The effect of pregnancy on survival in women infected with HIV a systematic review of the literature and meta-analysis


Correspondence: Dr P. Brocklehurst, Perinatal Trials Service, National Perinatal Epidemiology Unit, Radcliffe Infirmary, Woodstock Road, Oxford OX2 6HE, UK.


Objective To investigate the effect of pregnancy on disease progression and survival i? women infected with HIV by a systematic review of the literature and meta-analysis.

Methods Appropriate publications were identified using electronic and hand searching of relevant journals from 1983 to 1996. Studies were included in the review if they were cohort studies, either prospective or retrospective, or case-control studies which investigated disease progression of pregnant women infected with HIV and included a control group of non-pregnant women infected with HIV for comparison. Methodological quality was assessed for each study. Data were extracted for predetermined outcome measures. Sensitivity analyses were performed to explore the association between pregnancy and disease progression for the following study characteristics: clinical setting (developed or developing countries), methodological quality (high or poor) and whether studies had controlled for potential confounding.

Results Seven studies, all prospective cohorts, were eligible to be included in the rediew. The summary odds ratio for the risk of an adverse maternal outcome related to HIV infection and pregnancy were as follows: death 1.8 (85% CI 0.99–3.3); HIV disease progression 1.41 (95% C1 0.85–2.33); progression to an AIDS-defining illness 1.63 (95% CI 1.00–2.67) and fall of CD4 cell count to below 200×106/L 0.73 (95% CI 0.17–3.06). Sensitivity analyses showed that HIV progression in pregnancy was significantly more common in a developing country setting (odds ratio 3.71, 95% CI 1.82–7.75) than in developed countries (odds ratio 0.55, 95% 0.27–1.11) and also significantly more common in high quality studies when compared to low quality ones, odds ratios 3.71 (95% CI 1.82–7.57) and 0.55 (95% CI 0.27–1.11), respectively. However, there appears to be less progression of HIV disease and progression to AIDS when studies attempted to control for confounding by matching or restriction techniques, although this was not statistically significant in either case.

Conclusions The findings of this review have implications for women infected with HIV who are pregnant or are considering a pregnancy. There does appear to be an association between adverse maternal outcomes and pregnancy in women infected with HIV, although this association is not strong. The relation may be due to the result of bias including residual confounding. Further large scale observational studies with long term follow up are required before this issue can be fully resolved.


The objective of this study was to investigate whether pregnancy increases the risk of disease progression in women infected with HIV compared with women infected with HIV who do not become pregnant by systematic review of the literature. The adverse outcomes for these two groups which were pre-specified in this review are: maternal death; maternal death that is HIV attributed; HIV progression (defined as a progression from one Centres for Disease Control stage1 of infection to another); the development of an AIDS defining illness; and a fall in CD4 cell count to below 200×106/L.


The effect that a pregnancy may have on a woman's HIV disease state is an important consideration when counselling women infected with HIV about planned or future pregnancies. Although a woman's decision making process may be concerned with the effect on the baby, the effect a pregnancy may have on her own survival will be an important factor2. In addition, in developing countries infant survival has been shown to be adversely effected by maternal death3,4 with infants being over three times more likely to die if their mother had died.

Early reports of pregnancy in women infected with HIV suggested that pregnancy had a detrimental effect on the mother's immune system, hastening disease progression and shortening the time to death5. This association was almost certainly the result of selection bias whereby women infected with HIV were identified because they already had a child with AIDS6. This selected women with more advanced disease whose prognosis was already poor. In addition, the failure to recognise the early symptoms and signs of pneumocystis carinii pneumonia in pregnancy resulted in a very high mortality rate associated with this particular infection.

At present there appears to be a consensus that pregnancy does not have a major adverse effect on the survival of women infected with HIV7. However, many of the studies which inform this consensus have compared groups of pregnant women infected with HIV with groups of pregnant women who are not infected with HIV.8–12 These studies have tended to focus on immunological function, such as changes in T cell population subsets, rather than on more clinically relevant outcomes such as mortality or clinical disease progression.

Reviews in this area have not been systematic13–15 and therefore only may have found a proportion of the literature available, possibly resulting in biased conclusions. The benefits of systematic review have been extensively reviewed16, particularly in the field of randomised controlled trials. The issue of non-publication of data from observational studies has not been extensively explored. It seems likely, however, that small studies which demonstrate little or no association may be less likely to be submitted for publication, and those that are submitted for publication may be less likely to be accepted by journals with a large readership. If this is the case for trials then it is even more likely to be true for observational studie17. As a consequence, reviews which do not systematically search the literature are at a greater risk of finding a selected group of studies which may support a prevailing viewpoint. As a result, the conclusions of such reviews are likely to be biased and further support the prevailing viewpoint, which will encourage the selection of papers which continue to support this view being accepted for publication and soon.


