Study Design and Setting
The study was an individually randomized, group treatment, pilot trial with three randomization waves spaced 6 months apart (Fig. 1 and Supplemental Fig. 1). Enrolled participants were randomized to the SYV intervention or standard of care (SOC) within one month prior to the start of their wave. All participants randomized to SOC were later offered the SYV intervention in a crossover wave at a time that enabled randomized comparisons between SYV and SOC (for details see Supplemental Fig. 1). Importantly, although the study used a stepped wedge design and collected data after crossover, the current paper excludes participant data obtained from the post-crossover periods of the SOC arm in order to compare the durability of findings in the intervention versus SOC arm over time.
The study took place in Moshi, Tanzania at two adolescent HIV clinics that offer a Saturday clinic solely for adolescents 12–24 years of age. The adolescent HIV clinic includes group health talks and peer support networks as part of a routine visit. Kilimanjaro Christian Medical Centre (KCMC) is the northern zone referral hospital serving approximately 430 YPLWH, and Mawenzi Regional Referral Hospital (MRRH) is the regional government hospital for the Kilimanjaro region serving approximately 320 YPLWH. Antiretroviral therapy (ART) is free in Tanzania, and adherence counselling accompanies all clinical visits as part of the Tanzanian national guidelines . Currently, no mental health screening occurs in routine clinical care.
As previously described, YPLWH 12–24 years of age who were fully disclosed and aware of their HIV status and who attended the adolescent HIV clinic at KCMC or MRRH were approached to participate . Recruitment occurred from May to July 2016 through the announcement of the study during the adolescent HIV clinic. Those interested joined a study enrollment list and were grouped by age categories (12–14 years; 15–19 years; 20–24 years of age), sex, and clinic site in the order in which they appeared on the list. All young people attending the clinic who were receiving ART and could understand and fully participate in the study intervention and informed consent process were eligible. Enrollment was not limited to those with identified mental health or adherence challenges because the intervention content was assumed to potentially benefit all YPLWH, not only those who self-reported challenges. The target enrollment was 100 participants (50 per arm) in order to obtain sufficient pilot data for the establishment of feasibility and acceptability of the intervention, and to explore immediate and prolonged outcomes. The data were intended to be used to judge whether to proceed with a future definitive trial. Enrollment and randomization were increased to 130 during the study due to attrition.
Five evaluation visits occurred at: (1) baseline (pre-intervention), (2) 6 months, (3) 12 months, (4) 18 months, and (5) 30 months post-baseline, with baseline separated by 6 months between waves. Intervention was completed prior to the 6 month visit in all waves (Fig. 1 and Supplemental Fig. 1). Participants responded to a structured interview conducted by one of two research assistants in Kiswahili, the official language of Tanzania. The interview included questions about demographics, risk behaviors, perceptions of internal and external stigma, mental health symptoms, and ART adherence. A 5 mL blood sample was drawn to measure HIV RNA. The two research assistants conducting the study evaluation visits were blinded to the participants’ study arm assignment.
Randomization and Masking
Using the study enrollment list, participants were organized into one of three waves of males and females of similar age (Supplemental Fig. 1 for details) at each of the two sites (KCMC or MRRH). Then, within each wave and site, the males and females were separately randomized to SYV or standard of care (SOC) by coin flip just prior to the start of each wave in the presence of the principal investigator and group leaders. For the male MRRH group (Wave 2), only nine of the participants were reachable and able to participate at the time of randomization, thus there is no control group (Supplemental Fig. 1). Participants randomized to SYV received the intervention in a group of 9–11 participants, and participants randomized to SOC continued with usual attendance of the monthly adolescent HIV clinic. Participants in the SOC arm were later offered to participate in the SYV intervention as a cross-over wave (Supplemental Fig. 1). There was no time and attention-matched control group, meaning that the SOC arm did not meet as a study group.
The SYV intervention is a group intervention involving ten group sessions, two designed to be held jointly with caregivers, two individual sessions between the participant and a group leader, and a final celebration to summarize intervention content and distribute completion certificates . Sessions were designed based on formative research and youth advisor input [3, 13]. Components of evidence-based models—including Cognitive Behavioral Therapy, Interpersonal Psychotherapy, and Motivational Interviewing—were chosen based on the best fit for specific youth-identified needs and on whether they could be implemented with fidelity by lay group leaders . A description of the manualized session-by-session intervention protocol has been previously described [10, 15], and the full intervention protocol is available upon request to the corresponding author. Sessions lasted approximately 90 min and were held on consecutive Saturdays for a duration of 10 weeks at a non-clinical site (see Supplemental Fig. 1 for detailed dates). This resulted in approximately 18.5 contact hours in the intervention group. Fidelity to the intervention was marked on a fidelity checklist by the observing group leader. The six month study visit occurred after completion of the intervention in all waves. Enrolled young people who missed more than two consecutive intervention sessions were dropped from their assigned group to preserve group rapport and forward momentum, as the intervention content builds upon prior sessions. Young people who prematurely ended SYV were offered the possibility of joining a later cross-over control group.
