Development and Validation of an Automated Measurement of Child Screen Media Use: FLASH

Screen media use and TV viewing has been linked to child obesity through epidemiologic and randomized controlled intervention studies, resulting in excess TV viewing becoming a public health concern. Higher amounts of TV viewing among children have also been associated with decreased fitness and higher risk of metabolic syndrome. While TV viewing has often been used as a marker of sedentary behavior, it has been distinguished from inactivity (or lack of physical activity (PA)), by a meta-analyses that found only a small negative (-0.13) effect size between children's TV viewing and PA. A major limitation of the meta-analyses was most included studies relied on parent or self-report of screen viewing behaviors, potentially introducing errors and under or overestimating the true effect size. Screen media use and TV viewing is a common sedentary behavior among children, and accurate assessment of screen use is important to characterize the role it plays in children's overall sedentary behaviors. Accurate assessment of screen media use and TV viewing is also needed to delineate the causal pathways that screen use plays in children's physical activity, development of obesity, and other health outcomes.

Substantial problems and limitations have been identified in the current available measures of TV viewing, which reduces the likelihood of "high quality exposure assessment" of screen use. The most commonly used measure, child self-report or parent-report to estimate the minutes of TV viewing in a typical day has low correlation (0.27) with home video-observations. The current gold standard method to measure children's TV viewing, direct or video recorded observation, is too expensive and intrusive for most field studies. In addition, it may cause privacy concerns for participants and requires additional costs of coding the video recordings at a later time. TV diaries had higher correlations with home video recording (r=0.84) than parent estimates, but are very burdensome requiring documentation of child activity every 15 min for several days, which may introduce systematic errors. Lastly, children's TV viewing behaviors are quickly changing and many children now view TV shows and movies on other screen devices, such as smartphones and mobile tablets (e.g. iPads or Kindles). In summary, the current ability to measure and monitor how much TV a child watches is intrusive (observations), burdensome (TV diaries) or crude and fraught with error and bias (general estimates). This limits researcher's ability to identify causal associations with health outcomes. While automatic and objective measures of PA, such as accelerometers, have shown validity and reliability for quantifying children's PA, no system that is automatic, accurate and unobtrusive has been developed to assess children's screen use on traditional TV screens or tablets.

Imaging, computer vision and signal processing algorithms have made tremendous progress in the last decade, resulting in high-performance, reliable algorithms for many tasks such as face detection, face recognition, gaze tracking, human activity recognition, and 3D sensing. In collaboration with electrical engineers at Rice University, investigators at Baylor College of Medicine are leveraging these recent advances to build a first-of-its-kind 'Family Level Assessment of Screen use in the Home" (FLASH), a fully integrated, unobtrusive system for accurate, privacy preserving and reliable monitoring of TV and other screen usage by children and adults in their homes. In this study, investigators will focus on (a) Integrating the vision and signal processing algorithms and the imaging hardware required to develop FLASH (Rice investigators); (b) Refining the FLASH hardware and the software pipeline for TVs and mobile devices via alpha and beta laboratory studies (Rice and BCM investigators); (c) Validating the FLASH TV and mobile device against the gold standard (observations) among 6-11 year old children who are old enough to follow instructions in a controlled observation laboratory to ensure fidelity to the protocol (BCM investigators), and (d) Performing a preliminary investigation of the challenges for in-home use and adoption (home feasibility study) (BCM investigators).

The ultimate goal of this project is to develop a highly accurate system to measure the screen use by the index child, across diverse screens, including TVs, smartphones and tablets. Since smartphones and tablets are similar in software and hardware, investigators only need to develop two systems: FLASH-TV and FLASH-Mobile. Both systems will take a three step approach: 1) use detection, 2) index user identification, and 3) usage logging, to assess an index child's screen use. The goal is to merge a child's use across screen platforms into one database to be able to track their screen use across devices along with simultaneous use of devices.

FLASH will be developed collaboratively with electrical engineers at Rice University and behavioral researchers at Baylor College of Medicine in a series of four studies:

STUDY 1: Iterative alpha-tests with small samples of family diads or triads (n=3-5 families per trial) of the individual FLASH components and integrated FLASH systems in the CNRC Metabolic Research Unit (MRU) Observational Lab. Investigators plan for these alpha-tests to occur approximately every 3 months to assess the most recent advances of the FLASH technology. Approximately 3-4 separate alpha test with 3-5 triads per test, resulting in an anticipated sample of 42-60 participants (12-20 triads). The experiences from pilot studies we have conducted, identified that iterative testing of new components with a small sample, followed by refinement and retesting, allows for the ideal tight-loop feedback with input from both teams to capitalize on human subject data collection and allow advancement of the technology development.

STUDY 2: One to six beta tests (depending on results) of the integrated prototype FLASH system with 10 family dyads or triads per test will be conducted also in the CNRC MRU Observational Lab, with a maximum on 240 participants (80 triads or dyads). The experience from the pilot study suggests that data from 10 triads or dyads per beta test will allow adequate assessment of the functionality of the FLASH prototype.

STUDY 3: Lab based validation of the final FLASH-TV and FLASH-Mobile devices in the CNRC MRU Observational Lab with 32 parent-child triads. This number has been revised from our initial estimation based on our experiences in alpha and beta tests. Investigators will recruit 32 parent-sibling triads (96 participants) to complete the validation study. Three of the parent-sibling triads (9 participants) will participate in a pilot study prior to starting data collection on the remaining sample to refine the final validation protocol, if needed. To be conservative, we account for ICC of 0.6 in calculating a sample size required for this validation study. A random sample of 29 subjects who are each measured at least 100 times yields a two-sided 95% confidence interval with a width of 0.25 (margin of error = 0.25/2 = 0.125 = 12.5%) when the estimated ICC is 0.6 as assessed using a two-way mixed-effects ANOVA model.

STUDY 4: Home feasibility study to assess families' willingness to have devices installed in their home for a limited time study period and the ability of the FLASH devices to capture children's screen use across an entire day. With a power of 80% and an alpha of 0.05, to detect a moderate-to large effect (0.4) with bivariate correlation between minutes of TV viewing as measured by FLASH compared to TV diaries a final sample size of 37 families will be needed. A moderate-to-large effect size will help ensure high accuracy for FLASH to measure TV viewing in the home. Investigators anticipate having complete data on 80% of participating families and will therefore recruit 46 families (184 participants, estimating 4 family members per family) to achieve the final sample of 37 families.

Across the four studies, investigators therefore anticipate a total sample size of 448 participants (88 triads in studies 1-3 and 46 families in study 4). Each study is an observational study with no group assignments and no control/placebo.