Incidence of influenza virus infection among pregnant women: a systematic review

We conducted a systematic review in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) reporting recommendations [11]. The search strategy was developed and tested through an iterative process by an experienced medical information specialist (BS) in consultation with the review team. We searched PubMed, Embase, the Cochrane Library, and CINAHL Plus with Full Text on February 20, 2015. A second, add-on search utilizing keywords for “disability-adjusted life years” (DALYs) was performed in the same databases on February 28, 2015. Searches were updated biweekly through September 2, 2015. Strategies utilized a combination of controlled vocabulary (e.g., “Influenza A Virus”, “Pregnancy Complications”, “Incidence”) and keywords (e.g., influenza, prenatal, rate). Vocabulary and syntax were adjusted across databases. No language or date restrictions were applied but animal-only studies and opinion pieces (comments, editorials, interviews) were removed from the search results where possible. We did not conduct a review of grey literature. Specific details regarding the strategies appear in Additional file 1.

Following de-duplication, search records were uploaded into Abstrackr [12], an online systematic review tool, and screened by two independent reviewers (MK, JJ) to identify studies for full-text review. The two independent reviewers (MK, JJ) each read all of the full-text manuscripts and performed the final selection of included manuscripts. Discussion between the two reviewers was used to resolve disagreements.

We targeted studies that, regardless of methodology, included original data on any of the following outcomes in pregnant women: influenza virus infection of any kind; symptomatic influenza virus infection; influenza virus infection associated with hospital admission; influenza virus infection associated with intensive care unit (ICU) admission; influenza-associated death; and influenza-associated disability-adjusted life years (DALYs) lost. Only studies that had laboratory confirmation for influenza virus infection and included population-based incidence rates with denominator data were included. We excluded editorials, commentaries, opinion pieces, narrative reviews, clinical practice guidelines, conference abstracts, or literature not in peer-reviewed journals.

A data collection form was developed to abstract information from each manuscript regarding study country; study population; hospitalization status; time period; subject inclusion and exclusion criteria; laboratory methods used to confirm influenza diagnosis; whether the study evaluated pandemic or seasonal influenza;[13] methodology for determining numerators and denominators for calculations; and crude and adjusted incidence rates, with 95% confidence intervals, for each outcome. For the purpose of this review, we considered 2009 pandemic influenza A H1N1 to be seasonal influenza after August 2010, when WHO declared the pandemic over [13].

Two independent reviewers (MK, JJ) evaluated the quality of each study. Cohort studies were assessed using the Newcastle-Ottawa Scale [14]. Randomized clinical trials were evaluated using the Cochrane Collaboration Risk of Bias Tool [15], and case series were assessed using the JBI Critical Appraisal Checklist for Descriptive/Case Series [16]. For all of these scales, lower scores imply higher risk of bias.

We also evaluated the quality of the evidence for each outcome using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) assessment criteria – both the traditional criteria and the assessment criteria related to prognosis [17, 18]. GRADE was initially developed to provide methodological guidance in reviewing the quality of evidence for intervention research, and later adapted for use in evaluating diagnostic and prognostic studies [17]. The approach considers five factors that can influence our confidence in estimates of effects: (1) study design and limitations in study design, (2) inconsistency of results across studies, (3) indirectness of the evidence, (4) imprecision and (5) publication bias; and three factors related to observational studies: (1) large estimates of treatment, (2) a dose-response gradient and (3) plausible confounding that would increase confidence in an estimate [17]. In our review, beginning with cohort studies at a high level of evidence, according to the GRADE methodology, the level of confidence in the estimate of effects was downgraded according to the above criteria. Because we analysed influenza attack rates rather than effect estimates from interventions to prevent influenza, we could not apply the criteria relating to the magnitude and dose-response of the effect estimates. For each outcome, we applied a final GRADE assessment according to one of four categories-high quality, moderate quality, low quality or very low quality-based on the upgrading or downgrading of the quality of evidence for each outcome.

