Incorporating community partner perspectives on eHealth Technology Data Sharing Practices for the California Early Psychosis Intervention Network: Qualitative Study

Background: Increased use of eHealth technology and user data to drive early identification and intervention algorithms in early psychosis necessitate implementation of ethical data use practices to increase user acceptability and trust. Objective: Explore EP community partner perspectives on data sharing best practices, including beliefs, attitudes, and preferences for ethical data sharing and how best to present end user license agreements (EULA). Methods: We conducted exploratory, qualitative, focus-group-based study design and content analysis assessing mental health data-sharing and privacy preferences. Data was gathered for the California Early Psychosis Intervention (EPI-CAL) network project in 2020-2021. Focus groups were conducted via video conferencing and assessed three EP groups: clinical staff/providers (n=14), clients (n=6), and family members (n=4). Eligible participants were affiliated with one of 18 EPI-CAL network clinics (i.e., convenience sampling), English-speaking, and able to provide written informed consent and assent (minors). Clinic staff/providers were invited from EPI-CAL clinics. Thirty clients and family members were invited through referral; 20 did not participate. Key themes identified via content analysis of focus group discussions. Results: Clinical staff participants (N = 14) were adults (M = 37.9, SD = 9.0), predominantly cisgender female (71%), White (64%), and non-Hispanic (64%). Clients (N = 6) were young adults (M = 23.8, SD = 3.9), predominantly male sex (67%) and non-Hispanic (83%


Incorporating community partner perspectives on eHealth Technology Data Sharing Practices for the California Early Psychosis Intervention Network: Qualitative Study
Only make the preprint title and abstract visible.No, I do not wish to publish my submitted manuscript as a preprint.2) If accepted for publication in a JMIR journal, would you like the PDF to be visible to the public?
Yes, please make my accepted manuscript PDF available to anyone at any time (Recommended).
Yes, but please make my accepted manuscript PDF available only to logged-in users; I understand that the title and abstract will remain v Yes, but only make the title and abstract visible (see Important note, above).I understand that if I later pay to participate in <a href="http

Introduction
The past decade has seen an explosion in eHealth technology (e.g., smartphone/tablet applications, online web-portals) to facilitate improved outcomes for individuals with psychosis.Individuals with psychosis are willing and interested in using eHealth technology as part of their care [1][2][3][4] .eHealth tools promote treatment engagement 5 , symptom monitoring 6,7 , relapse prediction 8 , and enhance quality of life 9 and functioning 10 .Consequently, industry developers and academics are racing to implement eHealth technology at scale to improve outcomes for those experiencing serious mental illness.
As eHealth technology advances and we leverage user data to drive early identification and intervention algorithms 11 , it is imperative that we implement ethical data use standards.Typical software has long end user license agreements (EULAs) replete with legal jargon detailing the myriad ways user data are utilized and shared 12 with little or no user control.Consequently, users rarely read the EULA and may not understand what they are agreeing to.Users may unknowingly have their data shared or sold to third-parties, sometimes without encryption, rendering it vulnerable to data privacy breaches [12][13][14][15][16][17] .These issues may be particularly relevant in psychosis: the associated cognitive impairments could impact EULA comprehension and data breaches of sensitive and highly stigmatized psychosis diagnoses could be especially harmful.Users have varied attitudes about risk: some report skepticism of eHealth data 12,18 or feel cognitive dissonance around risks as a reality of using digital platforms, especially those that are "free" in return for data use 15,19,20 .However, health data is personal and private; researchers, providers, and industry partners alike have a duty to protect vulnerable individuals from data misuse.Moreover, an outcomes driven health care system (an agreed goal in the health care industry 21 ) relies on large, interagency data sharing; to do this, we must implement ethical data use practices to increase user acceptability and trust in eHealth platforms.
One such effort to build an outcomes driven health care system is the California Early Psychosis Intervention (EPI-CAL) Network.EPI-CAL is a multi-year project that connects early psychosis (EP) programs across California via an eHealth application, Beehive, in a learning health care network 22 .Beehive facilitates client-, family-, and cliniclevel outcomes data collection across EP programs using a battery of validated, low-burden measures.Data is immediately available to clinics via provider dashboard with data visualizations, clinical alerts, and treatment planning support.To date, 18 EP clinics participate in EPI-CAL, including both community-and university-based clinics.As more EP programs use Beehive to contribute consumer data, data analysis can identify which treatment components improve client outcomes.However, this relies on EP clients choosing to share their data for analysis.
The current study sought to develop a transparent EULA that clearly describes EPI-CAL goals and data sharing options.Prior research suggests EULAs should: 1) be relevant and understandable 23 by providing a supplemental video 24,25 , setting the reading level to 6 th -8 th grade 23,26 , and including comprehension checks 27,28 ; 2) offer explicit "opt-in" selections for levels of data sharing rather than "catch-all" agreement selections 18,26,29,30 ; and 3) include options to request ending data collection or delete data entirely 26 .
To explore community partner perspectives on data sharing best practices in an EP setting, we conducted focus groups with EP clients, family members, support persons, and clinic staff about their beliefs, attitudes, and preferences for ethical data sharing and how best to present EULA materials to EP consumers.Development of the Beehive EULA for EPI-CAL can be used as a case example of how a user-centered 31,32 , accessible, transparent, and flexible EULA can support participation in an eHealth technology facilitated learning healthcare network.

