Mobile Device Biomonitoring to Prevent and Treat Obesity in Underserved Youth

Part 1: Developing a Mobile Software Suite for Biomonitoring The software suite proposed in
this research will be implemented using a three-tier architecture as shown in Figure 1. The
front tier of this architecture is the data collection sensors coupled with mobile phones
that acts as data transmission device. The middle tier is a web server that receives and
processes information and sends it to a back-end database server that stores the information.

The sensor layer is a collection of off-the-shelf devices that measure the metabolic
activity. In particular, we propose to use heart rate, blood pressure, and blood glucose
monitors currently available from Alive Technologies25. We also propose to use BodyMedia WMS™
sensors to measure the Galvanic Skin Response (GSR) and motion estimation using
accelerometers. In addition to measuring the metabolic activity all these sensors are also
capable of wirelessly transmitting this data over Bluetooth interface. We propose to use
feature rich Nokia N95 as the mobile phone platform. N95 supports Bluetooth 2.0 +EDR for
quick pairing with external Bluetooth sensors, and has 3G and WiFi radios for high bandwidth
data transfer. All the external sensors listed above stream data to the Bluetooth enabled N95
mobile phone. In addition to the high bandwidth radio capabilities, the N95 mobile phone
platform has a highly accurate built-in assisted GPS unit that uses a combination of GPS
satellites, cellular tower and WiFi scanning to obtain a GPS position lock in less than 10
seconds. The stated location accuracy of GPS unit is 30 meters, while in practice the
observed accuracy is around 3 meters.

The Principal Investigator of this project has extensive experience using accelerometers in
minority youth, and is Co-Director of the Human Measurement Core of the USC Center for
Transdisciplinary Research on Energetics and Cancer (TREC) funded by NCI, which is
responsible for physical activity measures across all TREC center projects. Accelerometer
data for assessment of physical activity is one of the most prevalent measures (coupled with
dietary data) used to evaluate energy expenditure6, 26, 27. However, most research involves
the wear of a single accelerometer which does not yield truly accurate data 24. Therefore, in
this project we propose to synchronize multiple accelerometery data sources to estimate
energy expenditure more accurately28.

Part 1a: Testing and Initial Deployment of sensors in target populations: We will use unit
testing to extensively test our mobile software suite. We will recreate several usage
scenarios and environmental conditions that our deployment test bed is likely to encounter. A
total of 50 Hispanic and African American youth will be recruited to assist in technology
development: 1) an advisory group of 10 youth, 2) 20 youth to participate in laboratory
testing of the biomonitors and 3) 20 youth to wear the biomonitors and provide data and
feedback. The advisory group of 5 African American and 5 Hispanic youth will be retained
throughout the developmental phase to periodically visit our laboratory and test-run the
sensors (50% female, 12-17 years of age). We will recruit a separate group of 10 African
American and 10 Hispanic youth (50% female, 12-17 years of age) to participate in idea
building sessions to ensure that sensors will be attractive and wearable, and to test the
ease of usability of our mobile software suite. The initial deployment phase includes
training the children on how to wear and remove the sensors. In order to minimize errors in
configuring the software, we envision that our mobile software will have several in-built
checks to advise the user if any of the sensor readings do not match expected sensor
behavior, and suggest simple solutions to properly configure the sensors. In addition to
providing the wearable sensors we will also provide each child with a mobile phone platform
which interfaces with these sensors. Since the mobile device has to transmit the data to the
back-end servers, we will opportunistically use an open WiFi network that a phone can connect
to transfer data both efficiently and economically. In the absence of WiFi networks the
mobile software will automatically use the cellular data network to transmit the data. For
using cellular data network we plan to explore several available options that allow users to
send unlimited data for less than 10$ a month in the Los Angeles area from their mobile
phones. For laboratory testing of sensors, 10 Hispanic and 10 African American youth will
spend 3-6 hours wearing the sensors and following protocols for walking, sitting, standing
and doing various daily activities either at Dr. Spruijt-Metz's Physical Activity Observation
Laboratory at USC HSA, or to the Motion Capture Laboratory at the Viterbi School of
Engineering on USC Main Campus. Once the software and hardware is determined to be robust
enough for deployment we will conduct our initial monitoring study with 10 African American
and 10 Hispanic youth (50% female, 12-17 years of age). Children will wear the devices for
three periods of one week (7 days), after which they will participate in brief individual
interviews and surveys to ascertain ease of wear and to find ways to motivate and incentivize
teens for wearing the sensors. Data collected from these weeks of wear will be used for the
remaining data analysis and web presentation phases of the study.

