We conducted a two-arm, parallel, randomized controlled trial. Participants were individually randomized in a 1:1 ratio to either the SMART Connections intervention (intervention group) or standard of care (control group). This was an open-label study with no blinding of the study staff or participants. We conducted face-to-face structured interviews with participants and extracted data from medical records at enrollment (baseline) and again at the completion of the intervention period (endline), approximately 6-9 months after enrollment. We also conducted in-depth interviews (IDIs) at endline with 2 subsets of intervention group participants: participants with high group participation and those with low group participation (classified by group facilitators). We obtained informed consent from youth aged 18 years and older; for youth aged 15 to 17 years, we obtained parental permission and adolescent assent at enrollment. This study was reviewed and approved by FHI 360’s Protection of Human Subjects Committee, the University of Uyo Teaching Hospital’s Institutional Health Research Ethical Committee, and the Cross River State Health Research Ethics Committee.

SMART Connections was informed through workshops conducted with stakeholders and YLHIV in Akwa Ibom State, Nigeria, to gather input for design and content [12,13]. The intervention was designed to promote retention in HIV care by leveraging social support and improving HIV-related knowledge and treatment literacy [46,47]. The content of the structured support groups was adapted from an existing support group guide, Positive Connections, and delivered through secret Facebook groups [48]. The intervention was delivered over approximately 22 weeks (2 weeks per session) to groups of about 15 to 25 youths, with nearly daily activities (Figure 1). The intervention guide is attached as a Multimedia Appendix 1.

Two community-based organizations were engaged to recruit support group facilitators and assist study staff in training them to deliver the intervention. Selected facilitators had received prior training to facilitate in-person support groups and were living with HIV themselves. Facilitators underwent a 5-day training and received an implementation guide, smartphone and monthly data allowance. Facilitators met monthly with the study staff to debrief on challenges they had encountered and provided support to one another.

Once a sufficient number of participants were recruited and randomized to form a support group, each group began with an in-person meeting, during which participants met one another and the facilitator and agreed-upon ground rules for participation. Facilitators also instructed participants on how to secure their phones (using a password or passcode), turn off Facebook notifications so that messages and posts would not show up on their phones and to log out of Facebook after each use to keep others from accessing their Facebook accounts.

All study participants continued to receive standard services available to YLHIV in study facilities and communities. Standard health services included routine clinical care for HIV treatment, including viral load tests, active case management by community volunteers with intensive adherence support during the first 4 weeks of ART, and enhanced adherence counseling for patients with unsuppressed viral loads.

All study participants received a smartphone (valued at US $65) and a monthly data bundle (US $3.5 equivalent to 1 gigabyte of data) for the duration of the intervention.

The study was implemented in 14 health facilities located in Akwa Ibom State and Cross River State. The 2 states lie adjacent to each other in south-central Nigeria and share many sociocultural similarities. The population of interest was YLHIV aged 15 to 24 years who had been on ART for less than 12 months. In addition to age and ART eligibility criteria, participants had to demonstrate basic literacy for web-based chats, which was assessed at enrollment by asking the participant to read 3 short messages from intervention materials. We excluded individuals who were unable to attend the initial intervention group meeting if randomized to the intervention group, who were enrolled in another research study related to HIV retention or ART adherence, or who were severely ill and unable to provide informed consent at the time of recruitment. For IDIs, facilitators identified group members with high and low participation from whom participants were purposively selected to achieve representativeness by group participation, sex, and age.

The study was powered to detect at least a 0.125 difference in the cumulative probability of retention at endline (0.45 in the control group and 0.575 in the intervention group), corresponding to a hazard ratio of 0.69, with 80% power and 5% significance level for a two-sided comparison using the log-rank test. This resulted in a total sample size of 500 (250 per study arm). Calculations also assumed exponential times to event and a 10% loss to study follow-up. Our estimate of retention was based on program data for the prior year from the United States Agency for International Development (USAID)–funded bilateral HIV services delivery (Strengthening Integrated Delivery of HIV/AIDS Services; SIDHAS) project, which supported HIV services in the facilities selected for this study.

We originally planned to recruit YLHIV aged 15 to 22 years on ART for 6 months or less and expected to achieve our sample size within 3 months. Due to substantially lower than expected numbers of YLHIV enrolling in HIV treatment, we amended the study protocol and expanded the eligibility criteria to include YLHIV aged up to 24 years and on ART for 12 months or less. We also extended enrollment from 3 to 8 months and added 3 facilities (originally 11).

