This study used the data from a parallel-design randomized controlled trial (RCT). The CONSORT (Consolidated Standards of Reporting Trials) checklist was used. The study was conducted in Guangzhou, China, between September 2017 and October 2018. A total of 300 people living with HIV and depressive symptoms were recruited by the research staff from a designated hospital for HIV/AIDS treatment in Guangzhou in 2017. Guangzhou is the capital city of Guangdong province, and it is the third largest city in China. The Run4Love RCT protocol was approved by the Institutional Review Board at Sun Yat-Sen University (2018 No. 31) [29].

Participants were recruited if they (1) were aged 18 years or above, (2) tested HIV-seropositive, (3) had elevated depressive symptoms (measured by the Center for Epidemiologic Studies-Depression; CES-D score ≥16), (4) were WeChat users, and (5) were willing to provide hair samples (for the purpose of measuring cortisol as a biomarker of chronic stress). Participants were excluded if they were (1) currently on psychological treatment, (2) unable to complete baseline and follow-up surveys, or (3) unable to engage in the intervention (eg, unable to read or listen to the intervention materials on WeChat or perform physical exercise due to medical or other reasons). Participants who met the eligibility criteria and were willing to participate completed a baseline survey and were randomized to the intervention or the waitlist control group with a 1:1 ratio by the research staff (Figure 2).

The intervention protocol has been detailed elsewhere [29]. Briefly, the Run4Love WeChat-based intervention comprised two components: adapted CBSM courses and physical activity promotion [29]. The intervention adapted the evidence-based CBSM courses into China’s context and included 9 sessions and 3 review sessions with information on meditation, muscle relaxation, and coping skills. The courses were adapted into multimedia formats, including articles, audio clips, and posters. These materials were sent through an enhanced social media platform of WeChat 3 to 5 times a week for 3 months. Physical activity promotion consisted of information on instructions and benefits of regular physical exercise. The enhanced WeChat platform had extended functions, including automatic sending of materials, patient engagement tracking on course completion, and personalized feedback on a weekly basis. In the booster session, a total of 7 most read items were redelivered to the participants in the intervention group in the 3 months following postintervention. Participants in the control group received a printed brochure on nutrition guidelines in addition to the usual care for HIV treatment.

We calculated the sample size of the RCT based on the primary outcome of depressive symptoms measured by CES-D. Effect size was used to calculate between-group differences. We hypothesized an effect size of 0.4, with α=.05, β=.15, and a dropout rate of 20%, thus arriving at a sample size of 282 for the RCT. Finally, a total of 300 participants were recruited. For SEM, the recommended sample size of 5 observations per estimated parameters was considered [30]. Given a total of 46 estimated parameters in the SEM of our study, the sample size of 300 was sufficient for the analysis.

The research staff randomly assigned the participants to the intervention group or the control group 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 participants nor the research staff were blinded to the intervention.

First, descriptive statistics were reported on demographic characteristics, baseline HIV-related stigma, depressive symptoms, and QOL. Mean and SD values were used to describe continuous variables with normal distribution, and median and IQR values, for continuous variables with skewed distribution. Frequencies and percentages were used to describe categorical variables. Kolmogorov-Smirnov and Levene tests were used for normality and homogeneity tests.

Second, group differences were examined for the outcome and mediators at 4 assessment points by independent sample t test for continuous variables with normal distribution and Wilcoxon rank-sum test for continuous variables with skewed distribution.

Third, bivariate analyses were performed to examine relationships between baseline demographic characteristics and QOL. Independent samples t tests were used to examine differences in QOL by different demographic characteristics. Pearson correlation analyses were performed to examine relationships between QOL and continuous demographic variables, such as age. Significance level of potential associations was P<.10. Variables significant in bivariate analyses were considered as potential confounders and controlled in the SEM.