Part 2: Data Capture and Transmission to a Back-End Server: We propose to write a
comprehensive mobile software suite that will allow the mobile phone to use Bluetooth to pair
with wireless monitoring devices to collect vital health and behavioral data along with
reading the built-in GPS data. The BodyMedia and MemSensse units will provide accelerometry
data on physical activity and sleep. These measures will be validated in our physical
activity lab against the Actigraph accelerometer (which has been extensively validated in
youth) and Continuous Observation using the SOFIT system, a gold standard for physical
activity measurement in youth 32, 33. By using time stamps the sensor data from Accelerometer
can be correlated to the vital signs data collected from wearable sensors. We will use the
data collected from all these sensors to automatically classify the user's activity. In
particular, the software creates user specific movement signatures to account for differences
in user's gait, walking/running/bicycling speed, usual route taken between work and home etc.
The software will be able to use a combination of GPS and accelerometers to recognize the
differences between driving on road and walking. This sensor information will be recorded
continuously on the local storage on the mobile phone. For reference, our mobile device
platform has an 8GB in-built flash memory that can be used for storing sensor information.
Table 1 shows the approximate data rate of sensors. Using these data rates, we estimate that
our 8GB local storage can store approximately 1000 days worth of data.

While data may be stored locally on the mobile phone the real value of our approach is the
fact that these mobile devices can transmit the sensor information to any remote server using
cellular data network or even WiFi. The information collected from the sensors in the mobile
phone is sent to the web server for processing. The web server acts as data integrity manager
that prevents illegal data read/writes by using simple authentication mechanisms, such as
personal authentication. The web server utilizes HTTP/SMTP protocol to receive information
from the mobile phones. The web server also provides web enabled access to the data for
physician's around the globe.

Part 3: Simultaneous Data Analysis and Interpretation: The web server sends the data to the
back-end database server for storing and further analysis of the data. Sensor data is likely
to be noisy. Therefore, the experts in engineering and pediatric obesity in minority
populations gathered on this team will work together to create adaptive algorithms to distill
important health and behavioral information from noisy sensors. All data will be analyzed and
displayed in two modalities: 1) Average values over a week and 2) High and low points
according to geographical position. To accomplish the first modality, accelerometery data
will be reduced to display average time spent in sedentary, light, moderate and vigorous
activity and average daily energy expenditure. Average heart rate, blood pressure, blood
glucose and body temperature will be displayed along with the accelerometer data. We will use
current literature and existing software examples to develop systems that will allow us to
use Heart Rate and Galvanic Skin Response to measure and display stress and autonomic nervous
system (ANS) function appropriately for these pediatric populations. We will fine-tune models
for activity levels and correlate activity levels via accelerometry with our other sensing
modalities (heart rate, blood pressure, blood glucose, etc.). An objective is to develop new
signal processing tools based on multi-modal signal processing strategies to evaluate
activity levels with the potential of developing a continuum of activity measures (sedentary,
light, moderate and vigorous activity). The development of such algorithms will allow for
accurate measurements even when one sensor is not working properly. In addition, we shall
investigate the possibility of monitoring whether a child is properly wearing all sensors.

To accomplish modality 2), displaying high and low data points according to geographical
position, all data will be time-stamped and synchronized to the Geographic Positioning System
(GPS). This will yield high insight into the impact of the environment on physical activity
in this population. We will develop methodologies for data and device synchronization. These
methodologies may enable the turning on and off of key sensing devices after a particular
event has occurred (e.g. GPS data is interpreted and it is observed that a child has moved
into a playground or a convenience store - certain sensing devices are turned on or off, or
sample at a higher rate). This data will reveal where children are more active, where they
are sedentary and where they are experiencing stress. Finally, the web interface will be
equipped to signal health practitioners when average values fall above or below prescribed
health guidelines.

3a) Web interface capabilities: What, when and where:

Using the data acquired, health professionals will be able to visualize:

- Average amount of moderate to vigorous physical activity a child is getting (daily or
weekly), plus daily patterns of exercise

- Average number of steps a child is taking (daily or weekly), plus daily patterns of
steps

- Average amount of sedentary behavior a child is experiencing per day or per week, plus
daily patterns of sedentary behaviors

- Average amounts of sleep a child is getting per day or per week, plus daily patterns of
sleep

- Average blood glucose levels (daily or weekly), plus daily patterns of glucose levels

- Average levels of stress or arousal (daily or weekly), plus daily patterns of stress

- All patterns of behavioral and metabolic data will be time stamped. Therefore,
practitioners will be able to see when a child is active, sedentary, stressed, or
experiencing high blood glucose levels.

- All patterns of behavioral and metabolic data will be geo-coded. Therefore,
practitioners will be able to see where a child is active, sedentary, stressed, or
experiencing high blood glucose levels.

Providing notifications: The final product will be influenced by the advisory group of
medical professionals that we will assemble. We therefore provide a brief list of examples
below (thus not an exhaustive list) of notifications that we envision the web interface
supply to participating health professionals:

- Recommendations for physical activity range from 30-90 minutes of moderate to vigorous
physical activity per day and are somewhat age-specific34. Automated system
notifications could be generated when average daily physical activity moderate to
vigorous falls below 60 minutes, and again below 30 minutes.

