HIV and the Risk of Direct Obstetric Complications: A Systematic Review and Meta-Analysis

Background Women of reproductive age in parts of sub-Saharan Africa are faced both with high levels of HIV and the threat of dying from the direct complications of pregnancy. Clinicians practicing in such settings have reported a high incidence of direct obstetric complications among HIV-infected women, but the evidence supporting this is unclear. The aim of this systematic review is to establish whether HIV-infected women are at increased risk of direct obstetric complications. Methods and findings Studies comparing the frequency of obstetric haemorrhage, hypertensive disorders of pregnancy, dystocia and intrauterine infections in HIV-infected and uninfected women were identified. Summary estimates of the odds ratio (OR) for the association between HIV and each obstetric complication were calculated through meta-analyses. In total, 44 studies were included providing 66 data sets; 17 on haemorrhage, 19 on hypertensive disorders, five on dystocia and 25 on intrauterine infections. Meta-analysis of the OR from studies including vaginal deliveries indicated that HIV-infected women had over three times the risk of a puerperal sepsis compared with HIV-uninfected women [pooled OR: 3.43, 95% confidence interval (CI): 2.00–5.85]; this figure increased to nearly six amongst studies only including women who delivered by caesarean (pooled OR: 5.81, 95% CI: 2.42–13.97). For other obstetric complications the evidence was weak and inconsistent. Conclusions The higher risk of intrauterine infections in HIV-infected pregnant and postpartum women may require targeted strategies involving the prophylactic use of antibiotics during labour. However, as the huge excess of pregnancy-related mortality in HIV-infected women is unlikely to be due to a higher risk of direct obstetric complications, reducing this mortality will require non obstetric interventions involving access to ART in both pregnant and non-pregnant women.


Introduction
The substantial burden of HIV infection amongst women of reproductive age in sub-Saharan Africa and the maternal health risks that these women are challenged with has lead to HIV and maternal mortality being described as two intersecting epidemics [1,2]. Many pregnant women in this region face not only the threat of dying from the direct complications of pregnancy and delivery, but also from complications arising from advancing HIV disease. Given this intersection, it is important to understand whether and how HIV interacts with pregnancy. The biological interaction between HIV and pregnancy is not well understood. It has been argued that pregnancy may accelerate HIV progression as pregnancy is associated with suppressed immune function independent of HIV status [3,4]. However, the epidemiological evidence supporting this hypothesis is weak. A systematic review investigating the effects of pregnancy on HIV progression and survival found no evidence that pregnancy increased progression to an HIV-related illness or a fall in CD4 count to fewer than 200 cells per cubic millilitre. The same review showed weak evidence that pregnant women were more likely to progress to an AIDS-defining illness or death compared with their non-pregnant counterparts but this was based on only six studies [5]. Clinicians working in settings where HIV is highly prevalent have reported a high incidence of direct obstetric complications in HIV-infected pregnant women [6]. Some researchers have also hypothesised that HIV may increase the risk of direct obstetric complications, though the evidence was based on very few studies with small sample sizes [2,7]. There are several biological pathways which may explain such an association. Firstly, the compromised immune status and general poor health of HIV-infected women may leave them more vulnerable to infections, including puerperal sepsis [8]. Secondly, it has been suggested that HIV-related thrombocytopenia, where there is a low platelet count in the blood, may increase a woman's risk of haemorrhage [9]. Additionally, social factors such as poor access to healthcare increase a woman's risk of obstetric complications, and may be exacerbated in HIV-infected women due to the discrimination and stigma these women face in some settings [10].
To date there has been no effort to synthesise the empirical evidence on the association between HIV and direct obstetric complications. The aim of this study is to investigate whether HIV increases the risk of obstetric complications, by systematically reviewing literature which compares the risk of obstetric compli-cations in HIV-infected and uninfected women. The obstetric complications which were pre-specified for this review are obstetric haemorrhage, pregnancy-induced hypertension, dystocia and intrauterine infections.

Search Strategy
Pubmed, Embase, Popline and African Index Medicus were searched up to 6 th July 2011 using search terms for HIV, pregnancy and the following direct obstetric complications: obstetric haemorrhage, pregnancy-induced hypertension, dystocia and intrauterine infections (see Supplementary File S1 for the full search strategy). There were no language or publication date restrictions. All abstracts were reviewed by a single author (CC) and a 20% sample of abstracts was independently reviewed by a second researcher. Full text copies of potentially relevant papers were obtained and the reference lists of review articles and articles which were included in this systematic review were searched for further relevant publications.

