Ebola virus disease in pregnancy: a systematic review and meta-analysis

Abstract This review synthesises and appraises evidence on the effects of Ebola virus disease (EVD) in pregnancy. We searched bibliographic databases from dates of inception to November 2020, yielding 28 included studies. The absolute risk of maternal death associated with EVD was estimated at 67.8% (95% confidence interval [CI] 49.8 to 83.7, I2=85%, p<0.01) and the relative risk of death in pregnant women compared with non-pregnant women was estimated at 1.18 (95% CI 0.59 to 2.35, I2=31.0%, p=0.230). The absolute risk for foetal losses was estimated at 76.9% (95% CI 45.0 to 98.3, I2=96%, p<0.01) and neonatal death was 98.5% (95% CI 84.9 to 100, I2=0.0%, p=0.40). The gap analysis suggests limited or no data on the clinical course, non-fatal perinatal outcomes and EVD management in pregnant women. The review suggests that EVD has a high maternal and perinatal mortality, underscoring the urgent need for preventative and therapeutic solutions and improved screening and follow-up of pregnant women and newborns during outbreaks. There is not enough evidence to conclusively rule out pregnancy as a risk factor for mortality and there is limited evidence on the disease course, outcomes and management of EVD in pregnancy, and this supports the need for robust clinical trials and prospective studies that include pregnant women.


Introduction
Ebola virus disease (EVD) is caused by Ebola virus and was first reported in South Sudan and along the Ebola River in the Democratic Republic of Congo (DRC) in 1976. 1 EVD occurs sporadically, with all known outbreaks originating in sub-Saharan African (SSA), where an estimated 22 million people in 22 West and Central African countries are at risk of infection. 2 The reservoir host for Ebola virus is thought to be the fruit bat. 1 EVD has an abrupt onset with an incubation period of 2-21 d. 1 Clinical features range from non-specific symptoms to multisystem involvement with haemorrhagic manifestations and multiorgan failure. Disease outbreaks in humans have so far been due to three viral species: Zaire ebolavirus (EBOV), Bundibugyo ebolavirus (BDBV) and Sudan ebolavirus (SUDV). 3 A meta-analysis evaluating EVD mortality in the general population estimated a pooled crude case fatality rate of 65.0% (95% confidence interval [CI] 54.0 to 76.0, I 2 =97.98%). 4 Suggested reasons for the high amounts of study heterogeneity were study variability, differences in outbreaks and variability in outbreak responses. 4 In the DRC, the estimated crude case fatality rate for confirmed and probable cases was 66% in the 2018-2020 Kivu/Ituri Ebola outbreak and 42.3% in the 2020 Équateur outbreak. 5,6 The improvements seen in the Équateur outbreak were likely due to the availability of vaccines and drugs. 7 The prognosis of EVD is thought to be very poor in pregnant women, with maternal case fatality rates ranging from 74 to 100% and an aggregated maternal mortality of 84.3% in one review, 8 86% in another 9,10 and 72% in a more recent review. 11 Based on these reviews, foetal losses ranged from 23 to 100% 9,10 and neonatal losses were frequent at 99.1-100%. 9,10 While there are several reviews summarising evidence on EVD in pregnancy, the evidence is often aggregated or fails to account for differences in sample sizes and study variability. This is important because a quantitative synthesis that takes into account variability and samples size differences is likely to provide a more reliable summary estimate and specify areas of potential bias to facilitate further enquiry or formulation of hypotheses for future research. 12 N. D. Kayem et al. The goal of the review was to critically appraise and summarise evidence on the effects of EVD in pregnancy and to identify current gaps in the evidence reported in the peer-reviewed and grey literature. The review specifically appraises and synthesises evidence on the clinical characteristics, the maternal and perinatal outcomes of EVD during pregnancy and the clinical management of maternal Ebola virus infection.