A study protocol was written which detailed the process by which the review would be produced. This protocol included the search strategy to be employed, the criteria for including and excluding studies, the outcome measures to be included in the review and the methods of analysis including any sensitivity analysis to be used.

Search strategy

All publications describing studies of HIV infection and maternal outcomes were sought through computerised searches in MEDLINE, EMBASE and the Science Citation Index [MESH terms: HIV and pregnancy] from January 1983 to December 1996. All references quoted in these and other papers which were suitable were then sought. Hand searching was conducted of all the International Conference on AIDS abstract books to 1996; in addition, because abstracts may take some time to appear as full journal articles the first authors of abstracts from the XI International Conference on AIDS, Vancouver, 1996 were contacted and asked if they could provide any further data. The following journals were searched by hand back to 1983; AIDS, AIDS Care, Genitourinary Medicine, Journal of Acquired Immunodeficiency Syndrome and Human Retrovirology, Journal of Infectious Diseases, Annals of Internal Medicine, British Medical Journal, Journal of the American Medical Association, Lancet, New England Journal of Medicine, American Journal of Obstetrics and Gynecology, British Journal of Obstetrics and Gynaecology and Obstetrics and Gynaecology.

Study selection

Studies were selected for inclusion in the systematic review if they were cohort studies, either prospective or retrospective, or case-control studies which were investigating disease progression of pregnant women infected with HIV. In addition, the studies needed to include an appropriate control group for comparison, so that the effect of pregnancy in women infected with HIV could be compared to a group of women infected with HIV who have not been pregnant. Control of confounding variables was not necessary for inclusion.

The paperheport had to include appropriate numerator and denominator data so that odds ratios with 95% confidence intervals could be calculated. Proportions were acceptable if the denominator was provided as the approximate number could be calculated and the odds ratio derived.

Studies excluded

Studies without a control group and papers which failed to provide the raw data for analysis were excluded. Studies which did not provide information on the pre-specified outcome measures were also excluded regardless of methodological quality.

Data extraction and quality assessment

Copies of all selected publications were obtained. A pre-determined set of outcomes measures was defined as being clinically meaningful and data were abstracted from the publications onto specially designed data collection forms by each of the investigators independently. Differences were resolved by discussion and consensus.

Methodological quality was assessed by a predefined set of criteria designed to grade each report. These criteria differed depending on the type of study being assessed. The quality criteria included whether the paper included a sample size calculation, the quality of the selection of the cohort or cases and controls, the process of measurement of the outcome or exposure so that bias could be minimised, the extent of follow up and the methods used to control for confounding, in particular whether matching for time since HIV diagnosis, age or a measure of immune function occurred. Scores were awarded for each criteria met, the possible maximum score was 12. Studies were graded high quality if there score was 7 or above and low quality if there score was 6 or below. Abstracts were not graded for quality and were included in the low quality category for the sensitivity analyses (see below).

In order to standardise the results, all measures of effect were expressed as odds ratios and 95% confidence intervals calculated if this had not been done already.

Data synthesis

Odds ratio measures with 95% confidence interval are presented in graphical form. The summary measure of overall effect is a fixed effect model (Peto). The χ2 test for heterogeneity assesses the extent to which the included studies disagree with each other. Significant heterogeneity implies variation in the result which cannot be accounted for by chance and therefore summary measures are interpreted with caution.

Sensitivity analyses

The extent to which the summary measures are sensitive to variations in other characteristics of the studies were explored with three sensitivity analyses. These were: the clinical setting (developed or developing country), methodological quality (high or low) and whether the studies had employed restriction or matching techniques in their design. Sensitivity analyses were performed by repeating the meta-analysis but including only studies in developed countries and then only studies in developing countries. Where this analysis resulted in different results for the two settings this is reported. This process was then repeated for the quality of the studies and the confounding control issue. This final sensitivity analysis was undertaken because matching and restriction are techniques which can be used when the cohort is being assembled. As such the crude odds ratio at the end of the study is already adjusted for those confounding factors which were matched on or restricted for. Crude odds ratios can be incorporated into a meta-analysis. Alternative methods for controlling for confounding include stratification and regression analysis. These techniques are used during the analysis stage and result in adjusted odds ratios. Adjusted odds ratios cannot be incorporated into a meta-analysis.

