An approach for evaluating early and long term mother-to-child transmission of HIV (MTCT) in low and middle income countries: a South African experience

BACKGROUND: Eliminating mother-to-child transmission of HIV is a global public health target. Robust, feasible methodologies to measure population level impact of programmes to prevent mother-to-child transmission of HIV (PMTCT) are needed in high HIV prevalence settings. We present a summary of the protocol of the South African PMTCT Evaluation (SAPMTCTE) with its revision over three repeated rounds of the survey, 2010-2014. METHODS: Three cross sectional surveys (2010, 2011-2012 and 2012-2013) were conducted in 580 primary health care immunisation service points randomly selected after stratified multistage probability proportional to size sampling. All infants aged 4-8 weeks receiving their six-week immunisation at a sampled facility on the day of the visit were eligible to participate. Trained research nurses conducted interviews and took infant dried blood spot (iDBS) samples for HIV enzyme immunoassay (EIA) and total nucleic acid polymerase chain reaction (PCR) testing. Interviews were conducted using mobile phones and iDBS were sent to the National Health Laboratory for testing. All findings were adjusted for study design, non-response, and weighted for number of South African live-birth in each study round. In 2012 a national closed cohort of these 4 to 8-week old infants testing EIA positive (HIV Exposed Infants) from the 2012-2013 cross-sectional survey was established to estimate longer-term PMTCT impact to 18 months. Follow-up analyses were to estimate weighted cumulative MTCT until 18 months, postnatal MTCT from 6 weeks until 18 months and a combined outcome of MTCT-or-death, using a competing risks model, with death as a competing risk. HIV-free survival was defined as a child surviving and HIV-negative up to 18 months or last visit seen. A weighted cumulative incidence analysis was conducted, adjusting for survey design effects. DISCUSSION: In the absence of robust high-quality routine medical recording systems, in the context of a generalised HIV epidemic, national surveys can be used to monitor PMTCT effectiveness; however, monitoring long-term outcomes nationally is difficult due to poor retention in care. Cross-sectional national probability-based primary health care facility-based surveys, using an HIV biomedical marker (HIV antibody testing) to determine infant HIV exposure, and a virological test to measure perinatal HIV transmission amongst biomarker positive (HIV exposed) infants (HEI) were conducted at 4–8 weeks post-delivery from June–December 2010, August 2011–March 2012 and October 2012–May 2013 [15, 16]. The sampling frame for the primary sampling units (PSU) were public (government-funded) Primary Health Care (PHC) clinics and Community Health Centres (CHC) reportedly administering six-week immunisations (DPT1 or Pentaxim1 or the hexavalent vaccine), as documented in the South African National District Health Information System (DHIS). While other public facilities do administer immunizations (hospitals, mobile clinics), these PHC clinics and CHC are the primary locations for routine primary child health services. Data from the 2007 DHIS was used to obtain this sampling frame of eligible facilities and included the number of DTP1 administrations done in each facility for that year. The sampling design was stratified by the nine provinces. Eligible facilities were stratified on the number of annual immunizations reported with three strata based on reported annual DTP1 administration: < 130, 130–300 and > 300 immunizations per annum. The stratum of the small facilities (< 130 immunizations per annum) consisted of 951 facilities performing 67,514 immunizations per annum which represented 7.0% of the total immunizations done in 2007 in South Africa. For fieldwork cost-efficiency, a strategic decision was taken to exclude this stratum from the formal sample. The National Department of Health expressed an interest in the efficiency of the PMTCT in facilities with high prevalence and high birth rates. To accommodate this request, facilities with > 300 immunizations per annum in the sampling frame were also stratified in two strata (≥29% or < 29%) based on the 2008 HIV national antenatal prevalence of 29%, using the district HIV prevalence in which they were based. A two-stage stratified sampling was used. In the first stage, facilities (primary sampling units - PSUs) were randomly sampled proportional to size (PPS) within each of the 27 (9 provinces X 3 facility size/HIV prevalence) stratum (see Fig. 1). The method operated under the with-replacement-type selection [17]. At the second stage, a fixed number of infants per facility was systematically sampled. The fixed number was the median number of infants expected within the sampling window (3 weeks) across the population of facilities within the stratum as determined from the detailed information of the sampling frame above. The fixed number of infants sampled in each facility within a stratum ensured a self-weighting sample. A sampling window of 3 weeks was used to realize the required sample. Five hundred eighty facilities were sampled (approximately 21–22 per strata). More detail on sampling methods have been published by Goga, et al. 2014 & 2016 and Dinh, et al. 2015 [15, 16, 18]. Sampling Frame From the 2012–13 survey, HEU infants were invited to participate in a closed, prospective observational 18-month follow-up study. Establishing and following-up the cohort entailed