The study was a parallel-group RCT. It was conducted in Guangzhou, China, from September 2017 to October 2018. Participants were randomized into two groups in a 1:1 ratio: a WeChat-based mHealth intervention group or a usual care waitlist control group. The study design was detailed in the Consolidated Standards Of Reporting Trials–eHealth checklist in the Multimedia Appendix 1. The study protocol was approved by the Institutional Review Board of Sun Yat-sen University.

Participants were recruited from the outpatient clinic of the only hospital designated for HIV treatment in Guangzhou, the third largest city in China. The hospital treated over 14,000 PLWH. Patients in the waiting area were invited by the research staff to participate in the study. Patients first completed a brief screening questionnaire in a private space; those who met the eligibility criteria (see below) were provided with a pamphlet describing the Run4Love study and were then invited to join the study. Patients interested to participate were given further information about the study. After providing the written informed consent, eligible patients completed a baseline survey on a tablet and provided their hair samples.

The inclusion criteria were as follows: (1) being 18 years or older, (2) being HIV seropositive, (3) having elevated depressive symptoms (measured by the Center for Epidemiologic Studies-Depression Scale [CES-D] ≥16), (4) willing to provide hair samples, and (5) using WeChat.

The criteria for exclusion were as follows: (1) currently on psychiatric treatment, (2) unable to finish the screening or baseline survey, (3) unable to read or listen to the materials sent via WeChat (ie, short articles, audio, and posters), and (4) unable to engage in physical activities because of medical reasons.

The intervention protocol has been detailed elsewhere [15]. Briefly, participants in the intervention group received a 3-month Run4Love program, comprising two major components: the adapted CBSM course [16] and physical activity promotion. The adapted CBSM course included nine sessions and three review sessions on stress reduction management and coping skills such as muscle relaxation, breathing, and meditation, which was in a multimedia format and sent 3 to 5 times weekly. The articles were on average of 1300 words and took about 5 min to read; the audios took 5 to 10 min to listen to. The physical activity promotion program comprised goal setting and personalized feedback in addition to information on benefits of and guidance for regular exercise and healthy diet. The program was delivered via the enhanced WeChat platform with added functions of automated information sending, progress tracking on course completion and physical activity, and weekly personalized feedback. The most read articles were selected and sent to the participants weekly as a booster in the 3 months postintervention.

Participants in the waitlist control group received a brochure on nutrition in addition to usual care for HIV treatment. The design of the Run4Love intervention and the control condition was based on our previous research on PLWH and a pilot mHealth intervention [17-19].

Allocation to the treatment group was carried out by a computer-generated randomization list with a block size of 4, using SAS software version 9.4 (SAS Institute, Inc). By the nature of the trial design, neither the research staff nor the participants were blinded to the intervention.

We used multiple means for quality control and participant retention. The back end of the Run4Love account could track the course completion, which allowed us to send personalized feedback and reminders based on the participants’ progress. Participants in the intervention group also received up to 5 phone calls from the research staff at week 1 and month 1, 2, 5, and 8 after enrollment. The phone call in the first week was to confirm participation and ensure participants’ proper use of the Run4Love WeChat account. The other phone calls were to identify the barriers to intervention adherence, provide feedback, and remind the participants of regular medical checkups.

All outcomes were measured at baseline before randomization and at follow-ups. The self-report psychosocial measures were collected at baseline, and at 3-, 6-, and 9-month follow-ups using electronic questionnaires on a tablet. The hair samples were collected at baseline and the 3-month follow-up.

The intention-to-treat principle was applied to all analyses [31]. Baseline characteristics were summarized as means and SDs for continuous measures and as numbers and percentages for categorical measures in each group. Baseline characteristics were compared between the groups using two-sample two-tailed t tests for continuous measures and using chi-square (χ²) tests for categorical measures. For between-group differences, 95% CIs were calculated for continuous measures. Analyses for changes in outcomes between baseline and each follow-up were performed using multiple imputations for the missing data [31]. The R package mice (R Foundation, version 3.4.2) was used to obtain 80 imputed data sets. Variables used for imputation included age, gender, marital status, sexual orientation, education, BMI, family monthly income, household registration, duration of HIV infection, and outcome values.