In addition, the hypothesis that any association between pregnancy and maternal survival may be different depending on whether the mother is asymptomatic or has symptomatic HIV disease at the time of the pregnancy was explored.


Seven studies were eligible for inclusion in the meta-analysis (Table l). Although retrospective cohort and case-control studies were eligible for inclusion, none were found. The study reported by Schoenbaum et al. has two references as both of these reports are conference abstracts and they provide information about different maternal outcomes. Both abstracts contain information about HIV disease progression so the data from the later abstract is included in the tables and the meta-analysis for this particular outcome.

Eight studies were excluded from the review. The most common reason for exclusion was lack of an appropriate control group. A summary of the excluded studies is shown in Table 2. Case series are not included in the table as they were too numerous to list.

Table 2.  Characteristics of studies excluded from the review.
StudyReason for exclusion
Biggan et al. 198912Inappropriate control group: HIV infected pregnant women vs HIV uninfected pregnant women.
Burns et al. 199511Inappropriate control group: HIV infected pregnant women vs HIV uninfected pregnant women
Duarte et al. 199419[Abstract] insufficient data
MacCallum et al.198932[Abstract] insufficient data
Miotti et al. 199210Inappropriate control group: HIV infected pregnant women vs HIV uninfected pregnant women
Nachman 198733Inappropriate control group: HIV infected pregnant women vs HIV uninfected pregnant women
Schaeferetal.198834Inappropriate control group: HIV infected pregnant women vs HIV uninfected pregnant women
Temmerman et al.19959Inappropriate control group: HIV infected pregnant women vs HIV uninfected pregnant women

Summary odds ratios

A summary of the meta-analysis of the eligible studies is presented in Fig. 1 which shows the summary odds ratio for each of the pre-specified outcomes. Those summary odds ratios which fall to the right of the line of no effect suggest that pregnancy increases the risk of the outcome. However, all the 95% confidence intervals cross the line of no effect. Figure 2 details the studies contributing to each of the summary estimates of maternal death, HIV progression and progression to an AIDS defining illness. All of the maternal deaths were HIV-attributed. Only one study provided data on CD4 counts which satisfied the requirements of the review. One other study did provide data on CD4 counts, however they reported a fall in CD4 count to less than 100×106/L or 50×106/L.

Figure 1.

Summary odds ratio for each pre-defined maternal outcome measure.

Figure 2.

Summary and individual study odds ratios for each of the pre-defined maternal outcome measures.

Sensitivity analysis

Considerable heterogeneity is to be expected when observational studies of different population groups are included in one analysis. In order to reduce heterogeneity, studies were grouped and analysed by setting, either developed or developing, and quality, either high or low.

Clinical setting (developed and developing country)

Only one study was conducted in a developing country (Table 1). There was no difference in the conclusions concerning progression to an AIDS defining illness or in the number of maternal deaths. HIV disease progression of pregnant women, however, appeared to be significantly more common in a developing country setting (odds ratio 3.71, 95% CI 1.82–7.57) than in developed countries (summary odds ratio 0.55,95% CI 0.27–1.11).

Table 1.  Characteristics of studies included in the review and meta-analysis. IVDU = intravenous drug user; CDU = Centres for Disease Control.
StudyCountryParticipants (all HIV-infected)Mean follow upNotesQuality score
  1. *Also reported on group of infected women who had a spontaneous or induced abortion (not included in this review)

  2. † Additional data provided by the first author.

Berrebi etal. 199025France35 pregnant and 64 non-pregnant women*Range = 3–36 months (mean not specified)Restriction–Women either CDC stage II or III6
Bledsoeetal. 1990 (abstract)26USA55 pregnant and 51 non-pregnant women. Majority of women were IVDUs (no. not stated)Pregnant =16 months; non-pregnant = 24 monthsNo mention of control of confounding
Castellieta/. 1989 (abstract)Italy15 pregnant and 30 non-pregnant women. All CDC Stage II at enrolment. 82% of women were IVDUsPregnant = 9 months; non-pregnant = 13.5 monthsNo mention of control of confounding
Deschamps et al. 199328Haiti44 pregnant and 96 non-pregnant women. Both groups asymptomatic at entryAll women = 44 monthsLogistic regression for age and parity10
Hocke et al. 199529France57 pregnant and 114 non-pregnant womenPregnant = 61 months non-pregnant = 50 monthsMatched by age, CD4 count and year of diagnosis. Stratification by initial CD4 count11
Schoenbaum et al. 1987 (abstract)30USA15 pregnant and 19 non-pregnant women.All women = 12 monthsRestricted–IVDUs on methadone programme
Schoenbaum et al. 1989 (abstract)31USA85 pregnant and 56 non-pregnant womenAll women = 30 monthsStratification by number of live births
Weisseneta/. 199718Switzerland32 pregnant and 416 non-pregnant womenPregnant = 60 months; non-pregnant = 54 months (for mortality end point)Matched by age and CD4 at entry Logistic regression of baseline differences11