Design
We used an exploratory, qualitative, focus-group-based study design and content analysis to assess participants' mental health data-sharing and privacy preferences.See Table 1 for researcher demographics.

Recruitment
We recruited participants from three EP community partner groups: (1) clinical staff/providers, (2) clients, and (3) family members of clients.Eligible participants were (1) actively or formerly affiliated with an EPI-CAL network clinic (i.e., convenience sampling), (2) English-speaking, and (3) able to provide written informed consent and assent (minors).Participants received $30 compensation for participation.UC Davis IRB and relevant county review boards approved the study.Clinical teams from active EPI-CAL clinics invited one representative from each clinic to generate a maximum group size of 12.Some researchers had existing professional relationships with some participants due to prior research or contact at EPI-CAL focus groups.
Client and family participants were invited through clinician referral or by researchers who had permission to contact them for additional opportunities.We contacted 30 individuals about the focus group; 20 did not participate.Most did not respond to recruitment attempts; a few stated they were not available and 5 who agreed to participate ultimately did not attend the focus group.

Data Collection
We conducted two phases of ninety-minute focus groups in August 2020 and January 2021 via videoconferencing in compliance with COVID-19 restrictions.Each participant group was represented per phase.Each group had a facilitator (LT/SE), co-facilitator (SE/KN), and note-taker (KN/CH).We developed focus group guides for each phase (see Supplementary materials).After each group, we discussed salient points and refined the guide.Data saturation was not discussed as the number of focus groups was predetermined.However, high consistency in major themes identified at the participant level across groups indicates saturation.

Phase 2: EULA Focus Group
We incorporated Phase 1 concepts to create Beehive EULA materials.Phase 2 focus groups covered participants' opinions of these materials, including an informational whiteboard Beehive EULA video (created by CH/KN/LT) and visualization of opt-in choices of data sharing levels.Participants watched the video, facilitators elicited their opinions and feedback, participants watched the video a second time, and reviewed the opt-in data sharing screen.

Data Analysis
We used directed content analysis 33 to describe participant preferences for health data sharing.We coded and organized data using NVivo qualitative analysis software (QSR  International, 1999).Using an inductive approach, two authors (Phase 1: KN & SE, Phase 2: KN & VT) reviewed de-identified focus group transcripts and developed preliminary coding frameworks.All authors discussed the framework before coding began.Two authors independently coded each transcript then compared their responses and resolved any disagreements through discussion.From these codes, a set of categories were developed, then major and minor themes established.

Participant Feedback
Participants were emailed the major and minor themes, supported by key quotations, for the group(s) they participated in.Participants could provide feedback via a survey or via videoconference with researchers (KN, SE, VT).Eight participants (3 clients, 4 providers, 1 family member) provided feedback, 6 via survey and 2 (1 client, 1 provider) via videoconference.Overall, participants agreed with identified themes; no significant changes were made to results.

Phase 1: Data Sharing Preferences Focus Group
Prior experience, both positive and negative, influences understanding and willingness to share data.The purpose of data sharing also drives decision-making around willingness to share.In all scenarios where health data is being shared outside of the clinical environment, all participants preferred this data be de-identified.See Table 3 for quotes for each major theme.

