Birth outcomes in South African women receiving highly active antiretroviral therapy: a retrospective observational study

Single data entry was done in Microsoft Access and analysis performed with Intercooled Stata 8.0 (Stata Corporation, College Station, TX, United States). Characteristics of HAART-exposed and unexposed women were compared using chi-square tests for categorical exposure variables and Student's t-test for continuous variables. As we were interested in assessing the effects of drug regimen on LBW independent of the role of infant HIV infection, we initially restricted multivariate analysis of LBW to HIV-negative infants and thereafter included all infants in a similar analysis. A similar multivariate analysis was performed for variables associated with preterm birth in which we included all infants with a known gestation at birth, regardless of HIV status.

Variables associated with LBW or preterm birth in univariate analysis (p < 0.1) were included in the initial multivariate model and retained if their removal markedly altered the model fit. Model fit was assessed using the likelihood ratio test and judgement about the size of changes to the model with and without the variable [16]. The model analyzed the odds of LBW or preterm birth according to regimen type: lopinavir/ritonavir-based, nevirapine-based and efavirenz-based HAART; CD4 cell categories (cells/mm³) of 0-49, 50-99, 100-149, 150-199 and 200-250; maternal age; anaemia (haemoglobin <11 g/dl); hypertension (systolic blood pressure [BP] >160 mmHg or diastolic BP >90 mmHg on two occasions four hours apart, or one diastolic BP >110 mmHg); and infant HIV status. Women in the early and late HAART groups were analyzed in separate multivariate models due to the presence of interaction: associations between the exposures (drug regimen) and outcome LBW and prematurity) varied according to the duration of HAART.

The cohort comprised 1630 women, with a mean age of 30.1 years (sd 5.1) and median CD4 count of 159 cells/mm³ (IQR 105-200). Women who declined study participation were not included in the data base. Fourteen percent of included women (n = 233) were HAART unexposed. In the remainder (n = 1397) who received HAART during pregnancy, duration of therapy was known for 70%, with 54.5% of these (n = 533) initiating HAART before 28 weeks ("early HAART") and 427 at ≥ 28 weeks ("late HAART"; Figure 1).

Compared with unexposed women (Table 1), HAART-exposed women were older (30.3 vs. 28.9, p < 0.001), had a lower median CD4 count (154 cells/mm³ vs. 191 cells/mm³, p < 0.001), and a lower rate of anaemia (46% [416/912] vs. 57% [57/100], p = 0.03). However, no differences were detected between these groups for other risk factors for LBW or preterm birth, such as gravidity, hypertension, alcohol use, smoking, syphilis and previous miscarriage. Of note, diabetes requiring treatment with medication (oral hypoglycaemics or insulin) was uncommon, with only four women in the cohort meeting these criteria. Positive syphilis serology (rapid plasma reagin) was detected in 3.8% (38/1001) of women.

Compared with women at RHM, those at CMJH were less frequently primigravidas (10% [49/487] vs. 16% at RHM [77/467], p = 0.005), more likely to have advanced WHO HIV disease (46% stage 1 [165/359] vs. 73% [281/385], p < 0.001) and more likely to have had previous miscarriages (22% [67/305] vs. 14% [63/450], p = 0.002). Information on TB was not available from RHM and was only available for 33% (466/1397) of women at CMJH, with 17% (80/466) of these having either current or prior TB.

A total of 1019 infants attended their scheduled visit for HIV DNA PCR testing at six weeks. Risks of MTCT were 3.6-fold higher in the non-HAART than HAART group (19% [37/191] versus 5% [45/828], p < 0.001), with the largest reduction in transmission among the early HAART group (2% [8/350] versus 10% [31/316] in the late group, p < 0.001).

