Background

In 2021, 38 million people lived with HIV worldwide, of whom 1.3 million were pregnant women, and mostly in sub-Saharan Africa [1]. Maternal HIV infection among antiretroviral drug-naïve women is associated with a significantly increased risk of adverse perinatal outcomes, such as prematurity, low-birth-weight, small for gestational age and stillbirth [2]. Antiretroviral therapy (ART) during pregnancy has demonstrated a clear benefit for maternal health, and prevent the risk of HIV mother-to-child-transmission (MTCT) [3]. Since 2015, ART initiation is recommended in all people living with HIV, including pregnant women [4]. As a result, in 2021, 1.05 million pregnant women had access to antiretroviral drug regimen for their own health and prevention of MTCT (PMTCT), with 90% living in sub-Saharan Africa [1, 5]. In 2016, the World Health Organization (WHO) recommended a non-nucleoside-reverse-transcriptase-inhibitor (NNRTI)-based combination as the preferred first-line regimen, while protease inhibitor (PI)-based combinations were recommended as second or third-line regimen mainly due to incomplete information regarding its risk when used during pregnancy [6]. In 2019, the WHO recommended in all adults living with HIV a transition to dolutegravir (DTG)-based ART, despite a slightly higher but significant neural tube defect signal associated with DTG exposure in the pre-conception period compared to other antiretroviral combinations [7]. After all, a DTG-based combination was recommended in all adults, including pregnant women, as the preferred first-line regimen due to improved efficacy, better tolerability and durability compared to all previous ART regimens [3, 8]. Despite the beneficial effects of antiretroviral drugs during pregnancy on both maternal health and PMTCT, their use raises concerns on their potential embryo-foeto-toxicity. It remains crucial to fully assess their associated perinatal outcomes to optimise ART strategies in pregnant women worldwide, but more particularly in sub-Saharan Africa, where both maternal HIV prevalence and rates of adverse perinatal outcomes are high [5]. Several studies have reported an increased risk of adverse perinatal outcomes after prenatal exposure to antiretroviral combinations, depending on the antiretroviral drug classes used [9,10,11,12,13,14,15,16]. PI-regimens still remain an important alternative option for pregnant women in 2022 that still need to be fully understood, due to conflicting results. Indeed, several studies have reported an association between PI-based combinations and preterm birth, while other studies have not found similar results [12, 15,16,17,18]. Therefore, we conducted a systematic review and meta-analysis aimed to assess the risk of adverse perinatal outcomes associated with PI-based combination use during pregnancy compared to NNRTI-based combination.

Methods

Search strategy and selection criteria

We did a systematic review and meta-analysis according to the Preferred Reporting Items for Systematic review and Meta-Analysis (PRISMA) guidelines [19]. The protocol of this review was registered in PROSPERO, the International prospective register of systematic reviews (CRD42022306896). The bibliographic research was based on both published and unpublished studies from 01/01/2002 to 29/10/2021 relative to adverse perinatal outcomes in HIV women who received antiretroviral combination during pregnancy.

Searches were conducted on four electronic scientific literature databases: PubMed, Reprotox, Clinical Trial registry (clinicaltrials.gov) and the abstracts from HIV conferences (Conference on Retroviruses and Opportunists Infections, International AIDS Society, European AIDS Clinical Society, British HIV Association and International Workshop on HIV Pediatrics). We used the keywords and MeSH terms presented in the Table 1.

Table 1 Keywords and MeSH terms used in bibliographical researches
Full size table

Inclusion and exclusion criteria

To be eligible, studies must document population (pregnant women with documented perinatal outcomes) and exposure (antiretroviral combination based either on PI or NNRTI, initiated before or during pregnancy). We included all randomised controlled clinical trials, prospective and retrospective cohort studies using a comparative study design. Studies not eligible were those off-topic, not specifying the antiretroviral combination used, those where numbers of adverse perinatal outcomes according to antiretroviral combination were not detailed, those not comparing PI-based versus NNRTI-based antiretroviral combinations, studies with only one type of inhibitor used, and those where integral text was not available. For abstracts, we limited our search to studies in the English or French language. No restriction was applied to geographic area. Study investigators were contacted when the manuscript content was insufficient.

