Epidemiology Publish Ahead of Print DOI: 10.1097/EDE.0000000000000753 The relationship between 2009 pandemic H1N1 influenza during pregnancy and preterm birth: a population-based cohort study
EP TE D
bbreviated Title: Pandemic influenza and preterm birth Deshayne B Fell,a,b,c,d Robert W Platt,d Olga Basso,d,e Kumanan Wilson,c,f ,gJay S Kaufman,d David L Buckeridge,d Jeffrey C Kwongc,h,I,j,k a.
Children’s Hospital of Eastern Ontario Research Institute, Ottawa, Canada
b.
School of Epidemiology, Public Health and Preventive Medicine, University of Ottawa, Ottawa, Canada
c.
Institute for Clinical Evaluative Sciences, Ottawa and Toronto, Canada
d.
Department of Epidemiology, Biostatistics and Occupational Health, McGill University, Montreal, Canada
Department of Obstetrics and Gynecology, McGill University, Montreal, Canada
f.
Department of Medicine, University of Ottawa, Ottawa, Canada
g.
Clinical Epidemiology Program, Ottawa Hospital Research Institute, Ottawa, Canada
h.
Department of Family and Community Medicine, University of Toronto, Toronto, Canada
i.
Dalla Lana School of Public Health, Toronto, Canada
j.
Public Health Ontario, Toronto, Canada
A
C C
e.
k.
University Health Network, Toronto, Canada
Correspondence: Deshayne Fell, Children’s Hospital of Eastern Ontario Research Institute, 401 Smyth Road, Centre for Practice Changing Research, Room L-1154, Ottawa, Ontario, K1H 8L1, Canada. E-mail: [email protected] 1
Copyright © Wolters Kluwer Health, Inc. All rights reserved. Unauthorized reproduction of this article is prohibited.
Conflicts of Interest: The authors have no conflicts to declare. Source of Funding: This study was partially funded through an Operating Grant from the Canadian Institutes for Health Research (#MOP-119570). This study was supported by the Institute for Clinical Evaluative Sciences (ICES), which is funded by an annual grant from the
EP TE D
Ontario Ministry of Health and Long-Term Care. Data Access: The dataset from this study is held securely in coded form at the Institute for
Clinical Evaluative Sciences (ICES). While data sharing agreements prohibit ICES from making the data set publicly available, access may be granted to those who meet pre-specified criteria for confidential access, available at www.ices.on.ca/DAS.
Acknowledgments: DBF was supported by a Canadian Institutes of Health Research (CIHR) Doctoral Award while this study was being conducted. RWP is supported by a Chercheur-
national (National Scholar) award from the Fonds de Recherche du Quebec – Santé (FRQ-S) as
well as core support to the Research Institute of the McGill University Health Centre from FRQS. JCK is supported by a New Investigator Award from CIHR and a Clinician Scientist Award
C C
from the University of Toronto Department of Family and Community Medicine. We are
grateful to Better Outcomes Registry & Network (BORN) Ontario, Ottawa, Canada and the Institute for Clinical Evaluative Sciences (ICES) for providing data access. Disclaimer: The opinions, results and conclusions reported in this paper are those of the authors
A
and are independent from the funding sources. Parts of this material are based on data and information compiled and provided by the Canadian Institute of Health Information (CIHI). However, the analyses, conclusions, opinions, and statements expressed herein are those of the authors, and not necessarily those of CIHI. No endorsement by ICES, the Ontario Ministry of Health and Long-Term Care, or CIHI is intended or should be inferred. 2
Copyright © Wolters Kluwer Health, Inc. All rights reserved. Unauthorized reproduction of this article is prohibited.
Abstract word count: 242 Text word count (excluding references): 3242 Total number of pages (including references, but excluding supplementary files): 32 Number of text pages (including references, but excluding supplementary files): 20
EP TE D
Number of table pages (excluding supplementary files): 11
A
C C
Number of figure pages (excluding supplementary files): 1
3
Copyright © Wolters Kluwer Health, Inc. All rights reserved. Unauthorized reproduction of this article is prohibited.
ABSTRACT Background: Previous studies of maternal influenza illness and preterm birth have yielded inconsistent results. Our objective was to assess the association between 2009 pandemic H1N1 (pH1N1) influenza during pregnancy and preterm birth in a large obstetrical population.
EP TE D
Methods: We linked a province-wide birth registry with health administrative databases to identify influenza-coded hospitalizations, emergency department visits, or physician visits among pregnant women during the 2009 H1N1 pandemic (our proxy for clinical pH1N1
influenza illness). Using Cox regression, we estimated adjusted hazard ratios (aHR) for preterm birth and spontaneous preterm birth treating influenza as a time-varying exposure.
Results: Among 192,082 women with a singleton live birth, 2,925 (1.5%) had an influenzacoded health care encounter during the 2009 H1N1 pandemic. Compared with unexposed
pregnancy time, there was no association between exposure to the pandemic, with or without clinical influenza illness, and preterm birth (no pH1N1 diagnosis: aHR=1.0; 95% confidence
interval [CI]: 0.98, 1.1; pH1N1 diagnosis: aHR=1.0; 95% CI: 0.88, 1.2). Among women with
C C
pre-existing medical conditions, influenza was associated with increased preterm birth
(aHR=1.5; 95% CI: 1.1, 2.2) and spontaneous preterm birth (aHR=1.7; 95% CI: 1.1, 2.6), and these associations were strongest in the third trimester and when data were analyzed to allow for a transient acute effect of influenza.
