Factors associated with non-attendance at scheduled infant follow-up visits in an observational cohort of HIV-exposed infants in South Africa, 2012-2014

BACKGROUND: Since 2001 the South African guidelines to improve child health and prevent vertical HIV transmission recommended frequent infant follow-up with HIV testing at 18 months postpartum. We sought to understand non-attendance at scheduled follow-up study visits up to 18 months, and for the 18-month infant HIV test amongst a nationally representative sample of HIV exposed uninfected (HEU) infants from a high HIV-prevalence African setting. METHODS: Secondary analysis of data drawn from a nationally representative observational cohort study (conducted during October 2012 to September 2014) of HEU infants and their primary caregivers was undertaken. Participants were eligible (N = 2650) if they were 4-8 weeks old and HEU at enrolment. All enrolled infants were followed up every 3 months up to 18 months. Each follow-up visit was scheduled to coincide with each child’s routine health visit, where possible. The denominator at each time point comprised HEU infants who were alive and HIV-free at the previous visit. We assessed baseline maternal and early HIV care characteristics associated with the frequency of ‘Missed visits’ (MV-frequency), using a negative binomial regression model adjusting for the follow-up time in the study, and associated with missed visits at 18 months (18-month MV) using a logistic regression model. RESULTS: The proportion of eligible infants with MV was lowest at 3 months (32.7%) and 18 months (31.0%) and highest at 12 months (37.6%). HIV-positive mothers not on triple antiretroviral therapy (ART) by 6-weeks postpartum had a significantly increased occurrence rate of ‘MV-frequency’ (adjusted incidence rate ratio, 1.2 (95% confidence interval (CI), 1.1-1.4), p < 0.0001). Compared to those mothers with ART, these mothers also increased the risk of '18-month-MV' (adjusted odds ratio, 1.3 (CI, 1.1-1.6), p = 0.006). Unknown infant nevirapine-intake status increased the rate of 'MV-frequency' (p = 0.02). Mothers > 24 years had a significantly reduced rate of ‘MV-frequency’ (p ≤ 0.01) and risk of ’18-month-MV’ (p < 0.01) compared to younger women. Shorter travel time to health facility lowered the occurrence of 'MV-frequency' (p ≤ 0.004). CONCLUSION: Late initiation of maternal ART and infant prophylaxis under the Option- A policy and extended travel time to clinics (measured at 6 weeks postpartum), contributed to higher postnatal MV rates. Mothers older than 24 years had lower MV rates. Targeted interventions may be needed during the current PMTCT Option B+ (lifelong ART to pregnant and lactating women at HIV diagnosis) to circumvent these risk factors and reduce missed visits during HIV-care. Secondary analysis of data drawn from a nationally representative observational cohort study of HEU infants and their primary caregivers was conducted. The main study, conducted between October 2012 and September 2014, aimed to measure postnatal mother-to-child transmission of HIV (MTCT) between 6 weeks and 18 months postpartum and these primary results are being presented in a separate manuscript. The cohort was recruited from 6 weeks postpartum cross-sectional survey conducted in 2012 whose methods have been previously published [8]; briefly, the study was conducted at public primary health care clinics and community health centres nationwide offering immunisation services. A nationally-representative sample of facilities was selected through a multistage probability proportional to size sampling approach. HIV Enzyme-linked immunosorbent assay (EIA) was used to confirm infant HIV exposure at 6 weeks of age. Infants who were HIV-exposed but not HIV PCR positive (i.e., HEU infants) at 6 weeks old were eligible for postnatal follow-up. Ethics approval for the study was granted by the South African Medical Research Council Ethics Committee, and approval was also obtained from the Centers for Disease Control and Prevention. All caregivers signed informed consent. The 6 week and 18 month HIV tests in this study were routine tests. A final sample of 2650 eligible caregiver-infant pairs who provided signed informed consent to take part in the cohort study was used in this secondary analysis. The cohort was followed up from October 2012 until September 2014. Participants were given study inconvenience allowances at each visit of USD$2.50 in October 2012, increased to USD$6.00 in January 2014. PMTCT Option A policy (maternal AZT plus infant ARV prophylaxis to prevent MTCT and infant prophylaxis until 1 week after cessation of breastfeeding) was in use in South Africa during the start of the study to March 2013. Subsequently, PMTCT Option B (maternal triple ARV prophylaxis to prevent MTCT and infant prophylaxis for the first 4–6 weeks regardless of feeding status) was adopted in April 2013 and was in use during the remainder of the study period [9]. The recruited HEU infants and their caregivers were followed up at 3, 6, 9, 12, 15 and 18 months post-partum. Data collectors attempted to coincide these study follow-up visits with routine follow-up visits during the same time interval and drew blood for infant HIV testing at each of the visits. Socio-demographic background information were collected at 6 weeks postpartum (baseline). Since the primary outcome for the cohort study was HIV incidence and death in HIV-exposed infants, the length of follow-up for each infant was determined by one of the following three scenarios: (i) until study end-point at 18 months postpartum if they remained HIV negative; (ii) up-to-the time point of infant seroconversion where after they presented for one exit interview to assess access to care or lastly, (iii) up-to-the time of death if they died before the study end-point or before HIV seroconversion. This secondary analysis focusses on non-attendance of the six scheduled postnatal care study visits (at 3, 6, 9, 12, 15 