Outcomes

We studied the following adverse perinatal outcomes based on WHO definitions: preterm birth (PTB, < 37 weeks of gestation) [20], very preterm birth (VPTB, < 32 weeks of gestation), low birth weight (LBW, < 2500 g) [21], very low birth weight (VLBW, < 1500 g), small for gestational age (SGA, birthweight < 10th centile for gestational age) [22], very small for gestational age (VSGA, birthweight < 3th percentile for gestational age), stillbirth (foetus born with no sign of life after 28 weeks of gestation) [23], congenital abnormalities (alteration in embryonal development) [24] and spontaneous abortion (< 22 weeks of gestation) [25]. Gestational age was estimated based on the last menstrual period and confirmed by ultrasound when available.

Exposure variable

Pregnant women were considered exposed to antiretroviral combination if they started antiretroviral treatment before or during pregnancy, and continued at least until delivery. Antiretroviral combination was defined by at least three drugs: namely two nucleoside reverse transcriptase inhibitors (NRTI) associated with a PI (lopinavir/ritonavir, atazanavir/ritonavir, darunavir/ritonavir, fosamprenavir, saquinavir and nelfinavir) or a NNRTI (efavirenz or nevirapine). We categorised the exposure into three different periods: pre-conception, early pregnancy (first trimester) and late pregnancy (second and third trimester).

Data extraction and quality analysis

Two investigators, LSL and VL independently reviewed and identified the relevant citations. LSL performed data extraction including description of the studies, their populations, the adverse perinatal outcomes according to antiretroviral combination, and scores of methodological qualities. We used two scales for methodological quality assessment according to study design: the Oxford scale [26] for clinical trials and the Newcastle–Ottawa scale [27] for cohort studies. Methodological quality assessment was conducted by two investigators independently (LSL and JB) and any discordance was resolved by discussion with AS. Studies with an Oxford score lower than three [26] or a Newcastle–Ottawa score lower than four [27] were considered as low methodological quality and were excluded.

Statistical analysis

We first described the characteristics of the studies included. Then, we extracted data from individual studies to generate a relative risk (RR) of prenatal exposure to a PI-based combination compared to those with an NNRTI-based combination for each adverse perinatal outcome. We performed a meta-analysis when more than one study reported the same outcome, using a random effects model to estimate a weighted summary RR and corresponding 95% confidence intervals for each outcome [28, 29]. All pre-specified analyses were stratified according to country income: High-Income Countries (HIC) and Low-to-Middle-Income Countries (LMIC). We investigated between-study heterogeneity by reporting forest plots and using the I2 statistic, with a p-value significance of 0.10 (I² <0.10) [29]. The pooled summarised RR (sRR) was presented only when both LMIC and HIC RRs were consistent (I2 < 0.10). We searched publication bias using funnel-plot and asymmetric Egger tests [28]. We conducted sensitivity analyses excluding outlier studies by graphical research, and then including only studies with high score of methodological quality (Oxford score higher than five [26] and Newcastle–Ottawa score higher than seven [27]). For all analyses, we defined significance at an alpha level of 0.05 (p-value <0.05), except for heterogeneity analyses. Statistical analyses were performed using STATA (14.2).

Results

Study and population characteristics

Our search identified 49,171 citations: 1,885 published studies and 47,286 unpublished studies. Initial screening was from title and abstracts of studies in 48,650 records, after exclusion of duplicates. Overall, 208 full-text articles were selected for a complete reading. Finally, after excluding four studies of low methodological quality (two clinical trials [30, 31] and two cohort studies [32, 33]), 32 studies were retained for systematic review and meta-analysis (Fig. 1). These studies were published between 2002 and 2021, and included 45,427 pregnant women from 27 countries. Nineteen (59%) studies were conducted in HIC and thirteen (41%) in LMIC. Only one randomised controlled trial (3.1%) was selected, those remaining being cohort studies. Overall, thirteen studies (40.6%) had a high score of methodological quality (Table 2). Study sample size varied from 75 to 7,009 pregnant women, maternal age from 26 to 33 years and median CD4 + from 154 to 638 cells/mm3 (Table 3).

