Description of study setting

A protocol describing the care processes in the region states at which level of care the different patient groups are to receive treatment. In HCHC, nurses and midwives provide care for women with uncomplicated pregnancies. SRH, the secondary level of care is managed by midwives and medical officers who treat women with minor medical complications. At CMJAH all women are assessed by a doctor and there are medical specialists available for more complicated patients. Women are considered to have a complicated pregnancy if they were unwell in a previous pregnancy, or have, for example, co-morbid medical illnesses, multiple gestations or suspected congenital abnormalities. In HCHC and SRH, women do not require a referral letter and can present directly for services when pregnant. CMJAH only provides ANC for women with pregnancy complications and then refers healthy women to lower levels of care. Caesarean sections are only performed at the secondary- and tertiary-level facilities. There are no user fees for HIV, ANC or childbirth services. HIV testing, PMTCT and antiretroviral treatment are available at all facilities.

It is important to note that the study setting, region F (inner-city), has several distinctive features which influence the uptake of health services and programme implementation. The inner-city is one of seven regions within the district called City of Johannesburg. In turn, the City of Johannesburg is one of five districts of the Gauteng Province, the economic hub of the country. The inner-city is densely populated, consisting of the flatland areas of Hillbrow and Berea, as well as the Johannesburg Central Business District. It is estimated that the region contains about 15% of the 4 million people who live in the city of Johannesburg [21]. However, it is likely that this figure is well above 15% as the majority of the large transient population in the inner-city area are not included in those estimates. The inner-city is a uniquely complex and dynamic environment that has undergone major demographic, social and economic shifts over the last few decades. Many of its inhabitants are immigrants, both South Africans from other provinces of the country as well as foreigners. About a quarter of adults in the area are unemployed [21]. The inner-city houses several large taxi hubs and 800,000 commuters are said to pass through the city daily. Informal trading is the dominant economic activity. Unfortunately, the area is plagued with high levels of crime and drug dealing [22]. Not surprisingly, the area has a high burden of HIV: the antenatal HIV prevalence in Gauteng Province as a whole was 29.9% in the most recent national survey (2012), with a similar level in the City of Johannesburg (29.6%). Corresponding figures for Herpes Simplex Virus type-2 were 58.4% and 58.5% [23].

Data management and study variables

The following variables were manually extracted from the labour ward registers into a Microsoft Excel database: ANC attendance (having attended at least one ANC visit), maternal age (not available at SRH), gravidity and parity (only available at CMJAH), HIV testing uptake and status, mode of delivery, infant gender, gestation at childbirth and stillbirths. Patient names or other identifiers were not collected.

Adolescents were divided into younger and older groups (10–16 years and 17–19 years), which correspond broadly to the ages of those in primary or secondary school (10–16 years), and those in the final years of school or post-school (17–19 years). Adults were groups into 5-year age bands. As about 4% of data were missing on whether women had attended ANC, we performed a sensitivity analysis in which we reclassified the missing data in one of three ways. Firstly, in a worst case scenario, women whose attendance status was unknown were classified as not having attended ANC. In the second scenario, a ‘missing excluded’ analysis, we excluded all women where it was unknown whether or not they had attended ANC. The third scenario, the best case, classifies women with unknown attendance as having attended ANC. Findings across the three scenarios were compared for effect size and direction. The associations reported in the paper are drawn primarily from the worst case scenario, as we considered it most likely that where attendance was unknown, patients had not actually attended care. Also, many of the figures in the other two scenarios were implausible. Figures for HCHC were unchanged across the scenarios as data were not missing on ANC attendance at that site.

Newborns were classified as preterm (under 37 weeks gestation), term (37 to 41 weeks) and post-term (42 or more weeks). Stillbirths included some deaths that occurred shortly after birth as the birth records did not differentiate between those deaths and actual stillbirths.

Data analysis

Data were recoded and analysed using STATA version 12 (STATA Corporation, College Station, Texas, USA). Chi-square tests were used to detect associations between categorical variables and Wilcoxon Rank sum tests were used for analysis of continuous variables. Pearson Chi-square test for trend assessed associations between ordinal exposure categories and binary outcomes. Multiple logistic regression models examined whether ANC attendance was associated with two dependent variables assessing service access (HIV testing and Caesarean section delivery) and two birth outcomes (preterm births and stillbirth). These models included the variable site, and other potential confounding variables associated with the outcome in univariate analysis or in analysis stratified by site (P < 0.10). The exposure variable antenatal attendance was included in the multivariate model assessing factors associated with the outcome preterm delivery (attendance was associated with preterm delivery in all three facilities, even though not associated with the outcome when data from all sites were pooled). For similar reasons, the exposure variable delivery mode was included in the multivariate model assessing factors associated with the outcome stillbirths. Potential confounders varied by outcome variable, with, for example, infant sex not included in the outcome HIV testing, as there was no plausible causal pathway between these variables. Infant sex, however, and HIV status were considered potential confounders for the remaining three models.