Eligibility Criteria
Studies were eligible for inclusion if they compared the occurrence of direct obstetric complications during pregnancy, delivery and/or up to 365 days postpartum between HIV-infected and uninfected women using a cohort, cross-sectional or case-control design. Obstetric complications relevant for this review were categorised as: obstetric haemorrhage (including placenta praevia, placental abruption, antepartum haemorrhage, perior postpartum haemorrhage and retained placenta); pregnancy-induced hypertension (including eclampsia and pre-eclampsia); dystocia (including prolonged or obstructed labour, abnormal presentation and uterine rupture); and intrauterine infections (including puerperal sepsis, wound infection and endometritis). Studies were required to have a sample size of at least 30 women in each study group with no restrictions on country, dates or whether the study was population or facility based.

Data Extraction and management
Data were extracted by a single author (CC) on: study location, dates, design and population, definition and ascertainment of the obstetrical outcome (e.g. whether haemorrhage was ascertained through visual estimate or actual measurement of blood loss), the mode of delivery, gestational age at recruitment and length of postpartum follow-up, HIV prevalence in the study population, whether antiretroviral therapy (ART) was available, the number of women with the obstetric complication by HIV status, the type of denominator (pregnancy, live births or women) and the denom-inator. Study populations described in more than one paper were included only once, using data from the paper with the most detailed information. When more than one obstetrical outcome was evaluated in a single study, these were extracted and treated as separate data sets.

Assessment of risk of bias
The risk of bias for each data set was assessed using the component approach adopted by The Cochrane Collabora-tion [11]. All data sets were assessed on the definition and ascertainment of the obstetric complication, the completeness of data, adjustment for confounding and selection of the comparison group. Each of the quality criteria were classified as having a low risk or high risk of bias for each data set. For example, a data set was classified as having a high risk of bias for outcome ascertainment if methods which were likely to lead to cases being missed were used (e.g. hospital record review). Where there was insufficient information to assess the risk of bias, the data set was classified as at an unclear risk of generating bias.

Statistical Methods
All analyses were carried out using STATA 12.0. The association between HIV and each obstetric complication was estimated using odds ratios (OR). Summary measures of effect for each obstetric complication were obtained by conducting a random-effects meta-analysis of the best effect estimate available from each study. Where an adjusted OR was available from the paper, this was taken as the best estimate; otherwise the crude estimate was used. Articles do not generally state whether there is overlap between categories of obstetric complications, for exam-ple, whether the women who have puerperal sepsis are also the women who are included as having endometritis. We therefore only provide summary estimates for sub-categories within each broad obstetric grouping. As the effect of HIV on the obstetric complications may vary by the mode of delivery, studies which included either vaginal deliveries only or both vaginal and caesarean deliveries were considered separately from studies which only included caesareans. Publication bias was assessed using funnel plots and was formally tested using Begg's test [12]. Additionally, for data sets which included vaginal and caesarean section deliveries, a meta-analysis was conducted to assess whether HIV-infected women had increased odds of caesarean. ORs were computed for each study rather than each data set.

Risk of Bias Within and Between Data Sets
The assessment of the risk of bias is summarised in Tables 3 and  4. Only 23 of the 66 data sets provided a definition for the obstetric complication: from eight of the 19 data sets (42%) reporting on hypertensive disorders to seven amongst the 25 data sets (28%) for intrauterine infections. The risk of bias in the ascertainment of obstetric complications cases was judged to be high for 29 of the 66 data sets; most of which relied on medical records to ascertain the nature of the complication. Very few studies had sufficient information on the completeness of the data to enable the risk of bias to be assessed and only 17 of 66 data sets were classified as at low risk of bias. In particular, studies relying on medical records tended not to report how many records had to be excluded due to missing information (e.g. HIV status). Overall, 25 of 66 data sets either adjusted for confounders in their analysis or matched the HIV-infected and uninfected women with respect to some key confounders. The majority of the data sets (58 of 66) were judged to be at low risk of bias in the selection of the comparison group of HIV-uninfected women. There was no evidence of publication bias for any of the outcomes included in the analysis with the exception of pre-eclampsia (p = 0.01) (Supplementary material, Figure S1).

