This is an open-access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR mhealth and uhealth, is properly cited. The complete bibliographic information, a link to the original publication on http://mhealth.jmir.org/, as well as this copyright and license information must be included.
Health education and behavior change programs targeting specific risk factors have demonstrated their effectiveness in reducing the development of future diseases. Alzheimer disease (AD) shares many of the same risk factors, most of which can be addressed via behavior change. It is therefore theorized that a behavior change intervention targeting these risk factors would likely result in favorable rates of AD prevention.
The objective of this study was to reduce the future risk of developing AD, while in the short term promoting vascular health, through behavior change.
The study was an interventional randomized controlled trial consisting of subjects who were randomly assigned into either treatment (n=102) or control group (n=42). Outcome measures included various blood-based biomarkers, anthropometric measures, and behaviors related to AD risk. The treatment group was provided with a bespoke “Gray Matters” mobile phone app designed to encourage and facilitate behavior change. The app presented evidence-based educational material relating to AD risk and prevention strategies, facilitated self-reporting of behaviors across 6 behavioral domains, and presented feedback on the user’s performance, calculated from reported behaviors against recommended guidelines.
This paper explores the rationale for a mobile phone–led intervention and details the app’s effect on behavior change and subsequent clinical outcomes. Via the app, the average participant submitted 7.3 (SD 3.2) behavioral logs/day (n=122,719). Analysis of these logs against primary outcome measures revealed that participants who improved their high-density lipoprotein cholesterol levels during the study duration answered a statistically significant higher number of questions per day (mean 8.30, SD 2.29) than those with no improvement (mean 6.52, SD 3.612),
In this study, the ubiquitous nature of the mobile phone excelled as a delivery platform for the intervention, enabling the dissemination of educational intervention material while simultaneously monitoring and encouraging positive behavior change, resulting in desirable clinical effects. Sustained effort to maintain the achieved behaviors is expected to mitigate future AD risk.
ClinicalTrails.gov NCT02290912; https://clinicaltrials.gov/ct2/show/NCT02290912 (Archived by WebCite at http://www.webcitation.org/6ictUEwnm)
Health education programs have demonstrated their effectiveness in educating individuals with targeted knowledge relating to risk factors of various diseases [
Unfortunately, efforts to create a vaccine for AD have proven unsuccessful. Nevertheless, findings from clinical and epidemiological studies have suggested that behavioral, social, and environmental factors may delay or prevent the onset of AD [
The health education interventions that individually targeted such factors for other conditions exhibited positive results, suggesting that a similar effort targeting AD would be likely to result in the desirable adoption of healthy behaviors [
It is therefore hypothesized that a health education program that provides evidence-based information regarding AD risk factors and prevention methods may have the additional benefit of reducing risk for other health conditions, with particular improvement in cardiovascular functions.
Although the risk factors for cognitive decline and AD have been identified and are natural targets for a behavior change intervention, there is variation in the literature as to when such interventions should take place. It has become widely accepted that the neuropathological processes involved in AD begin decades before symptoms emerge [
It is therefore further hypothesized that an intervention targeted at those in midlife holds the greatest potential for reducing future risk of developing AD.
To appropriately distribute an education-based behavioral intervention program, a suitable method of delivery is required. This paper will describe the numerous empowering roles, for both end users and investigators, that the mobile phone has facilitated during an evidence-based multi-domain behavior change intervention, entitled “The Gray Matters study,” which aims to reduce the future risk of developing AD, while in the short term promoting cardiovascular health (ClinicalTrials.gov identifier: NCT02290912) [
This section details related works in the areas of behavior change interventions and public education programs, covering how intervention material is typically delivered to users, how engagement is maintained, and how various behaviors are tracked. This study also investigated the current use and potential unexplored capabilities of technology in each of these areas.
The term
There are a number of ways in which an individual's behavior change can be theoretically modeled [
This type of approach aims to empower an individual via education and introspection to create a positive feedback loop as behaviors are changed over time [
Using such an approach in a health care intervention would rely heavily on the ability to personalize the education material and the ability to set attainable goals. To achieve such an intervention in the physical world, using people, buildings, and paper, would require enormous resources and planning. Fortunately, Internet-based technologies can reduce this burden, by digitally delivering intervention material.
Mobile and Internet-based technologies have been accepted as suitable and sustainable methods to deliver intervention material in various studies. Mobile phone–based delivery, such as short message service (SMS), has been used extensively and successfully in the literature to support portable and widespread interventions [
Stages of behavior change within the Transtheoretical Model.