The median duration from HAART initiation to childbirth was 18.4 weeks (IQR 12.1-42.6) for the early HAART group and 5.8 weeks (IQR 3.3-8.5) for the late group (p < 0.001). Women in the early group more commonly received NVP-based HAART (46%, 254/553), while those in the late group were more likely to receive protease inhibitor-based HAART (68%, 290/427; Figure 1). No differences were detected between the early and late HAART groups in terms of baseline CD4 count, previous preterm birth rates, syphilis prevalence and smoking. However, women in the early HAART group had higher rates of previous miscarriage than the late HAART group (22% [70/313] vs.12% [38/310], p = 0.001).

The mean birth weight for infants in the cohort was 2.88 kg, with 22.4% (275/1228) having LBW. Among infants with known HIV status, LBW was more common in HIV-positive than negative infants (33% [24/73] vs. 22% [173/804], p = 0.04). In HAART-unexposed infants, 27% (60/224) were LBW compared with 23% (90/388) of early HAART-exposed and 19% (76/407) of late HAART-exposed infants (p = 0.05). No differences, however, were detected when comparing rates of LBW in HAART-unexposed versus HAART-exposed infants (27% [60/224] vs. 21% [215/1004], p = 0.08), or between the early and late HAART groups (23% [90/388] vs. 19% [76/407], p = 0.12). Among women with early HAART-exposure, higher rates of LBW were observed in women receiving efavirenz-based regimens (38% [36/95]) compared with nevirapine (20% [31/158]) and protease inhibitor-based regimens (17%, [23/135]; p < 0.001). There were no differences in rates of LBW by regimen in the late HAART group (Table 2).

Given that this initial analysis showed higher rates of LBW in women receiving efavirenz-based regimens from early in pregnancy, we compared characteristics of women in this group with women taking other regimens. Women taking early HAART and receiving efavirenz had higher rates of TB compared with those on nevirapine and protease inhibitor-based therapy (28% [25/88] vs. 14% [25/183] and 10% [1/10] respectively, p = 0.01), lower median CD4 counts (138 cells/mm³ vs. 155.5 cells/mm³ vs. 164 cells/mm³ respectively, p = 0.03), and longer median weeks on HAART (62.7 [IQR 33.1-86.4] vs.15.6 [IQR 10.7-25.8] and 17.1 [IQR 13.7-23.1] respectively, p < 0.001).

Overall, only 2% (26/1228) of all infants in the cohort were classified as very low birth weight (VLBW). More HAART-unexposed than exposed infants were VLBW (4% [10/224] vs. 2% [16/1004], p = 0.01). The mean CD4 cell count of the VLBW group did not differ from the remainder of the cohort (148 cells/mm³ vs. 153 cells/mm³, p = 0.65). Rates of VLBW infants were similar in the early and late HAART groups and for all HAART regimens. Seven of the infants had known maternal risk factors for VLBW, including maternal history of previous miscarriage, previous preterm delivery, or hypertension. One infant in this group was HIV infected.

In univariate analysis evaluating associations between HAART and LBW, including both HIV-positive and HIV-negative infants, no significant associations were detected in the late HAART group (Table 3). However, in the early HAART group, receipt of an efavirenz-containing regimen, lower CD4 cell count, and maternal hypertension were associated with LBW. Although rates were low, neither smoking nor alcohol use was associated with LBW. In multivariate analysis, large effects of immunological status remained, with each 50 cells/mm³ increase in CD4 cell count associated with a 57% reduction in the odds of LBW (95% CI 0.45-0.71, p < 0.001). There was a trend towards hypertension being associated with increased odds of LBW (AOR 2.1, 95% CI 0.92-4.82; p = 0.08). In multivariate analysis, the effect of efavirenz exposure was removed (AOR 1.02, 95% CI 0.46-2.25), and unexpectedly, nevirapine-based HAART was associated with a reduced odds of LBW compared with HAART-unexposed women (AOR 0.38, 95% CI 0.18-0.81, p = 0.01). Similar results were found in multivariate analysis of associations between HAART groups and LBW when modelling was restricted to only HIV-negative infants (data not shown).

