Using mobile apps to assess and treat depression in

Using mobile apps to assess and treat depression in Hispanics and Latinos: Results from a fully remote and randomized clinical trial Abhishek Pratap, MS2,3; Brenna N. Renn, PhD1; Joshua Volponi4; Sean D. Mooney, PhD2; Adam Gazzaley, M.D, Ph.D4; Patricia A. Areán, PhD1; Joaquin A. Anguera, PhD4

1

Department of Psychiatry & Behavioral Sciences, University of Washington, Seattle Department of Biomedical Informatics and Medical Education, University of Washington, Seattle 3 Sage Bionetworks, Seattle 4 University of California, San Francisco 2

Abstract Background: Most people with mental health disorders fail to receive timely access to adequate care. Digital technology has the potential to drive a sea change in the psychosocial treatment of mental health problems, especially in minority populations that have known barriers to accessing and utilizing traditional treatment approaches. Objective: The objective of the present study was to compare recruitment, engagement, and

treatment

outcomes

between

a

sample

of

depressed

Spanish-speaking

Hispanic/Latino individuals compared to a cohort of depressed English speakers using 3 different self-guided mobile apps (iPST (based on evidence-based therapeutic principles from problem-solving therapy [PST]); Project: Evolution™ (EVO; a cognitive training app based on cognitive neuroscience principles); and Health Tips (health information app that served as an information control)). Methods: Participants were recruited for this randomized controlled trial using social media platforms, internet-based advertisements, and traditional fliers in select locations in each state across the United States. Assessment and self-guided treatment was conducted on each participant's smartphone or tablet. We enrolled 646 English-speaking and 437 Spanish-speaking adults (≥18 years old) with mild to moderate depression as determined by a 9-item Patient Health Questionnaire (PHQ-9) score ≥5 or an endorsement of impaired functioning. Outcomes were depressive symptom severity (measured using PHQ-9) and functional impairment (assessed with Sheehan Disability Scale). Engagement in the study was assessed based on the number of times participants completed active surveys.

1

Results: We screened 4,502 participants and enrolled 1,083 participants from throughout the US over 6 months. The majority of the participants were recruited via posts on craigslist.org, with significant participant acquisition costs present when recruiting Spanish speakers ($31/ participant) compared to English speakers ($1.46/participant). Long-term engagement surfaced as a key issue with Spanish speaking participants, as these individuals dropped out two weeks earlier than their English counterparts. Furthermore, while no significant difference between Spanish- and English speakers were observed for treatment remission, Spanish-speakers consistently reported higher depression symptoms compared to their English-speaking counterparts. Conclusions: Our findings suggest that fully remote mobile-based studies can attract a diverse participant pool including people from lower socioeconomic backgrounds (in this case Hispanic/Latinos). However, keeping participants engaged in this type of ‘lowtouch’ research study remains challenging. Further research including user-centered approaches are needed to understand the culturally adequate incentive levels, notification strategies and effective mobile content formatting to enhance the recruitment and retention of minority populations. Trial Registration: Clinicaltrials.gov Identifier: NCT01808976

Keywords: mobile application, depression, Spanish-speakers, RCT, cognitive training, problem-solving therapy, mHealth, smartphone, minorities

2

Introduction Despite extensive evidence supporting the efficacy of psychosocial interventions for depression, access to these treatments has traditionally been limited. A minority of US adults with depression receive psychotherapy[1], despite a preference for psychosocial interventions over pharmaceutical approaches[2–5]. Access problems are further exacerbated by poor utilization and engagement, as the modal number of visits for behavioral interventions among depressed individuals is one[6,7]. Such inconsistencies between interest in psychotherapy and actual access/utilization suggest substantial barriers to engaging in such interventions. These types of barriers to psychosocial interventions for depression can be attributed to the burden of traditional models of psychotherapy (e.g., weekly in-person visits to a clinic), inconvenient location of services, stigma that interferes with health-seeking, poor implementation of evidencebased treatment, and actual costs and insurance barriers to care. For individuals with perceived barriers, such as those from underserved communities or rural areas, the concept of weekly, hour-long visits is untenable. Such barriers are particularly prominent in the Hispanic/Latino community, as evidenced by their low use of mental health services in general, and outpatient psychotherapy in particular, relative to other racial and ethnic groups. Hispanic/Latino adults with depression are approximately half as likely as their non-Hispanic White counterparts to receive mental health services[1]. Moreover, access to care is further affected by treatment barriers such as language and socioeconomic factors[8,9], as language and acculturation likely play critical roles in both learning about and navigating available resources. There is a glaring need to remedy this treatment disparity, given that

