Skip to main content
PMC home page
Journal of the American Medical Informatics Association : JAMIA logoLink to Journal of the American Medical Informatics Association : JAMIA
. 2018 Nov 16;25(12):1685–1695. doi: 10.1093/jamia/ocy130

Leveraging mobile health applications for biomedical research and citizen science: a scoping review

Hannah Schmitz 1, Carol L Howe 2, David G Armstrong 3, Vignesh Subbian 1,4,✉
PMCID: PMC7647150  PMID: 30445467

Abstract

Objective

This systematic review aims to analyze current capabilities, challenges, and impact of self-directed mobile health (mHealth) research applications such as those based on the ResearchKit platform.

Materials and Methods

A systematic review was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement. English publications were included if: 1) mobile applications were used in the context of large-scale collection of data for biomedical research, and not as medical or behavioral intervention of any kind, and 2) all activities related to participating in research and data collection methods were executed remotely without any face-to-face interaction between researchers and study participants.

Results

Thirty-six unique ResearchKit apps were identified. The majority of the apps were used to conduct observational studies on general citizens and generate large datasets for secondary research. Nearly half of the apps were focused on chronic conditions in adults.

Discussion

The ability to generate large biomedical datasets on diverse populations that can be broadly shared and re-used was identified as a promising feature of mHealth research apps. Common challenges were low participation retention, uncertainty regarding how use patterns influence data quality, need for data validation, and privacy concerns.

Conclusion

ResearchKit and other mHealth-based studies are well positioned to enhance development and validation of novel digital biomarkers as well as generate new biomedical knowledge through retrospective studies. However, in order to capitalize on these benefits, mHealth research studies must strive to improve retention rates, implement rigorous data validation strategies, and address emerging privacy and security challenges.

Keywords: mHealth, mobile applications, ResearchKit, smartphone

BACKGROUND AND SIGNIFICANCE

The increasing prevalence of mobile devices among the general population provides a unique opportunity to conduct social and biomedical research at-scale. Currently, there are more than 50 000 non-interventional, observational studies registered in ClinicalTrials.gov.1 The vast majority of these studies are coordinated in-person, require direct interaction with researchers, and the sample sizes are limited by time, geographic, and budgetary constraints. In order to break through many of these longstanding barriers to biomedical research studies, several open-source research platforms (eg, ResearchKit and ResearchStack) have leveraged smartphones as research tools to appeal to widespread, multi-demographic groups and enable any eligible person to consent and contribute data remotely through his or her own smartphone.2 In particular, the release of Apple’s open-source platform (ResearchKit) in March 2015 has led to widespread use of mobile health (mHealth) applications (apps) to collect and curate large amounts of data that will enable researchers to generate and test hypotheses of significance to public health, including chronic conditions, behavioral and mental health, and disability.

Compared to traditional in-person, biomedical studies, some of the benefits that mobile applications for medical research might provide include improved subject recruitment, comprehensibility of consent and data collection processes, cost effectiveness, elimination of geographic barriers to participation, broader participation from the community including individuals from marginalized subpopulations, and the potential to create large datasets for secondary analyses. On the other hand, there are several limitations to consider, including but not limited to, participant retention, self-selection bias, reliability and quality of patient-generated data collected through mobile applications, and ethical, legal, and social issues pertaining to citizen science and mHealth technology. In this paper, the term “citizen science” is typically used to refer to any large-scale, non-interventional study that solicits voluntary participation of general citizens in scientific research.3,4

Prior studies on mHealth have been focused largely on using mobile apps for personal health monitoring or as an intervention to improve care processes.5 However, studies related to use of mobile apps for large-scale health-related research are limited. The objective of this systematic review is to analyze the current capabilities, challenges, and impact of non-interventional, self-directed mHealth research applications.