Fig. 1
figure 1

Flow-chart of study selection process according to PRISMA guidelines

*Three studies had low score of methodological quality [30,31,32]. We could not assess methodological quality of one unpublished cohort study [33]: we considered this study of low methodological quality

Full size image
Table 2 Summary characteristics of the 32 selected studies and their methodological quality score
Full size table
Table 3 Description of the study population and summary results for studies included in the systematic review
Full size table

Out of the 32 studies, gestational age was estimated using ultrasound scan alone, last menstrual period alone or a combination of at least two methods, in 3.2% (1/31), 16.1% (5/31) and 67.7% (21/35) respectively. Information on gestational age estimation methods was not available for five studies (15.6%), reporting only congenital abnormalities [35, 38, 39, 43, 51]. The most common perinatal outcomes reported were PTB (22/32, 68.8%), LBW (13/32, 40.6%) and SGA (13/32, 40.6%) (Table 3).

Preterm birth

PTB was reported in 4,872 cases in twenty-two studies (15.2% [14.8–15.6]) [10, 18, 36, 37, 41, 42, 45,46,47,48,49,50, 52, 53, 55,56,57,58,59,60, 63, 64] (Fig. 2). In both LMIC and HIC separately, prenatal exposure to PI-based combination was not significantly associated with PTB compared to NNRTI-based combination (RR 1.17, 95%CI 0.91–1.49 and RR 1.11, 95%CI 0.97–1.26, respectively), but between-study heterogeneity was significant (I2 67.4% p = 0.002 and 41.6% p = 0.057 respectively). No global summary estimate was provided due to significant heterogeneity. More specifically, in LMIC, heterogeneity was only due to the van der Merwe study [41], the single outlier with opposite results compared to the eight others studies. When excluding this study, the RR became significant (RR 1.26, 95%CI 1.11–1.43) and homogeneous (I2 0%, p = 0.882). When including only studies with high score of methodological quality in LMIC and HIC [36, 42, 45, 55, 56, 59, 60, 64], the results were also homogeneous with a global significant increased risk of PTB after prenatal exposure to PI-based combination, compared to NNRTI-based combination (sRR 1.20 [1.08–1.32], I2 0% p = 0.653).

Fig. 2
figure 2

Forest-plot of preterm-birth risks in pregnant women receiving PI-based compared to NNRTI-based antiretroviral combination

Full size image

Very preterm birth

We found a total of 638 VPTB from nine studies (3.9% [3.6–4.1]) [36, 37, 41, 42, 47, 53, 55, 59, 60] (Fig. 3). In LMIC, prenatal exposure to PI-based combination was not significantly associated with VPTB compared to NNRTI-based combination (RR 0.77, 95% CI 0.26–2.27), but with significant between-study heterogeneity (I2 = 81.9%, p = 0.004). No association was either found in HIC (RR 0.96, 95%CI 0.72–1.27, I2 5.3%, p = 0.383). Results were homogeneous when including only studies of high methodological quality [36, 42, 55, 60], reporting a global not significant risk of VPTB (sRR 1.19 [0.89–1.60], I2 15.1% p = 0.318).

Fig. 3
figure 3

Forest-plot of very preterm-birth risks in pregnant women receiving PI-based compared to NNRTI-based antiretroviral combination

Full size image

Low birth weight

LBW was measured in thirteen studies for 1,902 cases (17.6% [16.9–18.3]) [10, 36, 37, 41, 47, 49, 50, 53, 55,56,57, 63, 64] (Fig. 4). In LMIC, prenatal exposure to PI-based combination was not significantly associated with LBW compared to NNRTI-based combination (RR 1.09, 95% CI 0.75–1.56), but between-study heterogeneity was significant (I2 = 74.5%, p < 10–4). In HIC, the RR was not significant (RR 0.93, 95%CI 0.80–1.08), and homogeneous (I2 16.5%, p = 0.307). Global results were homogeneous when including only studies of high methodological quality [52, 55, 59, 64], reporting a non-significant risk (sRR 1.04 [0.81–1.33], I2 0% p = 0.591).