Methodological quality

The quality scores which were allocated to each of the studies are presented in Table 1. There was evidence of considerable heterogeneity between the studies in terms of quality. In the sensitivity analysis high quality studies when compared with those of lower quality suggest that pregnant women were more likely to have progression of HIV disease stage (summary odds ratios 3.71, 95% CI 1.82–7.57 for high quality studies and 0.55, 95% CI 0.27–1.11 for low quality studies) and were more likely to progress to AIDS (summary odds ratios 2.03, 95% CI 1.16–3.35 for high quality studies and 0.71, 95% CI 0.24–2.09 for low quality studies), however these differences were not statistically significant. Study quality did not appear to affect the relationship between pregnancy and maternal death (summary odds ratios 1.80, 95% CI 0.97–3.33 for high quality studies and 1.9, 95% CI 0.10–35.21 for low quality studies).

Control of confounding

Some studies attempted to control for confounding by matching or restriction techniques (Table 1). When this factor was incorporated into a sensitivity analysis there appeared to be less progression of HIV disease and less progression to AIDS in those studies which had used these techniques compared to those which had not, although this difference was not statistically significant (Fig. 3).

Figure 3.

Summary odds ratio of studies which attempted to control for confounding by matching or restriction techniques compared with those which had not used matching or restriction.

One study stratified for parity. Schoenbaum et al. found a significant increase in risk of progression to AIDS when mothers had two or more live births. Twenty four percent of mothers with two or more live births progressed to AIDS, compared with 5% of mothers with one live birth. In this study, however, 32% of women with no live births progressed to AIDS, suggesting that if the women were comparable at study entry then pregnancy may prevent disease progression.

The study by Hocke et al. stratified the effect of pregnancy on survival, disease progression and fall in CD4 count to less than 200×106/L by initial CD4 count (less than 350×106/L and at least 350×106/L) and found that this made no significant difference to their conclusions which were that pregnancy appeared to have no effect on any of these three measures.

One study controlled for potential confounding at the analysis stage. Deschamps et al. controlled for age and parity through logistic regression and found that this had a negligible effect on maternal mortality outcomes, adjusted odds ratio 2–1 (95% CI 0.9–4.7) compared with the unadjusted odds ratio 2.3 (95% C1 0.9–5.7).


Although all of the studies included in the review were observational, and therefore generally more prone to bias, it is highly unlikely that the outcomes measured would be affected by the knowledge of whether or not a woman infected with HIV was or had been pregnant. There is the potential for bias if adverse events of HIV-infected mothers such as progression from one Centres for Disease Control stage to another is reported earlier as a consequence of closer follow up during their pregnancy and the post-partum period. In the studies included in this review, however, this seems unlikely.

There is great potential for selection bias when trying to measure outcomes of HIV-pregnant women infected with HIV with those HIV women infected with HIV who are not pregnant. There may be a variety of reasons why some women infected with HIV do not become pregnant, and this may be related to factors influencing their survival. For example, women with severe symptomatic HIV or AIDS may be too ill to become pregnant or if they do conceive may be more likely to chose to terminate the pregnancy, although if pregnancy does affect survival the relative effects of a termination or miscarriage and that of a full term pregnancy may be similar. There may also be differences with regards to socioeco-nomic status, ethnic origin, age and transmission group between pregnant and non-pregnant women. An American study2, monitoring a large cohort of HIV-infected women, found that those who were under 25 years of age and those who were black were more likely to become pregnant. Another source of selection bias is that pregnant women infected with HIV may be obtaining medical care for the first time because of their pregnancy rather than their HIV infection, and may have been unaware of their HIV status prior to antenatal screening. This may have two effects. Firstly, the non-pregnant group may be recognised because they have presented with some minor HIV related symptoms, whereas the pregnant group are recognised because they are screened in the antenatal clinic. Thus the two groups may not be comparable at study entry with respect to their HIV disease status (although this difference is likely to be more subtle than a difference in their Centres for Disease Control stage). Secondly, many women would have been enrolled into studies late into their pregnancy or even at delivery and therefore any progression of disease that had happened during the pregnancy would be unaccounted for.