Methods
The review was conducted based on the Preferred Reporting Items for Systematic Review and Meta-analysis (PRISMA) guidelines for conducting systematic reviews. 13 The review was conducted as part of a broader review on the effects of the viral haemorrhagic fevers in pregnancy and the methods have previously been published. 14

Search strategy and selection criteria
In summary, various bibliographic databases were searched in December 2017 and references from relevant reviews and included studies were also searched. There were no language restrictions (translation software was used where appropriate to facilitate screening and data extraction

Data extraction
Citations were screened using Rayyan, an online systematic review program. 15 Screening of titles and abstracts, assessment of full text for eligibility and risk of bias assessment were performed independently by two of the reviewers (NDK, CB, CA, MT or SB). Each article was preassigned randomly to any two of these reviewers and any discrepancies were resolved through discussion or with another (third) reviewer. 14 Studies were included if they presented data on EVD in pregnancy. All study designs, as well as original reports, briefs, letters, editorials and comments, were considered for inclusion. Exposure was defined by clinical or laboratory criteria. Outcomes were broadly classified as clinical characteristics, maternal and peri-natal outcomes and management practices. Outcomes were not predefined and definitions were extracted where available, from each citation. Data were extracted from eligible studies and odds ratios (ORs) or proportions were estimated depending on the available information. 14 Meta-analyses were performed if two or more studies reported on the same outcome and included at least five pregnant women or live births. 14 This was a pragmatic decision given that the disease is rare and we are evaluating pregnant women, who are a population that is often excluded from clinical trials or research. 16 Gestational ages were inconsistently reported. As such, all foetal deaths occurring before or during labour and delivery, including miscarriages, intrauterine foetal death and stillbirths, were labelled as foetal loss. Maternal death was defined as death of the pregnant woman during pregnancy, labour and delivery or within 1 month after delivery. Neonatal death was defined as death of a newborn within 1 month of birth.

Statistical analyses
All statistical analyses were done using R version 4.0.2 (R Foundation for Statistical Computing, Vienna, Austria). 17 The metaprop command was used for the proportional meta-analysis and the metabin command was used to summarise the effect estimate for ORs because they implement procedures specific to binary outcome data. The weighted summary proportion was calculated using the Freeman-Tukey double arcsine transformation. [18][19][20] A random effects model was used and Cochran's Q, τ 2 and Higgins I 2 were calculated. The degree of heterogeneity was interpreted as none (I 2 <25%), low (I 2 =25-49%), moderate (I 2 =50-74%) or high (I 2 ≥75%). 21

Sensitivity analysis and meta-regression
Post hoc sensitivity analyses were performed where possible, dropping studies with 0% or 100% proportions and studies with <10 pregnant women to assess if excluding studies with smaller sample sizes and with extreme proportions results in a statistically significant difference in summary estimates. 14 Where I 2 was >50%, meta-regressions were conducted to evaluate reasons for the observed heterogeneity. The explanatory variables evaluated were study design (cohort vs non-cohort), year in which outbreak occurred (before 2000 vs after 2000), sample size (<10 vs ≥10) 14 and country.

Publication bias
Peter's test was used in combination with funnel plots to assess for publication biases. 22 Funnel plots were derived using a mixed effects model with sample size as the predictor and the double arcsine transformed proportion. 14

Risk of bias assessment
Risk of bias assessment was based on Murad et al.'s 23 quality tool for case reports and case series studies, while the Newcastle-Ottawa scale was used for all cohort-type studies. 24 Although aggregated scores are provided, the risk of bias was colour-coded to allow for a better interpretation of the study Transactions of the Royal Society of Tropical Medicine and Hygiene quality by readers, given that an aggregated score fails to highlight where specific weaknesses in the reported study design are found. 25

Gap analysis
Research gaps were identified using Robinson et al.'s framework, 26 which allows not only for identification of gaps, but the reasons for the gaps. Gaps are classified under four categories: insufficient or imprecise information, biased information, inconsistency or unknown consistency and not the right information. Given that EVD is a rare disease and the selected population (pregnant women) is rarely included in clinical research, 16 and given that a meta-analysis will not be possible for all outcomes, we modified the framework such that evidence on metaanalyses were analysed separately. As such, where Robinson et al. 26 discuss gaps based on four categories, we had five categories: separating the options for insufficient and imprecise information into two distinct categories, for each objective we presumed certain outcomes and, as such, we assessed the evidence or lack thereof based on these objectives (detailed in a previous manuscript 14 ).

