through feedback provided via questionnaires, text-message and phone interviews.
The investigators developed a smartphone application, the DIAMANTE app, that uses machine learning to generate adaptive text messages, learning from daily participant data to personalize the frequency and type of motivation of messages. In the current study, the investigators will compare this application in undergraduate and graduate students at the University of Berkeley, to text-messaging chosen randomly. This study will provide insight into the effectiveness of this smartphone application for increasing physical activity in university students. Further, it will provide preliminary knowledge on the working mechanisms and variables that moderate the effectiveness of the intervention.
This study is characterized by a factorial design with a total of 3 factors representing Motivational Messages (M), Feedback Messages (F) and the Time Frame (T) when the message was sent, of 4, 5 and 4 levels each, respectively. One level of M and F corresponded to a control treatment, i.e., no message sent. Each participant received one different combination of M, F and T every day.
Both the adaptive and uniform random group will receive the same types of messages: feedback (4 active categories plus no message) and motivation (3 active categories plus no message). However, the message categories, timing and frequency will be optimized by a reinforcement learning algorithm in the adaptive group, and will be delivered with equal probabilities in the uniform random group (following a uniform random distribution).
For the reinforcement learner group, the algorithm training data consists of the historical data of all participants (contextual variables), which include which messages were sent previously and within which time periods, and select clinical/demographic data (such as age, day of the week and depression scores) to improve prediction abilities. Subsequently, the message is chosen based on the predicted effectiveness of messages, combined with a sampling method. As such, it frequently picks out from the most rewarding messages and occasionally explores the messages with uncertainty in their reward.
The aims of this study are:
University of California, Berkeley is the lead sponsor of 106 studies on the registry; 18 are open to participants now.
Of its 5 completed or terminated interventional studies of FDA-regulated products, 1 (20%) have results posted.
Counted across the registry records on this site, refreshed daily.
Inclusion Criteria: We will include currently enrolled undergraduate and graduate students ages 18 to 65.
-
Exclusion Criteria: Students that do not have a smartphone, are not able to exercise due to disability, or have plans to leave the country during the 6 week study will be excluded.
-
In this arm the types of messages were sent out randomly, i.e. with a uniform random distribution.
Behavioral: Uniform random message delivery
In this arm the types of messages were chosen by a reinforcement learning algorithm. The decision about which message to send was based on several contextual variables, including data for the pedometer app, and consecutive days since messages from different categories were sent.
Behavioral: Reinforcement learning message delivery
The uniform random intervention group receives feedback and motivational messages chosen from the messaging banks with equal probabilities.
The adaptive intervention group receives messages chosen from the messaging banks by a reinforcement learning algorithm.
Steps (measured by phone pedometer)
Change in daily step counts (today's steps count minus yesterday's steps count)
Time frame: 24 hours (measured for a period of 6 weeks)
Steps (measured by phone pedometer)
Mean change in daily step counts during the course of the study
Time frame: Change from baseline to 6 week follow-up
Depression scores
Patient Health Questionnaire 9 item (PHQ-9). The PHQ-9 has scores from 0 to 27. Higher scores mean a worse outcome.
Time frame: Change from baseline to 6 week follow-up
Anxiety scores
General Anxiety Disorder 7 item (GAD-7). The GAD-7 has scores from 0 to 21. Higher scores mean a worse outcome.
Time frame: Change from baseline to 6 week follow-up
Behavioral Activation
Behavioral Activation for Depression Scale - Short Form (BADS-SF). The BADS-SF has scores from 0-54. Higher scores mean better outcomes.
Time frame: Change from baseline to 6 week follow-up
Plan to share: Yes — Individual participant data that underlies the results reported in the articles will be made available to researchers on request after deidentification.
Supporting information: Study protocol, Sap, Analytic code
This study is completed, as verified in Jun 2020. You cannot join it, but the record below documents what was studied.
Get an email when the registry record changes — status, dates, results — or when someone posts here.
Sign in to followQuestions and observations about this study, from anyone following it. Not medical advice, and not a channel to the study team — their contact details are on the registry record.
Sign in to join the discussion. Reading takes no account; posting does. You choose a display name, and a pseudonym is the default.
Nothing here yet. If you are running this trial, taking part in it, or weighing whether to, this is the place to say so.
University of California, Berkeley