MATERIALS AND METHODS

Search strategy

A systematic literature review was performed using methods specified in the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement for reporting systematic reviews and meta-analyses.6 Both controlled vocabulary terms (eg, MeSH, EmTree) and key words were utilized to search the following databases for publications that focused on the development or use of smartphone applications to gather data remotely in the context of medical research: Ovid/MEDLINE, PubMed/MEDLINE, Elsevier/Embase, Elsevier/Scopus, Wiley/Cochrane Library, Thomson-Reuters/Web of Science, EBSCO/CINAHL, Elsevier/El Compendex, IEEE Xplore, and ClinicalTrials.gov. The Apple App store and Google Play were also searched for apps that were developed to gather medical research data remotely. Literature searches were completed on July 25, 2018. The complete PubMed/MEDLINE search strategy is available in Supplementary Appendix S1.

Inclusion criteria were 1) mobile applications were used in the context of large-scale collection of data for biomedical or clinical research, and not as medical or behavioral intervention of any kind; 2) all activities such as informed consent to participate in research and methods used for data collection were executed remotely without any face-to-face interaction between researchers and study participants; 3) publications and applications were in English; and 4) dates of publication were after January 1, 2006. All study and publication types were considered with the exception of reviews, systematic reviews, meta-analyses, and editorials/opinion pieces; conference abstracts and ongoing and completed clinical trial listings were included. During the first set of searches completed on April 17, 2017, 2 independent reviewers screened titles and abstracts of retrieved references for relevance and resolved disagreements by consensus. They similarly screened the full text of the articles selected from title and abstract screening to ensure that they met the full set of inclusion criteria and resolved disagreements by consensus. Searches were rerun on February 26–27, 2018, and again on July 25, 2018; titles and abstracts for the updated searches were screened by the primary author, as were the full text of selected studies.

RESULTS

Four open-source platforms were identified that enable development of mobile applications for research purposes: 1) EmotionSense, 2) Ohmage, 3) ResearchKit, and 4) ResearchStack. ResearchKit and ResearchStack are, respectively, Apple’s and Android’s open-source platforms, which enable rapid development of mHealth research apps that require data collection related to overall health and well-being such as cognition, fitness, hand dexterity, motor activities, and voice. EmotionSense, on the other hand, collects data from questionnaires, location, and voice recordings specifically for assessing emotions; as such, it is used primarily for social and psychological research.7 EmotionSense was validated as an individual app, and the literature search did not locate any additional studies conducted through the app.7 Ramanathan et al.8 states that, at the time of the publication, the Ohmage platform was being validated by 7 independent studies of different populations, including recent cancer survivors, men at risk for HIV, and individuals with ADHD; however, no additional literature was identified on any such studies. ResearchKit emerged as the predominant platform implemented in existing mHealth research apps; thus, we focus on ResearchKit-based apps and studies in this review.

Database searches resulted in the identification of 6643 studies. Of the 3703 articles that remained after duplicates, non-English, and pre-2006 publications were removed, 3645 were excluded because of irrelevance to the topic (Figure 1). Strict inclusion criteria, as outlined in the Methods section, were applied to the full text of 58 articles and an additional 19 mobile apps identified through the Apple App and Google Play stores. Of these 77 resources, 49 met the full set of criteria; this included 22 articles (relevant to 17 distinct ResearchKit apps),5,7–34 the additional 19 apps (without related publications) found on Apple App and Google Play stores,35–53 and 8 articles related to mHealth platforms and frameworks. Overall, 36 unique ResearchKit apps were identified. Three apps had been reproduced using ResearchStack and made available through Google Play.15,18,19 The ResearchKit platform has been available only since March 2015. Thus, the majority of ResearchKit-based studies cited here have not yet progressed to the point of published data; only 7 apps were identified to have published results.11,14,19,21,22,28,34

Figure 1.

Figure 1.

Open in a new tab

PRISMA flowchart illustrating the process of literature search and extraction of studies meeting the inclusion criteria.

