Digital health technologies, including app-based cognitive and psychosocial assessments and wearable sensor-based health monitoring devices, are pervasive and crucial for better understanding and predicting trajectories related to important public health outcomes such as Alzheimer’s disease and related dementias. The availability of detailed sensor data streams and data available via application programming interfaces creates new opportunities for scientific discovery, especially considering that passively collected data may be more robust for measuring aspects of daily life that can be influenced by observation, memory interference, or retrospective biases. The increase in easily shareable metadata (eg, Google Takeout) also creates a new fruitful avenue for research, however the extent public awareness and willingness to share these data and from other data streams remains unclear. Past studies have shown that in general, the overall acceptance for various smartphone-based tasks and sensing modalities is low, with variations in type of activity and sensor [1, 2]. However, repeated prompting for specific data streams has been shown to increase willingness to share [3]. In general, participants are interested in learning from their own data, and a proportion of these are also interested in sharing their information with a health care provider [4].

Critical to understanding mechanisms driving willingness to share various data streams are concerns about data privacy, security, and control over generated data [3, 5]Once the data are collected, different members of a research team, often across different institutions, work with deidentified versions of the datasets. In some cases, data are stored indefinitely for algorithm development—seeking new purposes or uses for the data that were not originally considered. In other cases, advances in machine learning have made it possible to reidentify individuals with enough precision to warrant security concerns over warehousing sensitive data without appropriate deidentification procedures [6]). As such, it is imperative to consider the most important stakeholder, the respondent and to assess their expectations regarding the data they generate for research and the level of controls they expect.

To answer important scientific questions, researchers are interested in collecting data across the spectrum of participant involvement and burden, ranging from active data collection activities (eg, completing a survey, cognitive assessments), to passive data collection (eg, extracting sensor data streams, reading existing metadata [7]; Figure 1). Passive data collection methods refer to data captured automatically without participant effort (eg, sensors, metadata), whereas active methods require participant involvement (eg, completing surveys or assessments). Active data collection also requires input from the respondent in a manner that makes the goal of the research more apparent. On the other hand, passive assessment does not requiring respondent burden or input, but rather, relies on data that is collected continuously in the background.

In general, the mechanisms driving willingness to contribute various data, irrespective of specific study demands, sponsors, and incentives [5] remain unclear. From the general patterns of attrition observed in longitudinal studies, it is apparent that a subset of participants may have a multitude of reasons for choosing to not engage in certain research activities.

The purpose of this study is to investigate the acceptability of active and passive data collection methods in mobile health research. This survey was designed to explore attitudes and opinions related to digital health technologies and willingness to share related data. The three research questions guiding this work are:

A total of 1509 adults were recruited from Prolific [8]. The Prolific platform allows researchers to post a ‘task’ with a short description for potential participants to review before beginning. In addition to the study background language, we the potential participants were informed: “This survey is designed to learn more about your attitudes and opinions related to digital health technologies and related data/data-sharing.” Two participants did not consent to participate, and 18 did not complete the survey and were dropped from the study, resulting in a total of 1489 participants with a mean age of 35.5 (SD 11.97) years.

This work was carried out in accordance with The Code of Ethics of the World Medical Association (Declaration of Helsinki). The Institutional Review Board (IRB) at the University of Central Florida approved the study protocol described in this manuscript (IRB No: STUDY00002281). The survey was pilot tested by members of the research laboratory conducting the study, for issues of usability and technical functioning. The IRB determined this study as exempt from regulation for human subjects research. Participants were provided an opportunity to accept or refuse consent before beginning this voluntary survey. The consent form indicated the expected time commitment, compensation, and nature of study. If consent was refused, the survey would close for that participant. Participants received US $2 after completing the survey on the Prolific Research Platform. All data was deidentified; the authors are accountable for all aspects of the work (including full data access, integrity of the data and the accuracy of the data analysis) in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Participants completed a battery of questionnaires via the Qualtrics survey platform, designed to assess sample demographic characteristics (eg, technology ownership, mobility), opinions about data control, and willingness to contribute various data for research. These modules were designed to assess various facets of attitudes and behavior relevant to longitudinal observational research, to determine heterogeneity in willingness to use devices or share data.

The participants completed a questionnaire via the Qualtrics survey platform, where the first page displayed the consent form and a radio button to select regarding consent. If participants consented, they were presented with modules on separate pages assessing: (1) demographic characteristics; (2) prior research experience; (3) mobility (ie, use of aids; driving frequency); (4) tech ownership (ie, own a tablet, smartphone, home Wi-Fi); (5) social media usage (ie, frequency of using Facebook, Twitter, Snapchat, Instagram, TikTok, WhatsApp); (6) willingness and extent of contributing various data (ie, across categories of activities, sensors, metadata); (7) opinions about data controls; and (8) willingness to use a robot for Instrumental Activities of Daily Living (IADLs). This survey totaled 52 questions over 12 pages presented in the same order for all participants (see Multimedia Appendix 1 for full survey). Questions for demographic reporting required responses, while all other items only suggested responses. Participants were only able to navigate forward and not backward through the survey.

This study reports findings related to the willingness and extent of contributing various data and opinions about data controls. All data were processed using R software (version 4.0.2; R Foundation for Statistical Computing) [9], using the tidyverse package [10], including ggplot2 for data visualization [11] and dplyr for data wrangling [12].