We summarized study characteristics in descriptive tables and outcome-specific forest plots. For each outcome of interest, we used incidence rates reported in the manuscript when they were available. When incidence rates were not provided, we computed incidence rates with 95% confidence intervals (CI) from raw data using the binomial exact method and the variance stabilizing transformation for proportions, as proposed by Freeman and Tukey,[19] using the Stata user command “metaprop” [20]. We generated summary incidence rates using random effects models and assessed statistical heterogeneity by visual examination of forest plots, and by χ ² and I² tests [21, 22]. We did not pool estimates if methodological or statistical heterogeneity was high. We considered statistical heterogeneity to be high if the I² value was ≥75%[23]. Forest plots were prepared using the R package “ggplot2.”

Our search results yielded 1543 records. After excluding duplicates, we screened 1041 titles and abstracts. We identified 31 potentially relevant articles, and, after reviewing the full manuscripts, we found 10 articles that met the study inclusion criteria (Fig. 1). The results of one study from Australia and New Zealand [24] were subsequently found to be included in another comparative study, [25] so the former study was excluded, leaving nine total articles. Three were cohort studies, [28‐30] five were case series, [26, 27] and one was a randomized controlled trial [5].

Four studies were from the United States,[26, 29, 31, 32] three were from the United Kingdom,[27, 28, 30] one was from South Africa,[5] and one included data from Australia, New Zealand and the United Kingdom [25]. All studies were from high-income countries, except for the one study from South Africa, an upper-middle income country [33]. Six studies included pandemic influenza data, [25‐27, 29‐32] and three studies included seasonal influenza data, [5, 28, 30] [5]. Among the nine studies, three studies reported rates of influenza virus infection, including asymptomatic infection;[28‐30] three studies reported rates of symptomatic influenza virus infection;[5, 26, 32] four studies reported influenza-associated hospital admission rates;[26, 27, 31, 32] four studies reported influenza-associated ICU admission rates;[25, 26, 31, 32] and four studies reported influenza-associated mortality rates [25, 26, 31, 32]. We did not identify any relevant studies that reported DALYs associated with influenza virus infection.

The methodologic approach to calculating denominator data varied among the studies (Table 1). Only one study defined the specific testing that was required for the determination of pregnancy [5].

Applying the GRADE framework, all outcomes were deemed to be based on a very low quality of evidence (Additional file 2: Table S1). We based this determination on a number of factors, which collectively determined the final assessment: in most of the studies, there were significant risks of diagnostic ascertainment bias related to the frequency of testing patients and the variability of reporting results. In addition, the reporting period varied in many of the studies evaluating the same outcome, making it difficult to compare the same outcome measure across studies. There was high heterogeneity in results across studies measuring the same outcome. Finally, many of the studies were case studies, and the sample size in a number of the studies was quite small (see study sample sizes in Additional file 2).

The Newcastle Ottawa scores for the five observational studies ranged from 6 to 9, out of a maximum of 9. The observational studies had suboptimal representativeness of study participants [28‐30], inadequate assessment of outcome [29], or important diagnostic ascertainment biases [26, 27]. Each of the three case series had a score of 6 out of 7 according to the JBI Critical Appraisal Checklist [16]. All but one [26] of the studies were carried out over relatively short time periods that captured only a portion of the period of activity of seasonal or pandemic influenza. For the five outcomes of interest that the studies addressed, the I² test demonstrated high statistical heterogeneity, ranging from 93.1% (mortality) to 99.7% (hospitalization) (Additional file 2). Because of this high heterogeneity for each outcome, and because most studies within each outcome employed different methodologies-different approaches to diagnostic ascertainment and different durations of follow-up-we did not pool incidence rates for any of the outcomes.