A summary of research purpose and mode of data collection for apps identified through the database searches and apps identified through the app stores is described in Tables 1 and 2, respectively. Additionally, information regarding the purpose of each framework can be found in Table 3. Of 36 apps included in the review, 16 of them (44%) investigate chronic conditions such as Parkinson’s disease, diabetes, COPD, and heart disease.9,11,14,20,21,28,36,37,44,46,49,50,52,53,56,57 Three of the applications (8%) focus on screening conditions such as mole growth, depression, and emotions.19,34,47 For example, the developers of Autism & Beyond are using facial expressions to determine the emotions being displayed. Four applications (11%) are intended to address behavioral issues such as impulsiveness, photo-sharing on social media, energy drink consumption, and relating gait to emotions.15,38,41,43 The most commonly addressed conditions are heart disease (4 apps)14,17,44,57 and cancer (3 apps).19,21,53

Table 1.

Apps identified through database searches

App Name Condition Purpose (To determine…) Data Typesa(Mode of Collection) 1st Author/ Sponsor, Year, Reference
Apps that collect both survey and sensor-based data
 mPower Parkinson's disease (PD) Potential to monitor PD symptoms and their severity and sensitivity to medication and to assess qualitative patient experience of eConsent Parkinson Disease Questionnaire 8 (survey), short-term spatial memory and dexterity and speed (screen tapping), gait and balance (accelerometer and gyroscope), sustained phonation (microphone) Abujrida et al., 201731; Bot et al., 201628; Doerr et al., 201730; Neto et al., 201629
 VascTrac Peripheral artery disease (PAD) Trends between patient-reported PAD outcomes and walking data Medical data (survey), physical activity (motion sensor) Ata et al., 201732
 6th Vital Sign General health Method to measure walking speed that is comparable across subjects Personal health (survey), walking speed (motion sensor) Bettger et al., 201633
 C Tracker Hepatitis C Feasibility of monitoring changes in medication therapy and associated effects on workforce productivity and physical limitations Changes in medication therapy (survey), physical activity (motion sensor) Boston Children's Hospital, 20176
 EpiWatch Epilepsy Whether EpiWatch can be validated to detect seizures and to monitor seizure experience, study medication side effects, and identify triggers Heart rate and movement (Apple Watch motion sensors), seizure experience (survey) Carmenate et al., 201710
 Asthma Health Asthma Asthma trigger data, quality of self-reported data, app user characteristics, and clinical impact Symptoms, triggers, medication adherence, healthcare utilization, and quality of life (survey), geographic conditions (GPS) Chan et al., 201711; Gowda et al., 201612
 SleepHealth Sleep Correlations between sleep habits and physical activity and alertness Sleep assessment activities (motion sensor) Deering et al., 201713
 Heart & Brain Heart attack/ stroke Feasibility of collecting quality data for heart and brain health Stroke risk factors and quality of life (questionnaire), pulse rate (Apple Watch), neurological data (gyro sensor, camera for face/limb/motor disorders) Kimura et al., 201614
 MyHeart Counts Cardiovascular health/physical activity Effect of daily activity (step count) measures on cardiovascular health, whether 4 different activity prompts impact physical activity, the correlation of genetic data Stepcount frequency and duration and sleep duration (motion sensors and survey), reported happiness and cardiovascular disease risk (survey) Stanford University, 201717
 AMWALKS or America Walksd Walking behavior U.S. walking behavior and the correlation between perceived and actual walking activity Stepcount (survey and motion sensor) TrialX, Inc, 201618
 Mole Mapperd Melanoma Potential to collect data pertinent to melanoma risk Melanoma risk factors (survey), mole sizes (phone camera) Webster et al., 201719
Apps that predominantly collect survey-based data
 Depression Monitor Depression Characteristics of depression app users, and if a depression intervention is necessary Risk of depression (patient health questionnaire, PHQ-9) BinDhim et al., 201534
 DMTbmPulsed Impulsive behavior Frequency of voluntary app use, quality of impulsivity assessment Impulsivity self-assessment (survey), performance tasks, passive data collection Northwell Health, 201815
 PARADEc Rheumatoid arthritis (RA) RA symptoms, mood, and medications over time Symptoms, medications, and comorbidities (surveys) Williams et al., 201720
 Prostrate PRO-Tracker Prostate cancer radiation therapy Potential to collect patient-reported outcomes throughout prostrate radiotherapy Urinary, bowel, sexual, and hormonal quality of life data (Expanded Prostrate Index Composite for Clinical Practice (EPIC-CP) survey Yang et al., 201621
 Back on Track Acute anterior cruciateLigament (ACL) tears Outcomes of different treatment methods for acute ACL tears Subjective experience of injury (survey) Zens et al., 201722
 WebMD Pregnancy Pregnancy Effect of weight on pregnancy outcomes Weight, pregnancy complications, birth outcomes (surveys) Radin et al., 201816
Open in a new tab
a

