Longitudinal adherence to maternal antiretroviral therapy and infant Nevirapine prophylaxis from 6 weeks to 18 months postpartum amongst a cohort of mothers and infants in South Africa

This study was a secondary analysis of data collected to evaluate the effectiveness of the national programme to prevent HIV transmission from mother to child in South Africa. The evaluation was a nationally-representative cross-sectional survey conducted in 2012/13 (October 2012–May 2013), with infant follow-up until September 2014. Public sector health facilities were sampled using multi-stage, probability proportional to size methodology using three strata per province based on 6 week client immunization load and antenatal HIV prevalence. The study was powered to produce nationally-representative results of MTCT at 6 weeks. More detailed information about the survey is available in previous publications [11, 13].

Mother-infant pairs were enrolled into the study during the infant’s first postpartum immunization visit at 6 weeks as national coverage for this first postpartum immunization visit is known to be over 60% in surveyed areas of South Africa [25]. If the infant was deemed HIV exposed but uninfected at the 6 week cross-sectional visit, mother-infant pairs [N = 2811] were eligible and consented for a further cohort survey involving three-monthly follow-up, from 3 until 18 months. The primary outcomes of this 3-to-18 month national cohort will be presented in a separate publication.

Trained nurse fieldworkers interviewed mothers or non-mother caregivers at the 6 week immunization visit and follow-up visits conducted at 14 weeks and six, nine, twelve, fifteen, and 18 months postpartum about socio-demographic information, infant health and feeding practices, maternal understanding of MTCT, HIV testing, postnatal care, and maternal and infant PMTCT prophylaxis and treatment. Approximately 375 μl of blood per infant was drawn and dropped onto five pre-printed circles on Munktell TFN filter paper to determine infants HIV exposure and infection. Visual aids such as pictures of ARV pills and ARV syrup bottles were used to assist mothers in identifying the type of regimen received. A feeding and medication diary was issued to mothers so they could document medication received, and to assist with recall of missed daily doses during interviews. Non-mother caregivers were not asked questions about maternal health and/or HIV testing and treatment.

Nonadherence to ART among mothers and NVP among infants was defined as reporting taking less than 95% of daily doses within each three month follow-up interval. For mothers and infants, this was operationalized in two scenarios where some mothers/infants were deemed non-adherent because they had failed to take ART/NVP for more than 5% of the daily doses in that interval, while others were non-adherent because they had ceased taking ART/NVP within the interval without professional advice. We assumed that medical advice to stop NVP was provided in compliance with national guidelines (therefore indicating adherence). We assumed other reasons for stopping NVP were not in compliance guidelines (therefore indicating nonadherence).

For mothers or infants who reported missing more than 5% of daily doses, failure date was calculated as the midpoint of the interview interval. For infants who stopped taking NVP during an interval, their age in weeks at NVP cessation was used to calculate date of failure.

Mothers were excluded from the adherence analysis at a particular time point if they had missed an interview during the interval of interest or a non-mother caregiver had responded to the interview, thus information about the mother’s adherence was not provided. Infants were no longer included in the analysis if breastfeeding had been stopped in the prior interval, thus removing the threat of transmission and the need for NVP per PMTCT guidelines.

We performed frequency analyses to describe characteristics of the mothers and infants and their care. To describe probability of ART or NVP adherence from 6 weeks to 18 months postpartum we plotted extended Cox model survival curves using the stcox and stcurve commands in Stata [26, 27] after preparing the data for Andersen-Gill time-to-event analysis for recurring events [28, 29]. The Andersen-Gill model is a counting process allowing assessment of multiple nonadherence events across follow-up intervals instead of censoring on the first instance of nonadherence [28, 30, 31].

We fitted multivariable Andersen-Gill extended Cox models to determine risk factors for ART nonadherence among mothers and infants separately. Andersen-Gill models generalize the Cox proportional hazards model, accommodating the non-proportionality of the effect of covariates in the case of recurrent failure events [31]. The resulting coefficients are a measure of all-cause effects across multiple recurrent events as demonstrated in other studies describing risk factors associated with recurring health events [32‐35].