We identified three articles that described rates of serologically confirmed influenza virus infection, including clinical and asymptomatic infection, in pregnant women (Table 2, Fig. 2a) [28, 30]. Two studies reported rates for serologically confirmed seasonal influenza [28, 30]. and the Hardy et al. study described serologically confirmed influenza data from the 1957 influenza pandemic [29]. The two studies reporting on seasonal influenza virus infection did not mention the women’s vaccination status [28, 30], while the pandemic study excluded the small number of vaccinated women from the analysis [29]. These three studies used serological testing to confirm influenza virus infection, while the other studies use RT-PCR to confirm influenza virus infection. In all three serological studies, influenza virus infection incidence ranged from 483 to 1,097 cases per 10,000 pregnancies. The three serological studies had Newcastle Ottawa scores of 6 to 7. Across studies both the testing methodology and the sampling frame varied (Additional file 2, Table S1).

Three studies-two observational studies describing 2009 pandemic influenza (Doyle, 2013, and Jamieson, 2009) [26, 32] and one randomized clinical trial-Madhi et al. [5]—that described seasonal influenza-reported rates for symptomatic influenza virus infection among pregnant women in South Africa (Table 3, Fig. 2b). The two observational studies together reported a total of 128 confirmed cases of symptomatic influenza virus infection, and had a combined denominator of nearly 3.7 million pregnant women. In contrast, the clinical trial from South Africa included a total of 2,310 HIV-infected and uninfected pregnant women and reported 80 cases of symptomatic influenza virus infection among vaccinated and unvaccinated women combined. The two observational studies [26, 32] reported rates of 0.10 and 6.6 cases of symptomatic influenza virus infection per 10,000 pregnancies. The trial from South Africa [5] found an annual incidence of 371 cases of symptomatic influenza virus infection per 10,000 pregnancies among HIV-uninfected women and 1,818 cases of symptomatic influenza virus infection per 10,000 pregnancies among HIV-infected pregnant women. Approaches to case ascertainment varied considerably among the three studies; in the two observational studies (Doyle, 2013, and Jamieson, 2009), providers had to initiate testing for 2009 pandemic (H1N1) influenza through local or state public health laboratories, which then reported the results to the state health department or the US Centers for Disease Control. In contrast, during the South African vaccine trial, all participants were questioned weekly by study staff about symptoms consistent with influenza-like illness (ILI), and patients with ILI were immediately sampled for influenza testing.

We identified four studies-all case series (Jamieson, 2009, Creanga, 2010, Yates, 2010, and Doyle, 2013)-that described influenza-associated hospital admission rates (Table 4, Fig. 2c) [26, 27, 31, 32]. All four included data from the 2009 influenza pandemic. Incidence rates of influenza-associated hospitalizations ranged from 0.04 to 7.7 cases per 10,000 pregnancies. While all four reported rates for 2009 pandemic (H1N1) influenza, the duration of surveillance varied among the studies, from one month [32] to 13 months [26]. In addition, methods of case ascertainment varied among the four studies. The three studies from the United States relied on passive reporting during the influenza pandemic, all to different government health authorities-local [31] state [26], and federal [32]—and all covered slightly different periods of the pandemic. As opposed to the United States studies, the Yates et al. study from the United Kingdom reflected the second wave of the pandemic [27]. The overall number of cases of influenza-associated hospital admission ranged from 14 to 271 in the four studies. Across studies, there was a high risk of diagnostic ascertainment bias related to the inconsistent frequency of testing patients and variability of reporting results to city, state, and federal public health authorities (Additional file 2, Table S3).