Finally, we intended to collect outcome data from participants after the intervention was completed and at 1 year from enrollment (approximately 6 months later) but were unable to collect a second round of outcome data because of prolonged recruitment—the funding project came to an end before the final round of data collection could take place.

Eligible participants were sequentially recruited from patients attending clinic visits at the study facilities. Health care staff informed eligible participants of the study and directed those interested to a data collector who was stationed in the facility. Data collectors met with potential participants in a private setting in the facility, provided additional details on the study, then proceeded to obtain informed consent and enroll participants. Data collectors assigned participants to study arms using randomization envelopes. Randomization groups were concealed in sequentially numbered sealed opaque envelopes. The allocation sequence was generated using permuted blocks and stratified by local government areas (LGA), which are administrative subunits of the state, by a biostatistician otherwise uninvolved in the study using a validated SAS macro (SAS version 9.4) [49]. Recruitment occurred at health facilities, but randomization was managed from a central location for all facilities within an LGA. This was an open-label study with no blinding of the study staff or participants.

We collected demographic information (sex, age, relationship status, education, occupation, and religion) and information on HIV history, including date of HIV diagnosis, ART start date, disclosure of HIV status to others, and viral load testing. Additionally, we collected data on secondary outcomes of ART adherence, HIV-related knowledge, and social support as well as psychosocial factors associated with poor retention, including social isolation, depression, and perceived and experienced stigma.

Retention: For our primary time-to-event analysis, we computed the time retained in care from study enrollment to the date the participant was no longer classified as active on treatment, consistent with the President’s Emergency Plan for AIDS Relief (PEPFAR) indicator definitions [50]. We recorded dates for all scheduled clinic visits from patient medical records for each participant from study enrollment until the end of the study. If a participant failed to return after a scheduled visit for more than 28 days, the date of the missed visit was the date of loss to care recorded, unless death or transfer of service was documented before the missed visit. For a small number of patients, their first missed scheduled visit was scheduled on or within 28 days before study enrollment. If the participant missed this first scheduled visit by more than 28 days, he or she was assigned a retention time of 0.

We also reported on treatment status at endline. To be considered active on treatment at endline, an individual must have attended a visit or had a follow-up visit scheduled within 28 days of the date of their endline questionnaire. For those who did not complete an endline questionnaire, an approximate endline date was used (based on the median time in the study of those enrolled the same month). Participants who died or transferred facilities were categorized accordingly.

Measurement of secondary outcomes and other psychosocial variables are summarized in Table 1.

Structured questionnaires programmed into password-protected computer tablets were used to collect data from participants at both baseline and endline. We also extracted medical record data (MRE) from patient charts at both time points into a separate form programmed into the tablets. Data from tablets were uploaded daily to a secure computer server. Trained interviewers administered a semistructured IDI guide. IDIs were audio recorded and transcribed verbatim.

All analyses used an intent-to-treat approach, in which all randomized participants were included in the statistical analysis and analyzed according to the group to which they were originally assigned, regardless of the treatment they received during the study. The primary hypothesis was that YLHIV enrolled in HIV treatment services who participate in SMART Connections will be more likely to be retained in HIV care than YLHIV enrolled in HIV treatment services who do not participate in the intervention.

Kaplan-Meier cumulative retention probabilities are reported with 95% CIs and plotted by study arm. Participants who were confirmed to have died or transferred to a facility outside the study facilities were considered censored. Participants who had elected to drop out of the study (n=8), or for whom all MRE were missing because of missing charts (n=14) or all visit data were missing data in charts (n=3), were excluded from the analysis.

The retention probabilities between the groups were compared using a log-rank test stratified by LGA with a two-sided α of .05. We also report on retention descriptively, examining lapses in care, and return to treatment over the course of study follow-up. The relationships between treatment exposure and secondary outcomes (ART adherence, HIV knowledge, and social support) as well as social isolation, depression, and HIV-related stigma were explored with t tests for continuous outcomes and chi-square test for categorical outcomes, using two-sided tests, with a significance level of 0.05.

We applied thematic analysis to the IDI data. A structured codebook was developed a priori based on the interview guide; emergent thematic codes were added during the analysis. To assess intercoder consistency, 2 analysts independently coded 3 transcripts, compared coding, and resolved differences through discussion. Coding was conducted using NVivo 12 [56]. Once all transcripts were coded, textual coding reports were produced.