Demographic surveys are not listed under “Data Types” because this feature is common to all cited apps.

b

DMT: Digital Marshmallow Test.

c

PARADE: Patient Rheumatoid Arthritis Data from the real world.

d

Available on ResearchKit and ResearchStack platforms.

Table 2.

Apps identified through Apple app and Google Play Stores

App Name Clinical Condition Purpose (to determine…) Data Type Catogorya Sponsor, Reference
Feverprints Diseases characterized by fevers How body temperature varies between individuals and the correlation of body temperature patterns with different diseases Survey, external service/feature Boston Children’s Hospital35
STOPCOPD Chronic obstructive pulmonary disease (COPD) Correlations between physical activity, peak expiratory flow, caloric intake, heart rate, oxygen saturation, and sleep patterns Fitness COPD Foundation, DatStat team36
Hand in Hand HIV associated neurocognitive disorders (HAND) Cognitive effects of HAND Cognition Digital Artefacts, UNMC46
Autism & Beyond Autism and other mental health disorders whether video collection can be validated as a means to identify emotions and behavioral characteristics Camera Duke Medicine47
TeamStudy Football, long-term effects Improved diagnostic and treatment interventions through analysis of memory, balance, heart health, pain, and mobility Survey, cognition, motor activities, fitness Harvard University48
COPD Navigator COPD Correlations between geographic data (ie air quality, weather, allergens) impact COPD, medications Survey, location LifeMap Solutions, MSSM49
GlucoSuccess Diabetes (type II) Correlations between blood glucose values, dietary choices, activity levels, and health behavior questionnaires (sleep quality, foot inspections, medication adherence) Survey, fitness Massachusetts General Hospital50
Concussion Tracker Concussion Whether Concussion Tracker can be validated to collect patient-reported outcomes (heart rate, physical and cognitive function) Survey, fitness NYU Langone Health51
Sarcoidosis Sarcoidosis Variables impacting quality of life for patients with sarcoidosis Survey Penn Medicine52
Share the Journey Health after breast cancer treatment Trends in fatigue, mood and cognitive changes, sleep disturbances, reduction in exercise for people who have been treated for breast cancer Survey Sage Bionetworks53
Neurons Multiple sclerosis MS symptoms through walking, dexterity, and Paced Auditory Serial Addition Test (PASAT) activities Fitness, cognition, hand dexterity Shazino37
Saker Gait and emotions When someone is scared based on motion sensors that monitor the user’s gait Motor activities Thomas Morrell38
uBiome Gut bacteria/ microbiome Relationship between weight management and the microbiome Survey, external service/feature uBiome39
PPD ACTb Postpartum depression (PPD) Why specific subpopulations experience PPD Survey UNC School of Medicine40
mTECH Energy drink consumption Correlations between energy drink consumption and blood pressure, heart rate, daily step count, and sleep activity Survey, fitness University of the Pacific41
PRIDE Study LGBTQ health Health correlations among the LGBTQ population Survey UC San Francisco, THREAD Research42
Biogram 2 Photo-sharing behavior Whether there is a relation between photo-sharing in social media and heart rate, weight, and step count Survey, fitness USC Center for Body Computing, Medable43
Yale Cardiomyopathy Index Cardiomyopathy How cardiomyopathies or risks of developing cardiomyopathies affect perceived quality of life and physical health (heart rate and walking limitations) Survey, fitness Yale University44
Yale EPV Pregnancy Correlations between estimated placental volume (EPV) and birth outcomes External service/feature Yale University45
Open in a new tab
a