Covariates of interest from the information collected at 6 weeks were identified a priori. We used the forward stepwise function to fit the models from these covariates, setting the significance level for addition to the model at 15% and removal from the model at 20% [30]. The proportional hazards assumption was upheld for both models which we assessed by running a model that included time-varying covariates. All analyses were performed in STATA 13.1. Because this is a subgroup analysis of the cohort of mothers followed up by the study, the results are not nationally representative. Therefore, results are not adjusted for the study design, non-response and not weighted for live-births.

The protocol was reviewed according to the Centers for Disease Control and Prevention (CDC) human research protection procedures and was determined to be research, but CDC was not engaged. The protocol was also reviewed by the institutional review board of the Medical Research Council of South Africa. Mothers and caregivers provided written informed consent prior to the onset of the interview and assent for infants to undergo blood collection.

The majority of the mothers in the study population were between 25 and 34 years old (ART analysis: 59.2%, NVP analysis: 56.3%). Three quarters of the mothers in both samples were educated up to grade 8–12. Most mothers were single (ART analysis: 70.8%, NVP analysis: 73.9%) (Table 1).

At the six week interview, most mothers reported a main source of income other than their own employment such as a child support grant, disability grant, a partner or husband’s support, or another family member’s support (ART analysis: 77.9%, NVP analysis: 80.6%). Over a third of infants (37.3%) were breastfed between birth and 6 weeks. Less than half of the mothers had planned pregnancies (ART analysis: 40.3%, NVP analysis: 37.1%).

Cumulative probability of adherence to ART until 18 months postpartum was 63.4% among mothers included in the ART analysis (95% confidence interval (CI): 60.7–66.0) and 74.5% among infants included in the NVP analysis (95% CI: 70.2–81.9) (Fig. 1, Table 2). From 6 weeks to 14 weeks postpartum, adherence to NVP was 96.4% among infants (95% CI: 94.3–97.8). From 6 weeks to 14 weeks postpartum, ART adherence was lower amongst mothers at 85.0% (95% CI: 82.8–87.0). Cumulative infant NVP adherence was consistently higher than mother ART adherence over the 6 weeks to 18 months postpartum period.

Among both mothers and infants, a higher proportion of nonadherence events were due to stopping treatment not according to national guidelines (range: ART analysis: 73.9–88.4%, NVP analysis: 58.3–100%) rather than missing at least 5% of daily doses (range: ART analysis: 15.2–26.2%, NVP analysis: 0–41.7%) (Table 2).

Maternal age influenced ART adherence from 6 weeks to 18 months postpartum with adolescents and young women (aged 16–24 years) having a higher hazard of nonadherence events during that period than women over 34 (age 16–24, adjusted Hazard Ratio (aHR): 1.9, 95% CI 1.4–2.5).

Failing to disclose one’s HIV positive status to anyone at baseline was a risk factor for nonadherence (aHR: 1.7, 95% CI: 1.3–2.1). Not knowing one’s CD4 result also increased risk of nonadherence among mothers. Compared to mothers with a known CD4 result, those who had a CD4 test done but results were not received (aHR: 1.3, 95% CI: 1.1–1.6) or it was not known if CD4 test was done (aHR: 2.1, 95% CI: 1.4–3.2) were at higher risk for nonadherence.

Mothers who initiated ART after delivery were at higher risk for nonadherence (aHR: 1.6, 95% CI: 1.3–2.0) than those who initiated ART before delivery. Mothers who delivered their infants outside of health facilities had almost double the hazard of nonadherence than those who delivered at a facility (aHR: 1.9, 95% CI: 1.2–3.0), however only 43 mothers delivered outside of the facility.

Parity and whether the pregnancy was planned were included in the model but were not significantly associated with nonadherence.

