All time references are in CEST
Wearables, Apps and Sensors for Data Collection 1 |
|
| Session Organisers |
Dr Heidi Guyer (RTI International) Professor Florian Keusch (University of Mannheim School of Social Sciences ) |
| Time | Thursday 20 July, 09:00 - 10:30 |
| Room | Martini U6-4 |
The recent and ongoing proliferation and development of mobile technology allows researchers to collect objective health and behavioral data at increased intervals, in real time, and may also reduce participant burden. Wearable health devices can measure activity, heart rate, temperature, sleep behaviors and more; apps can be used to track behaviors- such as spending, transportation use or health measures- as well as for ecological momentary assessment; smartphone sensors have been used to capture sound and movement, among others. The COVID-19 pandemic brought about additional uses of apps and sensors to measure population trends on a large range of topics including mobility, access, symptoms, infection, and contagion. Large national studies such as the UK Biobank study and the U.S. based NIH All of Us research program have demonstrated the scalability of integrating wearables in population-based data collection. Other studies, smaller in scope or sample, have developed innovative approaches of integrating apps and sensors in data collection.
However, researchers using these new technologies to collect data face many decisions about which devices to use, how to distribute them, how to process the data, etc. These decisions impact other components of the research design including selection bias and data quality. In this session, we invite presentations demonstrating novel uses of wearables, apps, and sensors for data collection as well as potential barriers or challenges. Presentations may be related to measurement, consent, data storage, data analysis and data collection.
Keywords: data collection, wearables, survey apps, sensors, measurement
Dr Ritika Chaturvedi (University of Southern California) - Presenting Author
Dr Marco Angrisani (University of Southern California, Schaeffer Center for Health Policy and Economics)
Dr Mick Couper (University of Michigan)
Person-generated health data (PGHD) from smartphones and wearables are invaluable to accurately assess social/structural determinants of health, which constitute 60-80% of the modifiable risk of adverse health outcomes, offering an opportunity to improve health equity. Extant PGHD research, which uses convenience sampling and “bring-your-own-device” (BYOD) designs, systematically underrepresents populations most affected by health disparities, including racial-ethnic minorities and individuals from lower socioeconomic status (SES). In this joint presentation, we introduce the American Life in Realtime (ALiR), the first longitudinal and sociodemographically representative registry of PGHD. ALiR includes 1,038 members of the Understanding America Study (UAS) – a probability-based internet panel representative of the U.S. adult population – who consented to contribute PGHD for one year. We gifted a Fitbit to all participants, regardless of whether they owned a device, and deliberately oversampled racial/ethnic minorities and low-SES individuals, who may face participation barriers beyond digital inclusion. As a result, the composition of the ALiR sample aligns with that of the population across sociodemographic and health outcomes (except for an underrepresentation of seniors). We show that, if we were to adopt a BYOD design and recruit among UAS wearable owners, the resulting sample would be highly selective and underrepresent groups experiencing health disparities. Relative to population benchmarks, it would include fewer males (32%), Blacks (48%), Hispanics (20%), individuals with only high school (33%), with less than $30,000 in yearly income (50%), and living in rural areas (25%). We analyze how previous device ownership drives the decision to participate in ALiR and document compositional differences between previous owners and non-owners within the ALiR sample. We also examine how engagement (e.g., frequency and duration of missing Fitbit data spells) and behavior (e.g., intensity of physical activity) vary with ownership status. Ultimately, we provide a methodological framework to achieve inclusivity in PGHD-based precision health research.
Dr Talip Kilic (World Bank) - Presenting Author
Dr Gayatri Koolwal (World Bank)
Dr Grey Seymour (International Food Policy Research Institute)
Dr Thomas Daum (University of Hohenheim)
Mr Hannes Buchwald (Independent Consultant)
Mr Wilbert Drazi Vundru (World Bank)
Design of effective policies to address gender disparities in time allocated to to unpaid domestic and care work versus activities that are tied to income generation require high-quality data on men’s and women’s time use, which, in low-income countries, are sourced from household and time use surveys. The requisite data, however, are often either unavailable or elicited through recall-based survey methods that are not widely validated but that are suspected to be subject to non-classical measurement errors that can bias subsequent statistical analyses. This study will report on a randomized experiment that is being implemented in Malawi and that will examine how use of innovative techniques in time use data collection could sidestep measurement concerns associated with the use of traditional recall-based time use survey methods. The experiment assigns random samples of households, and respondents within, to one of two treatment arms on time use: (1) a time use diary with a 24-hour recall, and (2) a self-administered smartphone-based pictorial time diary, namely the Time Tracker app, installed on a low-cost Android smartphone. Both treatment groups also receive a research-grade, waist-worn physical activity tracker, to collect high-frequency, objective measures of physical activity during the same period of time use data collection— to advance on-going research on the application of machine learning techniques to reliably predict time use patterns from physical activity records. The findings will advance the validation of scalable best practices to individual-level data collection on time use, which could in turn be considered for implementation as part of surveys implemented by national statistical offices.
