Purpose: Screen use among children is a public health concern. Children’s screen use assessment typically relies on self-report and suffers from biases. Our goal was to develop FLASH-Mobile, a background application (app) for Android Devices for tracking app usage with a user log-in to identify children’s device use, even when devices are shared.

Methods: Four design studies were completed with a parent-child dyad/triad (parent and siblings). Two, 2-hour lab observation studies (n=15) assessed the accuracy of FLASH-Mobile to capture participant use of device compared to staff coding of video data (gold standard). Two 3-day studies in the child’s home gathered real-world information on the feasibility to assess children’s typical device use (n=33). The FLASH-Mobile app (built on HealthSense) included three different iterative versions to detect child user: V1) pop-up prompt when unlocked (n=8); V2) notification banner when unlocked and every 15 minutes of use (n=9); and V3) notification banner only when unlocked (n=31). The user identified themselves as “child” or “other”. FLASH-Mobile collected time an app opened and closed. It estimated device use by summing duration of open apps for that user. Lab-based video data were coded by staff for device use by user (child, other or together). Agreement between FLASH-Mobile and actual child use is reported as percent agreement with gold standard for observation studies, and by parent report after reviewing usage logs during exit interview for real-world studies.

Results: 48 parent-sibling dyads/triads participated, 33 (69%) were able to download FLASH-Mobile on their device with all features functioning and 35 had complete data (2 used study device). Compliance with user identification improved across versions in home-based studies (56.1% V2 and 89.5% V3). Among 15 participants in lab observation, 3 had less than 20% compliance with user identification. Agreement of device use to gold standard among remaining sample was 73.6% (range 44.7-96.1%). The agreement of FLASH-Mobile to parent review of output log was high, with only minor discrepancies reported by two parents among real-world study with complete data (n=20).

Conclusion: FLASH-Mobile is a promising new tool to more accurately measure device use by children, including for shared devices.

Purpose: Most geospatial indicators of built and food environment exposures are static (e.g., buffer-based GIS measures), and thus, are prone to the Uncertain Geographic Context Problem (UGCoP). Some physical activity and dietary behavior researchers have begun collecting time-matched GPS and accelerometry data to overcome this issue. However, processing and analyzing these data in a way that yields meaningful insights for answering health and place questions remains challenging. We aimed to develop an open-source code that integrates Geographic Positioning Systems (GPS) and accelerometer monitor data for obesity-related behavior research.

Methods: We developed an open-source, Python code that integrates QTravel BT-10000 GPS and GT3X-wBT Actigraph device data via temporal matching. We implemented a series of rules to generate analytic datasets including variables about locations visited, trips between locations (distance, duration, travel mode), and spatiotemporally matched physical activity intensity categories. Output files include datasets at the following levels: 1) participant-level, 2) trip-level, 3) location-level, 4) visit-level, 5) fix-level (coordinates detected by the GPS monitor every 15 seconds).  

Results:  To demonstrate the utility, versatility, and types of analytic outputs generated by the code, we successfully applied it to GPS and accelerometry data from four independent studies. The first study examines the impact of an initiative to increase geographic access to non-traditional food stores (farmstands, mobile markets, healthy corner stores) on food purchasing and intake patterns among low-income urban residents. The second study examines the impact of a large-scale rollout of Safe Routes to School Program on active commuting to school and overall physical activity among a diverse sample of school-age children. The third study assesses the impact of rapid, large-scale implementation of new bicycling infrastructure in a middle-income country mega-city. The fourth study explores the role of natural and human-made shade infrastructure on physical activity levels during school recess among children, across varying weather conditions.

Conclusions: This open-source tool represents a novel, valid, and versatile approach for reducing the UGCoP and more comprehensively examining how individuals interact with their city as a whole, and how these interactions influence their physical activity and diet-related behaviors.