Four studies (Jamieson, 2009, Creanga, 2010, Doyle, 2013, and Knight, 2011) described influenza-associated ICU admissions among pregnant women (Table 5, Fig. 2d) [25, 26, 31, 32]. The four studies, all case series, were all conducted during the 2009 influenza pandemic and reported rates of 0.01 to 6.8 cases of influenza-associated ICU admissions per 10,000 pregnancies. The same four studies also reported rates of influenza mortality among pregnant women during the 2009 influenza pandemic, which ranged from 0.03 to 0.69 influenza-associated deaths per 10,000 pregnancies (Table 6, Fig. 2e). Methodologies for case detection and the duration of surveillance varied among the studies. Two studies from the United States reported findings from the initial month [32] or two months [31] of the pandemic, while Knight et al. reported rates from the United Kingdom during a surveillance period that occurred during the second, smaller peak of the pandemic [25]. The absolute number of laboratory-confirmed influenza ICU admissions and laboratory-confirmed influenza deaths was small; the total number of influenza-associated ICU cases in each of the four studies ranged from 3 to 59, while the total number of influenza-associated deaths in each of the studies ranged from 1 to 6. (Additional file 2: Tables S4 and S5).

In our systematic review of the scientific literature, we found few studies that described incidence rates of laboratory-confirmed influenza-associated outcomes in pregnant women. We identified a maximum of four studies for each of the outcomes of interest, and through an evaluation that used the GRADE criteria, we rated the overall quality of the evidence to be very low for all outcomes. The Madhi et al. study [5] was the only randomized controlled trial, and this had high quality data, while the other eight articles were based on cohort studies and case series and were considered to be of lower quality. Only three articles [5, 28, 30] described seasonal influenza, and two of these studies [28, 30] included information mostly on subclinical infection. Nearly all studies (5 out of 6) that reported symptomatic infection, hospitalizations, ICU admissions and death described data from 2009 pandemic (H1N1) influenza. There were no data from low-income countries.

The pandemic influenza studies, while more numerous, varied considerably in their approaches to case ascertainment. In particular, the three studies from the United States that described rates of 2009 pandemic (H1N1) influenza-associated outcomes among pregnant women during the pandemic relied on provider initiative to send samples from suspected cases for testing, [26, 31, 32] and also required that public health laboratories report cases to the city [31] the state [26] or the US Centers for Disease Control [32]. For this reason, the number of reported cases is likely a substantial underestimation of the true number of cases in these studies. The variability in surveillance for laboratory-confirmed influenza during the pandemic was likely in part due to the limited availability initially of laboratory testing, which is a challenge inherent to pandemics of novel viruses.

Additionally, studies reporting results from the 2009 influenza A (H1N1) pandemic were conducted at different time periods during the pandemic. Two studies captured less than two months of surveillance, [31, 32] and the studies from the United Kingdom captured only the second wave of the pandemic [25, 27]. Differences in timing and duration of surveillance periods make comparisons challenging. For example, previous studies have shown that the first and second waves of the 2009 pandemic varied considerably in many countries [34‐36]. Further, incidence rates of illness during the influenza season are likely to be substantially higher than incidence rates of illness observed over an entire calendar year.

We limited our search to studies with laboratory-confirmed influenza virus infection, which may have underestimated case rates because of variable testing and reporting of results. Other studies have reported estimated incidence rates of influenza-associated outcomes among pregnant women without laboratory confirmation of influenza disease. A study that used multiple data sources-not all of which contained laboratory-confirmed influenza disease-to quantify influenza-associated hospitalizations in Canada from 1994–2000 reported a rate of 150 influenza-associated hospitalizations (95% CI: 140–170) per 100,000 pregnant women per influenza season [37]. This rate was more than 10-fold higher than the highest rate reported by the four studies of laboratory-confirmed influenza hospitalizations among pregnant women in our study. A study of symptomatic probable and confirmed cases of 2009 pandemic influenza among pregnant women on Reunion Island found a rate of 3,568 cases per 100,000 pregnant women (95% CI: 3,267–3,868),[38] an incidence similar to the highest rates of symptomatic influenza reported by the Madhi et al. study in our review [5]. While these two studies may have overestimated the number of cases given their lack of laboratory confirmation, they provide an upper range of incidence for comparison.