For the primary outcome, group differences in CES-D scores over the 9 months of the trial were estimated using a linear mixed-effect model (LMM) with repeated measures, adjusting for baseline CES-D score, time, and other baseline characteristics, including age, gender, BMI, education, sexual orientation, family monthly income, marital status, duration of HIV infection, and employment [32]. In addition, interactions between group and time were also examined in LMM. R package nlme (R Foundation, version 3.4.2) was used to conduct the LMM analysis.

Similar analyses were repeated for secondary outcomes. In post hoc exploratory analyses, the effect of the intervention on the 3-month change in the CES-D score was evaluated using tests for interaction to determine statistical significance in subsets of participants grouped by baseline characteristics.

Analyses were performed using R version 3.4.2, and a two-sided P<.05 was considered as statistically significant. As multiple secondary outcomes were compared, a two-sided P<.005 was considered statistically significant for secondary outcomes.

Figure 1 summarizes the flow of participants through the study. Of the 1555 patients who were screened and provided information about the depressive symptoms measured by the CES-D scale, 1067 patients were excluded as their scores were lower than 16, and 488 patients were eligible for further interview. In the end, a total of 300 patients met the eligibility criteria and completed the baseline assessment before being randomized into the trial with 150 patients in each group (Figure 1). The mean (SD) age of the participants was 28.3 (5.8) years. Of the 300 participants, 277 (92.3%) were men and 245 (81.7%) were homosexual or bisexual or uncertain of their sexual orientation. The follow-up rates were 91.3% (92.7% in the intervention group and 90.0% in the control group), 88.3% (88.0% in the intervention group and 88.7% in the control group), and 86.7% (88.7% in the intervention group and 84.7% in the control group) at 3, 6, and 9 months, respectively. Moreover, participants in the intervention group completed, on average, 55% of the CBSM coursework at 3 months.

Except for the fact that the intervention group had a slightly higher proportion of participants with homosexual or bisexual or uncertain sexual orientation, the baseline characteristics were balanced between the two groups (Table 1). Those lost to follow-up were older than those who completed the trial; however, other characteristics were balanced between nonrespondents and respondents (see Multimedia Appendix 2).

The results of changes in depression are summarized in Table 2. At the 3-month follow-up, participants in the intervention group had significantly reduced depression severity (CES-D) compared with the control group (from 23.9 to 17.7 vs from 24.3 to 23.8; mean difference=−5.77, 95% CI −7.82 to −3.71; P<.001), with a standard effect size (Cohen d) of 0.66 in favor of the Run4Love intervention (Multimedia Appendix 2). At the 6- and 9-month follow-ups, between-group differences in the CES-D score remained statistically significant (6-month follow-up: −6.08, 95% CI −8.33 to −3.83; P<.001; Cohen d=0.63; and 9-month follow-up: −5.30, 95% CI −7.77 to −2.83; P<.001; Cohen d=0.51). LMM indicated significant interactions between the groups and at each follow-up time (at 3, 6, and 9 months), with statistically significant between-group differences in the CES-D score for mean change from baseline, controlling for baseline characteristics (P<.001; Table 2). Changes over time are presented in Figure 2. The results were not substantially different from data gathered before multiple imputations of missing data (Multimedia Appendix 2).

At the 3-month follow-up, participants in the Run4Love intervention group, when compared with the control group, had significantly improved QOL (WHOQOL-HIV BREF: from 77.4 to 82.6 vs 76.6 to 77.0; mean difference=4.79, 95% CI 2.72 to 6.87; P<.001), self-efficacy (GSES: from 24.4 to 26.6 vs from 23.3 to 23.4; mean difference=2.16, 95% CI 0.92 to 3.40; P<.001), and SWCQ positive coping (from 18.4 to 20.7 vs from 18.3 to 17.8; mean difference=2.91, 95% CI 1.39 to 4.43; P<.001; Table 3, for more details, see Multimedia Appendix 2). In comparison with the control group, participants in the intervention group also had significantly reduced perceived stress (PSS: from 20.0 to 15.7 vs from 20.7 to 18.9; mean difference=−2.45, 95% CI −3.63 to −1.27; P<.001) and depression severity (PHQ-9: from 10.2 to 6.8 vs from 10.7 to 8.9; mean difference=−1.56, 95% CI −2.63 to −0.50; P=.004). There were no significant between-group differences in changes in SWCQ negative coping, HIV-related stigma (HIV Stigma Scale), physical activity (METs; P>.005; Table 3; for more details, see Multimedia Appendix 2).