recruitment into the follow-up study at the time of the six-week interview, documenting detailed participant contact information (including address) in a protected web-based data base that only the allocated data collector had access to, and allowing home-based data collection and an inconvenience allowance of approximately 10 USD. HEU were followed up at 3, 6, 9, 12, 15 and 18 months to determine maternal and child access to care and MTCT. HIV infected children were followed up once after HIV positive diagnosis to determine access to HIV treatment. If the infant had not yet accessed care, appropriate referrals were made. The cross-sectional survey was conducted among all mother/legal caregiver-infant pairs (97% were biological mother) who presented at their local primary health care facility for their infant’s six-week immunisation (1st DTP dose) visit, regardless of their PMTCT status or HIV exposure. South Africa reports > 95% coverage of 6 week immunisation (1stDTP dose) [19], making these clinics an ideal catchment point for young infants. Infants accessing selected public primary health care clinics or community health centres were eligible for the cross-sectional surveys if they were: a) receiving their six-week immunization on the day of data collection, and b) 4–8 weeks old, c) did not need emergency medical care, and d) their mother/legal caregiver consented to participate in the survey. Infants were enrolled into follow-up if: a) their mother reported as being HIV positive and/or their HIV antibody test was positive, and b) they consented to follow-up. Infants remained eligible for on-going follow-up until 18 months if their PCR tests remained negative (i.e. HIV exposed uninfected infants). As noted above, if the infant PCR was positive at 6 weeks or became positive at any follow-up visit they received one additional follow-up visit to assess access to paediatric treatment, with appropriate referrals if infants had not accessed HIV care. To determine the sample size for each province, HIV prevalence was calculated based on the provincial antenatal survey prevalence and coverage of PMTCT ARV prophylaxis. Estimates of transmission rates for Sd-NVP and no treatment were taken from Rollins [12] while the transmission rate for dual therapy came from a provincial survey from KwaZulu Natal province (unpublished data, 2009). To balance samples across the nine provinces, absolute precision specified varied from 1 to 2%, with relative precisions of 22 to 60%. In general, provinces with a higher prevalence will have a lower (better) relative precision. These specifications resulted in better equity in sample size between provinces. Using this approach, the largest sample in a province was 1800 (Gauteng) and the smallest was 700 (Northern Cape) with a total sample size of 12,200 across all provinces (Table 1). Sample size calculation for the cross-sectional surveys Table 2 illustrates the sample size calculation for valid national estimates of 18 month MTCT and HIV-free survival. The red box in Table ​Table22 highlights the target sample size selected for 18-month MTCT, and the green box highlights the target sample size selected for HIV-free survival estimates. These calculations demonstrated that assuming a design effect of 2, 18-month MTCT of 5% (3.5–6.5%) and 10% (7–13%) HIV infection or death, 1620 infants were needed to estimate MTCT and 768 infants are needed to estimate HIV infection or death with the indicated precision. Assuming 30% loss to follow-up, 2314 infants needed to be enrolled into the follow-up study to estimate both MTCT and HIV-free survival. Sample size calculation for follow-up component Footnote: The bolded boxes highlight the sample sizes selected to estimate HIV infection or death (green box) or 18-month MTCT only (red box), with 30% precision, a design effect of 2 and assuming 20 or 30% loss to follow-up. These show that assuming a design effect of 2, 18-month MTCT of 5% (3.5–6.5%) and 10% (7–13%) MTCT or death, 1620 infants are needed to estimate MTCT and 768 infants are needed to estimate HIV or death. Assuming 30% loss to follow-up, 2314 infants need to be enrolled to estimate MTCT and HIV-free survival. Sample size calculations were made using standard nQuery Version 4.0 Trained nurse data collectors recruited mothers/caregivers from the PHC/CHC waiting room during immunisation days. Data collectors introduced themselves and the study verbally and in written form using a standardised information sheet. A screening questionnaire was administered to determine eligibility and then full informed consent was completed for eligible infants (see below). If an eligible mother/legal caregiver-infant pair agreed to be interviewed, the interview was conducted in a private location. Data were gathered using a questionnaire adapted from several validated tools [11, 12, 20]. The questionnaire included information on maternal age, parity, socio-economic status, antenatal care, HIV testing, maternal HIV status, PMTCT care during pregnancy and delivery, infant feeding counseling, birth information, infant feeding practices, infant weight; immunisations, postnatal visits and illness since birth. Legal (non-maternal) caregivers were administered a shorter version of the questionnaire that excluded antenatal care and PMTCT programme information. Trained research nurses also collected infant heel prick dried blood spot samples during infant immunization visits, from all consented infants, regardless of reported maternal HIV status or ARV exposure. The blood testing replaced the routine testing for