We recruited 356 youths between September 2018 and April 2019, 353 (99.2%) of whom enrolled in the study (Figure 2). A total of 4 participants were removed after enrollment, as they provided false eligibility information, and were deemed ineligible. Participants were then randomly allocated to the intervention group (n=177) or control group (n=172). At endline (June to November 2019), 108 participants were lost to follow-up from the study, including 4 who died, 8 who discontinued study participation, and 96 who were not reachable for an endline interview. Endline interviews and MRE data were completed for 241 participants. MRE data were collected for an additional 84 participants who were not reachable for an endline interview but who had not elected to END study participation. We conducted IDIs with 21 intervention participants following the endline survey, of whom 16 were female and ranged in age from 17 to 24 years; 13 were classified as high participation and 8 as low participation.

A total of 10 randomization errors occurred during the study: 6 participants randomized to the control group were erroneously recorded by data collectors to be the intervention group and took part in the intervention; 4 participants randomized to the intervention group were erroneously recorded as in the control group and did not participate in the intervention. Two other participants, randomized to and recorded as in the control group, were erroneously contacted to participate in the intervention. In addition, 14 people randomized to the intervention group chose not to participate in the intervention. Thus, a total of 167 participants joined an online support group during the study: 160 randomized to the intervention group, 5 due to randomization errors, and 2 control group participants erroneously contacted.

To assess the potential effects of loss to follow-up, we examined differences between those who did and did not complete an endline interview. There were no statistically significant differences in any background characteristic or baseline values of secondary outcomes between those who completed endline interviews and those lost to study follow-up (results not shown).

Both study arms were similar in all baseline demographic characteristics (Table 2). Most participants were female 87.7% (306/349) and had completed some secondary school or more; the mean age was 21 years (SD 2.3). More than half of the respondents reported being married or in a relationship. Among those in a relationship (including married), fewer than half reported that their partner knew their HIV status, and about half knew their partner’s HIV status. Most participants reported having access to a mobile phone in their household (303/349, 86.8%), with 67.1% (234/349) reporting that they owned their own mobile phone.

The probability of being retained in HIV care, defined as not having missed a scheduled appointment by more than 28 days, was similar between the intervention and control groups (Figure 3).

The probability of remaining in care without more than a 28-day gap was slightly higher in the intervention group than in the control group at each time point, except at 120 days; however, 95% CIs overlapped between the 2 study arms at all time points (Table 3).

Retention, measured as not having missed any scheduled visit by more than 28 days, did not reflect the participant’s treatment status at endline (Table 4). Almost half of all participants missed at least one scheduled visit by more than 28 days during the 6 to 9 months of follow-up and were classified as lost to care at some point; however, most participants were active on treatment at endline. A total of 40.0% (65/163) in the treatment arm and 46.0% (74/161) in the control arm were counted as having been lost-to-care in the time-to-event analysis but were active on treatment at the end of the study.

The average number of clinic visits during the study varied by participant, in part because of variation in the length of time between scheduled visits. The average time between schedule visits was 1 month for most, but one-quarter of participants had an average time between visits ranging from 2 to more than five months.

Barriers to clinic attendance, as described in IDIs, included traveling away from home, anticipating stigma at the clinic, and competing obligations. When asked how easy it was to attend scheduled appointments, one participant described it as follows:

A few participants described hesitation leaving school, work, or family to attend appointments if it required disclosing their HIV status. Findings were similar in endline questionnaire open-ended data; among 79 participants who reported missing an appointment by one week or more, reasons for doing so included traveling (33/79, 42%), transportation issues (20/79, 25%), or were busy (13/79, 16.5%).

Most IDI participants described that the intervention impacted their attitudes and behaviors toward attending scheduled appointments. Some described that the intervention reinforced the importance of attending appointments to prevent gaps in medication and taught them to “take your drugs so that you'll feel okay ” (21-year-old female, high participation). Others described learning strategies to avoid missing appointments, such as asking someone to collect refills if they could not. A few reported that the intervention served as a reminder to attend scheduled appointments; one participant mentioned receiving reminders from group members on their appointment day. Another described that being part of the group:

In contrast, a few participants, most categorized as low group participation, felt that the intervention did not affect their retention in care.

HIV-related knowledge was high in both groups at endline but significantly higher in the intervention group compared with the control group (Table 5). No other statistically significant differences between study arms were observed for ART adherence, social support, social isolation, HIV-related stigma, or depression.