Evidence from Internet-based interventions suggests that repeated visits are necessary to achieve sustainable change [
Gamification is the application of game design techniques and mechanics to nongaming domains [
For health-conscious adults, commercially available mobile phone apps and activity tracker companies, such as Strava, Fitbit, and Nike, use gamification elements extensively in their efforts to maintain and promote continual engagement. Although each platform has its own approach, they all record health-related data; examples include monitoring physical activity levels, tracking meals, and monitoring sleep quality. From these data, various performance metrics are calculated from which achievements are rewarded, such as badges and trophies. In addition, a user can view, typically at a high level via interactive graphs, their performances across time, allowing them to become informed of their behaviors and their resulting outcomes. Social sharing of recorded data also plays a role in enabling gamification elements, such as leaderboards, allowing users to compare their efforts with those of others. Apple and Google, whose mobile phone platforms combined account for 96.3% of the worldwide market share [
It is therefore hypothesized that the combination of constantly accessible, highly interactive, and individually tailored feedback, combined with gamification elements, such as rewards and leaderboards, would have the largest opportunity to maintain and encourage engagement with adults in a behavior change study [
To accurately assess the effect of a behavior change intervention, the validity of the reported behaviors must be accurate. There are numerous methods by which behaviors can be recorded within an intervention, including diaries, questionnaires, direct observation, and by proxy reports [
Diaries present a low-cost, easily maintained, and time-efficient method of recording behaviors; however, they are open to cognitive bias due to subjective self-assessment and rely heavily on the person’s ability to accurately recall past events [
Direct observation offers health investigators an accurate portrayal of behaviors within the given window of observation [
Self-reported questionnaires are commonly used in large-scale longitudinal studies because of their uniformity in questioning, repeatability, and ability to extract qualitative and quantitative information [
Proxy reporting is typically used when the subject in examination is somehow dependent on another adult, such as young children and the elderly. A study assessing the level of agreement between 6425 children and their parents regarding dietary, physical, and sedentary behaviors reported a mean agreement rate of 43% [
Although a variety of approaches can be employed to record behaviors, each has its own distinctive weaknesses relating to accuracy, repeatability, scalability, and cost [
The widespread public adoption of mobile phones, smartwatches, and wearable technology has enabled computing to become truly pervasive. Wireless digital devices can enable the digitization of individuals’ behaviors, often without the need for interaction. Wearable wrist-worn devices can be used to calculate an individual’s energy expenditure and step count [
To leverage this opportunity, the Gray Matters study has designed a clinically focused, technology-driven intervention program. An interdisciplinary team of computer scientists, biomedical engineers, mathematicians, psychologists, gerontologists, epidemiologists, and statisticians designed the Gray Matters mobile phone app: an app intended to deliver health education material, promote and monitor behavior change, and encourage the motivations of the participants via gamification elements [
This section details the study design and the technical development of the Gray Matters app, including study components, participant recruitment, eligibility criteria, outcome measures, and procedures.
The study was a randomized controlled trial (RCT) consisting of 144 subjects who were randomly assigned to either treatment or control group. The treatment group was not given a strict regimen and therefore a wide range of engagement levels were anticipated. A uniform random number generator (0,1) within SPSS v21 was used to randomize participants to treatment and control groups, with the aim of allocating 1/3 to control and 2/3 to treatment. The rationale for a 2:1 ratio for treatment and control was in consideration of the full autonomy given to each participant in the study. On recruitment, each participant was asked which behavioral domain or domains were of greatest interest to him or her to improve upon. In order to have a reasonably good power to study both the change in individual behavioral domains and its effects on those who wished to improve on particular domains, the ratio was adjusted to accommodate this. The intervention was delivered over a 6-month period, commencing in April 2014, with posttest collection performed at the close of the trial.
Recruitment of participants was achieved by emailing announcements to faculty, alumni, and staff of Utah State University and distributing flyers at health fairs and other venues, assisted by the local health department and their community liaisons. For those interested a prescreening eligibility survey was completed. Eligibility criteria included (1) age between 40 and 64 years, (2) BMI no higher than 41, (3) possession of a mobile phone or tablet (iOS or Android), (4) fluency in the English language, (5) residence in Cache County, and (6) not having any of the following medical conditions: pregnancy, dementia, unmanaged diabetes, or untreated major depression.
