Effect of HIV Infection and Antiretroviral Treatment on Pregnancy Rates in the Western Cape Province of South Africa

Abstract

Background

Previous studies suggest that untreated human immunodeficiency virus (HIV) infection is associated with a reduced incidence of pregnancy, but studies of the effect of antiretroviral treatment (ART) on pregnancy incidence have been inconsistent.

Methods

Routine data from health services in the Western Cape province of South Africa were linked to identify pregnancies during 2007–2017 and maternal HIV records. The time from the first (index) pregnancy outcome date to the next pregnancy was modeled using Cox proportional hazards models.

Results

During 2007–2017, 1 042 647 pregnancies were recorded. In all age groups, pregnancy incidence rates were highest in women who had started ART, lower in HIV-negative women, and lowest in ART-naive HIV-positive women. In multivariable analysis, after controlling for the most recent CD4⁺ T-cell count, pregnancy incidence rates in HIV-positive women receiving ART were higher than those in untreated HIV-positive women (adjusted hazard ratio, 1.63; 95% confidence interval, 1.59–1.67) and those in HIV-negative women.

Conclusion

Among women who have recently been pregnant, receipt of ART is associated with high rates of second pregnancy. Better integration of family planning into HIV care services is needed.

(See the Editorial Commentary by Marston and Gregson on pages 1919–21.)

In evaluating the demographic impact of human immunodeficiency virus (HIV), it is important to assess how fertility rates in HIV-positive women compare to those in HIV-negative women and how initiation of antiretroviral treatment (ART) affects subsequent fertility. Estimates of the effect of HIV on fertility are also important in estimating the need for services for prevention of mother-to-child transmission (PMTCT) and in assessing the coverage of PMTCT services, which in turn are important in estimating annual new cases of mother-to-child transmission. Knowledge of the effect of HIV on fertility is also important in assessing the extent to which HIV prevalence trends in pregnant women (the most widely used source of HIV prevalence data in many generalized HIV epidemic settings) reflect the HIV prevalence in the general population [1]. Most previous studies have noted that fertility is substantially reduced in HIV-positive women [2–7], with the extent of this reduction being proportional to the degree of immune suppression [8–11]. Some studies have found significant increases in fertility after starting ART [8, 12, 13], but other studies have found little or no effect of ART on fertility [14–16]. There is a lack of clarity on the extent to which ART restores fertility, and a recent review concluded that current data are insufficient to characterize the effect of ART on fertility [17].

Most of the research on the effect of HIV on fertility in Africa has been based on data from eastern and central African countries [4] and settings in which contraceptive use is low [18]. Some studies have suggested that HIV may have less impact on fertility in settings in which contraceptive use is high [4, 19], but this has not been well established. Most studies on the effect of HIV on fertility in Africa have relied on self-reported data collected as part of household surveys, rather than on health service data. Self-reported data could be biased if there is substantial misreporting of pregnancies and pregnancy outcomes [20].

The Western Cape province in South Africa provides a novel opportunity to assess the impact of HIV on fertility, using health service data. In this setting, 59% of sexually active women use modern contraceptive methods [21], and HIV prevalence in pregnant women attending public antenatal clinics has been high (19% in 2015 [22]). During 2007–2017, the number of women aged 15–49 years in the Western Cape increased from 1.58 million to 1.82 million, and the fraction of these women who were HIV positive increased from 9.0% to 13.1%, while ART coverage in women increased from 12.9% to 58.7% [23]. Each user of the public health service has a unique patient identifier, which can be used to link patient records across public services, thus allowing for comparison of pregnancy and fertility rates in women in different HIV stages. This study aims to assess the effect of HIV and ART on pregnancy rates in the Western Cape.

METHODS

Data Sources and Definitions

The Western Cape Government Health Department maintains a number of electronic record systems for the purpose of managing hospital and primary care administration, drug dispensing, and laboratory data [24]. Records for individual patients are linked across systems through a unique patient identifier, which was initially introduced in hospitals and has been rolled out to primary care clinics since 2007. This linkage process is managed by the Provincial Health Data Centre, which provided the data for this analysis.