Effect of HIV on obstetric haemorrhage
The prevalence of antepartum haemorrhage was higher in HIV-infected than uninfected women in four out of five data sets (Table 1) (Figure 2). There was no evidence for between-study heterogeneity (I 2 : 27.5%, p-value = 0.24). Based on three data sets, there was no evidence for an association between HIV and either placenta praevia (summary OR: 1.02, 95% CI: 0.33-3.14, I 2 : 0%, p-value = 0.70) or placental abruption (summary OR: 1.61, 95% CI: 0.12-20.79, I 2 : 76.1%, p-value = 0.02) ( Figure 2). Thirteen data sets compared the prevalence of postpartum haemorrhage in HIV-infected and uninfected women with ORs ranging from 0.25 to 11.18. The meta-analysis suggests there is no evidence that HIV increases the odds of postpartum haemorrhage (summary OR: 1.28, 95% CI: 0.69-2.38, I 2 : 53.4%, p = 0.01). Similarly, there was no evidence for increased odds of retained placenta with HIV infection (summary OR: 1.28, 95% CI: 0.80-2.06, I 2 : 0%, p = 0.50). One study looked at the association between HIV and postpartum haemorrhage amongst women undergoing a caesar-ean section ( Table 2). There was no evidence of an association

Effect of HIV on hypertensive disorders of pregnancy
Out of the 11 data sets with data on pregnancy-induced hypertension, eight found that HIV-infected women were at increased risk of pregnancy-induced hypertension ( Table 1). The meta-analysis showed some evidence for increased odds of pregnancy-induced hypertension with HIV infection (summary OR: 1.46, 95% CI: 1.03-2.05). However, there was strong evidence for between-study heterogeneity (I 2 : 79.3%, p-val-ue,0.001) (Figure 3). Nine data sets examined the association between HIV and pre-eclampsia; four of these found a higher prevalence in HIV-infected women than uninfected women. There was no evidence that HIV infection was associated with pre-eclampsia (summary OR: 1.04, 95% CI: 0.60-1.79, I 2 : 70.5%, p-value = 0.001). The ORs from the four data sets comparing the prevalence of eclampsia in HIV-infected and uninfected women varied from 0.39 to 38.47. The meta-analysis produced a summary OR of 2.56, however the confidence intervals were very wide (95% CI: 0. 15-44.11) and there was strong evidence for between-study heterogeneity (I 2 : 96.6%, p-value,0.001).
There was one data set from Nigeria which was restricted to caesarean sections. There was no evidence of an association between HIV and postpartum pregnancy-induced hypertension (OR: 0.33, 95%CI 0.01-8.21).

Effect of HIV on dystocia
There were only six data sets where the outcome could be broadly categorised as dystocia (Table 1, Figure 4). One data set from Rwanda found no association between HIV and dystocia (OR: 1.04, 95% CI 0.59-1.82), whilst a study from Thailand indicated that HIV-infected women have nearly eight times the odds of prolonged labour compared with uninfected women (OR: 7.86, 95% CI: 4. 64-13.33). Two data sets reported on abnormal presentation, and there was no evidence for an association between HIV and abnormal presentation in the meta-analysis (summary OR: 1.17, 95% CI: 0.68-2.03, I 2 : 0%, p = 0.50). Conversely, both data sets which compared the prevalence of uterine rupture showed an increased risk in HIV-infected women, giving a summary OR of 3.14 (95% CI: 1.51-6.50, I 2 :0%, p = 0.89).  HIV-infected women have over three times the odds of having puerperal sepsis compared with uninfected women (summary OR 3.43, 95% CI: 2.00-5.85, I 2 : 9.4%, p-value = 0.35). There was also evidence from eight data sets that HIV-infected women had over 2.5 times the risk of endometritis compared with uninfected women (summary OR 2.51, 95% CI: 1.50-4.21, I 2 : 19.6%, p-value = 0.27). The results of the meta-analyses for women who had a caesarean section are presented in Figure 6. The pooled OR from four data sets indicated that HIV-infected women had nearly six times higher odds of suffering from puerperal sepsis compared with their uninfected counterparts (summary OR 5.81, 95% CI: 2.42-13.97, I 2 : 0%, p-value = 0.93). Amongst the ten data sets which contained information on wound infection, the pooled OR was 1.75 (95% CI: 1.20-2.55) although there was weak evidence for between-study heterogeneity (I 2 : 30.1%, p = 0.17). Finally, there were 12 studies which looked at endometritis in HIV-infected and uninfected women; nine found a higher occurrence in HIV-infected women. The meta-analysis showed that HIV-infected women had over double the odds of endometritis than uninfected women (OR: 1.86, 95% CI: 1.28-2.71). There was good evidence for between-study heterogeneity (I 2 : 47.0%, p = 0.04).