A
Conclusion: In the general obstetrical population, there was no association between pH1N1 influenza illness and preterm birth, but women with pre-existing medical conditions known to increase the risk of influenza-associated morbidity were at elevated risk. Key words: Influenza; Preterm birth; Pandemic
4
Copyright © Wolters Kluwer Health, Inc. All rights reserved. Unauthorized reproduction of this article is prohibited.
Introduction Over the past century, descriptive reports from influenza pandemics have documented excess influenza-related morbidity and mortality in pregnant women.1–4 Epidemiologic studies during
EP TE D
seasonal influenza epidemics have also reported a higher risk of hospitalization for influenza illness among pregnant women;5–7 however, the potential risk that maternal influenza illness poses to the fetus remains unclear. As the influenza virus is rarely transmitted across the
placenta,8 any adverse fetal effects of maternal influenza are most likely mediated through other mechanisms such as fever8,9 or immunological responses10 that can influence placental function leading to adverse perinatal outcomes.11,12
With few exceptions,13–15 most descriptive studies from influenza pandemics have reported
higher rates of preterm birth among infected pregnant women;1,2,4 yet, a recent systematic review found inconsistent findings from comparative epidemiologic studies.16 Although the primary
goal of influenza vaccine recommendations is to prevent influenza disease in pregnant women,
C C
potential reductions in adverse birth outcomes would strengthen the case for investment in settings seeking to establish or expand influenza immunization programs.17 High-quality
evidence on influenza-associated risks of adverse perinatal outcomes is, therefore, essential for clarifying expectations for improved pregnancy outcomes following immunization.18 Given the
A
inconsistent evidence and public health importance of this topic, we designed this study to assess the relationship between 2009 pandemic H1N1 (pH1N1) influenza illness during pregnancy and preterm birth in a large obstetrical population, with a focus on addressing some of the evidence gaps identified in the recent systematic review.16
5
Copyright © Wolters Kluwer Health, Inc. All rights reserved. Unauthorized reproduction of this article is prohibited.
METHODS Study Design, Data Sources and Population We designed a population-based retrospective cohort study that traversed the 2009 H1N1 influenza pandemic (Figure). The cohort of births between 1 April 2009 and 31 March 2011 was
EP TE D
assembled using data from Better Outcomes Registry & Network (BORN) Ontario, a provincial registry containing all hospital births >500 grams or >20 weeks of gestation. Information collected in the registry included maternal demographics, pre-existing maternal medical
conditions, obstetrical complications, birth outcomes, and –– between November 2009 and
October 2010 –– whether women received monovalent 2009 pH1N1 vaccine during pregnancy. We linked the cohort with health administrative databases to ascertain influenza-coded health
care encounters among pregnant women during the 2009 H1N1 pandemic (our proxy for clinical pH1N1 influenza illness). The databases included the Canadian Institute for Health Information’s Discharge Abstract Database (hospitalizations) and National Ambulatory Care Reporting System (emergency department visits), and the Ontario Health Insurance Plan Claims Database
C C
(ambulatory physician visits). Further description of data sources and linkage is provided in eAppendix 1; http://links.lww.com/EDE/B268.
We restricted the study population to singleton live births and excluded women whose estimated last menstrual period (LMP) was more than 20 weeks before the start of the study period or less
A
than 43 weeks before the end, to avoid over-representing longer gestations at the beginning of the study and under-representing longer gestations at the end.19 We then applied an algorithm to exclude records with biologically implausible birth weight–gestational age combinations20 and excluded women with a health care encounter for pH1N1 influenza illness prior to pregnancy. Finally, we excluded women who were not continuously eligible to receive health care in 6
Copyright © Wolters Kluwer Health, Inc. All rights reserved. Unauthorized reproduction of this article is prohibited.
Ontario during the year before pregnancy and throughout pregnancy, and those whose birth registry record could not be linked to the administrative databases. Exposure and Outcome Measurement The H1N1 pandemic in Ontario involved an initial wave in spring 2009, followed by a second
EP TE D
wave that began in September 2009 and subsided in January 2010, with sporadic viral activity continuing until February 2010. Since cases continued to be detected between the two waves,21
we treated the pandemic as a continuous 269-day time period extending from 17 May 2009 to 6 February 2010. Pregnant women with a diagnostic code for influenza recorded during a
hospitalization or emergency department visit (ICD-10-CA codes J09 to J11) or a physician visit (ICD-9-CA code 487) during the pandemic were considered to have clinical pH1N1 influenza
illness (eTable 1; http://links.lww.com/EDE/B268), with the date of onset approximated by the date of first contact. These codes include both laboratory-confirmed influenza, as well as
suspected influenza based on symptoms (i.e., influenza-like illness). A recent validation study using Ontario laboratory data during the 2009 H1N1 pandemic found that sensitivity of
C C
influenza-coded health care visits ranged from 28% (physician billings database) to 76% (hospitalization database) ([Name suppressed], unpublished data, 2015; eTable 2;
http://links.lww.com/EDE/B268). We categorized the timing of the influenza diagnosis as first trimester (day 14 through 13 weeks +6 days), second trimester (14 weeks +0 days to 27 weeks
A
+6 days) or third trimester (28 weeks + 0 days to the end of follow-up). We defined preterm birth as a live birth prior to 37 completed weeks of gestation and spontaneous preterm birth as a preterm birth following spontaneous onset of labor or premature preterm rupture of membranes. Most pregnancy dating in Ontario during our study period was based on early ultrasound assessment.22 Follow-up commenced on the estimated LMP date and 7
Copyright © Wolters Kluwer Health, Inc. All rights reserved. Unauthorized reproduction of this article is prohibited.
