Neurodevelopment of HIV-exposed uninfected children in South Africa: outcomes from an observational birth cohort study

Background: HIV infection is known to cause developmental delay, but the effects of HIV exposure without infection during pregnancy on child development are unclear. We compared the neurodevelopmental outcomes of HIV-exposed uninfected and HIV-unexposed children during their first 2 years of life. Methods: Pregnant women (>18 years of age) at 20–28 weeks’ gestation were enrolled into the Drakenstein Child Health cohort study while attending routine antenatal appointments at one of two peri-urban community-based clinics in Paarl, South Africa. Livebirths born to enrolled women during follow-up were included in the birth cohort. Mothers and infants received antenatal and postnatal HIV testing and antiretroviral therapy per local guidelines. Developmental assessments on the Bayley Scales of Infant and Toddler Development, third edition (BSID-III), were done in a subgroup of infants at 6 months of age, and in the full cohort at 24 months of age, with assessors masked to HIV exposure status. Mean raw scores and the proportions of children categorised as having a delay (scores <–2 SDs from the reference mean) on BSID-III were compared between HIV-exposed uninfected and HIV-unexposed children. Findings: 1225 women were enrolled between March 5, 2012, and March 31, 2015. Of 1143 livebirths, 1065 (93%) children were in follow-up at 6 months and 1000 (87%) at 24 months. Two children were diagnosed with HIV infection between birth and 24-month follow-up and were excluded from the analysis. BSID-III assessments were done in 260 (24%) randomly selected children (61 HIV-exposed uninfected, 199 HIV-unexposed) at 6 months and in 732 (73%) children (168 HIV-exposed uninfected, 564 HIV-unexposed) at 24 months. All HIV-exposed uninfected children were exposed to antiretrovirals (88% to maternal triple antiretroviral therapy). BSID-III outcomes did not significantly differ between HIV-exposed uninfected and HIV-unexposed children at 6 months. At 24 months, HIV-exposed uninfected children scored lower than HIV-unexposed for receptive language (adjusted mean difference −1·03 [95% CI −1·69 to −0·37]) and expressive language (−1·17 [–2·09 to −0·24]), whereas adjusted differences in cognitive (−0·45 [–1·32 to 0·43]), fine motor (0·09 [–0·49 to 0·66]), and gross motor (−0·41 [–1·09 to 0·27]) domain scores between groups were not significant. Correspondingly, the proportions of HIV-exposed uninfected children with developmental delay were higher than those of HIV-unexposed children for receptive language (adjusted odds ratio 1·96 [95% CI 1·09 to 3·52]) and expressive language (2·14 [1·11 to 4·15]). Interpretation: Uninfected children exposed to maternal HIV infection and antiretroviral therapy have increased odds of receptive and expressive language delays at 2 years of age. Further long-term work is needed to understand developmental outcomes of HIV-exposed uninfected children, especially in regions such as sub-Saharan Africa that have a high prevalence of HIV exposure among children. Funding: Bill & Melinda Gates Foundation, SA Medical Research Council, Wellcome Trust. The DCHS is a population-based birth cohort study based in Paarl (a peri-urban area of the Western Cape, South Africa), investigating the early-life determinants of child health and development.11, 12 The antenatal HIV prevalence in this study population is 21%. Pregnant women who were older than 18 years of age, receiving antenatal care at a participating site, and intending to reside in the area for at least a year were enrolled into the study at 20–28 weeks' gestation while attending routine antenatal appointments.11, 12 Participants were recruited from two community-based clinics: T C Newman clinic (serving a mixed-ancestry community who speak Afrikaans) and Mbekweni clinic (serving a predominantly black African community who speak isiXhosa), both of which provide free maternal and child care. Written informed consent was obtained annually for mother–child pairs. A subgroup of randomly selected children underwent developmental assessment at 6 months, and all available children were assessed at age 24 months. This study was approved by the human research ethics committee of the Faculty of Health Sciences, University of Cape Town (approval numbers 401/2009 and 044/2017), and by the London School of Hygiene & Tropical Medicine observational and interventions research ethics committee (approval number 11903). Maternal HIV status during pregnancy was confirmed by routine HIV testing at booking, with retesting of HIV-negative mothers every 12 weeks, in accordance with Western Cape prevention of mother-to-child transmission of HIV guidelines. Additionally, maternal interviews and HIV status reviews of mothers and children were done by study staff at the child's birth, age 6 weeks, and every 6 months thereafter. All HIV-infected mothers were initiated on antiretroviral drugs according to prevention of mother-to-child transmission guidelines at the time: three-drug ART (the first-line ART regimen consisted of two nucleoside reverse transcriptase inhibitors plus a non-nucleoside reverse transcriptase inhibitor, commonly tenofovir plus emtricitabine plus efavirenz) or zidovudine from 14 weeks' gestation, depending on maternal clinical and immunological status (before May, 2013); or ART for life for all pregnant women (from May, 2013). All HIV-exposed uninfected children received prophylaxis (nevirapine alone or combined with zidovudine) from birth. HIV and ART data were collected by triangulating clinic and hospital folder information and maternal self-report interviews. Maternal CD4 cell count and viral load data were obtained from the online National Health Laboratory Service system. Where there were multiple results, the highest viral load during pregnancy was taken, and the lowest CD4 cell count within 1 year before childbirth and 3 months post-birth was used to maximise numbers. All HIV-exposed children received HIV testing as per local guidelines. HIV detection was done by PCR at age 6 weeks, and by rapid antibody, PCR, or