Fig. 4
figure 4

Forest-plot of low-birth-weight risks in pregnant women receiving PI-based compared to NNRTI-based antiretroviral combination

Full size image

Very low birth weight

VLBW was reported in 126 cases across seven studies (2.6% [2.2–3.1]) [36, 37, 41, 47, 53, 55, 63] (Fig. 5). Globally, prenatal exposure to PI-based combination was not significantly associated with VLBW compared to NNRTI-based combination (sRR 0.77, 95% CI 0.46–1.29), and without significant between-study heterogeneity (I2 15.4%, p = 0.313). Two studies [36, 55] had a high score of methodological quality and found similar results to those of the main analysis (sRR 0.82 [0.30–2.25], I2 0% p = 0.968).

Fig. 5
figure 5

Forest-plot of very low-birth-weight risks in pregnant women receiving PI-based compared to NNRTI-based antiretroviral combination

Full size image

Small for gestational age

The risk of SGA was reported in thirteen studies for 4615 cases (17.7% [17.3–18.2]) [18, 36, 37, 41, 42, 48, 52, 53, 55, 56, 60, 61, 64] (Fig. 6). In LMIC, prenatal exposure to PI-based combination was not significantly associated with SGA compared to NNRTI-based combination but significantly heterogeneous (RR 1.34, 95% CI 0.92–1.96, I2 77.8%, p = 0.001). In HIC, we found similar results (RR 1.11, 95%CI 0.92–1.34 and I2 54.2%, p = 0.033). When including only studies with high score of quality, we found similar results than those obtained in the main analysis: in LMIC [36, 42, 56, 64] (RR 1.18 [0.80–1.73], I2 56.5% p = 0.075) and HIC [55, 60, 61] (RR 1.35 [0.92–1.99], I2 45.5% p = 0.160).

Fig. 6
figure 6

Forest-plot of small-for-gestational-age risks in pregnant women receiving PI-based compared to NNRTI-based antiretroviral combination

Full size image

Very small for gestational age

The risk of VSGA, which was reported in two studies for 508 cases (9.9% [9.0–10.7]) [42, 61] (Fig. 7), was globally significantly increased for prenatal exposure to PI-based compared to NNRTI-based combinations (sRR 1.41, 95% CI 1.08–1.84), and consistent between-studies (I2 = 0%, p = 0.322). Only two studies with a high score of methodological quality contributed to this outcome, which did not allow to evaluate the publication bias.

Fig. 7
figure 7

Forest-plot of very small-for-gestational-age risks in pregnant women receiving PI-based compared to NNRTI-based antiretroviral combination

Full size image

Stillbirth

Stillbirth was reported in four studies for 269 cases (2.0% [1.7–2.2]) [42, 47, 58, 62]. Globally, prenatal exposure to PI-based combination was not significantly associated with stillbirth risk compared to NNRTI-based combination (sRR 1.06, 95% CI 0.74–1.50) (Fig. 8), and consistent between studies (I2 = 0%, p = 0.953). We found similar results when including only the two studies with high methodological quality [42, 62] (sRR 1.09 [0.75–1.57], I2 = 0%, p = 0.829).

Fig. 8
figure 8

Forest-plot of stillbirth risk in pregnant women receiving PI-based compared to NNRTI-based antiretroviral combination

Full size image

Congenital abnormalities

Congenital abnormalities were reported in nine studies for 279 cases (5.7% [5.0–6.3]) [35, 38,39,40, 43, 44, 46, 51, 63] (Fig. 9). Globally, prenatal exposure to PI-based combination was not significantly associated with congenital abnormalities compared to NNRTI-based combination (sRR 0.94, 95% CI 0.73–1.21). Between-study heterogeneity was not significant (I2 0%, p = 0.473). The three studies [38, 39, 43]] with a high score of methodological quality reported also a non-significant risk (sRR 1.22 [0.84–1.76], I2 0%, p = 0.832).

Fig. 9
figure 9

Forest-plot of congenital abnormalities risk in pregnant women receiving PI-based compared to NNRTI-based antiretroviral combination

Full size image

Spontaneous abortion

Spontaneous abortion was reported in only one study for 50 cases (20.4% [15.4–25.5]) [58]. In this study, prenatal exposure to PI-based combination was not significantly associated with spontaneous abortion compared to NNRTI-based combination (RR 1.26, 95% CI 0.77–2.08). No meta-analysis was conducted.