continued until either a preterm birth occurred or the pregnancy reached 37 weeks. Women with an influenza diagnosis between the LMP date and day 14 (approximate date of conception) or after 37 weeks were considered unexposed. Statistical Analyses
EP TE D
We used a time-varying approach to exposure classification in which ongoing pregnancies changed exposure status depending upon the timing of the pregnancy relative to the surrounding pandemic. Each pregnancy entered the study either during a time period when there was no viral activity in the community capable of causing H1N1 illness (i.e., outside the defined pandemic
time period) or when the ongoing pregnancy was at risk for H1N1 illness (i.e., within the defined pandemic time period). Exposure status changed when the pregnancy passed from non-pandemic to pandemic time, or vice versa. An additional change in exposure status occurred during
pandemic follow-up time if there was an influenza-coded health care encounter in any of the
administrative databases. In our main analysis, pregnancy time at-risk following an influenza-
coded health care encounter was classified as exposed even if the ongoing pregnancy passed into
C C
post-pandemic time (eFigure 1; http://links.lww.com/EDE/B268). To assess whether influenza
illness might have an acute transient effect, we re-defined the exposure variable such that status reverted to unexposed two weeks after a diagnosis of clinical pH1N1 influenza illness, if the pregnancy was still ongoing and at-risk for preterm birth. To accommodate the time-varying
A
exposure, we structured the dataset such that each observation corresponded to an interval of time representing a different exposure status, while all other time-fixed covariates remained constant. We then applied an extension of the Cox proportional hazards regression model23 using gestational age in days as the time scale and robust sandwich variance estimation to account for statistical dependence across repeated observations (i.e., to account for changes in a woman’s 8
Copyright © Wolters Kluwer Health, Inc. All rights reserved. Unauthorized reproduction of this article is prohibited.
exposure status). We found the proportional hazards assumption for all time-fixed covariates to be fulfilled based on examination of Schoenfeld residual plots and tests of time-by-covariate interaction terms.24 We estimated hazard ratios (HR) with 95% confidence intervals (CI) adjusting for the following
EP TE D
potential confounding variables in our models: maternal age, smoking during pregnancy, nulliparity (no previous live births or stillbirths), pre-existing maternal medical conditions
(asthma, chronic hypertension, diabetes, heart disease), obstetrical complications (pregnancy-
induced hypertension, preeclampsia, eclampsia, gestational diabetes, placenta previa, placental
abruption), history of preterm birth, rural residence, and neighborhood-level median household income, obtained from the 2006 Canadian Census. We additionally adjusted for 2009
monovalent pH1N1 influenza vaccination during pregnancy as a time-fixed covariate as we did not have the vaccination date. Multiple imputation methods were used to address missing
covariate information (eAppendix 2; http://links.lww.com/EDE/B268). To assess whether the
exposure may have had an influence on spontaneous labor onset or membrane rupture at preterm
C C
gestation,25 we fit separate models for spontaneous preterm birth in which medically-indicated
preterm births were censored at delivery. We additionally estimated trimester-specific HRs for overall preterm birth, but were unable to assess the influence of trimester when infection occurred among the subset of spontaneous preterm births owing to insufficient numbers.
A
Since the impact of influenza during pregnancy has been shown to differ according to underlying susceptibility to serious influenza morbidity,6,7 we performed a pre-specified subgroup analysis among women with pre-existing medical conditions known to increase the risk for severe influenza disease and influenza-related complications (i.e., asthma, chronic hypertension, diabetes, or heart disease6). We also carried out several sensitivity analyses to assess the 9
Copyright © Wolters Kluwer Health, Inc. All rights reserved. Unauthorized reproduction of this article is prohibited.
influence of data limitations and analytical approaches on our primary results. First, we assessed whether missing exposure information (i.e., live birth records that were unlinked to administrative databases) could impact the validity of our analyses (eAppendix 3; http://links.lww.com/EDE/B268). We also repeated our primary analyses with more restrictive
EP TE D
definitions of clinical pH1N1 influenza by excluding influenza-coded hospitalizations from our exposure measure if the birth occurred during the same hospitalization, or if the influenza
diagnosis was not recorded within the first three diagnostic code fields, and we adjusted our
models for season to ensure that temporal patterns in the timing of conceptions in Ontario26 did not confound our estimates. We carried out a post-hoc bias analysis to estimate the impact of
exposure misclassification, given a range of non-differential classification errors informed by validation studies.27 Finally, we assessed whether an alternate exposure definition —
characterized on the basis of having, or not, a diagnosis of clinical pH1N1 influenza illness
during pregnancy (as a time-varying exposure, but without further separating out follow-up time within the defined pandemic period) — would alter the conclusions from our main findings. All
C C
analyses were carried out using SAS Version 9.4 (SAS Institute Inc., Cary, NC) within the
secure network environment at the Institute for Clinical Evaluative Sciences. Ethical approval was granted by the McGill University Institutional Review Board, the Children’s Hospital of Eastern Ontario Research Ethics Board (REB), and the Ottawa Health Science Network REB.
