Introduction

Infection with influenza virus can result in a spectrum of illness ranging from asymptomatic infection to severe illness and death. Soon after recognition of the novel 2009 pandemic influenza A(H1N1) virus (pH1N1), pregnant women were identified as a high risk group for severe illness [1,2]. Severe illness may be due to respiratory, immunologic, or other physiologic changes that normally occur during pregnancy [3]. Numerous additional reports have further described increased risk for adverse maternal and neonatal outcomes among pregnant women infected with pH1N1[4–12]. Many of these reports called for additional prospective studies with larger sample sizes, robust comparison groups, longer follow-up periods, and additional control for potential confounding factors.

Florida is the fourth most populous state in the U.S. with an estimated population in 2010 of more than 18.8 million residents [13]. More than 200,000 live births occur in the state each year [14]. For each live birth, the birth record contains information regarding baseline demographic and health information of the birth mother, including pre-pregnancy body mass index (BMI), exposures during pregnancy such as tobacco and alcohol use, and maternal and neonatal morbidity indicators such as admission to an intensive care unit (ICU).

The 2009 influenza pandemic began early in Florida, with cases identified in April 2009 and more than 1430 cases reported by June 30, 2009 [unpublished surveillance data]. The summer of 2009 was characterized by persistent transmission across the state, amplified by localized outbreaks in group settings. Following resumption of school, peak transmission was observed in October 2009 [15]. In a national case series of pregnant women with confirmed pH1N1 illness in 2009, Florida had the second highest number of cases, following California [11].

In several of the initial case series of pregnant women with pH1N1 illness, data on birth outcomes was incomplete, and study designs and available data often did not permit control for other factors associated with adverse neonatal outcomes. Longer term follow-up was needed to ascertain pregnancy outcomes, as well as to fully characterize the impact on infant mortality beyond the perinatal period. Population cohort designs, derived from birth certificate data, may permit more precise risk estimates through a population-based comparison group and better control for confounding. In particular, obesity has been previously identified not only as a risk factor for severe pH1N1 illness[16–18], but also as a risk factor for maternal [19] and infant death [20,21]. Pregnant women were excluded from some past studies identifying obesity as a risk factor for severe pH1N1 illness, however, as data were not always available for pre-pregnancy BMI. The objectives of this study were to generate more precise risk estimates for adverse maternal and neonatal outcomes associated with pH1N1 illness identified during pregnancy, and to further explore the role of obesity in these outcomes. To do this, we analyzed data from a retrospective population-based cohort of live births in Florida occurring during the period 2009-2010.

Methods

Prior to the recognition of 2009 pH1N1, infection with a novel influenza virus was already a reportable condition in Florida [22]. Thus, as soon as the pandemic began, health providers in Florida were already required to report all individual cases to the Florida Department of Health (FLDOH). Initially, all laboratory confirmation of pH1N1 in the United States was conducted by the Centers for Disease Control and Prevention (CDC). In May 2009, Florida state public health laboratories became certified for confirmatory testing and later that summer, commercial labs also became certified. Case confirmation required positive test results from viral culture or real-time reverse-transcriptase polymerase chain reaction (rRT-PCR) assays, using primers specific to the novel virus.

On August 3, 2009, the state issued new guidance to health care providers and laboratories to test for influenza and, if positive for pH1N1, report people with influenza that was life-threatening or resulted in death, or resulted in hospitalization of a pregnant woman. Providers and facilities were also asked to report people who were part of clusters or outbreaks of influenza-like illness. Despite the change in disease reporting guidelines, some providers continued for several weeks to report non-hospitalized cases of pandemic influenza that were laboratory confirmed. All individual case reports received by the FLDOH meeting the case definition of an acute respiratory illness, with laboratory confirmation of pH1N1 by a certified laboratory, were maintained in the state’s notifiable disease surveillance database. Illness on the more severe end of the spectrum is more likely to be apparent, lead to specific laboratory testing, and be captured through passive notifiable disease surveillance [23]. Standard data contained in the case reports included patient identifying and demographic information, pregnancy status of individuals at the time of their illness, information regarding treatment with antiviral medication, presence of co-morbid conditions, and outcome of illness.