To achieve 80% statistical power to detect a medium effect size (Cohen’s
Primary outcome measures of the trial registration included a set of anthropometric measures, blood-based biomarkers, objective cognitive testing, and behavior in targeted domains. Secondary outcome measures included metacognition, motivation, readiness for change, sleep quality, social engagement, depression, and couple satisfaction (among married persons). Tables containing full summaries of all recorded values at the beginning of the study, for all 146 Gray Matters study participants, can be found in the study by Norton et al [
This subsection details the design of the Gray Matters mobile phone app and accompanying educational material, the development of the systems to support the collection of behavior data, and the method of deployment to the cohort within the Gray Matters study.
To enable the dissemination of evidence-based educational material relating to AD risk and prevention strategies, more than 130 peer-reviewed journals and papers relating to AD risk were analyzed. From the analysis, it was identified that risk factors and their prevention methods could be categorized into 6 domains: food, physical, cognitive, social, sleep, and stress. For these 6 domains, fact and suggestion pairs were produced (hereafter referred to as daily facts). An example daily fact from the food domain is as follows: “Consuming high amounts of processed foods is related to cognitive decline”; “Try a fresh salad for dinner instead of something from a box”. In total 164 succinct daily facts were produced across the 6 domains: physical (23), food (66), social (27), sleep (14), cognitive (24), and stress (10).
In addition to the daily facts, questions were designed for each domain to capture behaviors relevant to AD risk. All questions were quantitative in nature; however, they contained a mixture of subjective and objective questions. For example, a user may be asked to report the number of fruits and vegetables they consumed in a day (objective) and also rate their quality of sleep on a scale of 0-10 (subjective). In addition to the questions for the original 6 behavioral domains, a question was added to collect the activity data observed via a wearable device. In total 12 questions were designed for the domains: physical (2), food (3), social (1), sleep (1), cognitive (2), stress (2), and wearable activity monitor (1). For each question, a recommended value was extracted from external sources, such as the World Health Organization, the American Heart Association, the National Institutes of Health, and the Centers for Disease Control and Prevention’s (CDC) recommended daily targets (see
The questions presented to the user, showing their minimum, maximum, and recommended values.
Domain | IDa | Question | Minb | Maxc | Recommended (source) | Type |
Cognitive | 1 | How many minutes did you spend today doing “novel mental exercises”? | 0 | 120 | 30 minutes (NIHd) | Objective |
Cognitive | 2 | How many minutes did you spend today doing “cognitively stimulating activities”? | 0 | 120 | 30 minutes (NIH) | Objective |
Food | 3 | How many cups of fruits and vegetables did you eat today? | 0 | 10 | 5 cups (CDCe) | Objective |
Food | 4 | How many ounces of whole grains did you eat today? | 0 | 10 | 3 ounces (CDC) | Objective |
Food | 5 | How many servings of nuts, seeds, or legumes did you eat today? | 0 | 5 | 1 serving (CDC) | Objective |
Physical | 6 | How many minutes of “moderate” physical activity did you do today? | 0 | 60 | 30 minutes (AHAf) | Objective |
Physical | 7 | How many minutes of “vigorous” physical activity did you do today? | 0 | 60 | 20 minutes (AHA) | Objective |
Sleep | 8 | How would you rate your sleep promotion efforts over the past 24 hours? | 0 | 5 | 5 | Subjective |
Social | 9 | How would you rate your social engagement in the last 24 hours? | 0 | 7 | 7 | Subjective |
Stress | 10 | How much effort have you put into decreasing your stress over the past 24 hours? | 0 | 10 | 10 | Subjective |
Stress | 11 | On a scale of 1-10 how would you rate your stress level over the past 24 hours? | 1 | 10 | 1 | Subjective |
Wearable | 12 | How many Nike Fuelpoints did you earn today? | 0 | 5000 | 2000 (Nike) | Objective |
a ID: identification.
b Min: minimum.
c Max: maximum.
d NIH: National Institutes of Health.
e CDC: Centers for Disease Control and Prevention.
f AHA: American Heart Association.
The mobile phone app was developed natively for both Apple iOS and Google Android mobile phones. The decision to develop for both platforms was made based on rudimentary market analysis of mobile phone sales within the intended cohort’s location (Cache County, Utah, USA). Initially the app was developed for iOS 7.x devices, including iPhone and iPad, as the analysis showed a favoring for these devices in the area. As technology screening during the recruitment phase progressed, additional demand appeared for an Android version, which was subsequently developed. The functionality and visual layout of both versions are virtually indistinguishable, yet allowing enough flexibility to adhere to each platform's user interface design guidelines [
As the primary method to deliver health education material and track behavior change in the study, the app was designed to fulfill the following core functions:
1. Presentation of educational material relating to AD risk and prevention strategies.
2. Facilitation and recording of behavior self-reporting.
3. Calculation and presentation of personalized feedback based on reported behaviors.
Each function was presented in the user interface as a tab in the aforementioned order, allowing for easy and logical navigation. For the end user, the functions are displayed as the Tips tab, Log tab, and Performance tab.