A
RESULTS
Following exclusions, the study cohort consisted of 192,082 singleton live births (eFigure 2; http://links.lww.com/EDE/B268). Overall, 2,925 women (1.5%) had an influenza-coded health care encounter (Table 1), with the majority of diagnoses made in ambulatory clinical settings (2,780/2,925; eTable 3; http://links.lww.com/EDE/B268). Among women who received an 10
Copyright © Wolters Kluwer Health, Inc. All rights reserved. Unauthorized reproduction of this article is prohibited.
influenza diagnosis during a hospitalization, 70% were admitted during the third trimester, while most women with an influenza diagnosis in an ambulatory setting presented during the second trimester of pregnancy (43%; eTable 3; http://links.lww.com/EDE/B268). There were 11,968 (6.2%) preterm births in the study cohort, 72% of which were spontaneous (Table 2).
EP TE D
Compared with non-pandemic-exposed pregnancy time, there was no association between pregnancy exposure to the pandemic (with or without clinical pH1N1 influenza illness) and
preterm birth in the overall study population (no pH1N1 diagnosis: adjusted HR [aHR]=1.0; 95% CI: 0.98, 1.1; pH1N1 diagnosis: aHR=1.0; 95% CI: 0.88, 1.2; Table 3). The estimate for
influenza diagnosed during the third trimester was higher in magnitude than for first or second trimester influenza (compared with unexposed pregnancy time), but the confidence interval
included the null value (Table 3). In the sensitivity analysis aimed at evaluating a possible acute effect of influenza on preterm birth, the aHR for clinical pH1N1 influenza illness increased
slightly from 1.0 to 1.3 (95% CI: 0.73, 2.3; eTable 4; http://links.lww.com/EDE/B268). We saw no difference in the risk of spontaneous preterm birth by category of exposure status (Table 3).
C C
Although we excluded 1,742 women from our analyses due to missing exposure information arising from an inability to link their birth registry record with administrative databases, our
sensitivity analyses suggested that any bias in our primary results introduced by this exclusion was minimal and did not affect our interpretation (eTables 5 and 6;
A
http://links.lww.com/EDE/B268).
Among the 7% of women with a pre-existing medical condition, clinical pH1N1 influenza illness was associated with an increase in overall preterm birth (aHR= 1.5; 95% CI: 1.1, 2.2; Table 3) and spontaneous preterm birth (aHR= 1.7; 95% CI: 1.1, 2.6; Table 3), compared with unexposed pregnancy time. There was no association between pH1N1 influenza illness in the first or second 11
Copyright © Wolters Kluwer Health, Inc. All rights reserved. Unauthorized reproduction of this article is prohibited.
trimester and overall preterm birth in this subgroup of women, but influenza diagnosed during the third trimester was associated with a 3-fold increase in preterm birth (aHR= 3.1; 95% CI: 1.8, 5.5; Table 3). In a sensitivity analysis, the magnitude of the aHR increased from 1.54 in the original analysis to 3.6 (95% CI: 1.3, 10) when influenza was re-defined as an acute transient
EP TE D
exposure (eTable 4; http://links.lww.com/EDE/B268), similar in magnitude to that previously observed for the third trimester (aHR= 3.1). We observed a similar pattern when considering
spontaneous preterm birth as the outcome. The association between third trimester influenza and preterm birth persisted, but was attenuated, after we excluded influenza diagnoses documented only in the hospitalization during which the birth occurred or in instances where the influenza diagnosis was not recorded within the first three diagnostic code fields (eTable 7;
http://links.lww.com/EDE/B268). Additional adjustment for seasonal effects had no impact on the magnitude of any of the point estimates in our analyses. The sensitivity analysis using an
alternate exposure definition based on having, or not, a diagnosis of clinical pH1N1 influenza illness during pregnancy yielded point estimates that were almost identical to our primary
C C
findings (eTable 8; http://links.lww.com/EDE/B268). DISCUSSION
In our large, population-based study designed to accommodate the time-dependent nature of influenza illness and incidence of preterm birth, we saw no association between pregnancy
A
exposure to the 2009 H1N1 pandemic, with or without documented clinical pH1N1 influenza illness, and preterm birth in the general obstetrical population, irrespective of the gestational age of any influenza-coded health care encounter. In contrast, women at high risk for influenzarelated complications due to pre-existing medical conditions such as asthma had an increased risk of both preterm birth and spontaneous preterm birth, largely due to clinical pH1N1 influenza 12
Copyright © Wolters Kluwer Health, Inc. All rights reserved. Unauthorized reproduction of this article is prohibited.
illness during the third trimester. We found these estimated associations to be stronger when we analyzed the data to allow for a possible acute transient effect of influenza illness. Recent interest in the potential for immunization during pregnancy to protect against preterm birth and other adverse perinatal outcomes17,28 has followed reports from observational studies
EP TE D
suggesting influenza immunization is associated with risk reductions ranging from 14% to 37% for preterm birth.29–31 Although influenza immunization of pregnant women is efficacious in
preventing influenza disease32–34 when considering the multifactorial etiology of preterm birth,35– 37
low incidence of influenza during pregnancy,32,38–41 and lack of a consistent and strong
association between maternal influenza disease and preterm birth,42 it is unlikely that
immunization would reduce preterm birth in the overall obstetrical population to the magnitude observed by some studies.18,43 In fact, our findings in the same population and influenza season as an earlier study that reported a lower risk of very preterm birth (
EP TE D
bbreviated Title: Pandemic influenza and preterm birth Deshayne B Fell,a,b,c,d Robert W Platt,d Olga Basso,d,e Kumanan Wilson,c,f ,gJay S Kaufman,d David L Buckeridge,d Jeffrey C Kwongc,h,I,j,k a.