Core Concepts to Incorporate in Data Sharing Policies
Transparency is foundational in participants' data sharing calculus; knowing what, when, with whom, how, and why data is shared is paramount.This includes disclosure of conflicts of interest, and using laymen's and culturally appropriate terms.Receiving research results was one example of transparency that improved participants' understanding of how data is used.Clinic participants noted that explaining current data protection laws increases willingness to share data.Individuals want to know that the institution or entity to which they are entrusting their data is competent in upholding legal protections.Clients emphasized extra protections should be in place when individuals are in a vulnerable state (e.g., a mental health crisis).Participants emphasized the importance of having control over their data, including sharing the minimum data necessary, restricting access, having the ability to change one's mind to facilitate no regrets (including being able to opt in at a later time), and to delete data to give peace of mind.All participants noted that limitations of deleting de-identified data should be clear, especially if data has been shared with outside parties.
When researchers cannot be in direct contact with participants, they rely on established rapport between client and clinician as clinicians are often the individuals who provide information about research opportunities.One clinician stated understanding what the purpose of the research is and how it's helpful can be a conduit for transparency.A clinical research coordinator noted rapport alone is insufficient; clinicians must be able to explain the study.See Table 4 for Phase 1 concepts incorporated into Beehive EULA materials.

Phase 2: EULA Focus Group
In phase 2, we elicited feedback on Beehive EULA materials, including responses to the implementation of key themes from Phase 1. See Table 5 for quotes for each major theme.

Transparency, Data Protections, and Control
We presented Beehive EULA materials via a whiteboard video and a user-interface where users indicate their data-sharing choices.Participants noted the whiteboard presentation video increased their engagement with the material and gave them a clearer understanding than previous experiences of oral-only presentation methods.However, clinician group participants indicated that adjustments to visualizations, speed, and repetition could improve the clarity of information.Participants also identified aspects of the user-interface design (Supplemental Figure 1) which improved transparency, including having the decision to opt-in to data sharing for different institutions separately.However, multiple participants noted the lack of clarity regarding which check-boxes are optional versus required could prompt confusion and mistrust regarding whether data sharing was truly optional.Participants, especially clients and families, noted the EULA materials adequately explained and met their expectations for control over data and protections of data.

Willingness to Share Data
It became clear that rapport is part of a larger theme of "familiarity" under which the category of "reputation" also falls.Participants maintained the idea the relationship developed with researchers or clinicians is an important factor impacting their willingness to share data.The concept of reputation was identified as a sub-type of familiarity that may influence an individual's willingness to share their health data for research purposes.In the context of the EULA video, some participants mentioned that they would be interested in learning about the reputations of unfamiliar organizations (e.g., NIH) to help them make their decision.Participants gave positive feedback about the level of transparency present in the EULA materials.They also noted a need for more information around control over their data or how their data is protected.Multiple participants commented that they felt the direct benefits of using Beehive were not obvious.

Willingness to Use eHealth Technology
Participants (64.3%, n=9) indicated the EULA materials, with added information about direct benefits, would likely prompt engagement with Beehive, due to an increased understanding of how their data would be used in standard care and research to help others.Some clinic participants and family members (28.6%, n=4), however, raised concerns regarding willingness of their clients and families to use Beehive or share data with researchers at the beginning of treatment even after being presented with information about how their data would be used, citing psychosis symptoms and personality traits as potential reasons.

Phase 2 Outcomes and Actions
Prior to Beehive's launch in March 2021, we used Phase 2 focus group results to update the EULA video script and voice over to 1) clarify UC Davis team's access to de-identified data for quality management purposes, 2) clarify potential benefits to clients, and 3) simplify text and reduce rate of speech.We also updated the user inter-face by changing the "optin" data sharing choices to a forced-response (yes/no) regarding data sharing (Supplemental Figure 2).