Caesarean section
Of the studies which included vaginal and caesarean section deliveries, 13 did not provide information on the proportion of HIV-infected and uninfected women who had a caesarean (one stated that there was no difference in the mode of delivery in HIV-infected and uninfected women, one was restricted to only vaginal deliveries, three only followed women during pregnancy and eight did not provide any information on the proportion of infected and uninfected women having caesareans). Figure 7 shows the relative odds of having a caesarean for HIV-infected compared with uninfected women across the 19 studies which provided data. The ORs varied from 0.40 to 5.55 and there was no evidence that HIV-infected women were more likely to have a caesarean compared with uninfected women [pooled OR: 1.20, 95% CI: 0.81-1.78]. However, there was strong evidence for between-study heterogeneity (I 2 : 88.4%, p,0.001).

Discussion
Our systematic review suggests that HIV increases the risk of intrauterine infections during pregnancy, delivery or the postpar-tum. Studies including vaginal and caesarean deliveries indicated that HIV-infected women had over three times the risk of a puerperal sepsis compared with uninfected women; this figure increased to nearly six amongst studies only including women who delivered by caesarean. The evidence for an association between HIV and other direct obstetric complications was inconsistent. Whilst HIV was associated with an increased risk of antepartum haemorrhage, there was no evidence of an increased risk of placenta praevia, placental abruption, postpartum haemorrhage or retained placenta. Similarly, HIV did appear to increase the risk of pregnancy-induced hypertension, but not of pre-eclampsia and eclampsia. Finally, we found an association between HIV and both uterine rupture and prolonged labour, but not between HIV and other complications of dystocia. The higher risk of intrauterine infections in HIV-infected women is biologically plausible, as the immune suppression associated with HIV increases susceptibility to infection [57]. Caesarean sections may increase the risk of postpartum infection, but caesarean sections were equally common in HIV-infected and uninfected women, and the excess risk of intrauterine infections in HIV-infected women persisted among caesarean only deliveries. Whether the excess risk of endometritis and puerperal sepsis in the intraand postpartum period is directly attributable to the pregnancy or indirectly related to HIV or AIDS-associated infections is uncertain. Intrauterine infections were mostly ascertained from hospital records, definitions were lacking, it was not always clear whether the infection was diagnosed during pregnancy or the postpartum and microbiological examination was not done. The signs and symptoms suggestive of endometritis and puerperal sepsis in intraor postpartum women may have been a direct consequence of the increased prevalence of sexually transmitted infections associated with HIV [58,59]. In the 2008-2010 confidential enquiries into maternal deaths in South Africa, only 6% of maternal deaths in HIV-infected women were attributed to pregnancy-related sepsis, while 62% of deaths were attributed to non-pregnancy-related infections [60]. Without clear definitions misclassification of non-pregnancy-related infections as pregnancy-related, or vice versa, cannot be excluded. We did not find a consistent association between HIV and the risk of either haemorrhage, dystocia or hypertensive diseases of pregnancy. HIV-related thrombocytopenia affects around 10% of HIV-infected individuals and 30% of individuals with AIDS, [61- 63] but it rarely leads to severe bleeding [61,63]. The association between HIV and uterine rupture, concomitant with no associ-ation between HIV and other categories of dystocia, may suggest that delayed care seeking in HIV-infected women plays a role. However, this finding was based on two studies only, and caution is required in its interpretation. The observed associations between HIV and broadly defined categories of complications such as antepartum haemorrhage or hypertensive diseaseswhereas no association was found between HIV and more narrowly defined clinical diagnoses such as placenta praevia, placental abruption, pre-eclampsia or eclampsiasuggests that measurement errors may have occurred. Unfortunately, few studies provided informa-tion on the number of women with more than one diagnosis, and we were not able to pool findings within the broad obstetric categories. The lack of an association between HIV and caesarean section is surprising. Caesarean sections have been recommended to prevent the mother-to-child transmission (PMTCT) of HIV in many regions, [64,65] and the most common indication for caesarean section in HIV-infected