We conducted a population-based retrospective cohort study using the Nova Scotia Atlee Perinatal Database (NSAPD). The cohort comprised all Nova Scotia residents who delivered an infant weighing 500 g or more or who delivered at 20 weeks' gestation or later between Jan. 1, 1990, and Dec. 31, 2002. Information on dates of each trimester of pregnancy and on pre-existing risk factors for severe influenza (e.g., asthma and cardiovascular disease) was obtained from the NSAPD. The NSAPD contains extensive information related to maternal medical conditions, maternal risk factors and demographic factors, the prenatal period, labour and delivery factors and neonatal outcomes for all hospital deliveries among Nova Scotia residents. Data are abstracted from medical records by trained health records personnel using standardized data collection forms. An on-going data quality assurance program, which includes periodic data re-abstraction studies, indicates that the data are reliable.11
Our primary outcome measures — hospital admissions and physician office visits because of respiratory illness — were determined from the Canadian Institute for Health Information's Discharge Abstract Database (hospital admissions) and the Nova Scotia Medical Services Insurance Database (physician office visits). These databases capture all uses of the formal public health care system in Nova Scotia. There is no parallel private system in the province. In addition, a master registration file tracks individuals moving into, and out of, the province by their enrolment in the Nova Scotia Health Services Program. We estimated socioeconomic status from data in the Family Benefits Database, which records receipt of financial support through assistance programs of the provincial and federal governments on a yearly basis. The administrative databases were successfully linked to the NSAPD for 96% of the women in the cohort. The individual categories of respiratory conditions and the corresponding International Classification of Diseases (ICD) codes that we used to identify hospital admissions and physician visits of interest are shown in Appendix 1 (available online at www.cmaj.ca/cgi/content/full/176/4/463/DC1
). The ICD codes selected were adapted from those used by Neuzil and colleagues8
except for the code for asthma and the codes for respiratory conditions related to occupational exposures. To be conservative, we chose not to include asthma as a respiratory condition because not all exacerbations of asthma are due to influenza, and any cases of asthma that are due to influenza most likely would have an accompanying influenza-related code on the record.
We obtained information on the defined periods of influenza season from the Nova Scotia Department of Health for the study years (1990–2002). For the purposes of our study, the beginning of the influenza season was defined as the time when 2 or more isolates of influenza were identified in Nova Scotia in sequential weeks or 3 or more isolates were identified in a single week. The end of the influenza season was defined as the time when no influenza isolates were identified in the province in 2 or more sequential weeks unless the lapse was followed by a large number of isolates. The peri-influenza season was defined as the period between Nov. 1 and the start of the influenza season and the 2-week period following influenza season. The non-influenza season was defined as the end of the peri-influenza season until Oct. 31. Data from the Medical Services Insurance Database for October 1999 onward include a code for influenza vaccination; therefore, after this date, we were able to examine the rates of influenza vaccination administered in physician's offices.
Pregnant women were stratified according to the presence or absence of any of the following pre-existing conditions (all information obtained from the NSAPD): pre-existing diabetes, pulmonary disease (including asthma), heart disease, renal disease, and anemia (hemoglobin < 10 g) during the pregnancy. Women with 1 or more of these conditions were defined as having comorbidities. All other women were defined as having no comorbidities.
Trimesters were determined by the date of the birth and gestational age at the time of birth. Gestational age was assigned using last menstrual period, if known, or a clinical estimate of gestational age based on examination of the newborn. If both of these estimates of gestational age were missing, we imputed gestational age using the approach used by Neuzil and colleagues8
(gestational age calculated as the median gestational age for infants born in the same birth year and whose birth weight was within the same 500-g category). Gestational age was imputed for less than 2% of the infants.
We based rates of hospital admissions on the date of admission minus 4 days to attribute the event to the influenza season during which most of the exposure occurred. To be conservative, admissions that resulted in a delivery and during which a respiratory illness was diagnosed were not counted in the rate, and only the first non-delivery-related hospital admission because of a respiratory illness was considered.
Person-time analyses were conducted for the entire pregnancy and by trimester. The number of events (hospital admissions or physician office visits) that occurred during the influenza, peri-influenza and non-influenza seasons was divided by the woman-months in each season to determine the event rate for each of the 3 defined seasons. For the trimester-specific analyses, the number of events and woman-months were further divided into the amount of time a woman spent in each influenza season during each trimester. Poisson regression models were used to adjust for confounding factors, and rate ratios (and 95% confidence intervals [CIs]) were calculated. Potential confounders included maternal age, maternal smoking status, socioeconomic status (based on receipt of family benefits), and number and ages of other children in the home. Since the outcomes for women with more than one pregnancy during the study period are not independent, generalized estimating equations were used to generate unbiased standard errors and 95% CIs.12
Initially, we calculated unadjusted rate ratios. The potential confounding factors were entered into the model, with season as the independent variable of interest. Each potential confounding factor was removed, one at a time, and the model rerun. If removing the factor did not change the coefficient (for season) by 5% or more, the factor was removed and the process repeated. A change of 5% or more was chosen so that factors would be included in the final model if they produced a modest change in the coefficient for season. With this approach, the final model included factors that confounded the relation between season and rates of hospital admission.
Rates of hospital admissions (and physician office visits) during the influenza, peri-influenza and non-influenza seasons for each trimester of pregnancy were compared with rates of admissions (and physician visits) during these 3 seasons in the year before pregnancy for the same cohort of women. In addition, rates of hospital admissions (and physician visits) for each trimester of pregnancy during the influenza and peri-influenza seasons were compared with rates for each trimester during the non-influenza season.
Influenza-attributable risks in each trimester of pregnancy were estimated by subtracting the rate of hospital admissions during the peri-influenza and non-influenza seasons from the rate during the influenza season, as described by Neuzil and colleagues.8
We used the rates of hospital admissions during the peri-influenza season to determine the baseline risk to help quantify the risks specific to influenza-related complications rather than to complications that may be related to other viruses (e.g., respiratory syncytial virus).
This study received approval from the Research Ethics Board of the IWK Health Centre.