A pregnancy is identified if there is an antenatal clinic visit; a rhesus antibody test; an International Classification of Diseases, Tenth Revision, code indicating pregnancy or an abortive outcome of pregnancy; drugs dispensed for a termination of pregnancy; or a birth recorded on the birth register. Based on these data sources, a pregnancy outcome and pregnancy outcome date are inferred for each woman. In a substantial fraction of cases, it is not possible to determine a pregnancy outcome (Table 1), and a pregnancy confidence score is calculated to measure the degree of confidence that the evidence truly indicates a pregnancy. In the main analysis, all possible pregnancy events are considered, but in a sensitivity analysis we restrict the analysis to those pregnancies with a pregnancy confidence score of 0.7 or higher (termed “probable pregnancies”). These pregnancies had at least 1 evidence that was considered to be high confidence and indicative of pregnancy on its own (eg, rhesus antibody testing) or multiple moderate-confidence evidences (eg, antenatal clinic visits); further details are provided in the Supplementary Materials. For pregnancies with no outcome recorded, in most cases the outcome date is predicted to be 41 weeks after the first evidence of pregnancy. In cases where additional data are available, the outcome date is inferred by using the last menstrual period date, estimated delivery date, or gestational age.

A woman is identified as HIV positive if there is a positive enzyme-linked immunosorbent assay (ELISA), a CD4⁺ T-cell count, a viral load test result, a record that combination ART was dispensed, or registration on the TIER.net or equivalent database (ie, databases for the management of HIV-positive patients). The date of ART initiation is taken as the earlier of the first recorded date on which ART was dispensed and the recorded date of ART initiation. As rapid diagnostic test results are not electronically recorded, exact dates of diagnosis cannot be inferred in most cases.

Eligibility Criteria

Analysis is restricted to women who had evidence of at least one pregnancy, with the first pregnancy outcome date occurring between 1 January 2007 and 31 December 2017. Women were excluded if they were aged <15 or >49 years at the first pregnancy outcome date. In the analysis of times to conception after the first conception, the analysis was further limited to women whose first pregnancy outcome date was 9 months prior to the end of 2017, in order to exclude women who were unlikely to have had a second pregnancy outcome before the end of 2017.

Statistical Methods

The total number of pregnancies in the Western Cape public health sector in each year was compared to estimates of the Thembisa model (version 4.1), a combined demographic and HIV model fitted to Western Cape data [23]. The model is calibrated to antenatal and household survey HIV prevalence data, as well as to reported numbers of HIV-positive patients receiving ART in the Western Cape, and is the official source of Joint United Nations Programme on HIV/AIDS (UNAIDS) estimates for South Africa.

Consistent with the approach adopted in several previous evaluations of the effect of HIV on fertility [8, 9], we then assessed factors associated with the time from the first pregnancy outcome date to the second pregnancy conception date (estimated as 270 days before the second pregnancy outcome date), using Cox proportional hazards models. Follow-up was censored at the earlier of the second pregnancy conception date and 31 March 2017. ART and CD4⁺ T-cell count were treated as time-varying covariates. Women with no evidence of HIV infection in the database were assumed to be HIV negative for the entire follow-up, while women whose first date of HIV evidence occurred after their first pregnancy date were included in the follow-up only from the date of their first CD4⁺ T-cell count.

Because of concern that women who died or migrated out of the province might be incorrectly classified as residing in the Western Cape during follow-up, a sensitivity analysis was conducted in which women were censored 2 years after their last date of contact with the health system, if this occurred before the original censoring date of 31 March 2017 (HIV-positive women could be censored later, 270 days before the date of their last HIV-related laboratory test, if this occurred before the original censoring date). Because it was anticipated that hazards in different covariate categories would not be proportional, Cox proportional hazard models were also fitted separately for the following 4 durations after the index pregnancy: 0–1, 2–3, 4–5 and 6 or more years. All statistical analyses were performed using Stata, version 15.1 (StataCorp, College Station, TX).

Ethics

The analysis was undertaken as part of a broader evaluation of the impact of expanded access to ART in pregnancy, approved by the Human Research Ethics Committee at the University of Cape Town (HREF 541/2015). Informed consent was not sought, as the data were collected through routine health services and anonymized before they were shared with the researchers.