Characteristics of women and birth outcomes

Of those who had a test, 30% were HIV positive (8216). These 8216 women would have required PMTCT services (this reflects the burden of care relating to HIV in these three facilities). Considerable differences were noted in HIV prevalence between sites: it was highest in HCHC (37%, 1897), lower in CMJAH (30%, 5452) and lowest in SRH (24%, 867). At SRH and CMJAH, the prevalence of HIV was similar among women who had and had no attended ANC. At HCHC, however, HIV prevalence was 65% in the 144 non-attenders who had been tested, almost 20% higher than the prevalence in attenders.

Overall, for the sub-district, 36% of deliveries were by Caesarean section (11,288/31,179), with levels reaching 54% in CMJAH (10,304/19,048). Caesarean section was considerably less common in women with an unknown HIV status, even after controlling for other variables (Table 2). The odds of having a Caesarean section were 1.2 fold higher in preterm than term gestations, and similar in women with or without HIV infection. At CMJAH, 52% of women with a live newborn were delivered by Caesarean section (9027/17,270), compared to only 33.9% of stillbirth deliveries (156/460). Very negligible differences in infant gender were noted between facilities, with slightly more males delivered in higher levels of care.

In total, 17% of all newborns were classified as preterm babies (5198/30,678). Almost all deliveries categorised as preterm took place at CMJAH (99%, 5125/5198), where 27% of newborns were preterm. Rates of preterm delivery were 1.3 fold higher in HIV positive than negative women (aOR = 1.2–1.4), and 4 fold higher in women with an unknown HIV status, even after controlling for ANC attendance (Table 2).

Overall, the stillbirth rate was 18 per 1000 live births (535/29,004). Stillbirth rates were several fold higher in CMJAH than in HCHC and SRH (28/1000 in CMJAH versus about 4/1000 for HCHC and SRH). Independent of ANC attendance status and other potential confounders, having an unknown HIV status was associated with stillbirths (aOR = 4.2, 95% CI = 3.3–5.5). Stillbirths were also more common among female than male infants (aOR = 0.76, 95% CI = 0.64–0.92).

Overall antenatal attendance and variations among sub-groups

In CMJAH and HCHC, ANC attendance increased step-wise with age of the woman, though it declined from 35 years upwards at HCHC. In the worst case scenario, adolescents were 1.4 fold more likely not to have attended ANC than adults (95% CI OR = 1.3–1.6). The absolute percentage difference between adolescents’ and adults’ attendance was 3% at HCHC and 2% at CMJAH. Antenatal attendance in women above 30 years in CMJAH was higher than all other age groups in the other two facilities (94%, 6313/6700).

In HCHC only 95% of HIV-positive women had attended ANC (1803/1897), compared with 99% of negative women (3220/3270). HIV prevalence was similar in those who had or had not attended ANC in the other sites. Lastly, the proportion of women at CMJAH who had attended ANC rose step-wise with an increase in gravidity in the worst case, but not in the other two scenarios.

Associations between antenatal attendance and HIV testing, delivery mode and birth outcomes

Overall, in the worst case scenario, HIV testing rates were 37% in women who had not attended ANC (1308/3528), compared with 93% of ANC attenders (25,756/27,651). Moreover, when separated by site, HIV testing rates among women not attending ANC varied considerably: from 8% at HCHC (144/1730) to 59% at SRH (268/452) to 67% at CMJAH (896/1346). In the missing-excluded scenario, the corresponding figures for SRH and CMJAH are 49% (1895/3859) and 99% (25,756/26,120). The respective proportions were 52% (2151/4115) and 99% (26,700/27,064) in the best case scenario.

In univariate analysis, rates of Caesarean section were considerably lower in women who had not attended ANC (13%, 422/3360) than those who had (40%, 10,866/27,344). Multivariate analysis showed similar findings and was consistent across all three scenarios (Table 2; Additional file 3: Table S3 and Additional file 4: Table S4).

No association was detected between ANC attendance and preterm birth in univariate analysis. However, in bivariate analysis stratified by facility, in each site the proportion of births that were preterm was higher in women who had not attended ANC than in those who had. This finding persisted across all scenarios. In multivariate analysis of the worst case scenario, non-attenders were 1.6 fold more likely to have a preterm birth than those who attended (aOR 95% CI = 1.4–1.8). The point estimates for this effect were 2.7 in the other two scenarios.

