The BALANCE tool features the MSB, which measures the type, amount, and location of physical activity in real time under free-living conditions by capturing a number of signals, including acceleration, barometric pressure, light, and geophysical location. Expanding upon earlier proof-of-principle studies demonstrating that the MSB could classify 10 different common physical movements with 95% accuracy,19,20
we completed a study in which 53 participants performed a series of activities equipped with a MSB unit worn at the left hip under two conditions. In the laboratory, they walked and jogged on a treadmill over a range of speeds and elevations. In the field, they performed a series of activities under free-living conditions, including sitting, standing, walking on level and graded surfaces inside and outside of a building, riding up/ down an elevator, and walking up/down a flight of stairs. The accuracy of the unit for classifying physical activity type and amount was established using the recorded activity (“labeled ground truth”) as the standard and its associated energy expenditure against directly measured oxygen consumption using calorimetry as the standard. A portable calorimeter (Cosmed K4 b2
) was used in the field.
Our initial efforts focused on estimating speed of movement by determining step frequency from the accelerometry profile and estimated stride length from subject height. Although we were not able to measure grade changes during treadmill activity, we were able to detect movements that occured on a level or graded surface during the field experiments using the barometric pressure sensor output. Activity-based energy expenditure was estimated with activity type and subject body weight using either the American College of Sports Medicine's metabolic prediction equations21
for major forms of human movement, including walking, jogging, stair climbing, and cycling, or, for other detected movements, the metabolic equivalent values (where 1 metabolic equivalent value is equal to 3.5 ml O2
/kg body weight-1
) in the Compendium of Physical Activities.22
Using this unique approach, MSB energy expenditure estimates were 89% accurate in the lab and 76% in the field. Importantly, the device performed significantly better during both conditions than a commercially available accelerometer that was also worn during the experiments. The technical details of the MSB were presented at UbiComp 2009.23
The full study results were presented in part at the American College of Sports Medicine's 56th Annual Meeting (May 2009), and the final data are currently being prepared for publication.
In year 1, using a list of design goals developed and refined over several months, we built a prototype food diary with energy balance visualization on a mobile phone and integrated it with the MSB via wireless Bluetooth connection. We reviewed currently available food consumption tracking programs to create a new and innovative food diary offering improved features and usability compared to available products.
The first version of the program was built using Python programming language on a Nokia S60 smartphone. The BALANCE tool consists of three major components: (1) a food diary allowing users to manually enter the type and amount of food and beverage consumed (A), (2) an activity diary allowing users to see the list of activities “sensed” by the MSB (B) and to manually enter activities the MSB cannot measure, and (3) the personal “fuel gauge” visualization, which computes and displays the difference between energy intake from electronic food entry and energy expenditure from the MSB and activity diary (C). Because the MSB cannot measure all types of movements (e.g., water-based activities and those using primarily upper-body movements), we needed to build an activity diary into the mobile phone program so that subjects could manually enter specific “purposeful bouts” of activity to complement the continuous objective data being provided by the unit. Energy intake was estimated using food types and quantities entered in the food diary, based on the United States Department of Agriculture (USDA) National Nutrient Database for Standard Reference Release 20. The diary allows users to manually enter foods and beverages that are not in the database. Users can also enter the time of consumption, and each entry is automatically time stamped. A portion of the basal metabolic rate is “credited” to the user every hour based on the Harris–Benedict equation, a widely used, albeit imperfect, method for predicting metabolic rate.
Figure 1. The BALANCE tool. The food diary is shown in A, the activity diary and sensed activities in B, and the personal “fuel gauge” in C. The left side of the visualization (orange) depicts energy expenditure and the right energy intake (green) (more ...)
Next, we performed an initial evaluation of the prototype by defining and entering specific meals, a combination of simple common foods, prepared foods, and some foods that were not in the USDA database that required manual entry of nutrient content. We experienced difficulties with the food diary, including a lack of availability and variety of ethnic and commercial foods, limited units to select for portion sizes (i.e., grams only), and programming limitations. This work provided early feedback for improving the food diary and the energy balance visualization.
We also assembled an up-to-date geographic information system (GIS) database of King County, Washington, for linking types and amounts of measured physical activities and eating episodes with XY locations captured by the integrated global positioning system (GPS) receiver. The GIS and the mobile application program ArcPad work together to provide users with a visual display of food sources and their classification, and locations associated with physical activity, within their proximal environment (). The GIS data were compiled from several data sets, including King County tax-lot-level and building-level assessor's files, Public Health—Seattle King County food permits, InfoUSA, and Washington State Department of Transportation roads. Land uses were classified into several categories commonly used in public health literature (e.g., generalized land uses, transportation infrastructure, parks and open spaces, supermarkets, fast-food restaurant, and fitness facilities). Other classifications include sources known to be associated with more walking, size of establishment (including number of seats provided), establishment belonging to a national or regional chain, and availability of alcohol, among others.
Figure 2. The activity and eating resource-mapping program. Red circles depict convenience stores and purple triangles bakeries/delis (eating locations), while green areas show public parks (activity locations). The user can obtain detailed information about a (more ...)
Early in project year 2, we replaced the USDA database with the Nutritionist Pro Knowledge Base (Axxya Systems LLC, Stafford, TX). This database consists of over 32,000 foods and ingredients, including brand-name foods, fast foods, and ethnic foods. We also switched platforms to the HTC Fuze, a smartphone running Windows Mobile 6.1 Professional, because of its large touch screen, qwerty-type keypad, and more extensible operating system. Next, we completed a usability study of 12 participants that collected data on users' habits and preferences in order to inform software redesign. These participants were mostly young to middle-aged women (n = 9, 20–49 years) who were employed full time in a variety of professional fields. All had mobile phones and were well accustomed to using them. Results included users' level of interest in different software features and evaluation of various energy balance visualizations. The preferred visualization is shown in C.
The Design Feedback Iterative Cycle
We are currently implementing a series of design/feedback cycles to test and refine the latest version of the food diary. Each cycle, designed to obtain both quantitative and qualitative data about user experience, consists of small focus groups of three to five subjects who use the phones for three days to enter their food and beverage intake. Subjects are between the ages of 18 and 65. There are no restrictions on any other demographic characteristics, although we are aiming for each focus group to include people with varying levels of technological skills. Quantitative data are obtained from two validated user experience questionnaires24,25
and qualitative data from videotaped focus group sessions. These data will be analyzed to provide information for usability improvement. We anticipate five to six cycles of data collection before we reach a “final” version ready for validation in a larger sample. We have completed three focus groups to date, which have already provided valuable insight for system improvement.
As discussed previously, we validated the MSB unit in a concurrent study. In the upcoming project year 3, we will validate the BALANCE tool with its energy balance and environment-mapping features enabled using two approaches. In one experiment, subjects will complete an expanded, one-hour field test on the University of Washington campus and in the adjoining urban neighborhood while wearing a portable calorimeter and a Garmin GPS receiver, which provide criterion measures of energy expenditure and location to match against the tool's output. The test consists of an approximately three-mile loop of both level and graded walking (with some areas having greater than 10% slope), sitting, and standing. In another experiment, subjects will use the tool for three days to enter food and beverage consumption and specific physical activities not measured by the MSB while simultaneously completing a three-day food record and a physical activity questionnaire.