Professor Arie Kapteyn (Center for Economic and Social Research, University of Southern California ) - Presenting Author
Mr Htay-Wah Saw (Center for Economic and Social Research, University of Southern California & Michigan Program in Survey and Data Science, University of Michigan-Ann Arbor)
Unlike surveys that mainly rely on self-reports, wearables provide ecologically-valid, highly granular, and longitudinal data because, with wearables, in-situ behaviors can be measured repeatedly and frequently. Data are collected passively, eliminating the need to administer some survey questions. Additionally, use of wearables eliminates differential item functioning associated with the use of survey questionnaires.
Studies involving wearables employ various study protocols and require respondents to perform various tasks. The device and study characteristics (and their interactions with respondent characteristics) can affect sample selection into various wearable studies.
We implemented three wearable studies involving different wearables in the Understanding American Study (UAS). The first study, implemented between 2015-2017, recruited a random sample of about 600 UAS respondents and asked them to wear a GENEActiv accelerometer for 7 consecutive days. The second study, implemented between 2019-2020, recruited a random sample of about 100 UAS respondents aged 50 and above in which respondents wore a GENEActiv watch for 7 consecutive days, a Fitbit smartwatch for another 7 consecutive days, and both GENEActiv and Fitbit for 2 consecutive days. The third study, which started in 2021 and is still ongoing, recruited a random sample of about 600 UAS respondents and asked them to carry a pollution monitoring device for one year. In the first two studies, respondents also completed end-of-day surveys after 6 PM each day, while in the third study respondents answer a simplified time use diary once a month.
The UAS has many hours-worth of background information on respondents. This provides a unique opportunity to study selectivity of respondent participation in different types of experiments involving wearables. In the presentation, we will present and compare wear time, compliance with study protocols, sample demographic composition, substative variables, nonresponse, and attrition across the three studies mentioned above.
Mr Johannes Lemcke (Robert Koch-Institut) - Presenting Author
Mr Ilter Öztürk (Robert Koch-Institut)
Mr Daniel Grams (Robert Koch-Institut)
Dr Ronny Kuhnert (Robert Koch-Institut)
Background
With the Corona-Datenspende app, self-selected citizens provide the Robert Koch Institute with fitness tracker data that can help to better understand COVID-19. After the launch of the app, more than 500,000 users had registered. In the app users can release the retrieval of vital data from their wearables. This data can provide indications of symptoms of an infection with SARS-CoV-2. Users can also participate in further studies in the app. In these studies, users are asked to answer short questionnaires.
App users are self-registered. So, they might not be representative of the general population. In this study we investigate potential differences to the general population.
Methods
In order to better understand the potential systematic self-selection bias within this wearable study, we conducted a survey study to evaluate potential differences among app users compared to the general population of Germany in 2021. The study was launched in October 2021. 37,309 users took part in the survey. We compared the educational level, the subjectively assessed health status and the utilization of general medical services of the participants in the sub-study with reference data. Furthermore, we investigated the spatial distribution at the federal state level, gender, and age as potential determinants.
Results
Our results show significant differences between the group of users of the Datenspende app when compared to external benchmark data of the general German population.
We see different participation rates in the federal states, gender and the age distribution. Participants in the data donation are slightly underrepresented in the youngest age groups and strongly underrepresented in older age groups. Participants with low education are strongly underrepresented compared to the benchmark data. The participants in the study rate their subjective health status more frequently as very good or good compared to
Mr Joeri Minnen (hbits VUB) - Presenting Author
Mr Ken Peersman (hbits)
In a Time Use Survey (TUS) respondents are asked to keep a record of their activities in a time diary, usually for 2 days (one weekday, one weekend day). Besides the activities themselves, their context also plays an important role. In a TUS, the activity and its social and spatial dimension define an episode. If one of the elements changes, a new episode starts.
Time registration is however burdensome, and solutions are necessary both to support the respondent and to higher the data quality. A solution can come from the interrelation between time and space (Thörsten Hägerstrand). Today a time diary asks about people’s activities within a time frame, while the place is (just) an added context. However more accurate information on the place of being is often available to respondents via their Smartphones. These smart devices have sensors included that, amongst others, grasp the locality of the device (as a proxy of the respondent).
In order to include this information, the GeoService tool was/is developed under EUROSTAT-contracts holding a plugin and a back-end software to process, enrich and visualize the captured data. Via an API the GeoService is connected to the MOTUS software platform and location information is shown as ‘Tentative data’ to the respondent in the mobile and web application. In this way respondents are able to accept, correct and/or improve the location data coming from the GeoService. And use this in the keeping of their time diary.
This presentation will show how this process works, but will also show how the consent of the respondents is asked, and how the privacy of the respondents is preserved.