This tab displays the evidence-based daily facts regarding risk factors and preventative strategies. The tab also contains a sports coach avatar, designed to aid visual delivery and personification of the recommendations offered in the daily fact (refer to
This tab facilitates the collection of behavioral data via self-reporting. As seen in
In order to reduce subjectivity in the questions, a user who is unsure about the exact meaning of the question may tap on it to present an expanded and elaborated phrasing of the question, including examples. For example, the question “How many minutes of moderate physical activity did you do today?” may be considered subjective if the term “moderate” is not understood. To counter this, tapping on the question presents the description “The CDC recommends 2 hours 30 minutes of ‘moderate’ activity per week. Examples of moderate activity are walking, skiing, raking leaves, washing the car.”
By answering each question, the users can longitudinally track their behaviors across all 6 domains, including their wearable device metrics. Answering all 12 questions is not compulsory; however, it is advantageous for both the participants and the study investigators as it increases the granularity of the data for each user and the study cohort as a whole. Answers are uploaded to a remote server via http protocols, using the open-standard JSON format to package the data.
Tips tab main screen showing daily fact (fact and suggestion pair; left). Evidence-based literature reference and link are displayed when the fact section of the daily fact is tapped (right).
Log tab main screen showing list of questions ordered by domain (left). The sliders can be dragged to adjust logged amount. Dialog box containing additional information displayed for a question regarding servings of nuts, seeds, and legumes (right).
The performance tab is designed to present various summaries of the data collected from the log tab, while encouraging continual participation via rewards. The main mode of presentation is via star ratings (refer to
In the equation,
The stars are designed to encourage and reinforce a participant’s effort to change his or her behavior. Because all domains can be viewed on screen at the same time, it provides a fast method to deliver visual feedback on the domains that require more effort and those that are under control. Users may also tap on each domain to receive additional pertinent information and an additional graphical representation of their efforts. The users may also view their performance aggregated across the previous 7 days in the form of a spider diagram or a bar chart. Again, this serves to visually assist the participants in understanding their behaviors for the purpose of self-affirmation.
Performance tab showing star ratings for each domain, calculated by assessing a user’s reported values against recommended values (left). Expanded view of performance with a domain, containing additional information and helpful tips (center). A bar chart showing aggregated performance, as a percentage, from the previous week’s data (right).
Equation developed by the author to normalize users performance metrics to a 5 star rating for visual representation.
Participant data from the app are uploaded to a remote MySQL server located at Ulster University. This occurs in real time if a user has a valid Internet connection, via Wi-Fi or mobile network. This instant transmission of behavioral data offers health investigators in the study an opportunity to perform immediate analysis, at any given point during the intervention. Because the data are in a structured digital format, very little human processing or interaction is required to run queries or statistical analysis. This presents a huge advantage over studies that control their data collection and processing via paper-based postal services and questionnaires [
To exploit the opportunity and increase the granularity of available data, the app also monitors all in-app actions using proprietary and open-source analytical tools. These analytical data enable the investigators to examine the profile of the average user and provide insight into how the app is actually being used. Examination of the analytical tracking data also highlights features that fulfill their purpose, while also identifying problematic areas of the app, flagging them to be addressed in future updates. Components of the app that contain analytical tracking code include app launching, tab navigation, updating log values, changing notification times, question detail expansion, and performance analysis.
In addition to the aforementioned mobile phone app, participants in the treatment group had access to a number of components to encourage behavior change. These included a wrist-worn activity monitor, booster events, a personal coach, and a study website.
Each participant was given a Nike FuelBand SE activity monitor. This device is worn on the wrist and serves to collect information such as steps taken, stairs climbed, and minutes of activity. This information is then consolidated into Nike’s proprietary metric of “NikeFuel points.” This device not only serves to collect data, but also acts as a physical reminder and motivator to increase levels of activity. Participants were asked to manually enter their total number of NikeFuel points earned at the end of each day via the mobile phone’s log tab.