Children’s Hospital of Eastern Ontario Research Institute, Ottawa, Canada
b.
School of Epidemiology, Public Health and Preventive Medicine, University of Ottawa, Ottawa, Canada
c.
Institute for Clinical Evaluative Sciences, Ottawa and Toronto, Canada
d.
Department of Epidemiology, Biostatistics and Occupational Health, McGill University, Montreal, Canada
Department of Obstetrics and Gynecology, McGill University, Montreal, Canada
f.
Department of Medicine, University of Ottawa, Ottawa, Canada
g.
Clinical Epidemiology Program, Ottawa Hospital Research Institute, Ottawa, Canada
h.
Department of Family and Community Medicine, University of Toronto, Toronto, Canada
i.
Dalla Lana School of Public Health, Toronto, Canada
j.
Public Health Ontario, Toronto, Canada
A
C C
e.
k.
University Health Network, Toronto, Canada
Correspondence: Deshayne Fell, Children’s Hospital of Eastern Ontario Research Institute, 401 Smyth Road, Centre for Practice Changing Research, Room L-1154, Ottawa, Ontario, K1H 8L1, Canada. E-mail: [email protected] 1
Copyright © Wolters Kluwer Health, Inc. All rights reserved. Unauthorized reproduction of this article is prohibited.
Conflicts of Interest: The authors have no conflicts to declare. Source of Funding: This study was partially funded through an Operating Grant from the Canadian Institutes for Health Research (#MOP-119570). This study was supported by the Institute for Clinical Evaluative Sciences (ICES), which is funded by an annual grant from the
EP TE D
Ontario Ministry of Health and Long-Term Care. Data Access: The dataset from this study is held securely in coded form at the Institute for
Clinical Evaluative Sciences (ICES). While data sharing agreements prohibit ICES from making the data set publicly available, access may be granted to those who meet pre-specified criteria for confidential access, available at www.ices.on.ca/DAS.
Acknowledgments: DBF was supported by a Canadian Institutes of Health Research (CIHR) Doctoral Award while this study was being conducted. RWP is supported by a Chercheur-
national (National Scholar) award from the Fonds de Recherche du Quebec – Santé (FRQ-S) as
well as core support to the Research Institute of the McGill University Health Centre from FRQS. JCK is supported by a New Investigator Award from CIHR and a Clinician Scientist Award
C C
from the University of Toronto Department of Family and Community Medicine. We are
grateful to Better Outcomes Registry & Network (BORN) Ontario, Ottawa, Canada and the Institute for Clinical Evaluative Sciences (ICES) for providing data access. Disclaimer: The opinions, results and conclusions reported in this paper are those of the authors
A
and are independent from the funding sources. Parts of this material are based on data and information compiled and provided by the Canadian Institute of Health Information (CIHI). However, the analyses, conclusions, opinions, and statements expressed herein are those of the authors, and not necessarily those of CIHI. No endorsement by ICES, the Ontario Ministry of Health and Long-Term Care, or CIHI is intended or should be inferred. 2
Copyright © Wolters Kluwer Health, Inc. All rights reserved. Unauthorized reproduction of this article is prohibited.
Abstract word count: 242 Text word count (excluding references): 3242 Total number of pages (including references, but excluding supplementary files): 32 Number of text pages (including references, but excluding supplementary files): 20
EP TE D
Number of table pages (excluding supplementary files): 11
A
C C
Number of figure pages (excluding supplementary files): 1
3
Copyright © Wolters Kluwer Health, Inc. All rights reserved. Unauthorized reproduction of this article is prohibited.
ABSTRACT Background: Previous studies of maternal influenza illness and preterm birth have yielded inconsistent results. Our objective was to assess the association between 2009 pandemic H1N1 (pH1N1) influenza during pregnancy and preterm birth in a large obstetrical population.
EP TE D
Methods: We linked a province-wide birth registry with health administrative databases to identify influenza-coded hospitalizations, emergency department visits, or physician visits among pregnant women during the 2009 H1N1 pandemic (our proxy for clinical pH1N1
influenza illness). Using Cox regression, we estimated adjusted hazard ratios (aHR) for preterm birth and spontaneous preterm birth treating influenza as a time-varying exposure.
Results: Among 192,082 women with a singleton live birth, 2,925 (1.5%) had an influenzacoded health care encounter during the 2009 H1N1 pandemic. Compared with unexposed
pregnancy time, there was no association between exposure to the pandemic, with or without clinical influenza illness, and preterm birth (no pH1N1 diagnosis: aHR=1.0; 95% confidence
interval [CI]: 0.98, 1.1; pH1N1 diagnosis: aHR=1.0; 95% CI: 0.88, 1.2). Among women with
C C
pre-existing medical conditions, influenza was associated with increased preterm birth
(aHR=1.5; 95% CI: 1.1, 2.2) and spontaneous preterm birth (aHR=1.7; 95% CI: 1.1, 2.6), and these associations were strongest in the third trimester and when data were analyzed to allow for a transient acute effect of influenza.