3

this group accounts for half of US national population growth since 2000; in 2016, Hispanic/Latino individuals were the second-largest ethnic or racial group and comprised 18% of the nation’s population[10]. Despite this demographic prevalence, there is a gross shortage of Spanish-speaking psychotherapists, and Hispanic/Latino communities are notoriously under-represented in mental health research[11]. Finally, mobile health (mHealth) interventions are increasingly of interest to the research community, yet little is understood about how the Spanish speaking US community engages with such technology. Clearly, methods for increasing access to and sustained use of these treatments are needed, particularly those of relevance to Hispanic/Latino individuals. The widespread availability of digital technology has the potential to drive a sea change in access to psychosocial treatment for mental health problems[12]. Internet-based interventions have already demonstrated comparable treatment outcomes as traditional face-to-face psychotherapy[13]; this success has paved the way for testing the delivery of these interventions on mobile phones. Data from Pew Research Center suggest that smartphone ownership has penetrated a wide range of American demographic groups, with 75% of Hispanic/Latino owning a smartphone[14]. These same data suggest that Hispanic/Latino adults are increasingly smartphone dependent for online access, relative to White and Black Americans. Given the widespread usage of smartphones, mobile-based mental health apps have the potential to increase treatment accessibility and engagement. They offer a significant advantage compared to internet-based interventions because of the potential for users to use assessment tools and apply intervention principles in the

4

moment that they most need help with their mood. However, a strikingly small number of these apps have been tested to develop an evidence base for treatment[15–17]. Furthermore, there is relatively little information regarding app user access (do users download health apps and use them more than once?), app user engagement (do users follow the app protocols?) and the effect of app use on mood and daily functioning. These questions are of the utmost importance in minority groups who do not speak English, as these individuals do not typically have the same opportunity as majority groups to utilize mental health services or cutting-edge technologies. There is an immediate need for further research to develop and evaluate new solutions for mental health care that are economically viable, scalable, and focused on engaging users to inform timely and evidence-based clinical interventions. We previously demonstrated feasibility, proof-of-concept, and preliminary effectiveness of the first large-scale, fully remote randomized clinical trial (RCT) of three mobile depression apps in an Englishspeaking US sample[18,19]. The present study reports on an extension of this previous work that directly targeted Spanish-speaking Hispanic/Latino individuals as well as a new cohort of English-only speakers. The aim of this study was to report and compare recruitment, engagement, and treatment outcomes between three different self-guided mobile apps for depression between these groups.

Methods Ethical approval for the trial was granted by the UCSF Committee for Human Research. Recruitment: Three different types of recruitment approaches, including traditional, social networking, and search-engine strategies, were used. Traditional methods consisted

5

of craigslist.org postings throughout the United States, specifically posting to the ‘Volunteer’ and ‘Jobs etc.’ pages within Craigslist in at least one major city in every state. Social networking methods included regular postings on sites such as Facebook and Twitter, and contextual-targeting methods to identify and directly push recruitment ads to potential participants, based on their Twitter and other social media comments. This approach was led entirely by trialspark.com, which designed specific recruiting campaigns using machine learning approaches to create optimal advertising. Furthermore, we reached out to Hispanic/Latino Catholic Ministries in at least one city in every state to see if they would be willing to champion this study and post flyers in their communities. Each approach (described further in supplementary materials) provided potentially

interested

(www.brightenstudy.org),

participants which

a

was

link

to

translated

our

custom

entirely

for

study

website

Spanish-speakers

(www.brightenstudy.org/spa) and included a welcome video featuring Spanish-speaking researchers describing the goal of this study (http://bit.ly/2D06KKK). All translations were done by a combination of native Spanish speakers associated with this study and professionals at Babble-on (https://www.ibabbleon.com/translation.html).