Results
A search of the databases yielded a total of 3651 records that were screened. A total of 264 full texts were evaluated for eligibility and 68 met the inclusion criteria ( Figure 1). Twenty-eight of these were on EVD in pregnancy,  and the characteristics of included studies are summarised in Table 1.
The clinical course of infection during pregnancy was not described and most of the information was obtained from case reports or series. The length of hospital stay (evaluated as the time from admission to either discharge or death) was reported in 32 pregnant women in 12 studies and ranged from 0 to 32 d. 27,29,31,36,37,39,43,[45][46][47][48]50,54 The time from illness onset to admission was 1-7 d in 19 pregnant women from five studies 29,36,39,43,45 and the time from illness onset to treatment was 3-7 d recorded in four pregnant women in three studies. 29,36,45

Maternal outcomes of Ebola virus infection
The most-reported outcome in EVD-positive pregnant women was maternal death, with a weighted summary proportion of 67.8% (95% CI 49.8 to 83.7, I 2 =85%, p<0.01; Figure 2). The gestational age at which maternal death occurred was specified in 15 women and ranged from 14 weeks to term (≥37 weeks), with most deaths (9/15 [60.0%]) occurring in the third trimester. 36,39,44,47 Three studies provided sufficient information to estimate the odds of death in pregnant women compared with non-pregnant women. 34,39,44 The pooled OR was 1.18 (95% CI 0.59 to 2.35, I 2 =31.0%, p=0.230; Figure 3).

Perinatal outcomes of maternal Ebola virus infection
Perinatal outcome measures were not defined in the included studies, and most studies did not report gestational age at which the outcomes occurred. The weighted summary estimate for foetal loss was 76.9% (95% CI 45.0 to 98.3, I 2 =96%, p<0.01; Figure 5).

Clinical management practices for maternal Ebola virus infection
Most of the included studies did not report the use of antivirals, immunosuppressive therapy or immune modulators for the management of EVD in pregnant women. Favipiravir was used in a single case report, 36 while convalescent plasma (CP) was used in the non-randomised trial. 49 Of the eight pregnant patients who received CP, six of them survived. None of the studies assessed Obstetric management of pregnant EVD women was not described in most studies. Termination of pregnancy was offered in two studies, 39,54 but only one woman agreed to termination, which occurred at 7 weeks of gestation. 39 Induced delivery, reported in six studies, 32,39,[45][46][47]54 generally occurred as a result of foetal demise and not necessarily due to Ebola infection. Dilatation and curettage was performed in three pregnant women 44 due to incomplete abortion, but only one of these women survived.
Neonatal management was described in one neonate and included monoclonal antibodies (ZMapp), 36 antivirals (GS-5734) 36 and admission to the neonatal intensive care unit (NICU). 36

Sensitivity analysis and meta-regression
A post hoc sensitivity analysis dropping studies with 0% and 100% proportions and studies with <10 pregnant women did not have a substantial impact on the estimates for foetal loss and maternal death, which became 73.5% (95% CI 34.0 to 99.0, I 2 =97%, p<0.01) and 75.3% (95% CI 55.8 to 90.7, I 2 =89%, These studies were excluded from the meta-analysis because of possible risk of duplication with another included study. b These studies include the same group of pregnant women, however, one reports on clinical characteristics while the other reports on pregnancy outcomes.  p<0.01), respectively. Sensitivity analysis was not performed for other outcomes because only one study qualified. A meta-regression evaluating the combined effect of the sample size, the year in which the outbreak occurred, country and study design on the pooled estimates for maternal death suggests that these variables accounted for the observed heterogeneity (R 2 =100%, p<0.001), with most of the heterogeneity being as a result of the country (75.4%). Interestingly, for foetal loss these variables did not significantly account for the heterogeneity observed (p=0.629). Other potential confounders Transactions of the Royal Society of Tropical Medicine and Hygiene   such as viral strain were included in the review, but all studies in the meta-regression model investigated viruses of the Zaire ebolavirus strain, and there were no data on subtypes, so further analysis could not be performed.
Given the developments in therapeutics and vaccines for EVD, we performed a subgroup analysis accounting for the year in which the outbreak occurred, considering only studies with >10 pregnant women. We found that the weighted summary of maternal death due to EVD prior to 2000 was 90.4% (95% CI 83.