A
Conclusion: In the general obstetrical population, there was no association between pH1N1 influenza illness and preterm birth, but women with pre-existing medical conditions known to increase the risk of influenza-associated morbidity were at elevated risk. Key words: Influenza; Preterm birth; Pandemic
4
Copyright © Wolters Kluwer Health, Inc. All rights reserved. Unauthorized reproduction of this article is prohibited.
Introduction Over the past century, descriptive reports from influenza pandemics have documented excess influenza-related morbidity and mortality in pregnant women.1–4 Epidemiologic studies during
EP TE D
seasonal influenza epidemics have also reported a higher risk of hospitalization for influenza illness among pregnant women;5–7 however, the potential risk that maternal influenza illness poses to the fetus remains unclear. As the influenza virus is rarely transmitted across the
placenta,8 any adverse fetal effects of maternal influenza are most likely mediated through other mechanisms such as fever8,9 or immunological responses10 that can influence placental function leading to adverse perinatal outcomes.11,12
With few exceptions,13–15 most descriptive studies from influenza pandemics have reported
higher rates of preterm birth among infected pregnant women;1,2,4 yet, a recent systematic review found inconsistent findings from comparative epidemiologic studies.16 Although the primary
goal of influenza vaccine recommendations is to prevent influenza disease in pregnant women,
C C
potential reductions in adverse birth outcomes would strengthen the case for investment in settings seeking to establish or expand influenza immunization programs.17 High-quality
evidence on influenza-associated risks of adverse perinatal outcomes is, therefore, essential for clarifying expectations for improved pregnancy outcomes following immunization.18 Given the
A
inconsistent evidence and public health importance of this topic, we designed this study to assess the relationship between 2009 pandemic H1N1 (pH1N1) influenza illness during pregnancy and preterm birth in a large obstetrical population, with a focus on addressing some of the evidence gaps identified in the recent systematic review.16
5
Copyright © Wolters Kluwer Health, Inc. All rights reserved. Unauthorized reproduction of this article is prohibited.
METHODS Study Design, Data Sources and Population We designed a population-based retrospective cohort study that traversed the 2009 H1N1 influenza pandemic (Figure). The cohort of births between 1 April 2009 and 31 March 2011 was
EP TE D
assembled using data from Better Outcomes Registry & Network (BORN) Ontario, a provincial registry containing all hospital births >500 grams or >20 weeks of gestation. Information collected in the registry included maternal demographics, pre-existing maternal medical
conditions, obstetrical complications, birth outcomes, and –– between November 2009 and
October 2010 –– whether women received monovalent 2009 pH1N1 vaccine during pregnancy. We linked the cohort with health administrative databases to ascertain influenza-coded health
care encounters among pregnant women during the 2009 H1N1 pandemic (our proxy for clinical pH1N1 influenza illness). The databases included the Canadian Institute for Health Information’s Discharge Abstract Database (hospitalizations) and National Ambulatory Care Reporting System (emergency department visits), and the Ontario Health Insurance Plan Claims Database
C C
(ambulatory physician visits). Further description of data sources and linkage is provided in eAppendix 1; http://links.lww.com/EDE/B268.
We restricted the study population to singleton live births and excluded women whose estimated last menstrual period (LMP) was more than 20 weeks before the start of the study period or less
A
than 43 weeks before the end, to avoid over-representing longer gestations at the beginning of the study and under-representing longer gestations at the end.19 We then applied an algorithm to exclude records with biologically implausible birth weight–gestational age combinations20 and excluded women with a health care encounter for pH1N1 influenza illness prior to pregnancy. Finally, we excluded women who were not continuously eligible to receive health care in 6
Copyright © Wolters Kluwer Health, Inc. All rights reserved. Unauthorized reproduction of this article is prohibited.
Ontario during the year before pregnancy and throughout pregnancy, and those whose birth registry record could not be linked to the administrative databases. Exposure and Outcome Measurement The H1N1 pandemic in Ontario involved an initial wave in spring 2009, followed by a second
EP TE D
wave that began in September 2009 and subsided in January 2010, with sporadic viral activity continuing until February 2010. Since cases continued to be detected between the two waves,21
we treated the pandemic as a continuous 269-day time period extending from 17 May 2009 to 6 February 2010. Pregnant women with a diagnostic code for influenza recorded during a
hospitalization or emergency department visit (ICD-10-CA codes J09 to J11) or a physician visit (ICD-9-CA code 487) during the pandemic were considered to have clinical pH1N1 influenza
illness (eTable 1; http://links.lww.com/EDE/B268), with the date of onset approximated by the date of first contact. These codes include both laboratory-confirmed influenza, as well as
suspected influenza based on symptoms (i.e., influenza-like illness). A recent validation study using Ontario laboratory data during the 2009 H1N1 pandemic found that sensitivity of
C C
influenza-coded health care visits ranged from 28% (physician billings database) to 76% (hospitalization database) ([Name suppressed], unpublished data, 2015; eTable 2;
http://links.lww.com/EDE/B268). We categorized the timing of the influenza diagnosis as first trimester (day 14 through 13 weeks +6 days), second trimester (14 weeks +0 days to 27 weeks
A
+6 days) or third trimester (28 weeks + 0 days to the end of follow-up). We defined preterm birth as a live birth prior to 37 completed weeks of gestation and spontaneous preterm birth as a preterm birth following spontaneous onset of labor or premature preterm rupture of membranes. Most pregnancy dating in Ontario during our study period was based on early ultrasound assessment.22 Follow-up commenced on the estimated LMP date and 7
Copyright © Wolters Kluwer Health, Inc. All rights reserved. Unauthorized reproduction of this article is prohibited.