The data type category divisions were adapted from the ResearchKit Active Tasks categories (motor activities, fitness, cognition, voice, audio, hand dexterity),54 along with a few additional categories (surveys, camera, location, and external service/feature). The “External service/feature” category refers to the requirement of any service or technology that is not provided from within the app itself (eg, thermometer, external devices such as Apple Watch).

b

PPD ACT: Postpartum Depression Action towards Causes and Treatment.

Table 3.

Frameworks

Frameworks Purpose 1st Author/ Sponsor, Year, Reference
Fast Healthcare Interoperability Resources (FHIRframe) To establish data standards and an application program interface (API) for the exchange of health data Anwar, 201523
Ontology for Defining mHealth and Mapping Research Foci To establish an ontology to provide a common definition for mHealth and as a resource to quantify areas of mHealth research Cameron, 20175
Comparison of ResearchKit Consent to Standard Consent To validate the eConsent process used in ResearchKit Duke University, 201724
REDCap/ResearchKit Clinical Decision Support To design a clinical decision support tool compatible between ResearchKit and REDCap API Metts, 201725
C3-PRO: the Consent, Contact, and Community framework for Patient Reported Outcomes To design an open-source toolchain for connecting ResearchKit apps with the clinical research IT infrastructure i2b2 by implementing FHIR Pfiffner, 201626
Open Source Ecosystem Health Operationalization (OSEHO) To evaluate the quality and health of open-source software ecosystem Vulpen, 201755
Open in a new tab

Most apps enrolled participants 17/18+ years old; 3 allow children to participate with a guardian’s permission or co-participation.35,38,47 Two allow participation from 12+ year old.40,43,56 Several of the apps allow subjects who have not been diagnosed with the disease/condition of interest to participate as controls, while a few of the apps accept only subjects with self-reported, clinical diagnosis. The majority of apps are restricted to iPhone users only (with the exception of 4 apps available on Android),15,18,19,40 English only (3 exceptions),38,40,47 and U.S. only (1 exception).47

Most apps were conducting longitudinal, observational studies or simply amassing data that could be analyzed at a later time. Given the novelty of mHealth apps, a component of the purpose for all apps was to conduct preliminary research to validate the ability to recruit a large number of participants. Recruitment and consent data and demographic data for apps that had such data available are listed in Tables 4 and 5, respectively. Many studies relied heavily on survey data, whereas others utilized built-in phone sensors (see Table 1). Two applications required additional devices or services in order to utilize the app; EpiWatch required an AppleWatch, and uBiome required DNA-sequencing of gut bacteria.39,56

Table 4.

Available recruitment and consent data

App Number of Downloads # Participants Consent Rate (persons/month) Study Period (months)
Asthma Health 40 683 6470 1078 6
Back on Track 953 549 61 9
Depression Monitor 8241 6089 1522 4
Mole Mapper 11 056 2069 373 7.5
mPower 48 104 9846 75 12
PARADE 1170 399 399 1
WebMD Pregnancy 49 411 1889 210 9
6th Vital Sign 771 361 226 1.6
Open in a new tab

Table 5.

Available demographic statistics

App Gender (% female participants) Age in Yearsmean ± SD (range) % Participants in 18–34 age range % Participants with more than high school education
Asthma Health 39% NR(18–65+) 60% 89%
Back on Track 26% NR (18–54) 54% NR
Depression Monitor 57.8% 29.4 ± 10.8 (18–72) 70% 46.4%
Mole Mapper 41.7% NR (18–85+) ∼50% NR
mPower 21.6% 35.9 (NR) NR 57.5%
Open in a new tab

NR = not reported.