All participants had the option of attending organized booster events. Each booster event was designed to emphasize the link between a specific domain and the risk it posed to developing AD, accompanied by preventative measures that the participants could apply in their daily lives. For example, a booster event that focused on the food domain hosted cooking classes that promoted sustainable healthy eating choices, while educating attendees about the link between the ingredients and AD risk. In total 46 booster events were organized and delivered across the 6-month intervention period.
Participants also had access to a personal coach whom they could contact if they required assistance with any aspect of the behavioral domains. A team of 28 student interns with majors in the 6 behavioral domains volunteered to be personal coaches. Student coaches were trained in motivational interviewing and the TTM and provided a weekly email or text message exchange with their assigned participants to provide emotional support and encouragement for lifestyle change goals.
Participants also had access to a password-protected website [
An exit survey was designed to capture opinions of participants in the treatment group. The survey asked questions about app usage, motivations, their perceived behavior change, and social network usage. At the end of the study, 102 of the 104 participants completed this survey.
This section presents the results from the RCT, including analysis of the treatments group’s adoption, typical usage, and perceptions of the app. This section also examines the app’s observed effects within the clinical and behavioral domains.
In week 1 (April 10, 2014), the first iOS version of the app was released to the treatment group. This was performed through a launch event, in which attending participants were instructed how to sign up and download the app through the TestFlight platform. TestFlight is a platform by which developers can distribute apps to internal or external testers. This platform allowed the investigators to control visibility in the app marketplace, ensuring that only enrolled participants could see and install the app. By the end of week 1, a total of 31.7% (33/104) participants had installed the app on their iPhone and/or iPad. In week 3 (May 13, 2014), the first Android version of the app was released to the treatment group because of demand from Android users. Two weeks after this release, 19.2% (20/104) participants had installed the first version of the Android app. By week 10, a total of 86.5% (90/104) of participants from the treatment group had installed an iOS or Android version of the app on their mobile phone and/or tablet, with the remainder shortly afterward. Many users opted to install the Gray Matters app on both their mobile phone and tablet. Of the 104 users using the app, at the end of the study, 75.97% of all Gray Matters app sessions were on iOS devices (iPhone: 54.7%; iPad: 21.27%) and the remainder on Android devices (24.03%). Regarding self-reporting of behaviors, the average user answered 7.3 (standard deviation, SD 3.16) questions per day during their participation in the study. The average duration of each session with the app, across all devices, was 1 minute 55 seconds. This time is less than the originally specified goal of 2 minutes for a user’s session duration. For further information on app usage statistics during the initial 10 weeks of the study please refer to the study by Hartin et al [
The app was distributed with two default notification times. The first notification was issued in the morning at 8 am by default, which reminded the users to check their daily fact every day. The second notification was issued at 6 pm by default, which reminded the user to complete the questions in the app’s log tab. Analysis of app usage times (
Bar graph showing the mean number of times each domain’s questions were edited using the sliders in the log screen, using updated analytical code, from week 18 to study end. The wearable domain is updated almost twice as often as the other domains.
Bar graph showing the typical hours of use, with morning activity around the default daily fact notification time at 8 AM and activity peaking at 10 PM, 4 hours after the default log notification time.
Upon the close of the study, an exit survey was issued to those in the treatment group. A total of 41 participants completed the survey. The survey acted to gather users’ motivations for behavior change and thoughts on the various components of the study, how they used them, and where they felt improvements could be made. First, users were asked how often they used the app (
Respondents’ answers to survey question: “Over the six month Gray Matters intervention period (April 2014 – October 2014), how often did you use the App?”
Usage | N | Mean | Standard deviation |
Months used | 39 | 5.54 | 1.315 |
Days per week | 38 | 6.21 | 1.695 |
Times per day | 38 | 1.66 | 3.122 |
In addition, the survey acted to glean how the app altered motivations toward various parts of the intervention. The survey also revealed that the app motivated users to perform physical activity (never: 14.6%, rarely: 12.2%, sometimes: 24.4%, often: 31.7%, and all of the time: 17.1%) and make healthier food choices (never: 12.5%, rarely: 2.5%, sometimes: 17.1%, often: 48.8%, and all of the time: 17.1%). When queried about their past, current, and future behaviors, 46.3% said they definitely would continue with their physical activity changes and 31.7% wanted to continue and increase their activity; 46.3% wished to continue their improved eating habits, with 29.3% wanting to continue and improve. When asked if they would continue using the app, 46.3% said they would not, 29.3% said they likely would not, and 24.4% said they would continue.