In the worst case scenario, ANC attendance was 89% in women who had a live birth (25,813/29,004), while only 81% of women with a stillbirth had attended clinic (431/535). In multivariate analysis of the worst case scenario, ANC non-attenders were 1.4 fold more likely to have had a stillbirth than their counterparts (aOR 95% CI = 1.1–1.9), figures similar to the other scenarios.

Strategies to improve antenatal attendance

Clearly, in this setting, interventions to raise ANC attendance are a priority and critical to reducing maternal mortality, stillbirths and paediatric HIV, and to securing access to ART for HIV-infected women [12]. In the country as a whole, failure to address deficiencies in access to ANC, will substantially constrain the ability to further improve national maternal and child health outcomes.

Research identifying the specific demand-side barriers to access in this setting, especially within groups such as adolescents, HIV-positive women and foreigners, could provide actionable information. In other settings, these factors include lack of women’s empowerment and supportive partners [46], perceived poor quality of services [47], having an unintended pregnancy and low socio-economic status [48–50]. Demand for antenatal services could be raised through community involvement and activities to raise awareness around these services, including through mass media campaigns [17, 51, 52]. Further strengthening the antenatal care component of the government led MomConnect mHealth intervention might make an important contribution to such initiatives and to raising the number of ANC visits attended [53, 54]. Specific messages on MomConnect, or even a separate mHealth service, could be targeted at immigrants and younger poorer women. Cash transfers have been successfully applied in many settings to incentivise attendance at maternal health services [55]. Currently a cash transfer is given to women in South Africa once a child is born until they are 18 years. Beginning the transfer during pregnancy, as is done in a host of countries, (and not only restricting it to South African women) could raise demand for care (presentation of a ANC patient-held card could be required for enrolment in the scheme, for example) [56].

Supply side actions that have been found to raise attendance include improvements in service quality [57, 58]. More broadly, government should promote adolescent-friendly services and perhaps reserve specific time slots at the clinic for this group [51, 59]. Both within health services and the country as a whole, much more needs to be done to counter the high levels of xenophobia that mark the country and hinder access of migrants to health care [60]. Strategies to achieve this could include training and sensitisation of health workers, monitoring of health workers’ interactions with migrants and strengthening the coordination mechanisms between the multi-sectoral partners who work in this field [61]. Birth registers in areas with high numbers of foreign nationals could include measures of migrant status, such as country of origin and length of time resident in the country, in order to track outcomes of these women and identify any particularly vulnerable sub-groups to be targeted.

Study strengths and limitations

Deficiencies in data quality, especially the gaps in data on ANC attendance at SRH and CMJAH, hinder the ability to interpret the findings. These deficiencies may partly be explained by the data having been collected as part of routine services, where staff have numerous competing priorities. However, regardless of what assumptions were made about the missing data, the direction and size of effect were generally consistent, suggesting that study findings may be robust despite this potential bias. Also, noting these deficiencies in data and the usefulness of the data collected is a key step towards improving data quality in the facilities and other parts of South Africa. Also of note, levels of attendance measured through health service data can differ considerably from coverage measured using population–level data [41]. However, where utilisation of skilled birth attendance is high, health service data, such as used in this study, may approximate antenatal coverage measured in a survey [62]. In Gauteng province, 99% of women deliver with a skilled birth attendant [7], suggesting that our measures of ANC attendance reflect that of population-level coverage.

Number of visits was not recorded in the birth registers, and thus we were unable to differentiate between those who had and had not completed the recommended four visits. Even attendance fewer times than the recommended number offers considerable benefits: the use of point-of-care HIV and syphilis testing, and of same-day initiation of antiretroviral treatment means that women attending even one visit can still access several important services.

The data collection periods differed between facilities. This could have biased the study findings as differences observed in attendance rates between the tertiary facility and other levels may be due to systematic improvements or reductions in ANC utilisation across the facilities, rather than due to the differences between levels of care. Also, patterns of patient referral may have shifted during that period and it is even possible that some women gave birth at the primary or secondary level facility in 2008–2009 and then later at the tertiary centre in 2011. It is likely, however, that the background population living in the study area remained relatively constant across the two periods. Moreover, overall, despite the data gaps and quality concerns, this was the only available data that provides information on this important topic. The use of disaggregated data from this large database is a major strength of the study, allowing us to identify differential access to services and variations in birth outcomes by sub-populations, who can then be targeted.