Provincial variation in NVP nonadherence in the 6 week to 18 month postpartum period was seen in the infant population with those in Northern Cape, North West, Western Cape, and Eastern Cape experiencing the highest hazard of nonadherence compared to those in Free State (Northern Cape, aHR: 5.3, 95% CI: 1.9–14.8; North West, aHR: 4.8, 95% CI: 1.8–12.6; Western Cape, aHR: 4.1, 95% CI: 1.4–11.7; Eastern Cape: aHR: 3.2, 95% CI: 1.1–9.3) (Table 3).

Infants whose mothers had not heard of PMTCT prior to the six week interview had higher hazard of nonadherence within the 6 week to 18 month postpartum period than those whose mothers had heard of PMTCT (aHR: 1.6, 95% CI: 0.9–32.8; p value < 0.10), although this was not significant at 5%. Not knowing the mother’s partner’s HIV status showed increased hazard for nonadherence (aHR: 1.4, 95% CI: 10.9–2.1; p value < 0.10) but this was not significantly at 5%. Maternal parity was included in the model but was not significantly associated with adherence.

This study observed suboptimal ART adherence until 18 months postpartum among this cohort of HIV-positive mothers and infants, with below 65% of mothers and below 75% of infants cumulatively maintaining adherence over 95% from 6 weeks to 18 months postpartum. This is gravely concerning, given the global policy shift to lifelong ART amongst pregnant and lactating women, and the need for extended infant prophylaxis amongst mothers who are not virally suppressed. We found ART adherence to be higher at earlier time points, with both mothers and infants over 85% adherent by 14 weeks over or about 80% by 6 months. A similar analysis of infant ART adherence in Zambia showed a lower adherence until 6 months of about 70% (versus our result of 93.0%), and until 18 months of just above 0% (versus our result of 73.4%) [36]. A study from Malawi showed lower maternal ART adherence than our study, with 75% adherence by 3 months, about 50% adherence by 6 months, and under 40% adherence by 18 months [37]. While the study in Malawi used a pharmacy-based adherence measure, our study utilized self-reported adherence which tends to overestimate adherence due to recall bias and social desirability. This may account for the differences in adherence seen across these two studies. Results from a systematic review of adherence among women on ART in low- middle- and high-income countries reported adherence of 53% (95% CI: 33–73) during the postnatal period, although the timeframe was not specifically defined [4]. While our results from this cohort of mothers and infants indicate higher ART adherence than that seen in similar studies from the region, possibly because of the Hawthorne effect [38] as a result of three-monthly follow-up by nurses, adherence reported herein remains suboptimal to eliminate MTCT and highlights room for improvement on the implementation of PMTCT guidelines [5, 24].

Infant adherence was higher than mother adherence by about 10% across the entire 6 weeks to 18 months postpartum period. This may suggest that efforts made by the mother to ensure adherence within the mother-infant pair are prioritized toward the infant’s HIV prophylaxis adherence over her own HIV treatment.

We identified risk factors for nonadherence to inform further improvements to PMTCT programming. Adolescents and young women in this cohort (age 16–24 years) had a higher hazard of nonadherence than women older than 25, consistent with similar studies about PMTCT and maternal and child health seeking [39‐41]. A study from Uganda indicates that this is potentially due to lack of knowledge and experience with childbirth and health, as well as age-related discrimination from the heath system [42]. In South Africa, adolescent girls and young women have particularly high risk for acquiring HIV infection and low uptake of HIV services [43]. Targeted HIV interventions to increase HIV prevention and service coverage among this high risk group were initiated in South Africa in 2015.

Initiating ART after delivery was a risk factor for nonadherence among mothers, indicating that initiation on ART earlier in the continuum of care may be an important protective measure to improve adherence. The current national guidelines entitling all HIV positive pregnant and breastfeeding women to lifelong ART partially address this concern [5]. Additionally, messaging that encourages timely attendance to antenatal care visits among pregnant women in South Africa may increase early initiation on ART [44].