RESULTS

From the beginning of 2007 through the end of 2017, 1 042 647 pregnancies were identified in the Western Cape. Of these, 57.6% resulted in a live birth, 2.7% ended in stillbirth or miscarriage, and 2.9% ended in induced abortion; for 36.2% the pregnancy outcome was uncertain (Table 1). The quality of pregnancy recording improved over time, with more live births being recorded in recent years and fewer pregnancies with uncertain outcomes in recent years (Figure 1). After excluding pregnancies that were known to have resulted in miscarriage, stillbirth, or abortion, total pregnancies in the Western Cape were consistently lower than the total number of births estimated by the Thembisa model (Figure 1), reflecting underrecording and births occurring in the private sector. The proportion of pregnancies in which women were known to be HIV positive was 16.9%, similar to the levels of 16.1%–18.9% measured in annual antenatal surveys in the Western Cape over 2008–2015 [22].

After excluding women whose index pregnancy outcome date was after 31 March 2017, there were 697 456 who had at least 1 pregnancy (Table 1). Of these, 171 235 women were recorded as having a subsequent pregnancy, with >2.91 million years of follow-up, yielding a crude pregnancy incidence rate of 5.89 events per 100 woman-years. When stratified by age, pregnancy incidence rates were consistently lower in untreated HIV-positive women than in HIV-negative women but were higher in HIV-positive women who had started ART than in HIV-negative women (Figure 2A).

In multivariable analysis, pregnancy incidence remained lower in untreated HIV-positive women than in HIV-negative women, with the extent of the difference being greatest at low CD4⁺ T-cell counts (Table 2). HIV-positive women who were receiving ART had a higher pregnancy incidence than HIV-positive women who were untreated (adjusted hazard ratio, 1.63; 95% confidence interval [CI], 1.59–1.67). Women also had higher pregnancy incidence rates if their index pregnancy outcome was a pregnancy loss or their infant died in the first month of life. The year of the index pregnancy was also a significant determinant of the pregnancy incidence rate, reflecting improvements in the recording of pregnancy over time. In a sensitivity analysis, in which all pregnancies with confidence scores of <0.7 were excluded, the number of women decreased to 502 073, and the fraction of index pregnancies with unknown outcomes decreased to 9.3%, but similar estimates were obtained in the multivariable model, although the effect of ART was slightly weaker (adjusted hazard ratio, 1.42; 95% CI, 1.37–1.46; Table 2). Similar results were obtained when the analysis was limited to live births (Supplementary Table 11).

In the sensitivity analysis, in which women were censored 2 years after their last contact date, pregnancy incidence rates in untreated HIV-positive women and HIV-negative women increased substantially (Figure 2B), and the ratio of the pregnancy incidence rate in treated women to that in untreated HIV-positive women was 1.35 (95% CI, 1.32–1.38), which is lower than the ratio of 1.63 in the main analysis (Supplementary Table 1). Results were similar to those in the main analysis when ART effects were allowed to vary by CD4⁺ T-cell count, although the effect of ART appeared more substantial at low CD4⁺ T-cell counts than at high CD4⁺ T-cell counts (Supplementary Table 2). Pregnancy incidence rates were not found to be proportional across covariate categories, with the effects of age and index pregnancy outcome being particularly divergent at different durations after the index pregnancy (Supplementary Table 3). However, the effects of CD4⁺ T-cell count and ART on pregnancy incidence rates were roughly similar at different durations after the index pregnancy. Effects of HIV and ART were also similar to those estimated in the main analysis when considering separately the periods before and after the recommendation of ART for all HIV-positive pregnant women in 2013 (Supplementary Table 4) and when considering alternative specifications of time effects (Supplementary Table 5). The effect of ART was also similar when the analysis was limited to the period up to the end of 2010 (adjusted hazard ratio, 1.55; 95% CI, 1.41–1.70; Supplementary Table 6), a period in which South African guidelines recommended promotion of contraception to HIV-positive women receiving ART, owing to concerns regarding the potential teratogenicity of efavirenz [25].