Participant eligibility: Participants had to speak English or Spanish, be 18 years old or older, and own either an a) iPhone with Wi-Fi or 3G/4G/LTE capabilities or b) an Android phone along with an Apple iPad version 2.0 or newer device. iOS ownership was required as one of our intervention apps were only available on iOS devices at the time of the study. Participants had to endorse clinically significant symptoms of depression, as indicated by either a score of 5 or higher on the Patient Health Questionnaire ([PHQ-9,

6

21]), or a score of 2 or greater on PHQ item 10 (indicating that they felt disabled in their life because of their mood).

Procedure: Potentially interested participants were directed to either the Spanish or English version of the study website which explained the study goals and procedure to enroll directing the potential participants to a brief online eligibility screening assessment.

Initial Screening: Interested participants completed a brief online screening consisting of questions about mobile device ownership and their ability to speak Spanish.

Consent: We used a combination of a written consent and custom videos to explain the study. Participants had to pass a quiz that confirmed their understanding that participation was voluntary, was not a substitute for treatment, and that they were to be randomized to treatment conditions. Each question had to be answered correctly before moving on to baseline assessment and randomization. Individuals who indicated that they speak Spanish were given a 7-question multiple choice survey assessing language proficiency. Eligibility was established after consent was obtained. Upon being eligible, participants were sent a link to download their assessment application (Surveytory).

7

Figure 1: Overall BRIGHTEN V2 study schematic showing participant recruitment, consent, enrollment and randomization workflow along with weekly and daily data collection. Assessment: We collected baseline and 4-, 8- and 12-week outcome data. Our in-house custom mobile app, Surveytory, was used to collect self-reported mood, function, and passive analytics such as communication data (text logs including call/text time, call duration, text length, and screen usage) and mobility data (activity type and distance traveled using the phone’s accelerometer and GPS) (Figure 1). Participants were automatically notified every 8 hours for 24 hours if they had not completed a survey within 8 hours of its original delivery. A built-in reminder also prompted the participant to check for any surveys occurring on a daily basis in case they missed a new survey notification. An assessment was considered missing if it was not completed within a 24hour time frame. The baseline assessment included the collection of demographics including age, race/ethnicity, marital and employment status, income, education, smart device ownership, use of other health apps, and use of mental health services, including use of medications and psychotherapy. We collected information on mental health status using 8

the PHQ-9[24] for depression and the Sheehan Disability Scale (SDS) [27, 28] to assess self-reported disability. The PHQ-9 rates the presence and severity of depressive symptoms across nine items, with higher scores signifying more severe symptomatology (range 0-27). The SDS assesses perceived functional impairment across three domains (work/school, social life, and family/home responsibilities), yielding a sum score (0-30) in which higher scores represent greater disability. We also asked participants to rate their health on a scale of excellent to poor. Daily assessments were a combination of selfreport and passive data collection. Participants were sent notifications every morning to report their daily mood using the PHQ-2 survey. The Surveytory app collected passive analytics daily. Private information such as actual content of voice calls or text messages or emails was not collected. The 4-, 8- and 12-week assessments included the PHQ-9 to measure changes in mood and the SDS for changes in disability. The SDS was only administered post-enrollment so comparisons are with respect to week 1 after enrollment. Participants were also asked this question about treatment: “Since using this app, I feel that I am: (1) much worse (2) worse (3) no different (4) improved (5) much improved.”