Based on formative work we presented at ISBNPA 2021, this study aimed to psychometrically test novel self-administered measures of three aspects of a household’s resilience to financial shocks (e.g., job loss) that can increase food insecurity risk. These three measures assess Absorptive Capacity (i.e., on-hand resources to absorb a shock, short-term), Adaptive Capacity (i.e., knowledge/skills/barriers to adapt to a shock, intermediate-term), and Transformative Capacity (i.e., community conditions affecting long-term transformation in household resilience). These measures were intended to provide a multidimensional view of a households’ food insecurity risk resilience, as well as provide actionable data that complements the United States Department of Agriculture’s Household Food Security Survey Module (HFSSM).

In April 2021, we piloted the measures in a convenience sample of individuals at risk for food insecurity in the United States. The pilot survey included the three new measures, scales for validation (Conner and Davidson personal resilience to challenges scale (CD-RISC), CFPB’s Financial Wellbeing scale, and the HFSSM), and demographic questions. We used Classical Test Theory and Item Response Theory (IRT) approaches to assess model fit (confirmatory factor analysis (CFA)), Cronbach’s alpha, IRT parameters (discrimination, difficulty), test bias (moderation effects), and convergent validity (Spearman’s coefficients). 

Respondents (n=494) were 18-89 years old, 67% experiencing food insecurity, 47% with high school diploma or less, and 72% were women. Races/ethnicities: non-Hispanic White (48%), Hispanic/Latino (22%), non-Hispanic Black (17%), Asian (4%), Tribal/Indigenous (2%), and multi-racial/ethnic or not listed (7%). Acceptable metrics were seen for: CFA model fit (AGFI=0.963-0.992; Standardized RMR=0.039-0.084), Cronbach’s alpha (0.76-0.92), and IRT indicators (acceptable slopes, and thresholds spreads). No test bias was observed by race, gender, age, education, or test mode. Scores were negatively associated with food insecurity (-0.294 to -0.508) and positively associated with CD-RISC (0.302-0.336) and financial wellbeing (0.328-0.470).

These findings are encouraging and support reliability and validity of these new measures within similar samples. These measures can be used for needs assessments, program evaluation, clinical screening, and research/surveillance. We anticipate that widespread adoption will promote a more comprehensive understanding of the food insecurity experience and facilitate development of tailored interventions on upstream causes of food insecurity.

Purpose: The role of dieting in the etiology of obesity and disordered eating is controversial. “Dieting” during behavioral weight loss programs can lead to clinically significant weight and binge eating reductions, but “dieting” in population-based samples is a risk factor for obesity and disordered eating. This “dieting paradox” presents a challenge for developing strategies that address both problems. Innovative statistical tools are needed to investigate this paradox and better characterize the heterogeneity of “dieting”. In this presentation we will demonstrate the process and advantages of using topological data analysis (TDA) to identify meaningful sub-groups of adolescents based on their dieting status, weight-related behaviors (e.g., energy intake, physical activity, binge eating) and psychological factors (e.g., body satisfaction). TDA is a novel, data-driven analytic technique that pairs unsupervised pattern detection and network visualization to obtain the full taxonomy among a complex dataset of inter-related variables. TDA potentially captures sub-groups that could be missed with traditional statistical clustering methods.

Method. Baseline data from the Project EAT study were utilized for these analyses. Participants were adolescents (ages 11-17, n=4746) attending 31 public schools in an urban area in the State of Minnesota in the United States. In Python 3.6, we used the Kepler Mapper library from scikit-tda 0.0.4 to identify fuzzy subgroups of adolescents. TDA uses a network structure in a highly-dimensional space to group participants based on the similarity of their variable profiles. This network, or highly-dimensional data shape, creates participant sub-groups visualized as nodes with relations among sub-groups represented as lines between nodes.

Results. We used an iterative process of visual examination and comparison of descriptive characteristics across nodes to identify clinically meaningful participant subgroups for a series of TDA solutions with varying numbers of nodes. Preliminary results support identification of nine sub-groups of boys and girls, respectively. We will demonstrate how TDA-derived sub-groups differ on key behaviors at baseline and their association with different obesity and disordered eating trajectories over time.

Conclusion. TDA shows promise for identifying clinically meaningful sub-groups of adolescents that could inform how intervention approaches should be targeted and tailored to optimize weight and disordered eating outcomes.