Fourth, in SEM analyses, HIV-related stigma, depressive symptoms, and QOL were treated as latent variables and measured by their subscales [44]. SEM allows to control for measurement error of variables; thus, it yields unbiased estimates of latent variables [45,46]. To assess the goodness of fit of the measurement models, confirmatory factor analyses (CFAs) of HIV-related stigma, depressive symptoms, and QOL were performed, respectively. With satisfactory measurement models, the SEM was then conducted to examine the hypothesized mechanisms of long-term intervention effects on participants’ QOL.

Finally, a mediation model of SEM was performed to test the hypotheses of sequential mediating effects of HIV-related stigma and depressive symptoms by using longitudinal data. Such models were designed to explore the mechanisms, direct and indirect effects of the intervention, and sequential relationships [47]. Based on the associations between HIV-related stigma, depressive symptoms, and QOL in previous studies, we assigned sequential orders to the SEM variables [24,25,48-50]. Specifically, HIV-related stigma at 3-month follow-up, depressive symptoms at 6-month follow-up, and QOL at 9-month follow-up were estimated using the model. The pathways of intervention→QOL, intervention→depressive symptoms→QOL, intervention→HIV-related stigma→QOL, and intervention→HIV-related stigma→depressive symptoms→QOL were examined separately (dummy coded as 0=control group, 1=intervention group). Baseline QOL and sociodemographic characteristics that were significantly associated with QOL in bivariate analyses were controlled as covariates in the model. The statistical significance in SEM was defined at P<.05.

To evaluate the goodness of fit of the model, multiple indicators were used, including the chi-square statistic, Comparative Fit Index (CFI), Tucker-Lewis Index (TLI), root mean square error of approximation (RMSEA), and standardized root mean square residual (SRMR). A smaller chi-square value indicates better model fit. Moreover, CFI ≥0.95, TLI >0.90, RMSEA ≤0.06, and SRMR ≤0.08 indicate good model fit [51,52]. Descriptive statistics, bivariate statistics, and correlation analyses were performed using SAS version 9.4 (SAS Institute, Inc). CFA and SEM were tested by robust maximum likelihood method (estimator=MLR in Mplus) and performed using Mplus version 7.0 [53,54].

The Run4Love program recruited 300 participants; their baseline characteristics are described in Table 1. Participants’ mean age was 28.3 (SD 5.8) years. Most of the participants were male (277/300, 92.3%), nonheterosexual (245/300, 81.7%), not married (262/300, 87.3%), and well educated (high school and above: 182/300, 60.7%). The mean scores of HIV-related stigma, CES-D, and QOL were 37.5 (SD 7.6), 24.1 (SD 6.6), and 77.0 (SD 9.2), respectively, at baseline. A majority (213/300, 71.0%) of the participants had a moderate level of stigma, whereas 20.3% (61/300) had high level and 8.7% (26/300) had low level of stigma.

The proportion of homosexual, bisexual, or sexual orientation-uncertain participants in the control group was slightly lower than that in the intervention group (115/150, 76.7% vs 130/150, 86.7%; P=.03). Other demographic characteristics and psychosocial outcomes (HIV-related stigma, depressive symptoms, and QOL) were balanced between the two groups at baseline.

The proportion of participants who lost to follow-ups was less than 15% over 9 months. Of all participants, 91.3% (274/300; n=139 in the intervention group; n=135 in the control group), 88.3% (265/300; n=132 in the intervention group; n=133 in the control group), and 86.7% (260/300; n=133 in the intervention group; n=127 in the control group) completed the follow-up surveys at 3, 6, and 9 months. The characteristics of the participants who lost to follow-ups were not significantly different from the remaining participants.

As shown in Table 2, the Run4Love social media-based intervention had significant effects on QOL and potential mediating variables over 9 months. Repeated measurements of the outcome variable (QOL) and potential mediators, including HIV-related stigma and depressive symptoms at 4 assessment points for the intervention and control groups, are shown in Table 2. The intervention significantly reduced HIV-related stigma and depressive symptoms, and improved QOL in intervention group participants compared to control group participants at 3-, 6-, and 9-month follow-ups. No adverse events were reported.