Discussion
This study explored EP community partner perspectives on ethical data sharing practices, and what impacts their willingness to share data on eHealth platforms.Using this data, we developed a user-centered, accessible, transparent, and flexible EULA, which users reported would increase the likelihood they would engage in data sharing.These findings present a framework for eHealth platforms to engage EP consumers in a manner consistent with consumer preferences, facilitating increased consumer engagement in eHealth technology.
Three main findings emerged.First, regarding data sharing preferences, participants endorsed core themes of transparency, data protections and limitations, control, and rapport.This builds on prior research highlighting a range of privacy-adjacent concerns 19,23,25,29 , including transparency 23,34 , relevancy 23 , user-level control 34,35 , and comprehension 12,16,36 .This demonstrates users' desire to know the "what", when", "how", "why" and "with whom" to make informed data sharing decisions; eHealth platforms need to equip users with enough information to objectively assess the benefits and risks of sharing their sensitive personal information.
Second, users' concerns can be alleviated via implementation of an understandable and transparent EULA with user-level control for data sharing.To address comprehension concerns, we implemented a number of recommendations from prior research: the EULA was offered (1) in multiple modalities (animated video and text 24,25,27,28,37 ) to maximize engagement and visually present complex information; (2) all narrative and text materials were presented at the recommended 8 th grade reading level 23,26,27 ; (3) key terms (e.g., identified vs. deidentified data) were clearly defined and specific data examples were offered (age, zip code).To address control and access concerns, we clearly named and described third-party research entities (UC Davis researchers, NIH, Westat) and how they might use de-identified Beehive data.We then gave participants opt-in choice points for sharing their data with these entities and emphasized that users do not have to share their data to use Beehive as part of their care.In Phase 1, participants were acutely aware of widespread third-party sharing across digital domains, the risk of breaches, and the unclear motives hidden in the legal language of EULAs.Consequently, they reported feeling a loss of control and lack of protection over access to their data.In Phase 2, participants reacted positively to the novel approach implemented in the Beehive EULA, stating this would make them more likely to use Beehive and choose the data sharing level that best fit their comfort level.
Third, results demonstrate the value of partnering with community members to develop eHealth technology.Participants' wide range of experiences and perspectives emphasized their desire for control and protection over their data.Phase 2 participants upheld the importance of allowing clients to change their data sharing preferences every time they logged onto Beehive to support those who may have made data sharing decisions during times of symptom exacerbations that they later wish to change.Similarly, participants highlighted the impact of trust and rapport between client and provider on data sharing decisions, suggesting that providers review the EULA video with clients and families to provide time for questions before making selections on their data sharing options.This indicates that person-to-person discussion of the EULA also impacts comprehension, comfort using eHealth technology, and whether the user chooses to share their data.By centering the voices of users, we gained valuable insight into how best to balance user control over data and researchers' need for data.
To date, the community-informed Beehive EULA has resulted in high rates of data sharing in EPI-CAL: 88.2% (n=231) of users have chosen to share their data with UC Davis researchers, and 85.1% (n=223) have chosen to share their data with NIH.A minority of clients (0.8%, n=2) have chosen to change their data sharing preferences.These data potentially allay concerns that increased transparency regarding data sharing may lead to substantial increases in people electing not to share personal data to learning health care networks and national datasets.
This study has significant strengths, including the centering of perspectives from community partners, the use of a multi-phase approach to incorporate participant feedback, and the development of actionable steps to ensure ethical data sharing in eHealth technology.Limitations include the possibility of bias inherent to qualitative methods: facilitator age, social status, race, and participant involvement in the development of the EULA materials reviewed could bias their responses.Participants may have felt pressure to please facilitators (social desirability bias) and may have limited contributions due to discomfort (sensitivity bias).Finally, while there was high consistency at the participant level indicating saturation, this may be partly attributable to group dynamics; data from additional focus groups would be informative.Limitations were minimized where possible: to lessen dominant respondent bias, facilitators promoted less vocal participants.To avoid reference bias, questions were ordered logically minimizing swaying participants' perspectives.To minimize reporting bias, we employed codebooks, multiple coders, and participant feedback before finalizing themes.COVID-19 logistical barriers likely impacted provider recruitment of consumers.Relatedly, COVID-19 safety precautions necessitated videoconference focus groups, excluding participants without adequate internet access or electronic devices, and/or those uncomfortable with virtual participation.Although cross-clinic videoconferencing likely increasing the breadth of voices included in the discussion, this selection bias may be particularly relevant given the technology-oriented subject matter.Future research should examine attitudes to eHealth technology and data sharing with individuals with low comfort with technology and/or who prefer in person participation.
In sum, this study demonstrates the value of community informed research and indicates transparent EULAs with user-level control of data sharing are desired and result in high rates of data sharing.Future work includes collaborating with partners who speak languages other than English to determine best approaches for translating EULA materials in a culturally accessible and linguistically appropriate manner.

Table 1 .
Researcher demographics and background

Table 2 .
Demographic and clinical characteristics of enrolled participants (n = 24).Due to rounding, percentages may not equal 100.

Table 3 :
Quotes from Phase 1 Groups

of Data Sharing of Health Data
Right now it's presented where you have to sign this release or I'm not going to go through with …if I put my data in here and we do come up with a solution that actually would help my I feel like our clients would ask, "What is going to be shared and why?What's the

Table 4 :
Implementation of Phase I Themes into Beehive EULA video

Table 5 :
Quotes from Phase 2 Groups