The Florida birth registry is maintained by the FLDOH Office of Vital Statistics and includes records on all live births occurring in Florida. Records are submitted by hospitals and other medical providers and contain information on the newborn, as well as the mother and father. There are >330 variables contained in the comprehensive electronic birth record, including information on maternal and infant demographic variables; pre-pregnancy BMI, pre-natal exposures such as alcohol and tobacco use by the mother; obstetric history, medical care received during labor and delivery; and health status of the infant. For this analysis, we also used a variable indicating if the mother resided within the city limits, or in a more rural location outside the city limits. In addition, the electronic birth record is updated to indicate mortality for any child dying within five years following live birth.

Electronic records of live births from the Florida birth registry were cross-referenced to the state notifiable disease database to identify children born to women with pH1N1 illness during their pregnancy. All pH1N1 confirmed case-patients with onset of illness between April 24, 2009 and May 31, 2010, who were pregnant at the time of their illness, were included. The record matching included a combination of automated methods and manual review, in an iterative process. Firstly, records in the birth registry were matched to the disease surveillance database by the mother’s social security number and date of birth. Other combinations of matching variables subsequently used included mothers first and last name, alone or in combination with social security number, or date of birth. All matched records were manually reviewed to ensure the accuracy of the match.

Data were further compiled from de-identified records of a birth cohort of all live births occurring in Florida between June 1, 2009 and September 1, 2010. This period corresponds with the time interval of live births among women with reported pH1N1 illness. Records for women with multiple gestational (plural) pregnancies were identified and later de-duplicated to also permit analysis of the cohort of pregnant women giving birth during the same interval. Records were excluded for births among non-Florida residents, and in instances in which adoption or foundling was indicated on the birth record. The database was updated in January 2012 to reflect any infant deaths following birth, occurring up to that time. For secondary analysis, an alternative pregnancy and birth comparison cohort was also constructed from the pre-pandemic period to include all live births occurring between January 1 and February 28, 2009.

Illness due to pH1N1 infection during pregnancy was treated as the outcome variable when assessing disease risk and treated as the primary exposure variable for assessing risk of adverse maternal and neonatal outcomes. Women and children in the pregnancy and birth cohorts were categorized as pH1N1 illness in pregnancy (i.e. ‘ill’ for the first analysis or ‘exposed’ for the second) if the birth records matched to the pH1N1 notifiable disease surveillance database by the methods previously described. All women and their newborns who did not match to the disease surveillance database were defined as ‘unexposed’ or non-cases of pH1N1 illness during pregnancy. Measures of association between exposure and outcome were computed using univariate and multivariable logistic regression and expressed as odds ratios (OR) with 95 percent Wald confidence intervals (95% CI). Univariate relative risks (RR) were also computed for comparison with ORs, but were not presented in the results tables. Because minimal differences were generally noted between the univariate RRs and ORs and because outcome frequencies were generally uncommon and less than 10%, odds ratios were used throughout as approximate estimates of relative risk. To adjust for possible confounding, standard and stepwise multivariable logistic regression was used. Multivariable models were constructed using variables identified during univariate analyses as being associated with influenza, or known risk factors for adverse maternal and neonatal outcomes, while also attempting to minimize the exclusion of cohort members due to missing data for a particular variable. For all stepwise analysis, α=0.1 was the threshold used for variable entry into the model, and α=0.05 was used for retaining variables in the final model. Data analysis was conducted using SAS software, version 9.1 [SAS Institute; Cary, NC]. Statistical power computations presented in the results were obtained using the computer program Power v3.0 [National Institute of Health, National Cancer Institute].

Data on pre-pregnancy BMI was available for approximately 94% of records and was expressed as [weight in kilograms/(height in meters)²]. BMI data in the Florida birth registry has previously been found to be generally reliable and valid [24]. Nevertheless, the top and bottom 0.5% of BMI values were excluded as non-valid, resulting in the inclusion of BMI values only between 16.2 and 49.1. Standard cut points were used to categorize women based on pre-pregnancy BMI [25] and analyses were conducted using both 4 and 5 different categories. For analysis using 4 categories, the BMI cut points were as follows: underweight (<18.5), normal (18.5-24.9), overweight (25.0-29.9), and obese (≥ 30.0). For analysis using 5 categories of BMI, the obese grouping was further divided into obese (30.0-34.9) and very obese (≥ 35.0). Analyses were also conducted using BMI as a continuous variable. Since pre-pregnancy BMI was not available for approximately 6% of women in the cohort, analyses were conducted both with and without BMI in the models. Interaction between pH1N1 illness and obesity was also assessed in relation to the most severe maternal and neonatal outcomes by including in the logistic model, the first level interaction term between pH1N1 illness and obesity (yes/no), defined as pre-pregnancy BMI ≥ 30.0.