HIV PCR as part of the PMTCT programme during the period of the study as it was felt to be inappropriate to bleed a child twice during the same clinic visit. Trained research nurses worked from immunization clinics to facilitate standardization of procedure and data quality control. They offered routine HIV testing to all infants attending the clinic for immunization, thus preventing the testing of a potentially biased sample of infants: routine PMTCT EID clinics only test known HIV-exposed infants. Additionally, we found that maternal request for early infant diagnosis of HIV infection (EID) during the routine immunization services were low (47%) [21]. The mothers/caregivers were informed that the infant testing would also act as a biomedical marker for HIV status of the mother and that she may need to have further follow-up including HIV counselling and testing (HCT). For the cohort study, the same trained research nurses used contact details obtained at cohort enrollment to arrange to meet the mother/legal caregiver-infant pairs during routine follow-up for all HEU at the health facility at 3, 6, 9, 12, 15 and 18 months. Interviews were completed and iDBS were taken for HIV PCR at 3, 6, 9, 12 and 15 months as the national guidelines did not recommend routine blood testing during these visits. During the 18 month visit the routine in-clinic HIV rapid test was used to document 18-month infant HIV status. The iDBS were sent to a centralized accredited laboratory (National Institute for Communicable Diseases, a division of the National Health Laboratory services) for HIV enzyme immunoassay (EIA) testing, followed by HIV total nucleic acid polymerase chain reaction (TNA PCR) testing on EIA positive samples. All EIA positive samples were serially tested for HIV infection with a second HIV EIA test, and 10% of negative samples were re-tested using a second test. Discordant first and second EIA test results underwent additional testing using Western Blot. Details about laboratory procedures have been published previously [15, 16]. Questionnaire data were entered in a mobile phone platform and transferred electronically to a central server for data management and data analysis. This is described in detail in another paper in this supplement [Ref pending]. The Ethics Committee of the South African Medical Research Council (SAMRC), provincial Ethics Committees, and the Centers for Disease Control and Prevention, Atlanta, approved the study protocol. All eligible mothers/primary caregivers (14+ years of age) were taken through a written informed consent process. Pregnant adolescents age 14–17 are considered emancipated minors in South Africa and were eligible for study participation. The process was conducted in the language of the participant and information sheets were translated into all South African official 11 languages. Consent forms were completed in duplicate, and one copy was given to the participant and confidentiality discussed and the other sent to the SAMRC offices. At the end of the interview, if the mother reported being HIV-positive and hence eligible for the follow-up study, a second consent process was conducted, followed by the completion of a second consent form. A participant contact information form was completed for all participants who agreed to follow-up and this was stored in a password protected database and could only be accessed by the data collector allocated to that participant. All mothers and infants were referred into care (routine maternal HIV testing or CD4 cell count testing or ART) as appropriate using referral cards which stipulated the exact, most convenient and appropriate routine health care setting that could to be accessed for further care. These sites were determined following a situational assessment conducted prior to the surveys. Determining appropriate referral sites for each study clinic and establishing referral protocols for HIV-positive mothers and infants is an important ethical consideration for any PMTCT evaluation. If this information is not already established, then a situational assessment to map referral sites may be needed prior to PMTCT study implementation [21]. Sample weights were calculated for the cross-sectional surveys to adjust for differential sampling design across provinces and the sample realization. To achieve this, the data from provinces were weighted by using the proportional distribution of live births per province recorded in 2008. The realisation weights were done at various levels (district, provincial) depending on the sample size realisation within strata. All findings were adjusted for study design, non-response, and weighted for number of South African live-birth in each study round. For the cohort study, longitudinal data were weighted for ‘no consent’ and ‘loss to follow-up’ amongst HIV exposed uninfected infants eligible for follow-up. Follow-up analyses were to estimate weighted cumulative MTCT until 18 months, postnatal MTCT from 6 weeks until 18 months and a combined outcome of MTCT-or-death, using a competing risks model, with death as a competing risk. HIV-free survival was defined as a child surviving and HIV-negative up to 18 months or last visit seen. A weighted cumulative incidence analysis was conducted, adjusting for survey design effects. Specific details of each analysis will vary depending on analysis objective, but will generally include both descriptive (rates, means, 95% confidence intervals) and analytic analyses (chi-square, ANOVA or logistic regression, with 95% confidence intervals around adjusted estimates).