The proliferation of mobile devices boosts the generation of person-generated health data (PGHD). In 2018, more than 80% of Americans owned a smartphone [1,2], with modest growth to 85% by 2021 [3]. Wearable health care devices were used by nearly 28% of the US population by 2020 [4,5]. Among smartphone users, over 50% are collecting “health-associated information” [6]. We use the phrase “person-generated health data (PGHD)” for health-related data created, recorded, or gathered by or from individuals, family members, or other caregivers to help address a health concern inside and outside clinical settings.

PGHD from mobile devices have significant research implications. Mobile devices could facilitate access to large pools of study participant data in a granular, longitudinal, and personal way, potentially accessing high-frequency data at low costs [7]. In this study, “mobile devices” refers to smartphones and wearables, which include fitness trackers and smartwatches; “research” covers biomedical and behavior studies [8].

Because of the above advantages, researchers have made numerous efforts to collect PGHD for mobile device research. Those efforts are enabled by participants willing to share their PGHD for research [9] and by using the informatics infrastructure needed for participants to share their data. Examples of informatics infrastructure to enable PGHD sharing for research dates from 2015 and 2016, with the releases of ResearchKit (Apple Inc) [10,11] and ResearchStack (Cornell Tech and Open mHealth) [12,13] and the launch of mPower [14], Sage Bionetwork’s first major smartphone-based health research study. Research practices dealing with PGHD from mobile devices are, in part, guided by the European Union’s General Data Protection Regulation (published in 2016; in effect since 2018) and the Food and Drug Administration’s guidelines on clinical research, medical devices, and mobile health apps (released in 2013, 2015, and 2019). Between 2000 and 2020, more than 12,000 mobile device–related health research papers were published [15].

Despite the potential advantages of using PGHD for mobile device research, the wide variety of practices to collect PGHD may lead to issues with data quality. These issues may be due to emerging research strategies such as Bring-Your-Own-Device (BYOD; which uses participant-owned devices instead of provisioned devices) that can miss some patient demographics [16], thus leading to low generalizability. Similar issues are noticed in studies using crowdsourcing platforms to collect data [17]. Selection bias with wearables is a significant issue due to differences in access [18,19], which in turn can lead to differences in sharing data for research [20]. Both accessibility and self-selection could lead to lower study sample diversity for race and socioeconomic status. For example, National Institutes of Health’s All of Us program allows participants to choose to contribute Fitbit (Google LLC) data through its BYOD subprogram. The All of Us BYOD subprogram has noticed a higher proportion of participants who are White, earn >US $25,000/year, and have college or advanced degrees in the BYOD subgroup compared with all participants [21]. When such data are used in machine learning, the low representation of some patient groups may lead to poor prediction accuracy when applied to those groups [22,23].

To better understand methodological issues such as those related to data representativeness and quality, we aimed to review publications on research collecting PGHD from mobile devices and to characterize the reporting elements of that research thoroughly, particularly those elements relevant to BYOD and crowdsourcing. In doing so, we sought to capture aspects of this field not addressed by previously published reviews. Several reviews explore mobile health research. A review by Cao et al [15] covering the years 2000 to 2020 and a review by El-Sherif et al [24] covering 2020 to February 2021 identified 4 leading contributors (the United States, the United Kingdom, Canada, and Australia). The 2 reviews, however, focused only on bibliometrics. A scoping review by Fischer and Kleen [25] published in 2021 investigated data collection by smartphone apps in longitudinal epidemiological studies, and they found a limited number of studies that integrated apps in data collection. Studies with cross-sectional designs were excluded. A scoping review by Huhn et al [26] published in 2022 summarized wearables in health research and found that most studies were observational and that 93% of the participants were in global health studies. Smartphones were not included in their scope. Other reviews of interest include one that briefly mentioned data collection using mobile devices for health research [27] and another that discussed the application of sensors (eg, inertial measurement units) in health care [28]. Our review differs from these reviews by covering wearables and smartphones and their use to collect PGHD in research. We also attempt to be comprehensive in our characterization of reporting practices, our inclusion of a range of study types (eg, both longitudinal and cross-sectional studies), and our examination of bibliographic information. To the best of our knowledge, ours is the first review of publications on research collecting PGHD from mobile devices and the first to examine BYOD practices and the use of crowdsourcing platforms in this field.