Although the 2 study arms did not differ significantly on social support, nearly all IDI participants, including those with low intervention participation, stated they received social support from facilitators and other group members. These participants described receiving encouragement and advice, having people to “share my feelings with” or “someone to talk to,” and receiving answers to factual and personal questions. One participant described the following:

Some support was related to self-management, such as encouragement to adhere to ART. For example, one participant described group members as:

Participants frequently received multiple types of support and often supported other group members by sharing their own experiences or providing emotional support and advice.

Nearly all IDI participants stated they felt a sense of connectedness with the group, sometimes described feeling as if they were “a family” or that participants got along “as brother and sister.” This feeling was often attributed to having group members of the same age range and HIV status and conferred a sense of safety and confidentiality within the group. In endline questionnaires, when asked what they liked most about the intervention, the most common responses included receiving encouragement and support, the ability to share their problems, and feeling a sense of unity or belonging.

When asked about sources of social support outside of the intervention group, most said they did not receive social support outside of the group, sometimes elaborating that they preferred not to disclose their HIV status to family and friends due to fear of stigma:

Only a couple of IDI participants stated that they received support from family members; a few mentioned a health care provider.

Nearly all who took part in the intervention (as treated) and completed an endline interview agreed that the intervention was useful to them, they enjoyed the intervention and felt comfortable interacting with the facilitators and other group members, and they would recommend the intervention to other YLHIV (Table 6). Nearly all also reported that connecting to the groups on Facebook was somewhat or very easy.

IDI participants often conveyed appreciation for learning about practical aspects of managing HIV, such as taking ART at consistent times during the day and eating “a balanced diet” to support overall health. Many participants also enjoyed learning why medication adherence is important. Participants recalled learning that ART adherence would help them feel healthier and achieve a longer life. One participant described as follows:

Many IDI participants also expressed enthusiasm for the social and interactive elements of the intervention, such as riddles posed by the facilitator. Most felt the Facebook platform was acceptable, reporting that the “secret” groups ensured their privacy, and the web-based format allowed them to interact with the group at their convenience. One participant described as follows:

Despite many strengths, including the use of rigorous experimental study design, the study had a number of limitations. Several factors limited the study implementation and perhaps the interpretation of results. First, the results may not be generalizable to other youths outside our study areas. In addition, recruitment lagged, taking more than twice as long as planned because YLHIV enrollment in ART was substantially lower than estimated using HIV service data before the study. Despite protocol modifications, we were only able to enroll 69.8% (349/500) of the planned participants, reducing the study’s power. Slow enrollment also prolonged the time necessary to enroll enough intervention participants to form a support group in many cases. Thus, participants contributed different amounts of time to the study, and some intervention group participants waited months before their support group could begin. This may have added to their risk of loss to follow up before initiating the intervention.

The context of HIV service delivery also changed during the study implementation. A dedicated surge in PEPFAR-supported HIV services aimed to increase the number of PLHIV enrolled on treatment and retained in care began in the study sites partway through the study [68]. This surge entailed intensive efforts to better support HIV treatment services and PLHIV in their care and treatment. Strategies including community-based ART initiation, multimonth ART dispensing for stable patients, and community-delivered ART refills were implemented to increase ART initiation and to improve treatment adherence and retention [68]. These changes meant that the average time between clinic visits and the number of visits participants had during the study varied. Although participants in both study arms were exposed to this changing context, retention increased dramatically during the surge; retention at endline was considerably higher in both study arms compared with retention rates indicated by the presurge programmatic data used to calculate effect size estimates. These external efforts may have masked any possible impact that the intervention had on retention.

Our findings of improved HIV knowledge and high acceptability are encouraging despite a lack of measurable effect on retention. No single psychosocial intervention is likely to meet the varying needs of YLHIV, but the SMART Connections intervention appears to contribute to some of those needs. Digital health support groups may fill critical gaps in the services available for YLHIV. Given the increasing use of social media by youth, the platform may be a preferred resource for some aspects of HIV-related support. Web-based delivery of support group interventions through platforms such as Facebook can permit people to access information and other group members privately, when convenient, and without travel.

Moving forward, we suggest a few adaptations to the SMART Connections intervention and to continue to examine its potential effects. Expanding groups to include YLHIV on ART for more than 1 year may provide better support to those newly initiating treatment and help meet their own informational and support needs. Findings from the IDIs suggest that the intervention had a perceived effect on social support. We believe that the tool we used to measure social support may not have adequately distinguished between more general social support and social support related to HIV. Further work to develop and test measures of social support that better reflect the support given/provided in the context of HIV should be pursued. Finally, focusing on more reliable outcome measures, such as viral load, is strongly recommended for interventions attempting to improve HIV-related health outcomes.