To assess our hypothesis, we conducted a case-cohort study nested within a cluster randomized controlled trial (“parent study”) evaluating the impact of short-term economic support on viral suppression across four regions (Mwanza, Shinyanga, Geita and Kagera) in Lake Zone, Tanzania (clinicaltrials.gov identifier: NCT04201353) [14]. In the parent study, participants were recruited at 32 HIV clinics (8 clinics per region), among adult PLHIV (aged 18 and older) who initiated ART within the last 30 days. Trial participants (n = 1990) were followed for the primary outcome of viral suppression at 12-months (endline). This nested case-cohort study is reported following the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines [16].

ACEs are sensitive and distressing topics to discuss– which can result in triggering memories for participants with ACEs. Therefore, collecting ACEs data requires additional training for surveyors to minimize distress, prevent triggers, respond sensitively, respect privacy, and make mental health referrals for participants if needed. To protect participants and reduce the time required from clinic staff conducting the surveys, we aimed to ask ACE-related survey questions to as few participants as possible. Thus, we selected a case-cohort design for the gains in statistical efficiency compared to a traditional cohort design and the benefit that the odds ratio calculated from this design directly estimates the risk ratio, without a rare disease assumption [17].

On September 24, 2022, we identified 317 trial participants who met two inclusion criteria: (1) enrolled in the parent study between August 14, 2021 and October 5, 2021 and would therefore have reached 11–13 months on ART, ensuring that participants were either approaching or had just reached trial endline and 12-months of follow-up; and (2) had yet to complete the parent trial’s 12-month survey, where ACE information would be collected. From the parent trial [14]we had access to outcome data for all trial participants; however, exposure data on ACEs was only measured among participants sampled into the case-cohort study– through adding an ACE module into the parent trial’s 12-month endline survey, conducted in Swahili by HIV clinic staff (in-person or over the phone). Clinic staff were given training on sensitivity and privacy when asking potentially triggering questions, and they offered all participants mental health referrals.

Cases were individuals who disengaged from HIV care at 12 months after initiating ART, defined as > 90 days elapsing since their last scheduled clinic appointment anytime during the window of 11–13 months after trial enrollment, based on medical record review. Though recent PEPFAR guidelines define interruption in HIV treatment as no clinical contact for 28 days after the last scheduled appointment, 28 days has been widely used as an indicator of “lateness”, while > 90 days has been widely used as a threshold to represent true disengagement [18]. While the threshold of 28 days is crucial for programmatic purposes of increasing tracing practices to help PLHIV return to care promptly to avoid long ART interruptions, we use the definition of > 90 days to be consistent with the widespread threshold representing a longer period of disengagement from HIV care– to identify PLHIV who are disengaged from care and may have higher risk of both accelerated disease progression and onward HIV transmission. Additionally, we did not include in our case definition participants who missed a scheduled clinic appointment by > 90 days but then returned to care prior to 12 months, as the outcome was disengagement from care without return: the group perceived to be most at-risk of unsuppressed viral load and transmission to others. Case status was determined through medical record data, including both biometric mHealth system data and manually abstracted medical file data.

Consistent with a case-cohort design, we randomly selected a sub-cohort of participants from the sampling frame to represent the control group, regardless of their outcome status. In the sampling frame of n = 317 there were 83 people who met the case definition; thus, we randomly selected a sub-cohort of n = 250 to enable a case: sub-cohort ratio of 1:3. Cases and sub-cohort members were not matched.

Following case-cohort study methodology, we included all cases. There were 83 cases: 63 of whom met the case definition in the randomly selected sub-cohort of 250, and the remaining 20 cases were sampled additionally. In total, the study included 270 participants. Researchers were masked to exposure status when identifying case status.

We contacted all sampled individuals for their endline survey through at least 3 phone calls, 3 text messages, and 3 home visits through home-based care providers, who trace individuals who have disengaged from HIV care according to Ministry of Health guidelines in Tanzania. Participants who were verified as having transferred to another clinic via official clinic records were excluded, because we could not access appointment data at their transfer clinics to verify current care status.