DISCUSSION

In this analysis, we found that pregnancy incidence rates in recently pregnant HIV-positive women receiving ART were higher than in recently pregnant HIV-negative women. To our knowledge, the only previous study that directly compared pregnancy incidence rates in HIV-positive women receiving ART and HIV-negative women is a study from Zimbabwe, which found lower pregnancy incidence rates in women receiving ART, compared with HIV-negative women [26]. Other studies that compared pregnancy incidence rates in HIV-positive women receiving ART to those in untreated HIV-positive women have produced inconsistent results: some have suggested that ART increases fertility, even after controlling for the most recent CD4⁺ T-cell count [8], while others suggest that ART has no effect on fertility after controlling for the most recent CD4⁺ T-cell count (or equivalently, that any improvement in fertility after starting ART is due entirely to improved immune function) [16, 27]. Our data are consistent with the former, suggesting that there are mechanisms independent of physiological improvements in health that lead to increased pregnancy incidence after ART initiation. However, our data relate only to women who have been pregnant, and it is possible that the effect of ART may be different in women who have not previously been pregnant. For example, in an analysis of East African data, the incidence of pregnancy in HIV-positive women who were recently pregnant and receiving ART was 1.42 times that in those who were recently pregnant and not receiving ART (consistent with our analysis), but in HIV-positive women who were not recently pregnant the incidence of pregnancy during ART was 0.94 times that in HIV-positive untreated women [27]. This could potentially explain some of the inconsistencies between these studies.

There are several possible explanations for the high rates of pregnancy incidence in women on ART. Some studies show increased fertility desires following ART initiation [15, 28, 29]. In addition, one study found that there was an association between fertility desire and interest in early ART initiation [30], suggesting that HIV-positive women who wish to have further children may be more motivated to start ART. Other unmeasured confounders, such as marital status (which is associated both with ART uptake [31] and pregnancy incidence [8, 9]), could also partly explain the observed association. Although HIV-diagnosed individuals tend to report reduced levels of unprotected sex after ART initiation [32], it is possible that this represents switching from more-effective hormonal contraceptive use to condoms, which are less effective in preventing pregnancy [33].

We do not have data on pregnancy incidence rates in HIV-positive women before diagnosis. However, previous studies suggest that HIV-diagnosed women are likely to have reduced desire for further children relative to HIV-negative women and HIV-positive women with undiagnosed infection [34–36] or are more likely to use contraception than HIV-positive women with undiagnosed infection [35, 37]. Given this prior research, it might be considered surprising that we did not see a more marked reduction in pregnancy incidence in HIV-diagnosed women before ART initiation, compared with HIV-negative women. For example, at CD4⁺ T-cell counts of ≥500 cells/µL, HIV-diagnosed untreated women were only 6% less likely to become pregnant than HIV-negative women (Supplementary Table 2). It is likely that a number of factors compensate for the effect of knowledge of HIV status. First, levels of breastfeeding by HIV-positive women in the Western Cape have historically been very low, as a result of a PMTCT program that encouraged formula feeding [38]. As lactational amenorrhea is one of the major determinants of fertility, a higher pregnancy rate in nonbreastfeeding women might be expected. Second, the high rates of contraceptive use in the Western Cape could be a confounder, as women with the highest fecundity tend to be the women who use contraception most frequently [33], and thus HIV-positive women, who are more likely to experience temporary amenorrhea [39], may be less likely to use hormonal contraception. Finally, although we controlled for race, there may be behavioral and socioeconomic factors that are differentially distributed between HIV-positive and HIV-negative women and that are also important in predicting pregnancy incidence.