Treatment: After confirming completion of baseline assessments (or 72 hours after the initiation of these assessments, whichever came first), participants were sent an online survey which described each of the 3 possible treatment arms. Following this description, participants were asked to select which two apps they were most inclined to use in this study (thus, we employed an equipoise stratified design[20]). Participants were then randomly assigned to one of these two preferred conditions, sent a link to download the intervention app, and provided with a brief video explaining how to download and use

9

the assigned treatment app, as well as a custom dashboard to monitor their study progress. Participants were asked to use their assigned app for one month. The first app was a cognitive intervention video game (Project: Evolution™ [EVO]) designed to modulate cognitive control abilities, a common neurological deficit underlying depression[22]. The second intervention was an app based on problem-solving therapy (iPST), an evidence-based treatment for depression[23]. The final intervention app, an information control, provided daily health tips (HTips) for overcoming depressed mood such as self-care (e.g., taking a shower) or physical activity (e.g., taking a walk; see supplemental materials from Anguera et al. 2016 for further descriptions of each). Similar to the assessment notifications, each intervention app was equipped with built-in reminders asking the participant to use their app on a daily basis.

Payment: Randomized participants were paid a total of $75 in Amazon gift vouchers for completing all assessments over the 12 weeks. Participants received $15 for completing the initial baseline assessment and an additional $20 for each subsequent assessment at the 4-, 8-, and 12-week timepoints.

Online Focus Group: Participants who enrolled through the Spanish language portal (www.brightenstudy.org/spa) could access a brief survey assessing their impressions of the study and to interrogate why enrollment and participation differed from an Englishspeaking cohort. This survey was open for five months (10/1/2017-2/28/2017), and was available to Spanish speakers visiting the website or any Spanish-speaking participants who had previously accessed our eligibility portal. The purpose of this survey was to

10

better understand which aspects of this work Spanish-speaking individuals found most compelling, their general impressions on participating in this type of research, their underlying motivations regarding participation, and suggestions as to what they believed would improve both participation and recruitment efforts in the future.

Procedures to reduce gaming: “Gaming” is a situation where a user fraudulently enrolls in a study solely to acquire research payment. We utilized the following safeguards to prevent this: i) locking the eligibility or treatment randomization survey if a participant tried to change a submitted answer so that only the initial answer was utilized, ii) using study links that are valid for one user/device, and iii) tracking IP addresses to minimize duplicate enrollment. Statistical Analyses Sample demographics and clinical characteristics were calculated using appropriate descriptive statistics. Comparison between participant demographics were done using a chi-square test of independence for categorical variables (with continuity correction) and one-way analysis of variance (ANOVA) for comparing continuous variables across the groups. To assess the marginal effect (i.e., association in the entire sample) between longitudinal weekly PHQ-9 and SDS scores and treatment arms, we used generalized estimating equations (GEEs) [21]. Briefly, GEE models extend generalized linear models to longitudinal or clustered data. GEEs use a working correlation structure that accounts for within-subject correlations of participant responses, thereby estimating robust and unbiased standard errors compared to ordinary least squares regression [21,22]. Treatment response was further categorized into three groups based on a change of at

11

least 5 points on the PHQ-9[23] (the minimal clinically important difference), to comprise treatment responders (decrease PHQ-9 >= 5), non-responders (change in PHQ9 less than 5 points), and those that deteriorated over treatment (increase in PHQ-9 of >= 5 points). To assess participant engagement, we examined the proportion of participants that completed at least one activity in any given week. ANOVA method was used to compare the daily, weekly and overall participation differences between Spanish and English speakers. Univariate estimation of time to drop out from the study between Spanish and English speakers was done using a survival analysis method. The distribution of the ‘survival’ days (total days active in the study) and nonparametric estimates of the survivor function was computed using the Kaplan-Meier method[24], with curves tested using the log-rank test[25]. To compare drop-out rates among the three interventions, a non-parametric Kruskal-Wallis test was used. Finally, all analyses were carried out using R, statistical computing language version 3.4.2[26]. Results Recruitment and Enrollment The BRIGHTEN V2 study started recruiting in August 2016 with screening and enrollment continuing for seven months. A total of 4,502 people were screened; 39.52% of them identified as Spanish speakers. Of the total screened, 1083 (24.05%) adults met the eligibility criteria and enrolled in the study. As in BRIGHTEN V1 study[18,19], the use of craigslist.org was the most effective approach in recruiting, with more than 80% of our participants coming from this approach. An additional 9% were referred by friends or colleagues.