All publications and apps included in the review were analyzed for recurring themes and challenges related to research protocols and technology implementation. Prominent themes include drop-offs in recruitment rates and low patient retention, non-representative demographic distributions, lack of information on how variable mobile use patterns could influence data quality, need for data validation, and privacy and security concerns related to use of mHealth apps for biomedical research and citizen science.

DISCUSSION

Based on our review and how traditional clinical trials are designed and deployed, we discuss 3 major areas as they relate to mHealth research apps, specifically those that are ResearchKit-based: 1) data generation and sharing, 2) participant recruitment and retention, and 3) privacy and security challenges.

Data generation and sharing

One of the most promising features of mHealth research apps is the ability to generate large biomedical datasets on diverse populations that can be shared and re-used for purposes of secondary research. To this end, it is critical to strategically plan and implement infrastructure that can promote data sharing and data quality through findable, accessible, interoperable, and reusable (FAIR) principles58 and to implement safeguards to protect data privacy.

To address data quality, apps such as Depression Monitor and Back on Track relied on surveys that were previously validated for self-administration.22,34 Other apps that used mobile sensor data employed methods to validate specific tasks such as motion sensing in the mPower app28 and self-directed skin examination and documentation in the Mole Mapper app.19 Nevertheless, there is a need for provisions to validate and ensure high quality of mHealth data. Currently, multiple frameworks have been proposed to evaluate general mHealth apps, but a single recognized framework specifically for mHealth research apps has yet to be established.59

Vulpen et al. applied select metrics from the Open Source Ecosystem Health Operationalization (OSEHO) framework to evaluate the quality of the ResearchKit platform as a whole.55 This study showed that ResearchKit enables productive app development through predesigned features; however, the robustness is seriously threatened by links to the several external, third-party tools and systems, which are being further incentivized by potential selection bias of Apple users.55 While this indicates that market attention will likely shift away from ResearchKit in the future, the lessons learned and data generated from the first wave of ResearchKit apps will nevertheless be informative for the next generation of mHealth research apps.

Several collaborative research programs are actively developing tools to increase the quality of mHealth data. Supported by the Center of Excellence for Mobile Sensor Data-to-Knowledge (MD2K), mProv is an initiative focused on the development of cyberinfrastructure for annotating mobile sensor data with source, provenance, validation metrics, and semantics.60 MD2K also aims to develop and validate new digital biomarkers from mHealth data.61 Additionally, the Open Humans Project provides a centralized platform to deposit, share, and privately store personal health data, including datasets generated and curated through the ResearchKit apps.62

To promote interoperability, the Fast Healthcare Interoperability Resources (FHIR) framework seeks to establish standards and a common application program interface for the exchange of health data between apps and personalized health records.23 FHIR details the need for in-app embedded design and platform-agnostic services. This level of interoperability is yet to be widely implemented in mHealth research apps. For example, the Consent, Contact, and Communication framework for Patient-Reported Outcomes (C3-PRO) provides an open-source toolchain to connect ResearchKit apps with clinical research infrastructure such as Informatics for Integrating Biology and the Bedside (i2b2).26 These efforts, together, demonstrate a promising trend towards a comprehensive ecosystem of tools and resources to deploy mHealth technology for biomedical research.

Retention and recruitment

Participant retention and recruitment are major pitfalls of clinical research that mHealth studies may remedy to an extent. Relative to traditional in-person clinical trials, mHealth studies have the ability to recruit from larger and more diverse populations in a more cost-effective manner.2 mHealth apps such as those based on ResearchKit allow for recruitment of subjects with rare conditions, stigmatized disorders, as well as those from marginalized populations.63–65 However, mHealth recruitment does face challenges in promoting the study and fostering an understanding of consent remotely. Additionally, it should be noted that development, deployment, and maintenance of ResearchKit-based apps is complex, often requires support from third-party vendors, and adds significant cost overhead to the overall study. With more demand and availability of readily interoperable systems (eg, Medable), these costs may be reduced in the future.66

Recruitment strategies

Potential study participants typically learn about clinical trials through listings in the ClinicalTrials.gov database or physician recommendation. However, the majority of apps were not listed on ClinicalTrials.gov. Physician recommendation is not yet viable on the national scale relevant to mHealth apps. Alternatively, exposure to ResearchKit apps was possible in news articles, eg, Forbes67 and The New York Times;68 within social media, eg, Twitter;69 and listings within the Apple App store such as search results in the “Health and Fitness” category.