Exposure period of antiretroviral combination during pregnancy

We were not able to conduct subgroup analyses by exposure pregnancy periods because only one reported numbers of adverse perinatal outcomes according to the pre-conception period [53], and another one according to early/late pregnancy period (before or after 28 weeks of pregnancy) [41].

Discussion

While there is an increasing number of pregnant women living with HIV receiving antiretroviral therapy, especially in sub-Saharan Africa, the risk of adverse perinatal outcomes according to antiretroviral combinations exposure remains critical of fully assess. Our meta-analysis specifically assessed these risks associated with maternal PI-based antiretroviral combination still recommended by WHO as second- or third-line therapy, using a head-to-head comparison to NNRTI-based combination, based on a large sample size of pooled studies and originally stratified according to country income.

By pooling the estimates assessed from 32 studies, our meta-analysis comparing the risk of adverse perinatal outcome after prenatal exposure to PI-based compared to NNRTI-based combinations, in both LMIC and HIC countries, provided the following findings: firstly, we did not report any global significant pooled risk consistently for VLBW, stillbirth and congenital abnormalities. Secondly, despite significant heterogeneity between LMIC and HIC studies, there was no significant risk related to PI-ART exposure reported for VPTB, LBW and SGA. Thirdly, we found a global significant increased pooled risk of PI- based regimen for VSGA (+ 41% [8–84]), in two studies. Fourthly, we cannot formally conclude for the global risk of PTB with inconsistent findings according to sensitivity analysis. Lastly, no meta-analysis was conducted for the risk of spontaneous abortion, reported in only one study.

Three other systematic reviews previously published in 2018, 2020, and 2022 reported the effects of PI-based antiretroviral therapy associated with adverse perinatal outcomes [15, 16, 65]. Saleska et al. reported significant higher risk of LBW when only compared to zidovudine (ZDV) monotherapy, but no significant effect compared to NNRTI-based ART [65]. A network meta-analysis of randomized controlled trials conducted by Tsuivila-Matala, also reports that lopinavir/ritonavir (LPV/r) based regimens were associated with a significant higher risk of LBW compared to ZDV monotherapy, but this was not significant compared to NNRTI-based regimens [16]. In 2022, another meta-analysis reported the PI-ART-related adverse perinatal outcomes risks to be significantly higher for SGA (+ 24%,95% CI 8%-43%), and VSGA (+ 40%; 95% CI 9%-81%), but not with PTB or other perinatal outcomes [15].

Based on our results, prenatal exposure to PI-based ART is significantly associated with a increased risk for VSGA, as also recently reported by Cowdell et al. [15]. Some studies have shown that prenatal exposure to PI-based combination was associated with decreased progesterone levels during pregnancy, resulting in elevated estradiol levels [66,67,68]. Progesterone levels are correlated and estradiol levels are inversely correlated with birth weight. These hormonal changes, induced by prenatal exposure to PI-based combination, may be associated with fetal growth restriction and therefore with a higher VSGA risk [69, 70]. This hypothesis needs to be further investigated to better understand the potential effect of PI-based combination exposure on foetal growth.

