Setting
The study was implemented in the Gauteng Region of South Africa, one of the six administrative regions of the South Africa Department of Correctional Services (DCS). Three management areas (Modderbee, Kgosi Mampuru II, and Baviaanspoort) were selected to provide a range in size of the inmate population, level of security, men and women, those awaiting trial, and sentenced offenders. The daily census across the participating study sites was approximately 14,000 inmates; 29% were awaiting trial. The inmate population in correctional facilities in South Africa is 98% men and 79% black African [16]. Information on the median sentence duration and estimated HIV prevalence in each study facility was not available. All participating facilities had on-site routine ART programs available to all inmates meeting South African national eligibility criteria at the time of the study enrollment (2014–2016). Prior to release, correctional health staff prepared a referral letter for the reentrant to take to a community-based primary health care clinic. In addition, a supply of ART was provided to last between 15 and 90 days [17].
Study design
Observational study of linkage to care following corrections release.
Participants
Inmates were eligible to participate if they were receiving HIV-related care at one of the study sites. Additional inclusion criteria were: ≥18 years old, capable of providing informed consent, known or anticipated release date within the study period, anticipated release residence within Gauteng Province, and willingness to provide telephonic or residential contact details to facilitate follow-up activities. In this report we only included participants receiving ART at the time of release. This decision was made because the on-ART population has the most relevance for current South African HIV care models of providing ART to all people with HIV. Study staff based at each of the participating sites worked with the correctional facility health staff and inmate peer educators to recruit and identify inmates who met the eligibility criteria. Those who met eligibility criteria and were interested in participating were provided written and verbal information in their language of choice and enrolled into the study. All participants completed written informed consent prior to enrollment.
This study was conducted according to the principles expressed in the Declaration of Helsinki. Human subject research approval was obtained from the University of the Witwatersrand Human Research Ethics Committee (M120492), the South African Department of Correctional Services Research Review Committee, the Johns Hopkins Medicine Institutional Review Board (NA_83335), and the CDC Institutional Review Board.
Procedures
Baseline characteristics, including age, sex, duration of incarceration, whether HIV was diagnosed in the correctional facility, and whether ART with initiated in the correctional facility were obtained through structured interviews conducted by research team members while the participant was still incarcerated. The research team members also recorded locator information, including cell phone number, alternative numbers, and location of anticipated residence upon release. Correctional facility health service routine medical records were abstracted onto structured report forms to record the date and result of any CD4 count testing performed closest to the date of release.
Follow-up data collection occurred through structured interviews conducted by research team members. These interviews were conducted either telephonically or in-person. Participants were contacted telephonically at 15 days post-release to verify locator information and starting at 30 days to ascertain post-release care status. These calls were repeated at 60 days and again at 90 days if contact was unsuccessful or linkage not reported during the prior call. The primary purpose of the interviews was to ascertain linkage to care status. Telephonic contact was attempted at least three times, at different times of the day and on different days (including weekends). When telephonic contact was unsuccessful, home visits were attempted in collaboration with DCS community corrections staff. At least three attempts were made to visit the reported place of residence.
Among participants reporting linkage to care, the clinic at which a participant reported receiving care was visited; if no records were located, the participant was contacted again to ask whether they entered care and at which clinic, and to inquire regarding any different names that may have been used to register for care. Records from routine clinical care in public clinics were abstracted to verify self-reported linkage. Local clinics used a mix of electronic care registries, electronic laboratory results, and paper-based charts for patient care and statistical reporting. All three sources were used for verification of self-reported linkage to care. Paper and electronic clinical records were matched to a participant based on name, date of birth, and a range of service dates.
Measures
Outcomes
The primary outcome was the proportion of participants with self-reported linkage to care within 90 days of release. Secondary outcomes were the proportion of participants with verified linkage to care within 90 days of release and the proportion of participants without a lapse in ART possession following release.
Self-reported linkage to care: This was defined as a participant having reported, during follow-up contact, attending a clinic for HIV care subsequent to release. Participants who were unable to be contacted were considered not to have linked to care.
Verified linkage to care: This was defined as a confirmation of self-reported linkage to care. Linkage was confirmed using abstracted clinic-based paper or electronic records.
Lapse in ART possession: This was based on the self-reported date of first linkage to care and the self-reported days of ART provided at corrections release. A lapse was defined to have occurred if the time to first clinic visit from corrections release was greater than the reported medication supply received at corrections release. If there was no self-reported linkage to care a participant was also classified as having a lapse in ART possession.
Independent variables
We assessed independent variables for association with the primary outcome of self-reported linkage to care and the secondary outcome of verified linkage to care. We specifically included the following variables in bivariate modeling: sex, age group (< 35, ≥35 years old), CD4 count, duration of incarceration, location of HIV diagnosis, and location of ART initiation. The age group was divided at age 35 years based on a prior study from South Africa reporting generally lower linkage to care from community HIV testing among individuals < 35 years old [6]. Sex and age were selected as independent variables based on the evidence that each is associated with linkage to care. Incarceration duration and location of HIV diagnosis and ART initiation were included to assess for associations with corrections specific factors that may be associated with linkage to care. We were particularly interested in assessing whether place of ART initiation (while incarcerated or prior to incarceration) affected post-release linkage to care.
Statistical analysis
We excluded from the primary outcome analysis participants who were known to have been re-incarcerated (consistent with prior analyses [15]), known to have died, or withdrew consent. Descriptive statistics were used to characterize the population.
Log-binomial regression was used to assess for bivariable and multivariable associations between self-reported linkage to care and verified linkage to care and independent variables. Log-binomial regression was selected to estimate relative risks for a common outcome. All regression analyses included adjustment for facility effect. We assessed the following independent variables: sex, age group (< 35, ≥35 years old), CD4 count closest to time of release, duration of incarceration, location of HIV diagnosis, and location of ART initiation. Bivariable analyses were initially performed. Independent variables with a p ≤ 0.1 in bivariable analyses were included in multivariable model building. Age group and correctional facility were included in all multivariable models regardless of p value. Sex was not included, a priori, in adjusted models due to the relatively small proportion of women in the study. Variables with p > 0.1 in multivariable modeling were eliminated in a stepwise manner starting with the variable with the highest p value. This process was continued until all selected variables had a p < 0.1. Through this process we developed final models for self-reported linkage to care and for verified linkage to care. Stata 14 was used for all analyses (STATA Corp. College Park, Texas, USA).