Several publications commented on potential influences on study exposure. Multiple studies noted higher recruitment rates nearest the times of app release.11,34 The Depression Monitor research found that its initially high enrollment rate was well correlated to when the app was unintentionally featured in the Apple App Store.34 Asthma Health hypothesized that its initially high recruitment rate was a result of higher media attention upon release.11 Given that ResearchKit apps were relatively novel throughout the duration of these studies, media attention could be influencing the recruitment rates for all ResearchKit apps. Given time, passive recruitment rates may rise due to increased prevalence and common knowledge of self-directed mHealth research.

Retention

Both mHealth and standard clinical studies experience problems with patient retention. However, given the potential for large-scale data collection in mHealth studies, data trends and participant feedback can be utilized to improve retention. The main solutions that were proposed in the literature to combat low patient retention were increased participant engagement and feedback through 1 or more strategies such as gamification, monetary rewards, and utilization of push notifications.11,22,30 Until participant retention can be improved, mHealth studies can still be useful for developing large datasets that can be used as hypotheses-generating tools.11

In spite of retention difficulties, mHealth studies have demonstrated higher overall retention numbers than standard clinical trials. One of the first ResearchKit-based apps that was developed and deployed to study Parkinson’s Disease (PD) – mPower, recruited more than 9000 participants, of which 898 consented to broad data sharing and contributed 5 days’ worth of data.28 Traditional PD-related clinical trials, on the other hand, had sample sizes ranging from a few dozen to around 150 participants.70. Even with its steep dropoff rate, the mPower study was able to recruit and retain more subjects than any standard clinical trials for PD. Consent and recruitment rates for all ResearchKit apps are summarized in Table 4.

Electronic consent

The majority of people who downloaded any ResearchKit app did not consent and contribute data to the study (see Table 4). This suggests that many people were interested in the study but were deterred by the enrollment and electronic consent (eConsent) processes. Challenges in communicating an understanding of consent documents warrants careful consideration by apps necessitating self-administered consent because, unlike standard in-person studies, a researcher cannot be present during the consent process to answer any questions and aid in a participants’ understanding of the research and potential risks associated with it.

The mPower app attempted to improve the experience of the remote consent process by using visually engaging graphics and maintaining transparency through incorporating an explicit decision point to determine the extent of data sharing.28 Furthermore, several apps’ consent processes required that subjects pass brief quizzes to demonstrate their comprehension of the study principles, their own rights, and data sharing options.16,19,28 The mPower study team conducted a qualitative study using open-ended responses from their participants on the consent process. In doing so, they were able to demonstrate that despite their strategies to communicate consent, many of its control group participants (those without PD) misunderstood their role in the study.30 Further research is needed to demonstrate how comprehension of eConsent protocols compare to standard in-person consent.

Potential selection bias

All but 3 of the apps15,18,19 were available only to Apple iPhone users with additional restrictions on the operating system version, which may lead to participant selection bias. Even among those who own iPhones, network data requirements could prevent some users from participation. The population that tends to use iPhones most frequently is more educated, younger, and wealthier than the average American citizen.71 Five out of the 7 apps that have published results included demographic data.11,19,22,30,34 Three out of the 5 had a majority of male participants.11,19,22 All 5 indicated that the most common age for participants was less than 35. Multiple studies suggested expanding their app to Android operating systems, the user base of which better represents the American population.11,19 Alternatively, the Back on Track study justified that it was exempt from this bias because its app engaged a target population with demographics comparable to those of iPhone users.22 In addition to smartphone requirements, some apps were further restricted by the requirement of additional devices or services to utilize the app; the EpiWatch app required an AppleWatch, and the uBiome app required DNA-sequencing of gut bacteria.39,56 These requirements may further exclude low-income populations.