The study was implemented with Tanzanian group leaders who were trained to deliver the intervention and were closely supervised twice weekly. Six young adults (age 23 to 30 years) with a background of either living with HIV and/or having prior experience with mental health research were trained during an intensive two-week training on-site with the principal investigator and the US-based clinical psychologist. Two group leaders led the group-based intervention session, and one group leader marked the fidelity checklist and took session notes at each session. The fidelity checklist and session notes were used to guide the weekly supervision meetings that occurred both in person with the group leaders and principal investigator, and using telecommunication via Skype to communicate with the US-based clinical psychologist.
Demographic measures including primary caregiver, social support (if you are feeling sad, is there someone you can talk to?), home environment (availability of plumbing and electricity), livelihood (whether in school or working), cell phone ownership, sexual history, and alcohol and drug use were documented at every study visit. No changes were made to the study measures over the course of the trial.
Mental health measures were chosen based on published research and harmonization with other adolescent mental health projects in Africa. Justification for use in this trial has been presented in more detail previously [3, 10]. The Patient Health Questionnaire (PHQ-9) was used to measure symptoms of depression. There has been published research on use in adolescent populations in Kenya, Ghana, Zimbabwe, and the Republic of Congo [4, 16,17,18], in addition to our prior research in Tanzania . This measure includes nine questions with a response range of 0–27. A score of 10 or greater is commonly used as a screening threshold suggestive of moderate to severe depressive symptoms [19, 20]. The Strengths and Difficulties Questionnaire (SDQ) was used to measure emotional and behavioral symptoms. The scale includes 25 questions (including 5 prosocial behavior questions that are not included in the composite score). A cut-off of 17 or higher (score range of 0–40) signals symptomatology of mental health difficulties with mean Cronbach α of 0.73 as assessed by the tool developers , signifying the internal consistency of this tool. Trauma related symptoms were assessed using the UCLA Post Traumatic Stress Symptoms Exposure Screener and Reaction Index, and survey responses were modified to a four-question Likert scale (None, Some, Much, and Most) based on translation from Swahili [22, 23]. A cut-off of 18 or greater was suggestive of post-traumatic stress related symptoms . Prior use of the tool internationally demonstrated inter-rater reliability and criterion-related validity with children in Zambia , and internal consistency was demonstrated with its use among adolescent Somalian refugees (Cronbach α = 0.85) .
Stigma was measured using 10 questions from the Berger HIV Stigma Scale. Questions related to internal stigma or “negative self-image” (four questions) and external stigma or “public attitudes” (six questions) were analyzed in these two categories and as a total stigma score (scale range 10–40) [3, 10, 26].
ART adherence was measured by a validated three-question survey that asked participants to rate their adherence over the past 30 days, translated to a 0–100 scale . The mean of these three questions was the value used to measure self-reported adherence, with a higher number indicating better adherence. HIV RNA measurement was performed at the Kilimanjaro Clinical Research Institute Biotechnology Laboratory, which participates in international external quality assurance programs and uses the Abbott m2000 platform (Des Plaines, Illinois, USA). Virologic suppression was defined as plasma HIV RNA level < 400 copies/ml. HIV RNA lower than the minimal detectable amount of 40 copies/ml was imputed as half the minimum detectable amount (20 copies/ml).
We summarized continuous variables using means, standard deviations, medians, and quartiles, and categorical variables using counts and percentages. Data were analyzed as intention-to-treat according to the participant’s randomization assignment, regardless of whether the participant attended or participated in the intervention if assigned to the intervention arm. Data from all participants contributing data at each timepoint were utilized, except after the time of cross-over for participants randomized to SOC who were later exposed to the intervention (Fig. 1).
A constrained longitudinal linear mixed effects model with clustering in the intervention arm was used for continuous outcomes . This model takes into account clustering using a random intercept for the group in which participants in the SYV-arm received the group-based intervention, while the individual-level variability is allowed to vary between the SYV intervention and SOC arms . The longitudinal nature of the data was accounted for using non-independent structure on the residuals within person. Since many of the continuous outcomes are skewed, we used robust standard errors and checked residuals. All models were adjusted for wave and site.
The binary variable of virologic suppression was analyzed using the modified Poisson approach, namely a log-Poisson generalized estimating equations (GEE) model with robust standard errors . The log-link is used instead of the logit because of the well-known issue of misinterpretation of the odds ratio . The model took into account correlation within participants over time using unstructured covariance, and was also adjusted for wave and site to account for these design characteristics.
We examined pre-intervention predictors of missingness at each time point and as a sensitivity analysis adjusted for variables in the statistical model with a p-value of the difference between missing and non-missing of < 0.10 at any time point. As additional sensitivity analyses, the intervention effect was examined including those who crossed-over as remaining in the control arm. One participant in Wave 2 did not have baseline data collected, but did have study entry data six-months prior. These data are included in the summary table, but study entry data are not included in the regression models.
Due to the pilot nature of the study, all analyses are exploratory. Analyses were conducted using Stata version 16.1 (StataCorp, College Station, TX).