For simplicity, adult BMI categories were used throughout the analysis, despite the fact that some gravida in the analysis were <20 years old, and standard classifications for BMI are different for children and adolescents [26]. Maternal age at time of birth was alternatively treated as either a continuous variable, dichotomized as ≤ 25 and >25 years, or classified into 3 groups consistent with previous analyses using Florida birth registry data (<18, 18-39, ≥ 40) [21]. The last grouping is generally consistent with accepted risk categories for adverse maternal and fetal outcomes. Continuous variables such as birth weight and gestational age at birth were categorized using generally accepted cut points to assess adverse neonatal outcomes. Admission to an Intensive Care Unit (ICU) was used as an indicator of severity of illness for both adverse maternal and neonatal outcomes.

Data regarding influenza vaccination history and antiviral treatment for influenza was missing from the surveillance case reports for a large proportion of case-patients. Furthermore, the majority of case-patients had illness onset before a vaccine specific for pH1N1 became available. Therefore, we did not attempt in this analysis to assess the role of vaccination or antiviral treatment for influenza on maternal and neonatal outcomes.

Results

A total of 194 women with confirmed pH1N1 illness during pregnancy were reported to FLDOH. Of these, matching birth certificates were identified for 187 women (96%); 4 women gave birth to twins, resulting in a total of 191 live births (Figure 1). Of these 191 children, 8 died following live birth. Of the 194 pregnant women with pH1N1 illness, 8 (4%) women died; 6 following live birth of their infants. Seven of the 194 pregnant pH1N1 case-patients did not have corresponding birth certificates. Two of these women were lost to follow-up without a matching birth certificate, and a live birth did not occur for the other 5 women: 2 pregnancies ended in miscarriage, 2 ended in fetal loss secondary to maternal death, and 1 pregnancy was voluntarily terminated (Figure 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Distribution of case-patients by vital status and birth record matching.

https://doi.org/10.1371/journal.pone.0079040.g001

Among pH1N1 case-patients, the temporal pattern of cumulative onset of influenza illness and subsequent birth is shown in Figure 2. Of the 194 pregnant case-patients, approximately 24% had symptom onset and laboratory confirmation prior to August 2009, when disease reporting guidelines were changed. The number of cases declined rapidly after November 2009. Of the 194 women with pH1N1 illness identified through surveillance, 24 (12%) were not hospitalized for their influenza illness. Onset of illness for non-hospitalized case-patients ranged from June 2009 to February 2010, with more than two-thirds having illness onset after August 2009.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Cumulative pandemic influenza illness in pregnant women by illness and delivery date.

https://doi.org/10.1371/journal.pone.0079040.g002

Non-pH1N1 associated individuals in the birth and pregnancy cohort included, after exclusions, 300,398 children born to 295,747 women between June 1, 2009 and September 1, 2010. A total of 4,198 birth records were excluded from the cohort due to non-Florida residence of the mother (n=1,547), child adopted (n=2,648), or foundling (n=3). The alternative comparison birth and pregnancy cohort from the pre-pandemic period included, after exclusions, 24,008 children born to 23,679 women.

Discussion

We observed a consistent, and statistically significant association between pH1N1 illness during pregnancy and adverse neonatal and maternal outcomes, including low birth weight, preterm birth, neonatal ICU admission, infant death, and maternal ICU admission. The observed associations with pH1N1 illness remained significant after adjusting for other observed and known risk factors for adverse neonatal and maternal outcomes. As the disease surveillance data sources used for this analysis are more likely to detect case-patients with more severe influenza illness, our findings apply more toward outcomes associated with more severe influenza illness, rather than influenza infection in general.