The source from which participants learn from about the mHealth study is another potential source of bias in the demographic distribution of subjects. While none of the studies reported whether they included surveys or other strategies to track how participants learned about the studies, general media and social media were 2 channels of exposure to such studies. Also, the greater mistrust of clinical trials, in general, among underrepresented minority groups is likely to amplify the representation of majority groups in mHealth citizen science efforts.72 Contrastingly, monetary compensation could bias studies in favor of low-income populations. Most studies included in this review are currently not using monetary incentives (with the exception of DMT mPulse), although there are plans for implementing such practices in the future.11,73

Use patterns

The use of mHealth for biomedical research and citizen science is relatively novel, and so little research has been conducted as to how variable use patterns could bias research studies. For example, one of the Depression Monitor study findings showed that users who completed their survey more than once were most at risk of a higher depression severity score.34 This phenomenon has yet to be studied to determine whether those who completed the survey more than once were truly more prone to depression or if the higher score can be correlated to the use pattern of taking the survey more than once. In addition to repeating surveys, other examples of variable use patterns include varied interpretation of the app’s purpose, benefits, or instructions.

Studies may use open-response questions embedded in the app or other strategies such as instrumentation to monitor use patterns. An ancillary study of the mPower project to investigate participants’ experience in the eConsent process showed that some individuals use the app with the incorrect expectation that it will deliver some therapeutic benefit, and even those who understood there would be no therapeutic benefit were nonetheless frustrated with restricted access to their own data.30

Privacy and security

There are widespread concerns over privacy and security in mHealth, given the sensitive nature of the data as well as methods used for user and data authentication.74 While there have been attempts to mitigate such risks, all mHealth research apps pose several privacy challenges, including the ability to identify participants or their behavioral characteristics from imaging data,19 multimodal sensor data, or open-ended prompts in which some participants included their contact information, despite instructions to the contrary.30

Current federal policies have been demonstrated to be insufficiently equipped to secure electronic protected health information (ePHI). Apps that enable biomedical citizen science grant access of data generated by general citizens to the broader community for secondary research. Despite this intimate level of data sharing, these apps are generally exempt from the same degree of federal regulation as traditional clinical trials.3,4 Additionally, the Federal Trade Commission (FTC) has tried many cases against use of mHealth apps and mobile devices to collect users’ information, violating the Electronic Communications Privacy Act of 1986 (ECPA or The Wiretap Act).75 Furthermore, given threats posted by breaches of portable electronic devices75 and the susceptibility to privacy invasion,76 the emerging mHealth research apps need a deeper assessment of current privacy laws and their relevance to research and citizen science.

CONCLUSION

Self-directed mHealth research applications, particularly apps that are based on open-source platforms such as ResearchKit, hold significant promise for biomedical research and citizen science. This review of existing mHealth research studies and platforms highlights their ability to recruit, consent, and retain far greater numbers of participants than traditional clinical trials in a cost-effective manner, along with representation of diverse populations. In particular, the large datasets that are being generated from these studies will allow for development and validation of novel digital biomarkers as well as retrospective studies. However, in order to capitalize on these benefits, mHealth research studies must strive to improve retention rates, implement rigorous data validation strategies, incorporate FAIR data principles, and address emerging privacy and security challenges.

FUNDING

This work was supported in part by the Office of Research, Discovery, Innovation (RDI) at the University of Arizona.

SUPPLEMENTARY MATERIAL

Supplementary material is available at Journal of the American Medical Informatics Association online.

Conflict of interest statement. None declared.

Supplementary Material

Supplementary Data
Click here for additional data file. (14.1KB, docx)

REFERENCES

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Data
Click here for additional data file. (14.1KB, docx)

Articles from Journal of the American Medical Informatics Association : JAMIA are provided here courtesy of Oxford University Press

RESOURCES