Table 3 shows the correlations between demographic characteristics and QOL at baseline. Bivariate analyses indicated that gender, sexual orientation, education, marital status, and household registration were significantly associated with QOL among people living with HIV and depressive symptoms at baseline. Specifically, those who were male, nonheterosexual, and unmarried; had a higher education level; or had an urban household registration had better QOL at baseline. These factors were included as controlling covariates in the SEM.

CFA indicated that the measurement models of HIV-related stigma, depressive symptoms, and QOL yielded good model fit. Indices of the measurement models were reported in Table 4. All factor loadings of each scale were significant at P<.05 level. Results of the standardized factor loadings are shown in Figure 3.

The structural model showed a satisfactory model fit (X²₁₃₈=240.4; P<.001; CFI=0.93; TLI=0.92; RMSEA=0.06; SRMR=0.06). Demographic characteristics, including gender, sexual orientation, education, marital status, and household registration, were controlled as covariates in the model.

Except for the pathway from intervention to QOL, other pathways were significant, which means that the Run4Love intervention had complete mediation effects on QOL via stigma and depressive symptoms. Standardized regression coefficients for the model are reported in Figure 3 and Table 5. Results indicated that the intervention reduced participants’ stigma at the 3-month follow-up (standardized β=−.19, P=.001), which was positively associated with depressive symptoms at the 6-month follow-up (standardized β=.47, P<.001) and negatively associated with QOL at 9-month follow-up (standardized β=−.20, P=.005). The intervention also reduced depressive symptoms (standardized β=−.21, P<.001), which consequently resulted in improved QOL at 9 months (standardized β=−.58, P<.001). The pathway from depressive symptoms to QOL had the strongest effect size in the mediation model (standardized β=−.58, P<.001). There were complete mediation effects of the intervention on QOL via stigma and depressive symptoms. Specifically, the pathway through depressive symptoms alone accounted for the largest mediation effect size of 57.1% (0.12/0.21). The path through stigma→depressive symptoms and stigma alone accounted for 23.8% (0.05/0.21) and 19% (0.04/0.21) of the total effect size on long-term QOL, respectively.

To our knowledge, this study is among the first efforts to explore the potential causal mechanisms of long-term improvement in QOL in an mHealth intervention using longitudinal data. The findings revealed significant mediating roles of stigma and depressive symptoms in improving long-term QOL for the mHealth intervention participants. Specifically, the long-term (9-month follow-up) intervention effect on the improvement of QOL was entirely mediated by the reduction of stigma in the short term (3 months) and reduction of depressive symptoms in the mid term (6 months).

Although several studies have explored the mechanisms of psychosocial interventions, the mechanisms of QOL improvement in the long term remain understudied. For example, studies on mind-body therapies have found that mindfulness might serve as a potentially important mechanism for patients’ QOL improvement in such interventions [16,55]. However, literature examining the mechanisms of QOL improvement in psychosocial interventions are scarce, especially in the long term. In addition, methods employed in previous studies such as repeated measures analyses of covariance and latent growth curve model could not identify the sequential and causal relationships when exploring intervention mechanisms [16,17,56]. Our study contributed to the literature by affirming the sequential and potential causal relationships between HIV-related stigma, depressive symptoms, and QOL, which have not been reported in previous studies. Furthermore, only a small number of studies used longitudinal data, of which, to the best of our knowledge, none examined the sequential relationships [16,17,56]. In this study, the use of SEM with longitudinal data allowed us to assign mediators and the outcomes in chronological order, thus shedding light on the potential causal relationships of the intervention mechanisms.

Given the critical roles of stigma and depressive symptoms in improving long-term QOL, it is important to address stigma and alleviate depressive symptoms in interventions aiming for QOL improvement among people living with HIV. There are several effective ways to reduce HIV-related stigma and to alleviate participants’ depressive symptoms as suggested in the literature and shown in this study. First, the intervention content should be evidence-based in reducing HIV-related stigma and alleviating depressive symptoms [5,12]. The intervention materials of the Run4Love program were adapted from the evidence-based and theory-guided CBSM courses, which have been proved effective in reducing stigma and depressive symptoms among people living with HIV [12,57,58]. The core elements of the CBSM courses, such as stress management and coping skills, were preserved and adapted into multimedia formats. The participants in the intervention groups had a moderate level of patient engagement. The cumulative completion rates were 50.6%, 51.5%, and 50.8% at 1, 2, and 3 months, respectively, in the intervention group, which were comparable to other mHealth interventions [59]. The positive relationship between patient engagement and health outcomes has been confirmed in our previous study [59]. Therefore, evidence-based intervention content, rigorous design, and implementation might have contributed to the significant reduction of HIV-related stigma and depressive symptoms in the intervention group.