The sample was 44% female (n=652), with a mean age of 36 (SD 11.97, 20-84). A total of 99% (n=1479) of the individuals owned a mobile phone and 98% (n=1464) had access to home Wi-Fi (Table 1).

Participants were asked to consider a hypothetical study in which they were asked to complete three activities (queried independently), and rate how long in days they would be willing to complete each activity, for a maximum of 100 days (see Textbox 1 for full questions): (1) daily assessments of cognition; (2) wear and charge an air quality or pollution sensor; and (3) wear and charge a GPS location sensor.

Paired sample t tests were conducted to assess the differential willingness to share data from these three critical data streams, each previously implicated in cognitive health research (ie, mobile cognitive assessments, location data (GPS), and ambulatory air quality monitoring). On average, people were willing to complete ambulatory cognitive assessments for 56.7 (SD 36.2) days, air quality monitoring for 58.1 (SD 37.7) days, and GPS location monitoring for 37 (SD 39.0) days. Comparing mobile cognitive assessments and air quality monitoring—participants were willing to complete each for a similar period of time (ie, mean difference of approximately 1.4 d; t₁₄₈₈=−1.44; P=.14). Upon comparing mobile cognitive assessments and GPS location monitoring, participants were willing to complete cognitive assessments for close to 20 more days than GPS location monitoring (ie, mean difference of approximately 19.7 d; t₁₄₈₃=−17.89; P<.001). Comparing mobile air quality monitoring and GPS location monitoring – participants were willing to complete air quality monitoring for close to 21 more days than GPS location monitoring (ie, mean difference of approximately 21 days; t₁₄₈₃=−20.65; P<.001).

Participants were asked: “Please select the types of control you expect to have over the data submitted as part of a research study” Among the data control and sharing aspects assessed, 1042 (70%) participants were most interested in the ability to delete all their contributed research data, followed by the ability to select specific data streams for deletion (65%, 960) and deleting a specific moment in time (44%, 653). Respondents were least interested in the ability to share data with their insurance providers and caregivers (n=453, 30% and n=384, 25%, respectively; Figure 3).

When asked “Please rate how comfortable (5-point scale, from extremely comfortable to extremely uncomfortable) you would feel about each of the following situations related to your hypothetical data,” participant responses varied by data usage scenario (see Figure 4): (1) data used only for the purpose of the original study; (2) data used to develop algorithms; (3) data used by outside institutions collaborating on this research; (4) data used indefinitely for purposes unknown today; (5) data used to identify the participant; (6) data used to solve a public health problem; (7) data used to develop therapies and diagnoses. Unsurprisingly, participants were mostly uncomfortable with reidentification and with data being used for unspecified future purposes.. Respondents were mixed in regard to data being used by external collaborations. Participants were mostly comfortable with data being used for the original purpose, to develop therapies and diagnoses, and for algorithm development.

Although a large sample (ie, of approximately 1500 respondents) was recruited for this study, a major limitation is that participant demographics indicate a recruitment bias toward midlife, highly educated (69% have associates degree or higher), White participants (76%, n = 1134), and individuals active on online research platforms. As participation required familiarity with digital interfaces and online survey platforms, the sample may not adequately reflect the views of individuals with low technology adoption rates—particularly those in rural or socioeconomically disadvantaged communities [13]. Future research should consider probing similar research questions, leveraging a representative and more diverse sample of respondents, especially considering past research has shown differential willingness to share various data as a function of education and minority status. The extent to which older adults (ie, >60 years) are aware of and are willing to participate by sharing sensor data and metadata is unclear. Older adults were sparsely represented in this sample, limiting insight into the acceptability of sensor-based monitoring among individuals most likely to benefit from digital health interventions (eg, including those aimed at chronic disease mitigation or cognitive decline). Age-related factors such as cognitive load, sensory impairments, or comfort with technology may interact with perceived burden; however, these nuances were not explored in this study.

In addition, this study did not directly measure aspects of personality, or expectations for data privacy—two factors likely to explain variance in willingness to share data across the spectrum of participant burden. Participants’ decisions may have been influenced by recent media exposure or personal experiences related to data breaches or algorithmic discrimination, but these factors were not assessed. It is expected that awareness of specific events that highlight the reidentification potential of various sensitive data (eg, GPS) may moderate willingness to share sensitive data, along with the personality constructs of openness and consciousness [14]).

Importantly, the study relies on self-reported, hypothetical willingness to share data, which may not correspond with real-world behavior, particularly when actual study conditions involve real data collection amidst daily life schedules and hassles. Future work should validate these findings in live digital health trials to assess whether stated willingness translates to actual participation and data-sharing compliance. This study aimed to elucidate the types of data and control measures that participants expect in research. These findings can guide the optimization of study protocols to enhance participant engagement, data quality, and minimize participant burden. Ensuring participant privacy is paramount, and researchers have a duty to transparently communicate the rationale for data collection and its potential benefits. As researchers intend to collect rich data across the Active-Passive data continuum (see Figure 1), our participant’s privacy comes first, and it is our duty to inform them why we collect the data we do. This approach will foster trust and encourage greater participation in research involving potentially sensitive data streams.