Knowledge of one’s CD4 count protected mothers from nonadherence, indicating that compliance with CD4 testing and returning of test results to patients may have contributed to improved adherence to ART among mothers. At the time of these data, ART initiation was based on CD4 count< 350 under Option B guidelines, therefore for mothers who were not experiencing symptoms and therefore did not feel the pressing need for ART adherence, CD4 results may have offered a motivating factor [45, 46]. Counseling messages from health providers received with CD4 test results could have improved a mother’s understanding of the importance of ART adherence to PMTCT, thus reducing nonadherence [47]. This confusion around CD4 count is no longer an anticipated barrier at the present era of test and treat for life.

Disclosure of HIV status to family and friends was protective from nonadherence within this cohort, as seen in other studies from South Africa and other sub-Saharan African settings [48‐50]. These studies postulated that the psychosocial support gained through disclosure, as well as the accountability to one’s treatment, encourage adherence. PMTCT programmatic messages should encourage mothers to disclose their HIV status to improve ART adherence. Furthering the case for support as a protective factor against nonadherence is the result that married and cohabitating mothers showed lower nonadherence than single, divorced, and widowed women.

Mothers who delivered their infants in health facilities had lower hazard of nonadherence than those who delivered outside of a formal health facility. We postulate that this is due to additional support provided when a delivery is in a health facility. Other studies have found that individuals exhibiting high health seeking behavior have been found to have higher adherence to ART [51]. Messages about the importance of ART adherence for PMTCT are provided whenever a mother interacts with the health system along the continuum of care [12], thus those delivering in a facility receive a higher frequency of these messages.

Provincial variation in nonadherence may be indicative of inconsistencies in effectiveness of PMTCT, potentially explained by a complex network of differences in relative strength of the provincial health systems, messaging surrounding the importance of PMTCT, the general status of health, and geography. As a highly dynamic and migratory population, variation in health seeking behavior is expected, as noted in other studies [17, 43, 52].

By nature of the population followed until 18 months postpartum by the national PMTCT Evaluation study, the present study did not include adherence for HIV-exposed infants who had a positive PCR result at 6 weeks or their mothers. As a cohort analysis observing a subgroup from this follow-up population, the results were not adjusted for the study design, non-response and not weighted for live-births, thus limiting the representativeness of these results.

The Andersen-Gill extended cox proportional hazards model selected for this analysis requires a strong statistical assumption of independence across the increments in which nonadherence events can occur [35]. This assumes that a nonadherence event is not influenced by a prior nonadherence event within a study participant. Thus, event dependence is not included in the model. Without strong evidence to support dependence across nonadhernece events, we chose this model.

ART adherence was evaluated on self-reported responses, therefore issues of false reporting and recall likely biased the results, however we cannot determine whether these are under- or over-estimates. To minimize such information bias we checked road-to-health cards to verify maternal self-report of HIV status and ART uptake. The consistency between maternal self-report of these measures and the record from the road-to-health card increased our confidence in other self-reported measures.

A high proportion of mothers and infants included in the follow-up study missed interviews during the 18 month follow-up period. Due to the lack of information about these participants at intervals where the interview was missed, participants were not included in the analysis for any missed interval, likely introducing bias into the results. There was no date of failure captured for mothers who stopped taking ART during an interval, thus the midpoint was used which could have under- or over-estimated the time contributed.

A higher probability of adherence during later analysis time may been driven by loss to follow-up of non-adherent women and may not reflect an actual improvement in adherence, as women with poor adherence are at higher risk of loss to follow-up than those with good adherence [37, 53]. This mechanism of “informative censoring” biases analysis of adherence patterns over time.

The population studied herein is highly dynamic [17, 54, 55], exemplified by the fact that individuals could move in and out of the included study population at each follow-up interval. Such mobility likely contributes to the low adherence seen here, both by providing a barrier to adherence among mothers and their infants and by posing a challenge to measurement within our follow-up interviews. The high proportion of missed interviews underscores the importance of strengthening counseling to ensure that individuals understand the importance of ART adherence to viral suppression, reducing risk of MTCT, and reducing risk of drug resistance. Mothers were continually counseled on the importance of adherence to health outcomes throughout the course of the study, especially when non-adherence events were identified.