At the 6- and 9-month follow-ups, the between-group differences remained statistically significant for QOL (6-month follow-up: 6.6, 95% CI 4.24 to 8.87; P<.001 and 9-month follow-up: 5.84, 95% CI 2.76 to 8.31; P<.001) and SWCQ positive coping (6-month follow-up: 3.41, 95% CI 1.80 to 5.02; P<.001 and 9-month follow-up: 2.53, 95% CI 0.85 to 4.21; P=.003) but not for self-efficacy (GSES at the 6-month follow-up: 1.86, 95% CI 0.50 to 3.22; P=.007 and GSES at the 9-month follow-up: 1.55, 95% CI 0.19 to 2.91; P=.03). At 6 months, between-group differences remained statistically significant for perceived stress (PSS: −1.88, 95% CI −3.10 to −0.67; P=.003) and depression severity (PHQ-9: −2.01, 95% CI −3.20 to −0.83; P<.001) but not for the 9-month follow-up (perceived stress: −1.79, 95% CI −3.06 to −0.53, P=.006; PHQ-9: −1.17, 95% CI −2.46 to 0.13; P=.08). At 9 months, participants in the intervention group had significantly reduced HIV-related stigma compared with the control group (HIV Stigma Scale between-group difference: −2.87, 95% CI −4.71 to −1.03; P=.002), which did not occur at 3 months (between-group difference: −2.29, 95% CI −3.93 to −0.65; P=.006) or 6 months (between-group difference: −2.05, 95% CI −3.83 to −0.28; P=.02). There were no statistically significant between-group differences in change in SWCQ negative coping and physical activity (METs) from baseline to the 3-, 6-, and 9-month follow-ups (Table 3; for more details, see Multimedia Appendix 2). The results from LMM did not substantially change after controlling for baseline characteristics (Multimedia Appendix 2).

As reported in the Methods section, hair samples were not properly collected; therefore, we did not have cortisol data. No adverse events were reported.

In the post hoc exploratory analyses, the proportion reduction in depression measured by CES-D ≥16 was greater in the intervention group than in the control group (Table 2). The between-group differences in the proportion reduction in depression (CES-D ≥16) were 23.3%, 25.3%, and 27.3% at 3, 6, and 9 months, respectively, in favor of the Run4Love intervention (3-month proportion reduction: 44.0% vs 20.7%; P<.001; 6-month proportion reduction: 48.7% vs 23.3%; P<.001; 9-month proportion reduction: 51.3% vs 24.0%; P<.001; Table 4). In addition, patients in the intervention and control groups reported high levels of satisfaction (92%-97%) at all three assessments, and there were no significant differences between the groups (see Multimedia Appendix 2 for details).

Except for age and marital status, there were no statistically significant interactions in the post hoc exploratory analyses, including gender, education, sexual orientation, family income, duration of HIV infection, or baseline CES-D score (Table 5). Those who were younger and not married had statistically significant improvement (between-group differences) in the CES-D score than those who were older or married (Table 5).

The WeChat-based mHealth intervention Run4Love significantly reduced depression severity measured by CES-D by 5.77 points at 3 months compared with usual care, and the improvement sustained at 6- and 9-month follow-ups, with a medium-to-large effect size of 0.66 at 3 months [33]. Between-group differences in depression (CES-D ≥16) proportion reduction were consistently more than 20% at the 3-, 6-, and 9-month follow-ups, in favor of the intervention group. The intervention also improved QOL, self-efficacy, SWCQ positive coping, reduced perceived stress, and depression severity (PHQ-9), compared with the control group at 3 months. The improvements in secondary outcomes such as QOL and SWCQ positive coping remained significant at the 6- and/or 9-month follow-ups. In addition to good efficacy, the Run4Love intervention demonstrated good feasibility as all participants reported a high level of satisfaction.