Publication bias for EVD
Funnel plots showed some asymmetry (Supplementary Figure 1), however, Peter's test showed no evidence of publication bias (p>0.1). The p-value for maternal case fatality proportion was 0.485, for maternal OR for pregnant women compared with non-pregnant women it was 0.727, for the foetal case fatality proportion it was 0.569 and for clinical features the p-value ranged from 0.362 to 0.941.

Risk of bias assessment for EVD
The overall risk of bias among the included case reports and case series studies was low, while the risk of bias in cohort-type studies was moderate or good in 5 of the 11 cohort-type studies. The risk of bias scores are summarised in Figure 8.

Gap analysis for EVD
A formal analysis of potential research gaps in the grey and peerreviewed literature suggests that, particularly for non-fatal perinatal outcomes, as well as the clinical course of EVD in pregnant women and newborns and for management of EVD in pregnancy, data are limited or non-existent (Table 3).
While meta-analyses were conducted for most of the outcomes identified in the literature, many of these estimates were imprecise, with large between-study heterogeneity, with some outcomes only discussed in case reports or case series studies. Additionally, when the estimates were precise, the sample sizes were small, and most of the cohort-type studies had a moder-ate to high risk of bias, which limits the strength of the available evidence.
Gestational ages were mostly self-reported and the gestational ages at which different outcomes occurred were rarely reported, as such evaluating outcomes by the trimester of pregnancy was not possible for most outcomes. Evidence synthesis was further limited by a lack of outcome definitions.

Discussion
This review shows that while EVD is associated with a high maternal mortality rate (67.8%), the relative risk of mortality among pregnant women was not significantly different from that in nonpregnant women (OR 1.18 [95% CI 0.59 to 2.35], I 2 =31.0%, p=0.230); however, given that there are few studies in the metaanalysis and a single low-quality study 34 accounted for the highest weighting in the meta-analysis, the evidence is inconclusive. When compared with previous studies, the estimates for maternal mortality were not significantly different. Foeller et al., 11 in a review published in July 2020, estimated an aggregated absolute risk of 72% and suggested that the mortality in pregnant women was not higher than that in non-pregnant women. A subgroup analysis accounting for the year in which the outbreak occurred showed that the absolute risk of maternal mortality was lower after 2000 (64.8%) compared with before 2000 (>90%) and may reflect the changes in public health preparedness. Interestingly, there was a large amount of heterogeneity in studies conducted after 2000, and this heterogeneity could perhaps be explained by Transactions of the Royal Society of Tropical Medicine and Hygiene Figure 8. Risk of bias assessment of studies included in the systematic review and meta-analysis on EVD in pregnancy. For cohort type studies green indicates the study received a star ( * ) for the indicated or criteria while red denotes the study didnot receive any star for that criteria on the Newcastle-Ottawa scale; for case series or case reports: green, yes; red, no; yellow, unclear/unsure; black, not applicable. GA: gestational age. differences in national responses, given that our meta-regression analysis suggests that the country in which patients were found accounted for most of the observed heterogeneity. This finding is similar to a recent review by Kawuki et al. 55 that found the country was a possible explanatory factor for variability in EVD outcomes and underscores the need for sharing of lessons learned both within and between countries.
The absolute risk of foetal loss was high, at 76.9%, and consistent with previous reports. 9,10 The subgroup analysis showed that unlike maternal case fatality, foetal losses after the year 2000 (83.5%) were much higher than before the year 2000 (59.4%). This may be due to improved reporting, and this finding may have been clarified by a comparison with data from the 2018-2020 DRC outbreaks, but three attempts to get unpublished data on pregnancy outcomes from that outbreak were unsuccessful. The absolute risk of neonatal deaths was high, at 98.5%, and the estimate seems precise, with low heterogeneity (95% CI 84.9 to 100, I 2 =0.0%, p=0.40), which supports the consensus that EVD almost always results in neonatal death. 9,10 However, this metaanalysis had only three studies with an overall small size; as such, the estimate for neonatal death is not conclusive.
The high maternal and perinatal mortality seen among EVD patients supports the urgent need for preventative and therapeutic solutions, and improved screening, care and followup of pregnant women and newborns during outbreaks, both within and outside the treatment units, to ensure early identification and management of pregnant women and their newborns.
A lot of progress has been made in the last 6 y following the West African Ebola epidemic. 7 Nonetheless, the review shows that there are still major gaps in the epidemiology of EVD in pregnancy, particularly with respect to non-fatal perinatal outcomes, the clinical course and the management of EVD in pregnancy. This underscores the need for well-conducted prospective cohort studies and clinical trials with pregnant women as a 'subgroup of interest'. 14 Given the sporadic nature of Ebola outbreaks, these gaps will be best addressed during future epidemics and would benefit from pre-epidemic preparedness for clinical research with pre-prepared protocols, 56 improved capacity for conducting clinical research in outbreaks in at-risk countries and the inclusion of pregnant women in clinical research. Additionally, improved data quality through either harmonised databases or core datasets and outcome sets for Ebola in pregnancy will improve the evidence base for EVD. 57,58 A common limitation across all meta-analyses performed was that they all had <10 studies, and for most outcomes the summary estimates were associated with large amounts of heterogeneity. A random effects model was used and a metaregression and sensitivity analysis were performed to explain the heterogeneity. However, the meta-regression analyses need to be interpreted with caution, given that we have <10 studies in each meta-analysis. The reason for this is that small sample sizes (<10 studies) tend to decrease the statistical power of the metaregression analysis. Similarly, the tests for publication bias need to be interpreted with caution given that our meta-analysis had very few studies. Foetal losses due to EVD Meta-analysis shows wide CIs with high heterogeneity and insufficient data to assess other potential confounders B Neonatal deaths due to EVD Small sample sizes and very few studies included in meta-analysis; importantly, there is a need for evidence on mechanisms by which neonatal infection occurs B Clinical features of EVD in pregnancy Meta-analysis done for most clinical features; however, the estimates have high heterogeneity with wide CIs. There were very few studies included in the meta-analysis and some studies included had a moderate to high risk of methodological bias B Maternal mortality-absolute risk Meta-analysis showed wide CIs with high heterogeneity and some of the studies included had a moderate to high risk of methodological bias C Maternal mortality-relative risk Estimate was precise with narrow CIs and a low heterogeneity, however, studies included had a moderate to high risk of bias D Clinical course of infection in pregnant women The clinical course of EVD in pregnancy was rarely reported and most of the studies were case reports or case series studies D Clinical features and complications in newborns Clinical features of neonates born to mothers with EVD were described in four studies, but most of these were case reports and case series studies The gap analysis evaluates gaps in the literature based on expected outcomes of the review. The table summarises the gaps identified and provides a justification for each. A: no data or insufficient information; B: meta-analysis conducted, but there were few studies in the meta-analysis, studies included have small sample sizes or the meta-analysis is associated with high heterogeneity and/or wide or extremely wide CIs. C: some of the studies included in the meta-analysis have a low-quality rating (i.e. a moderate or high risk of methodological bias); D: most of the studies discussing a specific outcome are case report or case series studies.

Conclusions
Maternal deaths and foetal losses in EVD are high, with almost all births to EVD pregnant women culminating in death. This evidence underscores the urgent need for preventative and therapeutic solutions, enhanced screening to identify pregnant women during outbreaks and improved care and follow-up of pregnant women and their newborns in the event of outbreaks. There is not enough evidence to conclusively rule out pregnancy as a risk factor for mortality in EVD patients, and given the paucity of evidence in pregnancy, it supports the need for robust clinical trials and prospective studies with pregnant women as a subgroup of interest.

Supplementary data
Supplementary data are available at Transactions online.
Authors' contributions: NDK, SK, PA and PH designed the study. NDK was responsible for the search strategy, the literature search and the statistical analysis. NDK, CB, CA, SB and MT assessed article titles, abstracts and full texts and the risk of bias assessment. NDK wrote the first version of the manuscript and all the authors reviewed and approved the final version. All authors had full access to the data in the study and had final responsibility for the decision to submit for publication.