In this meta-analysis, some perinatal outcomes (VSGA and spontaneous abortion) are reported in few studies, limiting result interpretation. Spontaneous abortion was investigated in only one study [58], with no meta-analysis conducted for this outcome. For VSGA, one large sample-size study [42] in Botswana, conducted in women who delivered in maternity wards at the national level using a standardized definition of perinatal outcomes, reported an increased risk of VSGA after prenatal exposure to PI-based combination compared to NNRTI-based combination. A standardized definition of gestational age was used in this study, while it is not necessarily the case in others LMIC, providing confidence in pregnancy outcome data quality. Another study [61], in HIC, did not find a significant result of VSGA, but weighed only 3.5% in the overall analysis of this risk. The results obtained for the VSGA risk seem robust thanks to large sample size study and use of standardized definition of perinatal outcomes, but need to be further investigated. Secondly, despite the subgroup analysis stratified between HIC and LMIC, between-study heterogeneity remains significant for most of the perinatal outcomes (PTB, VPTB, LBW and SGA), which may limit the interpretation of our results. This heterogeneity is partly explained by the diversity of methods used to measure perinatal outcomes. We can suppose that estimation of gestational age was more accurate in HIC compared to LMIC, because ultrasound is usually performed at least once during pregnancy in HIC. The heterogeneity for PTB risk can also be explained by one outlier study, excluded in sensitivity analysis in LMIC. The van der Merwe study [41] was conducted in South Africa between 2004 and 2007, and included pregnant women with a median CD4 count of 155 cells/mm3. NNRTI exposure was preferred for women with advanced HIV infection or co-infected with tuberculosis. Tuberculosis can be associated with growth restriction and PTB [71, 72]. It can explain outlier data with high prevalence for NNRTI exposure. Without this study, we found a significant increased risk (+ 26%) for PTB associated to prenatal exposure to PI-based combinations, compared to NNRTI-based combinations. This result was also find in the sensitivity analysis conducted only on studies with high scores of methodological quality (+ 20%). Therefore, the result reported in the main analysis appears not robust and heterogeneous. It must be interpreted cautiously, considering the significant association found in the two sensitivity analyses. Moreover, we cannot conduct sensitivity analyses excluding outlier studies for others perinatal outcomes (VPTB, LBW and SGA) due to conflicting results. Indeed, results in the main analysis were heterogeneous, especially in LMIC. However, we found homogeneous and similar results in the sensitivity analyses including only studies with high scores of methodological qualities (VPTB and LBW), showing the robustness of the main analysis results. Last, despite the sensitive analysis conducted on high quality studies, the result of SGA remained heterogeneous but still not significant. As we found a significant increased risk of VSGA, we supposed that no significant risk was reported due to lack of statistical power in the subgroup analyses. Since these two perinatal outcomes are strongly correlated (SGA and VSGA), we suggest to conduct further investigations on the effect of PI-based combination exposure on overall foetal growth.

Our systematic review has some limitations. We could not disentangle effects of HIV exposure also associated with adverse perinatal outcomes [2] from those associated with antiretroviral exposure. The impact of the timing of ART initiation in pregnancy (before/after conception) on perinatal outcomes remains uncertain, as we were not able to conduct the sensitivity analyses due to the lack of information. In our study, we did not specifically explore the effect of NRTI backbone, but it would be relevant to investigate the potential association between PTB and VPTB risks and exposure to ZDV-3TC-LPV/r. Finally, we were not able to investigate the effects on perinatal outcomes according the different PI-based regimen.

Our systematic review has also strengths. Our search was exhaustive thanks to use of several bibliographical databases, abstracts of HIV conference and clinical trials registry. Studies included all available comparative study designs (randomized clinical trials and cohorts) to guarantee the representativeness. No restriction for geographical area and publication date ensured representative results. The use of an appropriate comparator defined as a NNRTI antiretroviral combination allowed us to estimate relative risks, strong indicators of risk. Methodological quality assessment was also performed by two investigators independently. Standardization of data collection and outcome definition raise many challenges in data quality assessment [73]. Our results were detailed according to country-outcome to consider data quality heterogeneity. We conducted sensitivity analyses including only studies with high methodological quality and results were mostly consistent with those of the main analysis. Most of the results obtained in the subgroup analyses were consistent with those obtained in primary analysis, with robust analysis for most of the outcomes, except for PTB.

Conclusion

Our study did not show a higher risk for most of the adverse perinatal outcomes after prenatal exposure to PI-based combination compared to NNRTI-based combination. However, our review suggests a significant increased risk of VSGA, similarly reported in another recent review [15]. The risk of PTB initially reported is not clearly demonstrated [10, 18, 30, 31, 36, 37, 41, 45, 48, 49, 52, 53, 55,56,57,58,59, 63, 64], with significant between-studies heterogeneity. Therefore, this result should be interpreted with caution. Our results should be considered to inform clinical guidelines, with appropriate messaging regarding the PI benefit-risk balance in pregnant women and those of childbearing potential living with HIV to improve their perinatal outcomes.