The good effect sizes in the primary outcome and most secondary outcomes could be attributed to the following reasons. First, the study design was informed by extensive previous work, including a pilot mHealth intervention among PLWH [17-19]. Second, we culturally adopted the evidence-based CBSM, which had proven effect on stress management and positive coping [21]. Third, the Run4Love intervention was built on the enhanced WeChat platform, with functions of automatic distribution of multimedia programs, tracking of CBSM completion and physical activity, and personalized feedback and incentives. Fourth, we built our Run4Love mHealth intervention on a popular social media platform WeChat, used daily by most participants; it is easy for participants to join and use it continuously. Fifth, we established trust with participants at recruitment and continuously engaged them via tailored reminders, feedback, and incentives. Finally, the electronic questionnaire used for data collection ensured minimum missing values [34].

We also noted that older participants were also more likely to drop out of the study despite the high retention rate, suggesting a possible digital divide [34]. Our subgroup analyses showed that younger and unmarried participants benefited more from the intervention than their counterparts. Further research is required on the effective strategies to deliver mental health services to older PLWHD.

The literature documents that elevated depressive symptoms experienced by the PLWH are associated with the deteriorated immune system, worsened disease progression, poor QOL, increased risky behaviors, and increased mortality [35,36]. CBSM has been effective in improving mental health outcomes in PLWH [37]. However, the effects of these interventions delivered via mobile tools are understudied. A recent study by Schnall et al [10] randomized 80 low-income PLWHD who experienced at least two out of the 13 prespecified symptoms in the week before the mHealth intervention of “mobile Video Information Provider” with self-care strategies or a control group. At 12 weeks, participants in the intervention group showed improvements in 5 symptoms, including depression. However, this was a small-scale feasibility study conducted in New York City without follow-ups to examine long-term effects. To the best of our knowledge, the Run4Love intervention is one of the first RCTs of mobile app–delivered mHealth interventions among PLWHD with long-term and multiple follow-ups. It is also among the first efforts that adapted CBSM to a social media app and delivered it via a social media app to PLWHD. The findings of this study suggest that mHealth interventions could provide a feasible therapeutic option for many PLWHD in resource-poor settings where mental health services are limited but smartphones are widely accessible.

This RCT has several major limitations. First, most participants were young men who have sex with men recruited from a large hospital in South China. The findings from this study might not be generalizable to other PLWH, especially older PLWH, those living in rural areas, or those not infected through homosexual transmission. Second, intervention contamination was possible, and some participants might have shared the Run4Love information via their private WeChat accounts. However, the effect of such contamination was limited and, if any, only might have diluted the observed effect. Third, the intervention completion rates were suboptimal, but these were comparable with other internet-based cognitive behavioral therapy interventions [38]. With improved completion rates, intervention effects might be more pronounced. Fourth, the research staff was not blind to group allocation; however, the use of an electronic questionnaire on a tablet may limit the bias introduced by assessors. Finally, the improper collection of hair samples led to insufficient weight of hair and invalid readings. We therefore did not have cortisol data, which were planned for measuring chronic stress in our study protocol. More rigorous training and laboratory procedures are needed to ensure quality data collection. Future studies also need to include a robust biomarker to measure changes in mental health outcomes.

The WeChat-based mHealth intervention Run4Love effectively reduced depressive symptoms in PLWHD, and the effect was sustainable at the 9-month follow-up. QOL and other psychosocial measures were also significantly improved at follow-ups. This RCT suggested that mHealth interventions to deliver mental health services to PLWH were feasible and effective, even in resource-limited settings, such as China. Further research is needed to assess the generalizability, biomarkers, and cost-effectiveness of such interventions.