We adapted the Adverse Childhood Experiences International Questionnaire (ACE-IQ) [19] to assess ACE exposure, which is a validated scale created for collecting data on ACEs in LMIC. ACE-IQ has been used extensively globally, especially in Africa and even specifically in Tanzania [20]. Our survey assessed the first 10 categories within the ACE-IQ scale which were relevant to the Tanzanian context: (1) physical abuse; (2) emotional abuse; (3) sexual abuse; (4) emotional neglect; (5) household violence; (6) parental death or divorce; (7) household drug abuse; (8) household member incarceration; (9) household mental illness; (10) community violence. We used the binary ACE-IQ scale, which collects data on whether participants ever experienced ACEs, not frequency of occurrence. For all questions, participants are asked to report on events they experienced prior to the age of 18, which would have occurred before they initiated ART and enrolled into our parent study, thus ensuring the exposure preceded the outcome. A participant’s ACE score is the summation of ACE categories they experienced (range 0–10) and is treated as a continuous variable in all analyses.

We first described the prevalence of all individual ACEs in our sub-cohort. Next, we described the mean and median number of ACEs experienced among participants in our sample, stratified by cases and sub-cohort members. To assess the relationship between ACEs and disengagement from HIV care, we initially fit an unadjusted logistic regression model to estimate odds ratios; when applying logistic regression to the case cohort design, the estimated odds ratios represent the relative risk of disengagement from HIV care at 12 months associated with a one-unit increase in ACE score [17].

For our primary analysis, we adjusted the model for potential confounders of sex, age, and region, as well as arm of the parent trial, to ensure our results are not influenced by the intervention designed to reduce disengagement from ART. Based on our primary model, we also used the beta estimate from the exposure variable (ACE score) and standard error to estimate the adjusted risk of disengagement from HIV care associated with experiencing two-, three-, and four-unit increases in ACE score. We explored using a cubic term, linear spline, quadratic spline, and restricted quadratic spline to model our continuous exposure variable; however, when using Bayesian Information Criterian (BIC) and Akaike Information Criterion (AIC) to compare these models, the more parsimonious model using a simple linear term for ACE score was favored (smallest BIC and AIC). Consistent with a case-cohort approach, [17]all participants sampled in the sub-cohort who developed the outcome were included in our sample twice: once in the sub-cohort and once as a case. Therefore, we use robust standard errors in all models to account for individual clustering.

Next, we applied inverse probability weighting to account for survey non-response [21]. Stabilized weights were estimated from predicted probabilities of missingness, based on logistic regression results regressing survey non-response on variables collected in the parent trial’s baseline survey, including both standard covariates and outcomes, which may be correlated with ACE score such as mental health, stigma, and perceived social status. Weights were then incorporated into the primary model described above. Confidence intervals were estimated through 1,000 bootstrap iterations, where we maintained the original ratio of cases, sub-cohort members without the outcome, and sub-cohort members with the outcome across all resampling iterations– then extracted the 0.975 and 0.025 quantiles. We used R for all data cleaning and data analysis.

Finally, we conducted an additional exploratory analysis controlling for mental health– using Patient Health Questionnaire 2 (PHQ-2) as a measure of depression and Generalized Anxiety Disorder (GAD-2) as a measure of anxiety– to estimate the association between adverse childhood experiences and disengagement from HIV care extending beyond the association between mental health difficulties and disengagement from HIV care. PHQ and GAD are both commonly used to measure mental health and have been specifically used among PLHIV in Tanzania [22, 23]. The purpose of this analysis is: (1) explore whether the pathway between ACEs and disengagement from HIV care is mediated by current mental health challenges; (2) investigate whether screening for ACEs would provide additional benefit in identifying PLHIV at-risk of disengagement from HIV care, beyond solely screening for mental health (as is currently used in some HIV clinics). For this analysis, we first constructed models controlling for mental health reported 12-months after ART initiation, at the timepoint when disengagement from HIV care was measured; then, we constructed separate models to control for mental health reported at ART initiation.