This review characterizes the reporting elements of included publications and discusses the relevance of reporting practices for supporting reproducible and responsible research. There are no current, ready-to-use guidelines for reporting research involving PGHD from mobile devices. For example, the Future of Privacy Forum’s “Best Practices for Consumer Wearables & Wellness Apps & Devices” dates back to 2016 [29]. More recently, the Office of the National Coordinator for Health Information Technology released a white paper in 2018 to conceptualize a data infrastructure for the capture, use, and sharing of PGHD in care delivery and research through 2024 [30]. The final document, however, is still under development. Also relevant are the Structured Template and Reporting Tool for Real-World Evidence (STaRT-RWE) (2021) [31] and the Mobile Health Evidence Reporting and Assessment (mERA) checklist (2016) [32], although both are limited. STaRT-RWE does not specify requirements for mobile device studies, and mERA is limited to health interventions. There are also some focused guidelines for app development (2016) [33] and data integration (2021) [34], but none fully cover what is needed for high-quality reporting of mobile device research. In fact, reporting deficiencies are widely noted in studies involving mobile devices. A systematic review by Olaye et al [35] in 2022 found that 30% of publications failed to report quantitative adherence; such missingness can impede a complete understanding of study data. Our review characterizes the magnitude and types of reporting problems in publications on research collecting PGHD from mobile devices.

A PubMed search was used to find instances of explicitly described participant data contributions (eg, through the use of language such as “data donation,” “data sharing,” etc) involving wearables, smartphones, or mobile apps (Textbox 1). A MeSH (Medical Subject Headings) term analysis of a preliminary sample of articles was conducted to support PubMed query development (Yale MeSH Analyzer [37]). We selected search terms to balance recall and precision in the search results. We ran the PubMed search on August 7, 2021.

Covidence [38] was used to screen publications for eligibility in 2 steps, first by eliminating publications based on a title and abstract review and then by reviewing the full text of the remaining journals. A pre-established set of eligibility criteria was used for screening (Table 1). As this an emerging field and our motivation to see the trends in full landscape, we did not set criteria for time range. Screening decisions were made only after 2 independent reviewers on the screening team (SS, MA, and RHY) agreed. All conflicts were resolved by group discussion. Library journal subscriptions and interlibrary loan were used to obtain full text, with the full text of 1 publication [39] not available from either source. All reviewers conducted pilots for each round of screening.

We charted publications for bibliographic details and along 5 reporting dimensions (“participant,” “device,” “data,” “study,” and “ethics”; double quoted here and in the text below). Because our focus is on health data contributions from participants through BYOD and crowdsourcing platforms, we modeled the reporting dimensions on existing work in the areas of citizen science data contribution [40], bioethical approaches to using personal health data [41], and mobile health apps [42]. Upon reviewing the keywords of included publications (eg, “ethical code,” “consent,” and “informed consent”), we summarized frequent themes of keywords (eg, consent). Furthermore, we refined our charting template by selecting factors related to those frequent themes to improve data adequacy. The final factors included under the 5 reporting dimensions are listed in Table 2. A data charting pilot with a random sample of one-fifth of the final pool of publications was conducted before finalizing the data charting template [43]. A team of 4 reviewers (SS, MA, RHY, and ZL) charted the data using the template built in Covidence [38]. In addition, 2 reviewers independently charted each publication. Consensus was reached by group discussion.

To determine the leading countries and their contributions to various publication types, we compared the counts and percentages of publications by type and location. In addition, we visualized the number of publications by year to evaluate publication trends. We also visualized publication counts by year and location for original research publications. Reporting dimensions, including factors concerning “participant,” “device,” “data,” “study,” and “ethics” (Table 2), were analyzed for original research publications only. We charted the frequency and percentage of original research studies that included these 21 reporting dimension factors.

In addition, 4 factors with multiple-choice options were visualized using Venn (if options ≤3) or alluvial (if options >3) diagrams after excluding publications that did not report these factors. These 4 factors included 1 “device” dimension factor (device owner), 2 “data” dimension factors (data type and data owner), and 1 “study” dimension factor (research scenario). We also analyzed publications according to the frequency of data types collected. We assessed missing factors among original research publications to evaluate changes in reporting deficiencies for each reporting dimension. We compared the percentages of publications missing factors between 2 temporal groups (2010-2016 and 2017-2021). All analyses and visualizations were performed using R (version 4.2.1; R Core Team).

Our search strategy yielded 3206 results. A total of 100 publications remained following screening [50-149]. All 100 articles are included in our data analyses (Figure 1; bibliographic information in Multimedia Appendix 2).