Another source of bias in this review is that of publication bias. Studies which find no association or negative associations between HIV infection and maternal outcomes may be less likely to be published either because they were not submitted for publication or because journals are less likely to publish them. This phenomena has been well documented with trials, and there is no reason to believe it is not the case with observational studies also16. Publication bias cannot be easily demonstrated but the possibility of publication bias should be considered in interpreting the results of any review.

Women in developed countries are more likely to be on anti-retroviral therapy and opportunistic infection prophylaxis, which may be discontinued during pregnancy and which may lead to an increased risk of disease progression in this setting when the cases and controls are compared. Weisser et al.18 noted that pregnant women infected with HIV, when compared with non-pregnant women with HIV had a significant increase in risk of recurrent bacterial infection, hazard ratio 7.98 (95% CI 1.73–36.8). The authors concluded that this was most probably due to pregnant women discontinuing their prophylactic treatment for pneumocystis carinii pneumonia due to fear of its toxic effects on the fetus.

The poor maternal prognosis in developing countries may be due to factors other than just HIV infection, such as lack of antenatal care, malnutrition and a greater risk of infection after delivery19.

The differences between findings of the high and low quality studies could be explained by two factors. Follow up of both cases and controls and control of confounding were included in the quality scoring system. Therefore these studies are more likely to reflect the ‘true’ underlying relationship more than the lower quality studies.

The control of confounding in the study design may not necessarily eliminate confounding. For example, in developed countries drug use is probably a major confounding factor and restriction of entry to the study to drug users would seem an appropriate way of controlling its effect on the outcome. Drug users may not always admit to using drugs and pregnancy can effect drug taking patterns: for example, there is a decline of intravenous drug use reported during pregnancy20. In addition, drug users are not a homogenous group of women. Drug using women may occasionally use marijuana or regularly inject heroin. The pregnancy outcomes of these two categories of drug users is likely to be different.

Immunological changes during pregnancy have been well documented. The percentage and absolute number of CD4 cells have been reported to decrease, returning to their normal range three to four months post delivery5. There was concern that if an HIV-infected woman's CD4 cell count dropped substantially during pregnancy, she would be at a greater risk of opportunistic infections and other AIDS defining illnesses. The effect that this has on pregnant women with HIV is still unclear, with some studies reporting a detrimental effect of pregnancy on immune function11,12 and others reporting no significant effect8,10. This review was unable to determine the effect pregnancy had on the immune system of women infected with HIV.

To date, studies measuring the association between HIV infection and maternal outcomes have involved relatively small numbers and will have been unable to detect modest differences in survival or disease progression. This review has reported the available data from all the available studies and is still unable to determine with any certainty the effect of pregnancy on HIV disease progression and survival, although it seems unlikely that any major effect exists. There is a need for further large observational cohorts of pregnant and non-pregnant women infected with HIV to be undertaken which include long-term follow up. Data on immune function should be available before pregnancy. Ideally, a known date of seroconversion or at least an estimate should be established as date of diagnosis or CD4 count, by themselves, may not accurately reflect disease stage.

The majority of women who become infected with HIV are of reproductive age22. More of these women may be deciding to become pregnant due to the availability of interventions which decrease the risk of mother-to-child transmission23. If clinically indicated, mothers need to be encouraged to continue taking any relevant opportunistic infection prophylaxis unless this is specifically contraindicated in pregnancy24. Pregnant women and those women thinking about becoming pregnant need to be counselled and advised about the possible risks to their own health as well as any transmission risks before they can make an informed decision on how to proceed.

In summary, there may be an association between HIV disease progression and pregnancy, although this association is not strong. The observed relationship may be due to the result of bias including residual confounding and further, larger studies will be necessary before this issue can be resolved.


We would like to thank Ms L. Kumiega for her help with the literature searching, and Dr C. Hocke29 and Dr C. Rudin18 for providing us with extra information from their studies which was relevant to the review. In addition we would like to thank Dr J. Stephenson and Ms A. Petruckevitch.