Second, another effective way to reduce stigma is through social contact; this is one of the most effective strategies to reduce stigma, provide social support, and alleviate depressive symptoms [5,60,61]. Comparatively, mHealth interventions provide varied forms of social contact with better convenience and privacy than face-to-face interventions [62,63]. For example, anonymous virtual communities, such as community message board and web-based forums, serve as easy and secure access for people living with HIV to interact with peers and receive social support [62,63]. Earlier studies suggested that mass media communication or supportive virtual groups could be incorporated in interventions to mitigate stigma among people living with HIV [62,64]. Such web-based communities and connections are much needed by people living with HIV because of persistent stigma—both perceived and internalized—against HIV and people affected by HIV, as well as the consequent fear of discrimination and social isolation [65,66]. In addition, mHealth interventions allow alternative web connections (eg, human-machine interactions) to incentivize and engage participants. For example, the Run4Love program incorporated automatic weekly feedback on each participant’s completion status of the intervention materials, with personalized verbal incentives. Such personalized and patient-centered human-machine interaction is a unique advantage of mHealth interventions. Future mHealth interventions for people living with HIV need to incorporate social contact in various forms as an essential component to address HIV-related stigma and decrease social isolation, so as to reduce depressive symptoms and achieve long-term improvement in people living with HIV’s QOL.

Third, since HIV-related stigma is a multidimensional aspect, public policies and coordinated efforts are needed to reduce HIV-related stigma at multiple levels [67,68]. At the societal level, public policies and effective interventions targeting on stigma reduction, HIV testing, ART provision, and social empowerment are needed to address HIV-related stigma among people living with HIV [69]. Community-based support services and educational campaigns should be prioritized to reduce stigma and discrimination [70]. For example, in China, many celebrities, including the First Lady Peng Liyuan, joined the public campaign to reduce discrimination against people living with HIV and promote HIV testing and treatment [71,72]. At the health care level, Voluntary Counseling and Testing program has been instituted at the Centers for Disease Control and Prevention (CDC) and local hospitals [73]. At the individual level, a number of intervention programs focused on stigma reduction, problem-solving skills, and psychological support have been implemented [68].

There are some limitations to this study. First, we only focused on HIV-related stigma and depressive symptoms as two important mediators; however, other variables, such as stress, social support, and self-efficacy, might also be potential mediating factors and were not included in this study. Future studies are needed to explore the mediating effects of these factors to better understand the mechanisms of social media-based mHealth interventions. Second, the measurements of stigma, depressive symptoms, and QOL were self-reported and thus might introduce potential recall bias. However, SEM analysis has already considered the measurement errors of these psychosocial variables and therefore provided more reliable model estimations. Third, since the participants were recruited from one hospital in an urban setting and the sample comprised mostly male participants, the generalizability of the study findings to other patients or geographical locations needs to be taken with caution, especially for women and those living in rural areas.

In conclusion, this study is one of the first efforts to examine the potential causal mechanisms of an mHealth intervention in long-term QOL improvement by using longitudinal data and SEM. We found that the long-term intervention effect on QOL improvement was entirely mediated by the reduction of stigma in the short term and reduction of depressive symptoms in the mid term. The findings underscore the critical roles of reducing HIV-related stigma and depressive symptoms in mHealth interventions targeting QOL improvements among people living with HIV. We call for targeted mHealth interventions to improve long-term QOL among people living with HIV, especially social media-based interventions that can reduce HIV-related stigma and alleviate depressive symptoms.