In addition, users were asked about features that they wished were included in the app. A total of 68.3% of users wished that guidelines were based on their “current” health status, 34.1% wished they could set their own target goals, 53.7% wished they could focus the daily facts on specific behavior goals of interest, and 51.2% wished to receive text feedback if they had made good progress or no progress. A total of 53.7% wished they could compare their behaviors with others relative to their age, gender, and initial fitness status. Regarding the wearable device and app interaction, 70.7% of poststudy survey respondents wished that their wearable device automatically synched to the app. Such a feature would greatly reduce user burden of data entry.
During the duration of the study, 122,719 behavioral logs were uploaded to the central database. These logs have been analyzed for trends and correlations with clinical and biological markers recorded at the beginning and end of the intervention.
Logically, it is hypothesized that increased exposure to the app and its material would result in favorable outcomes, both in behavior change and in clinical markers. First, the number of times that the app was launched per week was calculated and categorized into groups (<1, 1-3, 3-5, 5-7, 7+ per week). These groups were then evaluated with various clinical and biometric measurements taken from the participants at the start and end of the study, along with the control group.
From a high level, it is evident that increased app exposure had an observable effect on various clinical measurements, in particular for BMI (
It can be seen that the control group had undesirable increases over the intervention period, whereas the treatment group had sustained or reduced the measurements. Notably, those who looked at the app
Boxplot showing no app (control) and grouped app launches per week (treatment) against observed changes in body mass index (BMI). Outliers are plotted as individual points.
Boxplot showing no app (control) and grouped app launches per week (treatment) against observed changes in systolic blood pressure (BP; mm Hg). Outliers are plotted as individual points.
The average number of logs completed per day was analyzed for correlations to the clinical changes observed in the study, suggesting the following hypothesis:
H0: There is no supported relationship between daily log and clinical or biological markers.
H1: The number of logs completed each day will correspond to greater change in clinical and biological markers.
Analysis shows that daily log completion rates show no relationship between pre-post BMI scores (
Using domain knowledge, it was possible to group the clinical and biological markers into dichotomous groups (improvement or no improvement), which allowed for further analysis to be carried out. Independent samples
Participants’ self-reported behaviors were analyzed to find the frequency and percentage of times that they achieved the recommended daily goal value for each question. The following hypothesis is tested:
H0: There is no supported relationship between achieving recommended values and clinical or biological markers.
H1: The higher the number of recommended goals achieved, the greater the degree of change in clinical and biological markers.
Correlation analysis between a participant’s mean percentage of recommended goals achieved, across the study duration, and observed clinical measurement changes showed the following: no relationship for systolic (
Once again, each pre-post clinical and biological reading was categorized as either improvement or no improvement. For each individual, a baseline performance level was calculated from his or her self-reported behaviors in the first week of enrollment. Because there were a number of individuals within the treatment group who were highly active and maintained a healthy lifestyle, to reduce the ceiling effect on the data the first quintile (n=20) of participants were removed from the analysis. Using the dichotomous groupings of improvement and no improvement, significant correlations were found between daily goal percentage achieved and BMI reduction (
Odds ratio and relative risk analysis for participants who achieved more than 60% of their recommended daily targets (mean) and body mass index change outcome.
Value | 95% confidence interval | ||
Lower | Upper | ||
Odds ratio for recommended targets achieved >60% (achieved/did not achieve) | 0.288 | 0.107 | 0.774 |
For cohort BMIa change = increased | 0.507 | 0.274 | 0.936 |
For cohort BMI change = decreased | 1.762 | 1.164 | 2.667 |
N of valid cases | 82 |
a BMI: body mass index.
Participants’ reported their levels of physical activity via 3 self-reporting questions:
1. Number of minutes performing moderate physical activity
2. Number of minutes performing vigorous physical activity
3. NikeFuel points earned via wearable device.
Each participant’s results were analyzed for correlations between these values and clinical observations. The following hypothesis is tested:
H1: The higher the number of minutes performing physical activity/higher the NikeFuel points, the greater the degree of change in clinical and biological markers.
H0: There is no supported relationship between achieving physical activity levels and clinical or biological markers.