Other studies that did not rely exclusively on laboratory-confirmed influenza virus infection have found rates of influenza-associated deaths in pregnancy that were similar to those that we found. A recent publication on confirmed or possible 2009 influenza A (H1N1) influenza virus infection from April to June 2009 in the United States [39] reported a rate of 2.2 influenza-associated maternal deaths per 100,000 live births, which was in the range of 3 of the 4 studies reported in our review, all of which had wide confidence intervals. A study that used diagnostic billing (ICD-9) codes to determine influenza-associated deaths among pregnant women during the influenza season from 1998 to 2005 found rates that ranged from 0.01 deaths per 10,000 live births to 0.059 deaths per 10,000 live births per influenza season,[40] which corresponded with the low end of reported incidence rates from the four pandemic influenza studies in our review.

Studies are lacking that describe rates of laboratory-confirmed influenza in non-pregnant women of childbearing age, which makes it difficult to assess pregnancy as a risk factor for severe influenza illness. A study from the United States that evaluated rates of influenza-attributable hospitalizations for and deaths from pneumonia, influenza, and other selected acute cardiopulmonary conditions (identified by ICD-9 codes) found that among non-pregnant women aged 15 to 44 without high-risk conditions, an estimated 4 influenza-attributable hospitalizations and deaths (combined) per 10,000 person-years occurred. Among women aged 45 to 64 years without high-risk comorbidities, 6 influenza-attributable hospitalizations and deaths (combined) per 10,000 person-years occurred [41]. A modelling study from South Africa estimated mean annual seasonal influenza-associated mortality to be 0.9 per 100,000 person-years among HIV-negative non-pregnant women. The estimated mean annual influenza-associated mortality in 2009, when the pandemic circulated, was 9.4 per 100,000 person-years among HIV-negative non-pregnant women [42]. In comparison, the rates for hospitalizations associated with laboratory-confirmed influenza among pregnant women described in the articles in our literature review were in a similar range to the rates for non-pregnant women in the US study. The mortality rates for pandemic influenza in the studies that we included were slightly lower than the estimated rate in the South African study for non-pregnant women. However, these comparisons are considerably limited by the differences in how cases were defined. In addition, the US studies described seasonal influenza only, while all four studies included in our review described rates associated with pandemic influenza.

For this report, we employed a comprehensive literature search strategy and assessed the quality of the evidence for each article; both are strengths of the study. We had a small number of studies for each outcome, with the exception of DALYs, for which we did not identify any studies. We found very few relevant studies that described seasonal influenza, no studies from a low-income country, and only one study from a middle-income country. The diverse methodologies, particularly regarding case ascertainment, and the variable durations of the studies made it methodologically inappropriate to pool results across studies for any of the outcomes. Measures of disease according to severity may have been affected by contextual factors. Decisions to seek care as an outpatient for influenza-like symptoms, and decisions to admit patients to a hospital and to an ICU may vary geographically [25]. Furthermore, decisions to test patients, when not standardized, likely vary by physician, region, and country. In addition, while we identified events associated with influenza virus infection, we were unable to discern events that were causally linked to influenza virus infection. Pregnant women may have complications of influenza virus infection but have no evidence of influenza at the time of hospitalization, and others may have coincidental influenza virus infection that is unrelated to their primary disease process.

Our literature review did not capture grey literature, including unpublished reports of influenza surveillance in pregnant women. While some unpublished surveillance reports describe important trends in the outcomes of interest in this study, denominator data may be lacking,[43] and influenza cases in these reports are often not confirmed by laboratory testing [44]. Therefore, many of these reports would not likely have met our study’s inclusion criteria.

Additionally, there are no consensus scoring tools to assess quality of the included influenza surveillance studies. The Newcastle-Ottawa Scale, the Cochrane Collaboration Risk of Bias Tool, the JBI Critical Appraisal Checklist for Descriptive/Case Series, and GRADE are all helpful tools for evaluation data quality, but they may not have identified all of the quality elements of included studies.