Our analysis has several limitations. Most significantly, we lack information on when women leave the Western Cape population, and we have therefore censored women at the analysis closure date. This may be a reasonable approximation in the case of HIV-negative women, as mortality rates in HIV-negative women of childbearing age are low in the Western Cape [40], and rates of migration out of the province have been estimated at only 0.6% per annum [41]. However, it is likely to be problematic in the case of HIV-positive women—particularly those who are untreated—as HIV-positive women experience relatively high mortality rates and may be more likely to migrate out of the province [42]. For this reason, we conducted a sensitivity analysis in which we censored women with reference to the date of their last contact with the health service. Results were broadly similar in this analysis, but the higher pregnancy incidence rates in untreated HIV-positive women and the smaller ART effect (Supplementary Table 1) are consistent with what would be expected if untreated HIV-positive women experience high mortality rates and are thus more likely to be censored. However, censoring HIV-negative women with reference to their date of last contact with the health service is problematic, as many HIV-negative women of reproductive age only attend health services when receiving antenatal or postnatal care, and excluding women with no healthcare contacts therefore introduces a bias toward women who have a higher pregnancy incidence.

A related limitation is that we lack data from the private sector. In the Western Cape, approximately 25% of the population are members of medical schemes (private healthcare) [43], and these results are therefore not representative of the whole province. Another limitation is that we lack information on pregnancy outcomes for a substantial proportion of patients, especially in the early years. The assumption of HIV negativity in the absence of evidence of engagement in HIV care could also introduce bias, although almost all HIV-positive women would have tested positive at the time of their index pregnancy, and the low HIV incidence rates in women in the Western Cape (0.7%–1.0% per annum over 2007–2017 [23]) suggest that misclassification of women who seroconverted after the index pregnancy would not substantially distort estimates of pregnancy incidence rates in HIV-negative women.

These results may have important implications for mathematical models that are calibrated to antenatal HIV prevalence data. In the Thembisa model for the Western Cape, the odds of HIV infection in pregnant women attending public antenatal clinics is estimated to be 1.66 (95% CI, 1.59–1.75) times the odds of HIV infection in women of reproductive age (the HIV infection prevalence among women is weighted by age-specific fertility rates) [23]. This is consistent with the finding of a significantly increased pregnancy incidence in HIV-positive women receiving ART in the present analysis. However, the Thembisa model assumes implicitly that this increased prevalence is due to a bias in the antenatal survey (due either to the sampling design or to problems with test specificity) and assumes that fertility rates in HIV-positive women receiving ART are lower than those in HIV-negative women. As a result, the Thembisa model may be underestimating the numbers of births to HIV-positive mothers. Similar assumptions about bias in antenatal survey data and levels of fertility after ART initiation are made in other HIV models, such as the EPP/Spectrum model used by UNAIDS [44]. A number of African countries have recently found that this model estimates lower numbers of births to HIV-positive mothers than recorded by PMTCT programs, which suggests that the EPP/Spectrum model may also be underestimating fertility rates in HIV-positive women in some settings (Robert Glaubius, personal communication). Further investigation is required to avoid misinterpreting antenatal HIV data, and it would be helpful to validate model estimates of the fraction of HIV-positive pregnant women receiving ART before conception against empirical data.

Finally, these results point to the need for better access to family planning services and counseling for HIV-positive women. Rates of unintended pregnancy are high in HIV-positive African women (as well as in HIV-negative African women), despite the integration of family planning into many HIV care services [45–47]. This is particularly concerning in light of the planned rollout of dolutegravir for HIV-positive individuals in many African countries; recent analysis suggests that infants born to women who receive dolutegravir are at increased risk of neural tube defects [48], and women receiving dolutegravir are therefore especially in need of counseling on family planning. We found that ART use was associated with increased pregnancy incidence even before 2011, when South African guidelines recommended the promotion of contraception to women receiving ART, owing to similar teratogenicity concerns. New strategies to increase the effective use of hormonal contraception in HIV-positive women are urgently needed [49].

Supplementary Data

Supplementary materials are available at The Journal of Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.

Notes

Acknowledgment.We thank the Western Cape Provincial Health Data Centre for collating and sharing these data.

Financial support. This work was supported by the National Institute for Child Health and Human Development (grant R01HD080465-01 to A. B.).

Potential conflicts of interest. All authors: No reported conflicts.

All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.

Presented in part: Meeting of the UNAIDS Reference Group on Estimates, Modelling and Projections, Bern, Switzerland, 17–20 September 2018; 10th International AIDS Society Conference on HIV Science, Mexico City, Mexico, 21–24 July 2019.

References