Using the dichotomous variables (improvement or no improvement), each physical activity feature was analyzed. Again, using the baseline performance metric calculated in the first week of observation, participants in the last decile (bottom 10%) were excluded from the analysis to reduce ceiling effects. An independent samples
Participants’ self-reported stress reduction efforts were analyzed for their effect on clinical measures. Participants' SBP were recorded before and after intervention and categorized into low (<90), ideal (90-120), prehypertension (120-140), and hypertension (>140). Those with nonideal SBP at their preintervention recording (n=50) were analyzed to observe if a change of category occurred during the intervention. Changes observed in these participants were categorized into 3 groups: improvement (n=13), no improvement (n=14), and deterioration (n=23). One-way analysis of variance of their category changes showed a significant correlation between efforts to reduce stress (effort rated 1-10, where 10 is high effort) and SBP category change as a whole,
The percentage of recommended values achieved for the entire treatment cohort was categorized into quintiles (1=highest, 5=lowest). These performance quintiles were then compared with a number of demographic variables collected at the start of the study. Analysis of these data showed relationships between a participant’s achieved percentages and whether that participant knew of a family member having dementia. This relationship is apparent between the second and fifth quintiles (
Percentages achieved (0%-100%) and gender (male or female) were also analyzed (
It would appear that users who have family members with dementia are motivated to reach their recommended daily targets, therefore performing better, perhaps because of first-hand experience with the condition. In addition, analysis shows a visible correlation between gender and the ability to reach the recommended daily target values. The reasons behind this observation are currently unclear and require additional analysis; however, they could relate to motivations, occupation, and education level.
Bar chart showing the log performance quintiles against the number of users who report to have known of a family member having/had dementia.
Boxplot showing distributions of male and female mean percentages of achieved targets across all domains.
The mobile phone app provided a novel method to remotely monitor participants in a behavior change intervention, while also facilitating the delivery of intervention material. In addition, analysis of exit survey shows that the app facilitated stages 3-5 of the TTM, preparing participants for change, allowing them to accurately monitor and assess their actions, and encouraging continued maintenance and improvement of their desired behaviors. Results from the exit survey showed that most users wished to continue their behavior change efforts, which if maintained, are expected to yield superior outcomes in AD prevention.
In this trial, the recommended values for each behavior played a key role in the uniform assessment of participants’ performance. Analysis of pre-post measurements from the treatment group showed clear physiological changes in those who achieved the highest in their attempts to meet recommended values. This was especially apparent in those who were previously underachievers in certain behavioral domains, before the study (based on the first week of observed behavior logs). Effects observed included a desirable lowering of BMI, improvements in HDL and low-density lipoprotein cholesterol levels, improvements in SBP, lowering of resting heart rate, and improvements in perceived stress levels.
Regarding user experience, most app users stated that they wished to alter their recommended values to be based on their “current” health status, whereas others wished to manually set their own target goals. Such a feature could improve engagement with the app, at the detriment of a true representation of progress. A compromise would be to present the user with their efforts against both personal and global targets (
Half of the users wished that their educational material was focused on a specific domain of interest, rather than evenly spread throughout all behavioral domains. Such a focus may be beneficial if the user requires extensive change in one particular domain, but for the purpose of a multidomain intervention the investigators decided it was of great importance to educate across all domains.
A prospective graphical representation of a user’s efforts against globally recommended values and his or her personal goals.
The findings of the study may be biased toward the study cohort’s locale and ethnic group. The study cohort was predominantly white (96.6%) and the participants resided in a county that is classified as 96.23% rural [
Within this larger study, additional work would be required to accommodate and account for the cultural, regional, and religious differences across groups; for example, adjusting dietary recommendations based on religious practice.
Through direct communications with participants and survey analysis, various aspects of the app and supplementing technology have been identified for improvements for a future version of the study.
The Nike FuelBand’s proprietary and nondisclosed metric of NikeFuel points is rather ambiguous for the purpose of a scientific study. Many users reported that the device did not accurately award them with points during activity and, conversely, awarded them with points when they were performing sedentary tasks, such as when they were driving their car. These false positives removed the opportunity to use the data to validate reported physical activity with the FuelBand’s NikeFuel metric. In agreement with the participants' comments, a recent study assessing the validity of commercially available activity monitors found the FuelBand to be one of the weakest performers overall, undercounting daily step count, on average, by 2529 steps [
As discussed earlier, the users also had the burden of repeatedly entering their NikeFuel points via the log screen. This user burden of data entry can be greatly reduced by enabling the transfer of data from relevant wearable devices directly to the app, greatly increasing the convenience of the solution.