women in the studies reviewed was PMTCT of HIV (data not shown). It is possible that clinicians, particularly in low income countries, weigh the health risks associated with caesarean sections against those of PMTCT, and are perhaps more cautious about performing caesarean sections in HIV-infected women. Although recent guidelines recommend that women with very low viral loads who are on ART do not need a caesarean for PMTCT of HIV [66] we would expect a higher rate of caesareans amongst HIV-infected women in high-income countries given the time period in which the studies were conducted. Stratifying the meta-analysis by high and low income countries did not alter the findings (data not shown). We did not systematically review the literature to assess the association between HIV and caesarean sections, however, and some studies may have been missed. This review was comprehensive covering a long time period with no restriction on language, world region or type of study. The studies found were predominantly conducted in tertiary health facilities, however, resulting in the enrolment of a higher risk group of pregnant women. While this will lead to an overestima-tion in the frequency of obstetric complications, it is unlikely this will have affected the relative odds comparing HIV-infected and uninfected women. The main limitation of this review is the poor quality of included studies, none of which were classified as at low risk of bias across all the quality components. Notably, only 25 data sets controlled for key confounders, either through matching HIV-infected and uninfected women or through adjustment in the analysis. Furthermore, very few studies provided information on how the obstetric complications were defined or ascertained. Whether the risks and stigma associated with HIV may result in health professionals reporting complications differentially in HIV-infected and uninfected women is not known, but information bias certainly needs considering. Unfortunately, due to the limited number of studies included in this review, it was not possible to tease out the effect of ART by restricting analyses to studies conducted when ART was available. HIV-infected women are thought to be eight times more likely to die in pregnancy or the postpartum than HIV-uninfected women, [67,68] and the excess mortality attributable to HIV among HIV-infected women is about 994 per 100,000 pregnant women [67]. While the increased risk of puerperal sepsis and endometritis in HIV-infected women contributes to this, direct obstetric causes only explain a tiny fraction of the excess mortality. Studies on the causes of death in pregnant or postpartum women by HIV status are scarce, except for the South African confidential enquiries, the most recent of which cover 2,756 and 1,149 maternal deaths in HIV-infected and uninfected women respec-tively [60]. The 2008-2010 confidential enquiries suggest that most deaths in HIV-infected pregnant and postpartum women are due to non-pregnancy-related infections, including pneumonia, tuberculosis and meningitis [60]. Although anaemia is thought to be exacerbated by HIV, [7] the confidential enquiries find a similar proportion of maternal deaths attributable to severe anaemia in HIV-infected and uninfected women (8% and 10% respectively) [60]. It is essential to ensure that both HIV-infected and uninfected pregnant woman have ready access to high quality antenatal and delivery services to correctly diagnose and manage direct obstetric complications when they occur. HIV-infected pregnant women will also benefit from prophylactic antibiotics during labour to reduce their risk of intrauterine infections [69]. However, given that most of the excess mortality associated with HIV in pregnancy is directly related to HIV rather than to a higher risk of obstetric complications, the greatest impact on pregnancy-related mortality will come from ensuring that HIV-infected pregnant women have adequate access to ART [70]. The World Health Organization recommends the provision of lifelong ART treatment for all HIVinfected pregnant women with a CD4 count below 350 cells/mm 3 , but many countries are still transitioning to these guidelines [71,72]. Scaling up Option B+, where all pregnant mothers start ART regardless of their CD4 cell count and then continue taking it for life, has been proposed as an additional strategy to benefit maternal health; however, any benefit must be carefully measured against the potential pitfalls which include the high financial costs of such a programme and possible poor adherence to ART amongst women who perceive themselves to be healthy [73]. Figure S1 Funnel plot illustrating potential publication bias for data sets which look at the association between HIV and pre-eclampsia. (EPS) File S1 Search Strategy. (DOCX)