12

Enrolled participants lived throughout the US, with all the metropolitan areas represented (Figure 2). Only 363 (33.51%) of the initially enrolled participants were active in the study, as defined by completing at least one post-enrollment weekly PHO-9 assessments and/or providing passive phone usage data within the first 12 weeks. The remaining 720 (66.48%) participants did not respond to any post-enrollment surveys or provided passive data and as a result, were considered dropped-out from the study (Figure 3). Table 1 presents the comparison of baseline characteristics between the active and dropped-out participants stratified by Spanish and English speakers. Income, education, working status, and race were significantly different between the two groups. Of the 363 active individuals, 77 received the treatment randomization survey but did not complete it, and thus were never assigned to an interventions. However, they continued to answer the PHQ-9 throughout the study period. These individuals were placed in a separate enrolled, not randomized (EnR) category (Figure 3). All further analyses were restricted to active individuals consisting of those in treatment or EnR arms. We also observed that a significant portion of the participants did indeed attempt to “gamify” (i.e., cheat) their treatment options following the random assignment to one of their preferred 2 intervention. Of those who accessed the randomization survey, 31.8% attempted to change their assigned intervention during the survey experience by hitting the ‘back’ button, while an additional 10.4% participants returned to the survey a second time to select a different potential intervention (3.1% of these individuals used each of these methods). Note that these attempts to “gamify” the system were unsuccessful because participant randomization

13

was determined by the first answer given by a participant, not any of the subsequent attempts made.

Figure 2: US map showing the location of people who were screen (gray) and enrolled(red) in thein the BRIGHTEN 2.0 Study.

Figure 3: CONSORT diagram

14

Sample Demographics The study participants were predominantly young (71.1% less than 40 years old; M =34.62, SD=10.84), female (75.9%), and non-Hispanic white (70.0%). The majority (70.8%) reported some form of employment. Our data shows notable disparities between Spanish and English speakers, such that Spanish speakers were more likely to report low income (e.g., earning between 0-$20,000 a year) compared to English speakers. Likewise, English speakers were more likely to be employed and more likely to have obtained a university education relative to Spanish speakers (Supplementary Table 1). Finally, our study had clear selection bias with >80% of participants being iPhone users. This resulted from one of our interventions, EVO™, only being available on iOS devices. If a user had an Android phone, they were only eligible to participate if they also owned an Apple iPad version 2 or newer iOS tablet device. The likelihood of owning this combination of devices is unusual, and likely spurred this bias toward iOS users in the sample. English speakers

Spanish speakers

Active

Dropped-out

Active

Dropped-out

N

241

405

122

315

Baseline PHQ-9 (mean (sd))

13.05 (5.18)

14.85 (5.38)

14.75 (5.88)

14.65 (5.68)

Female

180 (74.7)

260 (64.2)

94 (77.0)

234 (74.3)

Male

61 (25.3)

144 (35.6)

26 (21.3)

81 (25.7)

35.69 (11.00)

34.56 (10.91)

32.44 (10.22)

39.45 (112.21)

18-30

89 (37.2)

170 (42.6)

58 (49.6)

139 (45.9)

31-40

74 (31.0)

121 (30.3)

32 (27.4)

93 (30.7)

41-50

53 (22.2)

71 (17.8)

21 (17.9)

54 (17.8)

51-60

18 (7.5)

29 (7.3)

5 (4.3)

11 (3.6)

61-70

4 (1.7)

7 (1.8)

1 (0.9)

6 (2.0)

70+

1 (0.4)

1 (0.3)

0 (0.0)

0 (0.0)

Device = iPhone (%)

216 (89.6)

336 (83.0)

104 (85.2)

260 (82.5)

0.09

Working = Yes (%)

179 (74.3)

300 (74.3)

76 (63.9)

166 (53.9)