- Recommendations for steps per day are currently 11,000 to 12,000 steps per day for girls
and 13.000 to 15,000 steps per day for boys. Automated system notifications will be
generated when children fall below these recommendations. New recommendations for both
physical activity and steps are expected to be made public soon35. Advances in the field
will be scrupulously monitored and incorporated.

- Blood pressure values are specific to height and weight of children. This data will be
fed into the database for analyses and blood pressure notifications will be posted when
a child is at or above normal blood pressure according the Fourth Report on the
Diagnosis, Evaluation, and Treatment of High Blood Pressure in Children and
Adolescents36.

- According to the American Diabetes Association, casual plasma glucose concentration ≥200
mg/dl (11.1 mmol/l) is indicative of type 2 diabetes in children. Casual is defined as
any time of day without regard to time since last meal. Furthermore, concentrations of
≥140 mg/dl are considered to indicate pre-diabetes or Impaired Glucose Tolerance
(IGT)37. Therefore, we envision that the web interface will provide system notifications
at both of these values.

- There are currently no gold standards or threshold measures for stress or arousal in
youth or in adults. However, Firstbeat technologies
(http://www.firstbeattechnologies.com/) has been working to develop stress and recovery
analysis in collaboration with several research projects in the area of stress
measurement and autonomic nervous system (ANS) function, applying also research in the
fields of exercise physiology, behavioral sciences, and applied mathematics. We will
work with Firstbeat to explore development of guidelines for measurement and definition
of cut-points for stress in youth. We will begin the design of multi-modal signal
processing algorithms for determining personalized cut-points for individual youths.

Bio-monitoring for developing interventions and intervention strategies: Because
practitioners will be able to attach time of day and geographic location to metabolic and
behavioral indices and patterns, health practitioners will be able to pinpoint causes and
effects as well as develop targeted intervention strategies. Furthermore, for researchers,
the wealth of time and location stamped data will allow for much broader understanding of the
effects of the environment and time of day of metabolic and behavioral events. Here follow
some examples:

- Example 1: Adolescent gets less than 30 minutes of physical activity of any kind per
day. The participating health practitioner finds that adolescent is particularly
sedentary in the hours after school. A targeted approach would be to assist in finding
accessible and attractive after school programs.

- Example 2: Adolescent stress levels rise steeply and physical activity levels plummet
each Sunday at approximately noon. The participating health practitioner asks the
adolescent what is occurring on Sunday afternoons, and finds that the adolescent is
being teased at the park where the family typically spends Sunday afternoons. A targeted
approach would be to suggest more adult supervision or another venue for physical
activity to the family, or explore avenues for dealing with this kind of stress with the
child.

- Example 3: Adolescent blood glucose levels have been high for a month. The participating
health practitioner notices that this last week, blood glucose levels are normalized. A
text message could be sent to the adolescent letting her know that she is making
progress. Alternatively, a call or email asking how she changed her behavior in order to
pinpoint and reinforce health behavior might be appropriate and timely.

Part 4: Developing a User-Friendly, Web Enabled and Secure Interface for Health
Professionals: This part of the research will be ongoing and begin at the onset of the
project. An advisory team including pediatricians, family doctors, pediatric
endocrinologists, specialists in pediatric diabetes, African American health outreach
professionals and promotoras (outreach workers in the Hispanic community) will be assembled
to guide us through the process of developing a web interface that will ensure that the right
information will be displayed in an easily interpretable fashion. The advisory group will
participate in regular meetings to develop and test the web interface. At the onset of the
project, they will be invited to participate in an idea-building session in order to develop
specific guidelines for web interface development. Idea-building sessions are one of the
nominal group techniques that are used to structure small group meetings in such a way that
individual judgments can be effectively pooled. Typically, four steps are involved: 1) silent
generation of ideas; 2) group recording of ideas; 3) serial discussion of ideas; and, 4)
voting to order the ideas as to their salience. Idea writing is particularly helpful in
developing general ideas into more specific ideas using group interaction. This is also a
four-step process: 1) organization into small groups; 2) initial written response; 3) written
interaction; and, 4) analyses and report. Idea writing is an attractive technique because the
group produces a written product. Analyses of the idea-building data includes recording,
transcription, and examination of the data for emergent themes. The information from these
sessions will be used for several purposes: 1) to define the form and content of final data
that the web interface will provide, 2) to develop a system of notifications or alarms for
specific metabolic and behavioral values, such as blood pressure that is too high or physical
activity that is too low as discussed in the previous paragraph, 3) define important pathways
within the data that health practitioners will need to navigate easily and 4) to ensure web
interface user friendliness, accuracy and appeal. Ideas will also be generated on use of the
web interface to develop interventions for the targeted groups immediately and in the future.

Inclusion Criteria:

- Self-identification as either Hispanic or African American, between the ages of 12-17,
and without any disabilities that would disallow wear of sensors and normal physical
activity

Exclusion Criteria: