Complex pregnancy-induced physical and physiological changes [26], including immune system alterations, are thought to render pregnant women more susceptible to severe illness once infected with influenza virus [27–30]. Numerous descriptive reports from influenza pandemics documenting influenza-related morbidity and mortality in pregnant women [31–35] have contributed to the recognition of pregnancy as a risk factor for severe influenza illness. A recent systematic review assessed whether pregnant women are at higher risk for severe influenza-associated health outcomes compared with non-pregnant women of reproductive age [36]. The review was carried out as part of the WHO Taskforce to Evaluate Influenza Data to Inform Vaccine Impact and Economic Modelling, a working group convened to inform efforts by the WHO Initiative for Vaccine Research to promote evidence-based implementation research for influenza vaccine programs [37]. In comparative observational studies included in the review, pregnant women had a higher risk of laboratory-confirmed influenza (LCI) hospitalization than non-pregnant women (pooled odds ratio [OR] 2.94; 95% confidence interval [CI]: 1.58–5.47), but no differences were seen for severe outcomes such as LCI intensive care unit admission or death. The authors concluded that pregnant women may be hospitalized at a higher rate than non-pregnant women of reproductive age on a precautionary basis, since severe influenza outcomes did not differ between the two groups, although the low quality of evidence across studies [38] was acknowledged as a limitation of the review. A list of study quality issues and evidence gaps identified by the review is provided in Table 2.

High rates of pregnancy loss and preterm delivery were documented in case series from the 1918–1919 influenza pandemic, particularly among influenza-infected pregnant women who developed secondary pneumonia [32]. Increases in other adverse outcomes such birth defects have also been reported in case series from 20th-century influenza pandemics [39], but not consistently [40]. With few exceptions [41,42], descriptive studies from the influenza A (H1N1) pandemic in 2009 reported increased rates of adverse birth outcomes among infected pregnant women, especially among hospitalized cases [34,43–45]. However, the absence of a comparison group in these studies limits their ability to causally attribute adverse birth outcomes to influenza disease [46]. Moreover, many of the reports were limited to information collected at the time of an antepartum hospital admission and lacked information on the final outcome of the pregnancies delivered at a later time [34].

In early 2015, the WHO Taskforce assessing influenza immunization during pregnancy [37] initiated a systematic review of influenza-associated risks of preterm birth, SGA birth and fetal death [47]. The review identified 21 comparative studies of adverse birth outcomes and maternal influenza disease. The quality of the evidence was considered low to very low for all outcomes, owing to the limited number of studies (particularly for SGA birth and fetal death), inconsistency of results and concerns about ascertainment of influenza disease. A small number of higher-quality studies suggested that severe 2009 pandemic H1N1 (pH1N1) influenza disease was associated with an increased risk of preterm birth [48,49], but no association was found for mild-to-moderate 2009 pH1N1 influenza [50–52], nor for seasonal influenza [51,53]. Five studies of SGA birth suggested a small increase in risk associated with influenza virus infection (pooled OR: 1.24; 95% CI: 0.96–1.59) [47]. The highest quality evidence for fetal death outcomes (miscarriage and/or stillbirth) was limited to two studies, both of which reported a significantly increased risk of fetal death following maternal 2009 pH1N1 influenza disease [48,50]. The remaining studies were considered at high risk of bias [54] or had too few fetal deaths for meaningful interpretation. The systematic review found little evidence that mild maternal influenza illness was associated with adverse birth outcomes but acknowledged many limitations in the available evidence (Table 2).

Several potential pathways through which maternal influenza disease could lead to adverse birth outcomes were discussed: direct viral injury to the placenta or fetus, maternal immunological responses affecting placental function or temperature regulation and medical intervention. Human fetal membranes act as defensive barriers to protect the developing fetus against infectious agents [55]. Although maternal influenza viremia with transplacental transfer to fetal tissues has been documented in human cases of severe infection or highly pathogenic strains [56–58] and in lab studies [59], direct viral infection of the fetus is considered rare [55,60,61]. The fetal membranes additionally play an important role in the regulation of processes that initiate normal parturition through cytokine production [55]. Systemic maternal infection elicits inflammatory responses such as elevated proinflammatory cytokine levels [62], a known pathway to spontaneous preterm birth [20,63] and the main hypothesized mechanism by which influenza virus infection could lead to spontaneous preterm birth [56,60,61,64,65]. Another immune response, transient maternal hyperthermia [60,66] has also been associated with an increased risk of congenital anomalies [67]. Other medical interventions, such as pharmacologic therapies (e.g., antiviral or antipyretic medications), could also theoretically be associated with some adverse birth outcomes, but distinguishing between potential pathogenic effects of the influenza virus and iatrogenic effects of medications used to manage the infection is complex and requires further study [68]. Infection-related maternal inflammation and hyperthermia can also be elicited by other pathogens, making it unclear whether the influenza virus itself, or its sequelae (e.g., associated symptoms, secondary infections), or both are harmful during pregnancy.

The gestational age at which influenza virus infection occurs is highly relevant when considering possible biological effects of influenza disease. Complex physiological changes that increase with advancing gestation [29] may explain why pregnant women are at highest risk for influenza-related complications during the third trimester [69,70]. From the perspective of birth outcomes, only infection during early pregnancy can plausibly be associated with congenital anomalies or miscarriage; however, for outcomes such as preterm birth, stillbirth or SGA birth, the most susceptible gestational time windows for fetal exposure to maternal influenza are unknown. It is theoretically possible for early pregnancy maternal influenza disease to predispose to a preterm birth or stillbirth at a later gestation, representing a delayed effect of exposure on outcome. But it is more likely that such adverse birth outcomes are precipitated by an acute effect of influenza infection. Conversely, fetal growth restriction resulting in an SGA birth results from pathological processes that take place over a longer duration of time, making it unlikely that influenza disease would have a short-term acute effect.

The results from a systematic review of comparative studies of influenza immunization during pregnancy published up to April 2014 were discussed [7]. The review included 27 studies, 14 of which exclusively evaluated 2009 monovalent pH1N1 vaccines. Three studies of fetal death (at any gestational age) reported non-significant relative protective effect estimates in the range of 0.56–0.79, while four of five studies of early fetal death (<20 weeks’ gestation) reported estimates between 0.89 and 1.23 with confidence intervals including the null value [7]. Four of five studies of stillbirth (fetal death at ≥20 weeks’ gestation) reported estimates ranging from 0.44 to 0.77, including two with statistically significant results [71,72]. Although heterogeneity prevented meta-analysis of the 19 studies reporting preterm birth, studies generally reported either no association between influenza vaccination and preterm birth or decreased risks, some of which were statistically significant [73–76]. The authors inferred no increased risk of preterm birth or fetal death associated with influenza vaccination in pregnancy but emphasized the need for cautious interpretation of the decreased risks reported by several studies, given important methodological limitations of many of the primary studies.

Meeting participants discussed a number of methodological issues concerning observational studies of influenza vaccination in pregnancy, details of which have been published [4,77,78]. The two most important issues identified were: (i) confounding bias due to a more favorable health profile among pregnant women who select vaccination, and (ii) asymmetry between studies of potential harm and potential benefit of vaccination. In observational studies of influenza vaccination in elderly populations, healthier seniors have been shown to preferentially receive influenza vaccination. This selection makes immunization appear strongly protective against a range of health outcomes to an implausible degree, given influenza-attributable risks for these outcomes [79–82]. Statistically significant risk reductions for non-specific outcomes during time periods with no circulating influenza viruses and during seasons with a poor match between the vaccine and circulating virus have demonstrated the extent of this confounding bias [79–83]. The impact of such bias in the obstetrical population has not been extensively studied, but significantly reduced risks of preterm birth and fetal death reported by some observational studies [7], including some with methodologically strong study designs (e.g., survival analyses with time-varying exposures) and multivariable statistical adjustment [71,84], may implicate similar issues.

The importance of distinguishing between potential harm and benefit of vaccination during pregnancy was stressed [4]. To date, most observational studies of influenza vaccination in pregnancy were designed and analyzed to address the former; yet many of these studies have inferred beneficial effects of vaccination on birth outcomes. When assessing a potential harmful effect of vaccination on birth outcomes, all vaccinated women are at risk from the time of vaccine administration, irrespective of the timing of the influenza season. Conversely, an assessment of potential benefit from influenza vaccination must assume that influenza prevention is on the causal pathway and account for temporal features of influenza circulation [4]. Moreover, any beneficial effect of influenza vaccination can be achieved only among the small subset of women whose influenza illness was prevented by vaccination and in whom influenza illness would otherwise (i.e., counterfactually) have resulted in an adverse birth outcome (Fig. 1). To illustrate using preterm birth as an example, in a population with an 11% baseline risk of preterm birth [12], 5% influenza attack rate [85] and 50% vaccine efficacy [86], influenza illness would be prevented in 2.5% of vaccinated women. Assuming a moderate association between preterm birth and influenza virus infection (e.g., risk ratio [RR] = 1.5), the expected risk ratio for preterm birth comparing vaccinated to unvaccinated women would be 0.99 [77]. To obtain a risk ratio for preterm birth of 0.80 in the overall obstetrical population, similar in magnitude to reports from some observational studies [73–76], influenza virus infection would have to result in at least a 2-fold increase in risk of preterm birth, attack rates would have to be in excess of 20%, and vaccine efficacy would have to be at least 70% [77], conditions that are unlikely to be seen in the overall obstetrical population.

The first RCT to assess infant protection against laboratory-confirmed influenza virus infection following maternal influenza immunization was conducted in Bangladesh in 2004–2005. This trial randomized 340 pregnant women to receive either trivalent inactivated influenza vaccine (control arm) or pneumococcal polysaccharide vaccine (experimental arm) in the third trimester of pregnancy. Influenza vaccine was efficacious in preventing respiratory illness with any fever in women during the follow-up period, which extended from two weeks post-randomization to 24 weeks post-delivery (vaccine efficacy [VE]: 35.8, 95% CI: 3.7–57.2). Efficacy against maternal respiratory illness with fever above 38 °C was not statistically significant [2]. Overall, no statistically significant differences were observed in adverse birth outcomes such as SGA birth or preterm birth between the two treatment groups (Table 3) [2]. In post hoc analyses of 116 infants born during the influenza season, however, mean birth weight was significantly increased among infants born to women who received influenza vaccine, compared with those who received pneumococcal polysaccharide vaccine (adjusted mean difference: +190 g, 95% CI: 9–378), while the risk of SGA birth was significantly reduced (adjusted OR: 0.44; 95% CI: 0.19–0.99) [88]. These effects were not observed among 211 infants born when influenza was not circulating.

The second published trial, conducted in South Africa in 2011–2012, demonstrated a benefit of influenza vaccination during pregnancy in preventing laboratory-confirmed maternal influenza virus infection between randomization and 24 weeks post-partum [3]. Among the 2116 non-HIV-infected pregnant women randomized to receive either influenza vaccine or placebo, laboratory-confirmed maternal influenza virus infection was significantly reduced (VE: 50.4, 95% CI: 14.5–71.2) [3]. No differences were seen between the treatment groups in mean birth weight, mean gestational age or risk of preterm or SGA birth [3], even after accounting for maternal pregnancy exposure to time periods when influenza virus was circulating (Table 3) [89].

The group discussion focused on interpretation of trial findings on birth outcomes, as well as potential explanations for the divergence in some findings on birth outcomes between the two published trials. First, unlike therapeutic trials in which all participants have the disease and stand to benefit from the intervention, the expectation in these preventive trials is that any potential beneficial effect of influenza immunization on birth outcomes would occur through prevention of maternal influenza illness during pregnancy. Since the influenza attack rates in these trials were low, even in unimmunized women, and vaccine efficacy in pregnant women was approximately 36–50%, a very large adverse effect of influenza infection on birth outcomes would be required to produce a protective effect of influenza vaccination on birth outcomes in the overall study population. Much of the discussion on this issue revolved around interpreting the 190-g adjusted increase in mean birth weight in the influenza vaccine group (compared with the pneumococcal polysaccharide vaccine group) from the Bangladesh trial. These findings were limited to analysis of the sub-group of 116 infants (58 per treatment group) born during the time period when influenza virus was circulating [88]. For perspective, differences in mean birth weight between smoking and non-smoking mothers are typically around 150–200 g [90], which is of the same order of magnitude as the mean birth weight differences observed between the vaccine groups. Yet, maternal influenza virus infection has not been recognized among the leading risk factors for poor fetal growth, while maternal smoking is one of the strongest known risk factors [90]. Results from the two additional forthcoming BMGF-sponsored clinical trials [87] were recognized as critical for more robust interpretation of possible beneficial effects of influenza immunization on infant birth weight, particularly the Nepal RCT, which had the assessment of birth weight differences as the primary study outcome [87].

Participants also raised general concerns about post hoc analyses. Even when nested within an RCT, analyses of sub-groups defined post hoc are prone to type 1 errors [91]. Moreover, differences in gestational age at vaccination and in the number of days from vaccination to delivery were documented in the Bangladesh trial between the post hoc comparison groups, raising the possibility of confounding bias [88]. On average, influenza-vaccinated women who gave birth during the influenza season were at a slightly more advanced gestation than the pneumococcal-vaccinated women during the same time period, and thus had less gestational time in which to develop influenza-mediated effects on fetal growth. The investigators recognized this and included statistical adjustments for gestational age at immunization and interval from immunization to delivery, but post hoc sub-group testing remains a problem. Stratified analyses by influenza time period were also conducted, but not extensively reported, from the South Africa trial [89].

Finally, several possible explanations for the divergent findings for birth weight between the two published trials were proposed, including local differences in influenza season characteristics or season-specific influenza vaccine components, differences in the comparison groups (active control [2] versus placebo control [3]) and differential effects of influenza virus infection according to underlying characteristics of the maternal-newborn population (e.g., nutritional factors, presence of maternal co-morbidities such as tuberculosis). Again, the importance of additional data from the two forthcoming BMGF-sponsored clinical trials was emphasized, as was the planned pooled analysis of data from the three BMGF trials [87].