ELISA at age 9 months and 18 months. HIV-exposed uninfected children were confirmed to have a negative HIV test result at age 18 months or a negative test after cessation of breastfeeding if this occurred at more than 18 months of age. HIV-unexposed children were defined as children born to mothers without HIV infection. Sociodemographic data were collected between weeks 28 and 32 of gestation by trained study staff using structured interviews and questionnaires adapted from the South African Stress and Health study.11, 12 Detailed birth data were obtained at delivery. Gestational age was calculated using the best estimated delivery date based on antenatal ultrasound, the last menstrual period, or the symphysis-fundal height. Prematurity was defined as birth at less than 37 weeks' gestation. Infant feeding method and exclusive breastfeeding duration were documented by maternal report at age 6–14 weeks, 6 months, and 9 months. Maternal psychosocial data were collected antenatally between weeks 28 and 32 of gestation.12 Maternal alcohol use during pregnancy was assessed using the Alcohol, Smoking, and Substance Involvement Screening Test.12 Material tobacco exposure during pregnancy was assessed by use of the IMMULITE 1000 nicotine metabolite kit (Siemens Medical Solutions Diagnostics, Glyn Rhonwy, UK) to measure antenatal urine cotinine concentration. Maternal depression was assessed with the Edinburgh Postnatal Depression Scale (with a score of ≥13 considered to indicate depression).12 Neurodevelopment was measured with the Bayley Scales of Infant and Toddler Development, third edition (BSID-III),13, 14 which is widely used internationally and has been validated in South Africa with reported values similar to those of the BSID-III US-based reference population.15, 16 The objectively measured assessment consists of five subscales: cognition, receptive language, expressive language, fine motor, and gross motor. Trained assessors masked to HIV exposure status alternated testing between clinics, assessing equal numbers of children at each site and offering language prompts in the child's preferred language. Any child with significant developmental delay was referred to the relevant health-care service. Inter-assessor reliability was assured through training and supervision. Assessors were monitored by a paediatric neurodevelopmental specialist throughout, who periodically observed assessments to ensure standardised data collection across sites, accuracy, and continued agreement between assessors. A second level of external quality control was done centrally before data capture. Scores were entered into a specialised BSID-III software programme. 260 children were randomly selected from the original cohort as a sampling frame for developmental assessment at 6 months of age. If a mother–child pair was unavailable, another child aged 6 months was invited to attend from those with study follow-up visits that week, based on child age at the time (a convenience sample). All available children were assessed at age 24 months. Mothers who were unavailable at initial contact were tried again up to three times, and children who were unwell on the day of assessment were rebooked where possible. Raw scores, scaled scores, and developmental delay are reported. Raw scores represent the sum of individual items the child passes on each subscale. Raw scores were converted to age-adjusted scaled scores standardised using normative data derived from a US reference population, with a range of 1–19 and mean of 10 (SD 3), with correction for prematurity at 6 months.13 A BSID-III score of less than −2 SDs from the BSID-III reference mean was used to define a clinically significant delay in any domain.7, 13 Maternal and child sociodemographic characteristics were expressed as mean (SD) for continuous data or absolute frequencies (%) for categorical data, and were compared between HIV-exposed uninfected and HIV-unexposed children with BSID-III results at 6 months and 24 months using descriptive statistics (t tests or χ2 tests). The associations between HIV exposure and each BSID-III subscale (cognitive, receptive language, expressive language, fine motor, and gross motor) at 6 months and 24 months of age were compared with use of linear regression models for raw scores and logistic regression models for developmental delay. Standardised effect sizes were reported as Cohen's d values. A directed acyclic graph (DAG) was constructed using DAGitty, according to previously published literature, to delineate assumptions regarding the causal pathway between HIV exposure and neurodevelopment. Multivariable models were then created using all potential confounders of the exposure–outcome relationship, as determined a priori by the DAG (household income, maternal education, maternal age, and sex and age of child). Residuals were checked for normality through quantile-quantile plots to confirm the linearity assumption for the models. Adjusted mean differences, adjusted odds ratios (ORs), and 95% CIs are presented. Additional analyses were done by adjusting the models for key potential mediating variables identified from the DAG (prematurity, maternal depression, and breastfeeding [exclusive breastfeeding duration and exclusive breastfeeding to 6 months]) to assess any changes to specific domain associations. Sensitivity analyses were done to examine the effect of site and home language in a restricted subanalysis of Mbekweni clinic, where the majority of HIV-exposed uninfected children reside. A separate analysis limited to HIV-exposed uninfected children whose mothers were on first-line ART was also done. Finally, the association between maternal CD4 cell count (dichotomised into ≤500 and >500 cells per μL, as per previous guidelines,17 with similar group sizes) and developmental outcomes was explored in BSID-III language subscales to provide direction for future work. Statistical analyses were done with STATA software (version 14.0). p values of less than 0·05 (two-tailed) were considered to indicate statistical significance. The funders had no role in the study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.