continued until either a preterm birth occurred or the pregnancy reached 37 weeks. Women with an influenza diagnosis between the LMP date and day 14 (approximate date of conception) or after 37 weeks were considered unexposed. Statistical Analyses
EP TE D
We used a time-varying approach to exposure classification in which ongoing pregnancies changed exposure status depending upon the timing of the pregnancy relative to the surrounding pandemic. Each pregnancy entered the study either during a time period when there was no viral activity in the community capable of causing H1N1 illness (i.e., outside the defined pandemic
time period) or when the ongoing pregnancy was at risk for H1N1 illness (i.e., within the defined pandemic time period). Exposure status changed when the pregnancy passed from non-pandemic to pandemic time, or vice versa. An additional change in exposure status occurred during
pandemic follow-up time if there was an influenza-coded health care encounter in any of the
administrative databases. In our main analysis, pregnancy time at-risk following an influenza-
coded health care encounter was classified as exposed even if the ongoing pregnancy passed into
C C
post-pandemic time (eFigure 1; http://links.lww.com/EDE/B268). To assess whether influenza
illness might have an acute transient effect, we re-defined the exposure variable such that status reverted to unexposed two weeks after a diagnosis of clinical pH1N1 influenza illness, if the pregnancy was still ongoing and at-risk for preterm birth. To accommodate the time-varying
A
exposure, we structured the dataset such that each observation corresponded to an interval of time representing a different exposure status, while all other time-fixed covariates remained constant. We then applied an extension of the Cox proportional hazards regression model23 using gestational age in days as the time scale and robust sandwich variance estimation to account for statistical dependence across repeated observations (i.e., to account for changes in a woman’s 8
Copyright © Wolters Kluwer Health, Inc. All rights reserved. Unauthorized reproduction of this article is prohibited.
exposure status). We found the proportional hazards assumption for all time-fixed covariates to be fulfilled based on examination of Schoenfeld residual plots and tests of time-by-covariate interaction terms.24 We estimated hazard ratios (HR) with 95% confidence intervals (CI) adjusting for the following
EP TE D
potential confounding variables in our models: maternal age, smoking during pregnancy, nulliparity (no previous live births or stillbirths), pre-existing maternal medical conditions
(asthma, chronic hypertension, diabetes, heart disease), obstetrical complications (pregnancy-
induced hypertension, preeclampsia, eclampsia, gestational diabetes, placenta previa, placental
abruption), history of preterm birth, rural residence, and neighborhood-level median household income, obtained from the 2006 Canadian Census. We additionally adjusted for 2009
monovalent pH1N1 influenza vaccination during pregnancy as a time-fixed covariate as we did not have the vaccination date. Multiple imputation methods were used to address missing
covariate information (eAppendix 2; http://links.lww.com/EDE/B268). To assess whether the
exposure may have had an influence on spontaneous labor onset or membrane rupture at preterm
C C
gestation,25 we fit separate models for spontaneous preterm birth in which medically-indicated
preterm births were censored at delivery. We additionally estimated trimester-specific HRs for overall preterm birth, but were unable to assess the influence of trimester when infection occurred among the subset of spontaneous preterm births owing to insufficient numbers.
A
Since the impact of influenza during pregnancy has been shown to differ according to underlying susceptibility to serious influenza morbidity,6,7 we performed a pre-specified subgroup analysis among women with pre-existing medical conditions known to increase the risk for severe influenza disease and influenza-related complications (i.e., asthma, chronic hypertension, diabetes, or heart disease6). We also carried out several sensitivity analyses to assess the 9
Copyright © Wolters Kluwer Health, Inc. All rights reserved. Unauthorized reproduction of this article is prohibited.
influence of data limitations and analytical approaches on our primary results. First, we assessed whether missing exposure information (i.e., live birth records that were unlinked to administrative databases) could impact the validity of our analyses (eAppendix 3; http://links.lww.com/EDE/B268). We also repeated our primary analyses with more restrictive
EP TE D
definitions of clinical pH1N1 influenza by excluding influenza-coded hospitalizations from our exposure measure if the birth occurred during the same hospitalization, or if the influenza
diagnosis was not recorded within the first three diagnostic code fields, and we adjusted our
models for season to ensure that temporal patterns in the timing of conceptions in Ontario26 did not confound our estimates. We carried out a post-hoc bias analysis to estimate the impact of
exposure misclassification, given a range of non-differential classification errors informed by validation studies.27 Finally, we assessed whether an alternate exposure definition —
characterized on the basis of having, or not, a diagnosis of clinical pH1N1 influenza illness
during pregnancy (as a time-varying exposure, but without further separating out follow-up time within the defined pandemic period) — would alter the conclusions from our main findings. All
C C
analyses were carried out using SAS Version 9.4 (SAS Institute Inc., Cary, NC) within the
secure network environment at the Institute for Clinical Evaluative Sciences. Ethical approval was granted by the McGill University Institutional Review Board, the Children’s Hospital of Eastern Ontario Research Ethics Board (REB), and the Ottawa Health Science Network REB.