The overall rate of preterm birth in the cohort was 13.3% (145/1093). Rates were higher among women exposed to HAART during pregnancy than those unexposed (15% [138/946] vs. 5% [7/147], p = 0.002). In the analysis of early versus late HAART, infants whose mothers initiated treatment before 28 weeks had a higher rate of preterm birth compared with after 28 weeks (21% [81/389] vs. 5% [21/427], p < 0.001). HAART exposure was not associated with increased rate of extremely premature birth (6% [58/946] vs. 4% in unexposed women [6/147], p = 0.43).

In univariate analysis evaluating predictors of preterm birth in all infants (regardless of HIV status), in the early HAART group, every 50 cells/mm³ rise in maternal CD4 cell count was associated with a 31% decrease in the odds of preterm birth (95% CI 0.58-0.83, p < 0.001, Table 4). This association remained significant in multivariate analysis (AOR 0.68, 95% CI 0.55-0.85, p = 0.001). In the multivariate analysis of specific HAART regimens, early exposure to any regimen was associated with preterm birth compared with HAART-unexposed infants, with receipt of early efavirenz and nevirapine associated with the higher odds of preterm birth (AOR 5.6, 95% CI 2.1-15.2, p = 0.001, and AOR 5.4, 95% CI 2.1-13.7, p < 0.001, respectively), than protease inhibitor-containing regimens (AOR 3.0, 95% CI 1.1-8.4, p = 0.04). Neither smoking nor alcohol use was associated with preterm birth in this analysis. Drug regimen was not associated with preterm birth in the late HAART group.

The rate of LBW in this cohort (22%) was higher than the rate of 15% reported by UNICEF for all South African infants, regardless of HIV exposure [17]. Many of the factors associated with LBW in this study, such as CD4 count and hypertension, have been reported in previous studies [18, 19]. Despite findings from a variety of other settings, in our cohort, lopinavir/ritonavir exposure was not associated with LBW in early or late HAART-exposed women compared with untreated women [5, 20]. We detected an association between LBW and efavirenz, but this did not persist after controlling for stage of HIV infection and other potential confounding variables.

Unexpectedly, in the early HAART group, rates of LBW were lower in women on nevirapine-based HAART than in untreated women. Nevirapine has previously been associated with increased risk of LBW [12], as well as found to be neutral with respect to LBW [21]. These data are the first to suggest a decreased rate of LBW. This finding might be due to the fact that untreated advanced HIV infection (CD4 count <250 cells/mm³) confers increased risk for low birth weight, and initiation of treatment reduces this risk. There could also be site-specific reasons why non-nucleoside reverse transcriptase inhibitors (NNRTIs) and not PIs were significant in this analysis. Women who initiated NNRTIs tended to be from CMJH, which handled more complicated pregnancies [22, 23].

HAART-unexposed women had a high rate of LBW (27%) despite a relatively low preterm birth rate (5%). This might be explained by potential selection biases (discussed later). Additionally, HAART-unexposed women likely had a high risk of intra-uterine growth retardation due to their untreated advanced HIV infection with resultant immunological dysfunction (median CD4 cell count 191 cells/mm³).

Only 26 study infants were VLBW. In the small group of VLBW infants identified, no association was detected with maternal immunological status. Small numbers of VLBW infants limit this analysis.

In this cohort of HIV-infected women with CD4 counts ≤250 cells/mm³, early HAART exposure was associated with an increased risk of preterm birth compared with untreated women. This is consistent with previous studies demonstrating an association with preterm birth and in utero HAART exposure [[4, 5, 10, 24]]. A unique element of our study is the focus on African women with low CD4 counts who are initiated on HAART for their own health. Additionally, reported rates of smoking (3.7%) and alcohol use (3.9%) across all our groups were low, unlike many European and North American studies, where smoking prevalence has ranged from 7.5% to 55% [[6, 7, 25, 26]].