and 18 months postpartum) and non-attendance of the 18-month postpartum visit. We studied non-attendance at scheduled visits, operationalised as ‘missed visit’, (abbreviated here as MV), for an infant who was eligible for that visit. Infants eligible for a visit were those enrolled at baseline who were alive and free of HIV at the previous visit. Repeated MV was observed for some participants; therefore, we also present simple proportions with 95% confidence intervals (CI) of MV at each follow-up study time point, i.e., per time-point MV. Two scenarios of MV were of primary interest: The first, (i) ‘MV frequency’ defined as the number of MV across a participant’s exposure time over the study period. Exposure time is simply the number of months linked to scheduled eligible study visit points up to the end of the study (18 months postpartum) or time of death or HIV-infection. In this case, MV was a count variable, counting the number of missed eligible visits from 0 (attended all of them) through to 6 (was eligible for all study visits and never attended any of them). Exposure time ranged from 1 (if death or HIV-infection occurred at the 3-month point) to 6 (if either HIV-infection or death occurred after 18 months or never occurred during the study period. The second, (ii) 18-month MV, was a binary variable of failure to attend or not among those still eligible to attend this visit. The 18-month MV was of special interest because this visit was scheduled to coincide with the routine 18-month postpartum rapid HIV testing of the enrolled infant and the exit point from the PMTCT programme. Purposively selected baseline socio-demographic characteristics, and health-care information was assumed to have some direct or indirect influence on the uptake of scheduled visits. The independent variables were evaluated for potential association with MV including relationship between the caregiver and the infant at the six-week enrolment, mother’s age, mother’s highest education level, marital status. We also evaluated whether infant was ever breastfed, disclosure of maternal HIV status to family or friends, ever facing discrimination due to HIV status, knowledge of MTCT modes, maternal ART use at 6 weeks postpartum, whether infant had been given nevirapine after birth, infant birth weight, infant hospitalization, means of transport used and time taken to access the facility, socio-economic status (SES) ranking and province. SES was extracted from a calculation using the full cross-sectional sample available at 6 weeks postpartum. In the full cross-sectional dataset, SES was calculated using principal component analyses from household characteristics (which included the type of housing, sanitation, water and fuel), household possessions (such as TV, stove, radio), any food shortage and source of income [10]. A Chi-squared test was used in the descriptive statistics of these independent variables at baseline and by ‘per time-point MV’. None of the chosen independent variables had more than 10% missing data. Of those with some missing data, if excluding missing data changed the 95% CI of observed estimates on known categories of the variable, then an additional category for ‘unknown’ responses was added to the variable in the statistical and descriptive analyses in order to minimise deviation from actual observed estimates and to keep the survey structure and sampling weights accurate on the rest of the data. Two sets of analyses were conducted, to assess factors associated with ‘MV frequency’, and ‘18-month MV’, separately. To assess the former, a regression model for a count outcome variable allowing varying exposure time periods was most suitable for the data. However, the assumption of a Poisson regression model for a count outcome (i.e., data are not overly dispersed), was not met. Considering that the outcome was expected to have zeros from participants who missed all their eligible visits (exposed), a zero-inflated binomial regression model was used to confirm whether the data were zero-inflated. The assumption of zero-inflation was rejected (z = 6.44, p < 0.0001). The final model used for identifying predictors of MV frequency was, therefore, a negative binomial regression, with survey set functions and exposure times specified [11, 12]. Bivariate analyses were first run between each independent factor and the primary outcome and a Wald test (corrected for survey structure and including the exposure time variable) performed to see if the coefficients were significantly non-zero using a p-value cut-off of < 0.25. Independent factors with a Wald test p < 0.25 were then included in a multivariable analysis model. Those factors which had overall non-significant Wald tests but with p < 0.05 for the coefficients of sub-categories were then included in the multivariable model and retained if they caused a large change (a shift of the 95% CI) on the coefficients of any variables which were already in the model. The final multivariable model included seven of the 15 independent variables; maternal age, caregiver status, the disclosure of maternal HIV status, infant ever being on nevirapine, the mother being on ART, common means of transport and province. Adjusted point estimates are presented here. To identify factors associated with ‘18-month MV’, a logistic regression model, with survey settings specified, was used. For ‘18-month MV’, ten factors (maternal age, education level, whether infant ever breastfed, disclosure of maternal HIV status, being discriminated against, knowledge of MTCT modes, mother being on ART at 6 weeks, infant having taken nevirapine by 6 weeks, SES and province) were eligible for inclusion in the final multivariable model. All analyses were performed in STATA SEv13. Sampling weights were applied to all reported proportions and statistical analyses to adjust for sample ascertainment in the original cohort and non-consent for postnatal follow-up amongst all those eligible for postnatal enrolment.