Participants had informed us that they wished that the app were more socially engaging. For future development we have identified that a social element is required, allowing users to add friends with whom they can publicly compare their efforts. Integrating the app with existing social networks, such as Facebook and Twitter, can facilitate this feature. Social network integration will allow the users to find friends already in their network, who are also using the app. From here they may compare their own accomplishments with those in their friend group, thus offering an opportunity to heighten motivations for change. In addition, integration with these networks will also allow users to post their accomplishments to their public pages, allowing those outside the study to view their efforts and provide an opportunity for additional peer support, while boosting the public profile of the study.
There is a huge opportunity for personalization in all aspects of the app. Users of the Gray Matters app have suggested that they would like to set their own targets and behavior change goals. This includes adding or removing domains based on a user’s motivations. Daily fact delivery could also be revised to prioritize daily facts from a domain of interest to the user.
Within the study, participants were asked to report behaviors that were reasoned as favorable by the investigators because of their role in AD prevention. However, the participants were not asked to report behaviors that should be avoided. For example, although participants were encouraged to consume fruits and nuts, they were not asked to report how many refined sugars or processed foods they consumed. Using solely the measure of desirable food intake, without observing the undesirable food intake, results in a skewed representation of diet macronutrients and overall calories consumed.
Smoking cessation was not included in the original study, as there is an extremely low rate of smokers in the Cache County area [
On an ongoing basis, we will strengthen and expand the daily fact database, adding new facts and suggestions, with each vetted using a modification of the rating system developed by the Grading of Recommendations Assessment, Development and Evaluation working group [
A number of suggestions were provided by users of the app informally via email during the duration of the study. A familiar complaint included improving the distribution method of the app. The TestFlight platform, although useful for maintaining control of distribution, was developed for tech-savvy users, not for clinical interventions. As such, many users had problems registering with the platform and subsequently approving certificates and downloading the app. In the next iteration, all distribution will take place via the platform’s official app repositories, iOS App Store and Google Play Store.
The prevailing theme of this paper has been to express the benefit of using a mobile phone app as a core component of a behavior change intervention—to yield the advantages offered by the pervasive nature of the mobile phone within an individual’s daily life and routines. In this study, the mobile phone offered the opportunity for clinical effect to occur through behavior change. The app excelled as a delivery platform for the intervention, enabling the dissemination of educational intervention material, while simultaneously monitoring and encouraging positive behavior change. Although the effect of behavior change in midlife, observed during the 6-month RCT, on future AD risk is still relatively unclear, it is evident that participants in the treatment group had favorable improvements across numerous physiological domains, suggesting that a sustained effort would yield superior outcomes in the future.
Alzheimer disease
body mass index
Centers for Disease Control and Prevention
Health Belief Model
high-density lipoprotein
randomized controlled trial
systolic blood pressure
short message service
Theory of Planned Behavior
Transtheoretical Model
This work has been supported by the Vice President for Research seed grant, Utah State University, and the Department for Employment and Learning, Northern Ireland.
PH, doctor of computer science, designed and developed the Gray Matters apps and supporting server architecture; facilitated distribution of the software; provided one-to-one user support; created instructional videos for users; collected app and user behavioral data; performed statistical analysis of app, clinical, behavioral, and booster data; and authored the manuscript.
CN, professor of biomedical engineering, supervised PH; directed the technical development of the Gray Matters apps; and provided critical feedback on the manuscript.
SM, professor of mathematics, supervised PH; oversaw data processing and statistical analysis of behavioral data from the Gray Matters app; and provided critical feedback on the manuscript.
IC, research associate in pervasive computing, assisted with data preprocessing and analysis; provided technical recommendations on app development and data processing; prepared and carried out statistical analysis of self-report data from the app; and contributed to the writing of and feedback on the manuscript.
JT, neuropsychologist, directed the selection, collection, and interpretation of cognitive tests; organized and delivered several booster events in the domain of cognitive stimulation; and provided critical feedback on the manuscript.
CC, doctoral student in gerontology, assisted the principal investigator in all aspects of the study including recruitment, eligibility prescreening of participants; setting up and maintaining laboratory, cognitive testing, and survey data collection procedures; supervising undergraduate research interns; Web design; technical support for activity monitors assigned to participants; coordination of participant and intern appointment calendar; and provided critical feedback on the manuscript.
MN, gerontologist, was principal investigator, directing all aspects of the study; supervised the collection of the AD prevention “fact and suggestion” database; designed the functionality of the Gray Matters mobile phone app, trained study participants in the installation and use of the app; contributed text to the manuscript and provided critical feedback on the manuscript.
None declared.
CONSORT-EHEALTH Checklist (v1.6.1).