While several previous case series have described neonatal outcomes of pH1N1 infected women [5,8,11,12,27–30] few past reports used robust comparison groups or cohort designs to permit estimation of risk for adverse neonatal outcomes. Our estimates of risk for adverse neonatal outcomes are generally consistent with those previously reported by Pierce, et al, using a similar cohort design from the United Kingdom [31]. Methodological differences likely account for the slight differences in the magnitude of risk estimates between the 2 studies. However, in both studies, illness with pH1N1 during pregnancy was consistently and significantly associated with low and very low birth weight, late and early pre-term birth, and infant death, after adjusting for other factors.

In addition to the adverse neonatal outcomes we observed, the pandemic had a striking impact on maternal mortality in Florida, as elsewhere. Data from the Florida Pregnancy-Associated Mortality Review [32] indicate 58 pregnancy related deaths during 2009 [33]. The pregnancy-related mortality ratio (PRMR) in Florida for 2009 was 26.2 deaths per 100,000 live births, representing the largest annual PRMR in the previous ten years. This contrasts to the most recent national PRMR estimate available of 15.1 for the period 2006-2007 [34]. Of the 58 pregnancy related deaths in Florida in 2009, 15 (25.9%) were related to infection. This contrasts to the previous 10 year period in which infection accounted for approximately 10% of pregnancy related deaths. Of the 15 infection related deaths, 13 of these were associated with acute respiratory illness, and 8 were confirmed pH1N1 cases identified in our notifiable disease surveillance database, yielding a pH1N1 cause-specific PRMR of 3.6 in Florida during 2009. This compares to a pH1N1 cause-specific maternal mortality ratio of 4.3 reported from California [10] and 3.0 reported from Turkey [7]. Of the 8 pH1N1 case-patients in our analysis that died, only 2 received treatment with neuraminidase inhibitors within 48 hours of illness onset. Delayed antiviral treatment has repeatedly been identified as a risk factor for maternal ICU admission and death [5,8,10,11].

We identified younger age, racial and ethnic minorities, indicators of lower socio-economic status (SES), and obesity as risk factors for pH1N1 illness severe enough to be detected through our surveillance. Others have noted both an individual and ecological-level association between hospitalization for pH1N1 and indicators of low SES [35]. One possible explanation for the association with lower SES observed in our study is that surveillance case reporting was somehow biased toward reporting illness among those on public assistance. We have no evidence that surveillance case reporting was biased in this way. We also found that urban residence and previous live births were risk factors for illness. Increased risk associated with urban residence and presence of young children in the household are consistent with what is known of influenza transmission and what was observed in the United States during this pandemic [15].

Obesity was both an observed risk factor for recognized and reported pH1N1 illness, as well as a risk factor for poor maternal outcomes among pregnant women, as indicated by ICU admission. Obesity may have important effects on immune response which leave obese individuals more susceptible to infection [36–38]. The observed association between obesity and pH1N1 illness in this cohort, however, should be interpreted with caution. Due to the nature of surveillance and case reporting used to ascertain cases, our data likely underestimate the total number of pregnant women with pH1N1 infection and overestimate the proportion of pregnant women with severe illness. Thus the observed association with obesity likely relates at least partly to risk of more severe illness detected by our surveillance, rather than risk of infection. Coleman, et al previously concluded that obesity was not an independent risk factor for medically attended influenza, among adults with acute respiratory illness in a population based study [39].

Others have previously identified obesity as a risk factor for severe illness with pH1N1 [16–18,40,41]. Both Louie and Morgan reported that hospitalization and death from pH1N1 were associated with morbid obesity, however, both studies excluded pregnant women. Kwong et al, noted hospitalizations for respiratory illness during influenza season were consistently associated with obesity over several years, however this analysis also excluded pregnant women. Morbid obesity is associated with reduced fertility[42] making the condition less common among pregnant women than the general, reproductive-aged female population. There were a very small number of women with pH1N1 illness in our cohort with BMI ≥ 40, limiting our analysis of the effect of morbid obesity on maternal and neonatal outcomes. Obesity is also associated with certain complications and adverse outcomes of pregnancy such as gestational diabetes, preeclampsia, macrosomia, still birth, infant death, cesarean section delivery, and possibly certain birth anomalies and elective pre-term delivery [20,21,43,44]. Therefore, an important strength of our study was the availability of data on pre-pregnancy BMI for most cohort members, which allowed us to control for BMI when estimating risk for adverse neonatal outcomes associated with pH1N1 illness. Nevertheless, we observed no interaction between pH1N1 illness and pre-pregnancy obesity, as it relates to risk of maternal ICU admission, early preterm birth, or infant mortality.