A
RESULTS
Following exclusions, the study cohort consisted of 192,082 singleton live births (eFigure 2; http://links.lww.com/EDE/B268). Overall, 2,925 women (1.5%) had an influenza-coded health care encounter (Table 1), with the majority of diagnoses made in ambulatory clinical settings (2,780/2,925; eTable 3; http://links.lww.com/EDE/B268). Among women who received an 10
Copyright © Wolters Kluwer Health, Inc. All rights reserved. Unauthorized reproduction of this article is prohibited.
influenza diagnosis during a hospitalization, 70% were admitted during the third trimester, while most women with an influenza diagnosis in an ambulatory setting presented during the second trimester of pregnancy (43%; eTable 3; http://links.lww.com/EDE/B268). There were 11,968 (6.2%) preterm births in the study cohort, 72% of which were spontaneous (Table 2).
EP TE D
Compared with non-pandemic-exposed pregnancy time, there was no association between pregnancy exposure to the pandemic (with or without clinical pH1N1 influenza illness) and
preterm birth in the overall study population (no pH1N1 diagnosis: adjusted HR [aHR]=1.0; 95% CI: 0.98, 1.1; pH1N1 diagnosis: aHR=1.0; 95% CI: 0.88, 1.2; Table 3). The estimate for
influenza diagnosed during the third trimester was higher in magnitude than for first or second trimester influenza (compared with unexposed pregnancy time), but the confidence interval
included the null value (Table 3). In the sensitivity analysis aimed at evaluating a possible acute effect of influenza on preterm birth, the aHR for clinical pH1N1 influenza illness increased
slightly from 1.0 to 1.3 (95% CI: 0.73, 2.3; eTable 4; http://links.lww.com/EDE/B268). We saw no difference in the risk of spontaneous preterm birth by category of exposure status (Table 3).
C C
Although we excluded 1,742 women from our analyses due to missing exposure information arising from an inability to link their birth registry record with administrative databases, our
sensitivity analyses suggested that any bias in our primary results introduced by this exclusion was minimal and did not affect our interpretation (eTables 5 and 6;
A
http://links.lww.com/EDE/B268).
Among the 7% of women with a pre-existing medical condition, clinical pH1N1 influenza illness was associated with an increase in overall preterm birth (aHR= 1.5; 95% CI: 1.1, 2.2; Table 3) and spontaneous preterm birth (aHR= 1.7; 95% CI: 1.1, 2.6; Table 3), compared with unexposed pregnancy time. There was no association between pH1N1 influenza illness in the first or second 11
Copyright © Wolters Kluwer Health, Inc. All rights reserved. Unauthorized reproduction of this article is prohibited.
trimester and overall preterm birth in this subgroup of women, but influenza diagnosed during the third trimester was associated with a 3-fold increase in preterm birth (aHR= 3.1; 95% CI: 1.8, 5.5; Table 3). In a sensitivity analysis, the magnitude of the aHR increased from 1.54 in the original analysis to 3.6 (95% CI: 1.3, 10) when influenza was re-defined as an acute transient
EP TE D
exposure (eTable 4; http://links.lww.com/EDE/B268), similar in magnitude to that previously observed for the third trimester (aHR= 3.1). We observed a similar pattern when considering
spontaneous preterm birth as the outcome. The association between third trimester influenza and preterm birth persisted, but was attenuated, after we excluded influenza diagnoses documented only in the hospitalization during which the birth occurred or in instances where the influenza diagnosis was not recorded within the first three diagnostic code fields (eTable 7;
http://links.lww.com/EDE/B268). Additional adjustment for seasonal effects had no impact on the magnitude of any of the point estimates in our analyses. The sensitivity analysis using an
alternate exposure definition based on having, or not, a diagnosis of clinical pH1N1 influenza illness during pregnancy yielded point estimates that were almost identical to our primary
C C
findings (eTable 8; http://links.lww.com/EDE/B268). DISCUSSION
In our large, population-based study designed to accommodate the time-dependent nature of influenza illness and incidence of preterm birth, we saw no association between pregnancy
A
exposure to the 2009 H1N1 pandemic, with or without documented clinical pH1N1 influenza illness, and preterm birth in the general obstetrical population, irrespective of the gestational age of any influenza-coded health care encounter. In contrast, women at high risk for influenzarelated complications due to pre-existing medical conditions such as asthma had an increased risk of both preterm birth and spontaneous preterm birth, largely due to clinical pH1N1 influenza 12
Copyright © Wolters Kluwer Health, Inc. All rights reserved. Unauthorized reproduction of this article is prohibited.
illness during the third trimester. We found these estimated associations to be stronger when we analyzed the data to allow for a possible acute transient effect of influenza illness. Recent interest in the potential for immunization during pregnancy to protect against preterm birth and other adverse perinatal outcomes17,28 has followed reports from observational studies
EP TE D
suggesting influenza immunization is associated with risk reductions ranging from 14% to 37% for preterm birth.29–31 Although influenza immunization of pregnant women is efficacious in
preventing influenza disease32–34 when considering the multifactorial etiology of preterm birth,35– 37
low incidence of influenza during pregnancy,32,38–41 and lack of a consistent and strong
association between maternal influenza disease and preterm birth,42 it is unlikely that
immunization would reduce preterm birth in the overall obstetrical population to the magnitude observed by some studies.18,43 In fact, our findings in the same population and influenza season as an earlier study that reported a lower risk of very preterm birth (