The association between preterm birth and NNRTI exposure was larger than with PI exposure. Reasons for this may include bias related to site, as women exposed to NNRTIs were more likely to have attended the clinic at CMJH, which cares for women with more complicated pregnancies, increasing the likelihood for women to have multiple risk factors for preterm birth. Efavirenz-exposed women had more advanced HIV than women on other regimens, and early efavirenz-exposed women had been on HAART longer than other groups (62.7 weeks vs. 16 weeks); however, in multivariate analysis, there was only a small attenuation in the odds ratio (from 6.4 to 5.6), and residual confounding may be present, particularly given the lack of data on TB among women in the cohort.

The finding of an association with early PI exposure during pregnancy and preterm birth is similar to many previous studies. In these studies, PI-based regimens were often prescribed for women with more advanced HIV disease, making it difficult to disentangle the role of HAART regimen and stage of disease on infant outcome. In our study, all women had CD4 counts <250 cells/mm³ _, reducing the confounding caused by large differences in the degree of immunosuppression.

Additionally, the only PI used in our cohort was lopinavir/ritonavir as compared with previous studies evaluating unboosted PIs [4, 10]. Our results are consistent with recent randomized data from Botswana, in which a median of 11.3 weeks of lopinavir/ritonavir exposure during pregnancy was associated with a significant risk for preterm birth compared with women on non-PI regimens (21.4% vs. 11.8%, p = 0.003) [27]. Specific type of PI and the use of ritonavir may be important determinants of the overall risk for preterm birth.

Information was only collected from women who were willing to be included in the clinics' databases. Characteristics of women declining participation may differ from those giving consent, thereby introducing selection bias. Also, bias was potentially introduced by the selection strategy for the control group (HAART-unexposed women), as these women received less antenatal care and had poorer health-seeking behaviours than their HAART-exposed counterparts. These factors could, in part, account for the higher rate of LBW in the HAART-unexposed group.

Also, as the HAART-unexposed group were recruited postpartum, it is possible that a high mortality among infants born very preterm in this group partly explains the relatively low preterm birth rate (5%) noted in these women. Similarly, differences in how gestational age was assessed might have biased estimates of foetal age. HAART-unexposed women were more likely to have missed antenatal care, and have gestation determined by examination of infants at birth.

A considerable amount of data are missing for some study variables, mostly as the study was conducted within routine clinical settings and in two sites that collected different data. TB information was only available from CMJH women, and those exposed to early efavirenz-based HAART had the highest rates of TB as this regimen was specifically used with TB co-infection to minimize drug interactions. Because TB has been associated with intrauterine growth restriction and preterm birth [28, 29], incomplete data on TB co-infection may confound our findings. Similarly, HIV viral load is not routinely collected in the South African public system and thus was unavailable. HIV viral load prior to delivery is a useful marker of adherence to HAART regimen, the pre-eminent risk factor for MTCT, and may have a role in predicting other poor infant outcomes. Additionally, maternal body mass index was unavailable and may be an important factor for LBW and preterm birth [12, 22].

Since haemodilution during pregnancy is associated with spuriously low CD4 cell counts, women whose CD4 counts were measured postpartum (mainly HAART-unexposed women) were potentially more immunologically suppressed than the HAART-exposed group.

Differences in HAART prescribing protocols at CMJH and RMH could have affected the study outcomes. Women from CMJH tended to have more complicated pregnancies, for example, more frequent use of alcohol, higher rates of maternal hypertension or diabetes, and previous miscarriage. Baseline CD4 cell counts, however, were similar between the two hospitals. We were unable to include clinic site in the multivariate models due to its co-linearity with drug regimen.

It is hard to differentiate the effects of maternal HIV disease and that of HAART exposure on infant outcomes. Indeed, the improvement in immune status of the HAART-exposed women probably contributed to increasing infants' birth weights. However, improved maternal immune status would not explain the higher preterm birth rate in the HAART-exposed as compared with the unexposed women, which seems independent of HIV disease stage and might be due to HAART exposure or to the potential biases we have described. Finally, it is important to note that the findings of this study are likely to pertain only to women with marked immunological suppression, and may not extend to women with less advanced HIV disease.