Pierce et al, used a similar approach to ours, however, the comparison cohort they used was smaller and taken exclusively from the pre-pandemic period [31]. The disadvantage of using a comparison group from the pre-pandemic period is that, while there is no chance any of those in the comparison cohort were infected with the pandemic virus, it fails to account for women who may have been infected with seasonal influenza during pregnancy, and the impact this may have had on maternal and fetal outcomes. As imperfect as disease surveillance was in the initial months of the pandemic, it was better at detecting individual cases of influenza in Florida than during the pre-pandemic period, when individual cases of seasonal influenza were not reportable to the health department. For this reason, we chose the primary comparison group for our analysis to coincide with the period of infection and births among the pH1N1 case-patients in our analysis. Nevertheless, results of secondary analysis using a cohort from the pre-pandemic period did not substantially alter our findings.

By the end of 2009, an estimated 25% of residents in some areas of Florida were believed to have been infected with pH1N1, with the highest rates of infection noted among school-aged children 5-17 years, and infection rates approaching 50% among young adults 18-24 years [45]. Seroprevalence studies conducted elsewhere suggest similar findings [46–48], however, work from Australia suggests that infection rates may have been lower in pregnant women [46]. If the Florida estimate of 25% infection by the end of 2009 also holds true for pregnant women, then far more than the 194 women we observed would have been infected with pH1N1 during pregnancy. We attempted to overcome limitations in the case ascertainment by also searching text fields for H1N1 in the birth record, and this did not substantially alter our results. Nonetheless, the results we present are likely to represent those associated with the severe end of the spectrum of influenza illness, as women with mild illness during pregnancy may have gone undiagnosed and unrecognized in our analysis.

The study is limited by imperfect disease surveillance and case ascertainment of all women with pH1N1 illness during pregnancy. Therefore, our findings are only generalizable to the more severe pH1N1 illness detected through our surveillance, not to pH1N1 infection in general. Imperfect disease surveillance may have led to both random and differential misclassification bias of pH1N1 illness, which was both an outcome and exposure variable in this analysis, possibly resulting in some bias in the risk estimates. This study was also limited by lack of comparative data for women without a live birth and missing data on pre-pregnancy BMI for approximately 6% of the women. Data available in the birth record regarding maternal morbidity, (e.g. ICU admission), also appears to be incomplete, somewhat limiting our conclusions regarding maternal outcomes. Prospective studies may help address all of these limitations. However, the advantages of our study over past reports include a very large comparison group with extensive information available from the birth record to facilitate control for confounding factors, as well as longer follow-up to allow ascertainment of child mortality, through the birth record, several months after birth.

Conclusions

The present study uses a population-based cohort design to add to and further characterize the previously described risk of adverse maternal and neonatal outcomes associated with pH1N1 illness. In conclusion, we found increased risk of severe adverse maternal and neonatal outcomes associated with pH1N1 illness during pregnancy. Pregnant women should continue to be regarded as a high risk group for adverse outcomes of influenza virus infection and should continue to be targeted for enhanced prevention activities through vaccination and appropriate chemoprophylaxis and treatment for influenza, consistent with current recommendations [3,49,50].

Acknowledgments

Disclaimer: the findings and conclusions in this report are those of the authors and do not necessarily represent the official position of the Centers for Disease Control and Prevention or other funding sources.

We gratefully acknowledge the many people working in the medical and public health sector in Florida who contributed to the data sources used in this analysis. We thank Richard Hopkins for advice regarding study design and critical review of the manuscript; Daniel Thompson for assistance with data matching; and Kristopher Arheart for statistical and programming advice. Finally we thank the anonymous peer-reviewers for volunteering their time to review and help strengthen the report.