Publication year and location were extracted from the 100 included articles and organized by publication type (Figure 2, Multimedia Appendix 3). The majority of publications had a first author from the United States (46%, 46/100), European Union (20%, 20/100), United Kingdom (11%, 11/100), or Canada (6%, 6/100). Publications designated as “Research” (original research) made up the largest proportion of publications (45%, 45/100). These were followed by publications designated as “Tech” (20%, 20/100), “Other” (17%, 17/100), “Review” (13%, 13/100), and “Protocol” (5%, 5/100). Refer to Table 2 for definitions of these publication types.

There was a spike in total publications in 2016, with 11 publications in 2016 compared with 3 or fewer annually beforehand. Since 2016, a minimum of 9 publications have been published annually. There was another spike in output in 2019, with 19 publications. Other trends observed were US-dominant “Tech” publications and diversely geolocated “Review” publications.

We have characterized the reporting practices of published research studies that collected PGHD using mobile devices. Our findings show that the United States has gained momentum in publishing in this area over the past decade (Figure 2). In addition, a high proportion of the original research articles we identified used participant-owned devices or a BYOD strategy if they reported device ownership (75%, 21/28; Figure 3A), with few studies collecting data using crowdsourcing platforms (7%, 3/45; “Used crowdsourcing platform to collect data” in Table 3). We also found reporting deficiencies in the “data” and “ethics” dimensions among original research articles, where most original research articles were deficient in at least 1 factor concerning “data” (96%, 43/45) or “ethics” (96%, 43/45; Figure 4).

Our findings highlight several areas to consider when reporting research with PGHD collected using mobile devices. Recommendations for reporting based on findings from this review are summarized in Table 4. Further work is needed, however, to gain consensus on a more comprehensive set of reporting items. Such a consensus on best practices would greatly facilitate responsible and reproducible research and improve research quality and impact [172]. For example, a 2012 systematic review indicated a positive association between journal endorsements of a reporting guideline and the quality of reporting in randomized clinical trials published in those journals [173]. Another study looking at the publications from one journal found that those publications whose papers received editorial interventions designed to ensure adherence to reporting guidelines were more highly cited, with citation counts 43% higher than those publications that had not received the editorial interventions [174].

Our review has limitations related to how we gathered evidence from the literature, our choice of reporting factors, and one of our data analysis methods.

Our PubMed search strategy (Textbox 1) has concepts with explicit data donation and data sharing terms to retrieve publications that describe participant data contributions. Because of this design, the search strategy only retrieved publications whose titles and abstracts had data donation and data sharing terms. Possibly, our search missed relevant publications that described participant data contributions only in the full text. The articles we analyzed were only collected from PubMed. Articles were not retrieved from other databases (eg, Embase or Web of Science), from the reference lists of included articles, or from the gray literature. Thus, it is possible that articles only available through these other sources were overlooked. Our preliminary searches, however, suggested that PubMed provided literature coverage sufficiently comprehensive for our research objectives.

While we selected reporting factors relevant to our goal of reviewing research that collects PGHD using mobile devices with the perspective of an individual’s data contribution, we may have excluded relevant factors. For example, we did not include the demographic factors of study populations, which may be more broadly relevant to wearable PGHD research. Detecting the low representation of some groups when using wearables for PGHD collection can indicate where biases exist that could lead to downstream inequities. This need is apparent based on findings that aging populations were underrepresented in studies with wearables [175] and that studies of cardiovascular disease that involve wearables [175]. Such a need requires domain-specific examinations, and we regard it out of the scope of this review, which aims to explore the landscape as the first step. As an example of a second-step effort, we have investigated the studies of cardiovascular disease that involve wearables, and we noticed that they did not reflect the demographics of patient populations in terms of age, race, education, cigarette smoking status, and hypertension status [152].

We lowered the granularity of some factors to avoid issues of imbalance in categorical analysis. For example, when analyzing research scenarios, we consolidated academic institutions, for-profit commercial companies, corporations or foundations, independent research organizations, patient-led groups, citizen scientists, and so on, into 3 categories. Thus, our findings do not necessarily represent the full range of possible ways of describing this research.

Despite these limitations, we are confident that having used strict procedures to reduce bias (eg, 2-level screening, independent charting, and a predefined charting template), this review offers a solid foundation for future research. We also feel that the full PubMed query we provide (Textbox 1) is a valuable contribution to other researchers since it was developed and tested for wearables and smartphones and since publications in the emerging field of mobile devices and PGHD are not well indexed.

We identified trends and patterns in the reporting of research that collects PGHD using mobile devices. Given the growing interest in using a BYOD model among researchers, there is an opportunity for a broader use of crowdsourcing platforms for data transfer from participant-owned devices. There is also the opportunity to develop best practices that address observed reporting deficiencies in this research and make it more reproducible and responsible. Based on our findings, we recommend 9 items for enhanced reporting.