This study demonstrates an association between immunization with the inactivated influenza vaccine during pregnancy and reduced likelihood of prematurity during local, regional, and widespread influenza activity periods. For births during the 8 wk of widespread influenza activity, the odds of prematurity were approximately 70% lower among the newborns of the vaccinated mothers compared to mothers who did not receive the influenza vaccine. During the period of widespread influenza activity there was also an association between maternal receipt of influenza vaccine and reduced likelihood of SGA. The magnitude of association between influenza vaccine and prematurity (as measured by the values of ORs) increased with the increase in the intensity of influenza activity and was higher for the 2004–2005 season than for the 2005–2006 season. Based on laboratory and epidemiologic criteria, the 2004–2005 influenza season was more intense than the 2005–2006 season in the US 
. Although the SGA-related ORs were not statistically significant for influenza activity periods except for the period of widespread activity, the overall “gradient” of effect in the point estimates of the ORs was qualitatively similar to that of prematurity. The increase in the impact of maternal influenza vaccines on birth outcomes by influenza activity, both in terms of ORs and population prevented fractions, supports the validity of our findings.
The possibility of confounding due to differences between vaccinated and unvaccinated individuals included in observational studies of influenza immunization is well recognized 
. The most significant type of confounding in influenza studies is due to a higher likelihood of individuals with high functional capacity (i.e., healthier individuals) to get vaccinated—a phenomenon often known as the “healthy user effect.” However, most observational studies where significant confounding has been documented were conducted in the elderly and included a relatively nonspecific end point of all-cause mortality. It is reasonable to assume that, compared to older individuals, women of reproductive age may be less likely to have significant functional limitation even in the presence of co-morbidities. Therefore, influenza vaccine studies in this age group may be less likely to suffer from bias due to the healthy user effect. Moreover, we found no statistically significant difference between the vaccinated women and the unvaccinated women in terms of gestational age at which they sought antenatal care. On the other hand, the possibility of other confounders cannot be discounted in studies involving pregnant women. In order to address confounding, we used the pre-influenza period (i.e., the season where vaccine was available but there was minimal circulation of influenza virus) as the “control” period. The use of the pre-influenza period for selecting confounders from a broad set of covariates is an approach suggested by Nelson et al. 
and Jackson et al. 
that takes advantage of the seasonality of influenza circulation. The associations observed in our study were robust to adjustment for confounders identified using this approach (and the more traditional approach)—supporting the validity of our findings.
Influenza virus, particularly seasonal influenza virus, rarely crosses the placenta 
. However, the effect of infection on prematurity is thought be at least partially mediated through inflammatory pathways 
. Increase in pro-inflammatory cytokines (e.g., IL-1, IL-6 and TNF-α) and reduction in anti-inflammatory cytokines (e.g., IL-10) have been linked to preterm labor 
. IL-1 stimulates the amnion and the decidua to produce prostaglandins and can stimulate myometrial contractions 
. Prostaglandins are known to play an important role in the initiation and progression of labor 
. Moreover, in animal models, administration of IL-1 results in preterm birth 
. Similarly, TNF-α induces the amnion, the decidua, and the myometrium to produce prostaglandins, and administration of TNF-α to pregnant animals can induce preterm labor 
. Recent studies have shown that influenza virus infection induces gene expression of pro-inflammatory cytokines including IL-1β, IL-6, TNF-α, interferon (IFN)-β, IFN-α, and granulocyte macrophage colony-stimulating factor (GM-CSF) 
In addition to biological plausibility, there is epidemiological evidence of an association between maternal infection and preterm birth 
. The association is strongest for intrauterine viral infections and systemic and intrauterine bacterial infections 
. Viral infection outside the reproductive tract, including influenza infection, may also play a role in the etiology of prematurity. For example, in an analysis of 1957–1958 data, newborns of women who had serological evidence of “Asian” (pandemic) influenza during pregnancy were 50% more likely to be premature compared to newborns of uninfected women 
. Moreover, a recent literature review found that SARS infection in the second or third trimester of pregnancy may be associated with spontaneous preterm delivery and early cesarean sections for deteriorating medical condition, although only 16 such cases were identified in the literature 
. Moreover, in studies in China and Hungary, birth defects were associated with history of influenza 
However, a few observational studies have failed to demonstrate an association between influenza infection and birth outcomes 
. The lack of observed effect in some studies could be due to a true lack of association, small or difficult to measure effect size, challenges related to the study population (e.g., administrative datasets), or non-differential misclassification due to challenges in retrospectively identifying influenza infection. Although less than ideal, modeling receipt of influenza vaccine as the exposure/independent variable reduces the likelihood and the intensity of non-differential misclassification bias.
Preterm births represent a significant burden to health care and society 
. Like several developed countries, there has been an increase in the rate of preterm births in the US, which rose from 9.5% in 1981 to 12.8% in 2006 
. Although the etiology of prematurity is complex 
and not completely understood, our results suggest that at least a fraction of preterm births may be preventable through maternal influenza vaccination.
The association between maternal influenza vaccination and SGA was only statistically significant (and the highest in magnitude) for the period of widespread influenza activity. Possible reasons for the effect being limited to the period of highest influenza activity include the following: (a) in a developed country setting, the effect of maternal influenza infection on fetal growth is milder than the effect on prematurity; (b) SGA represents fetal compromise resulting from infection that is insufficient to trigger early parturition, but may result in the delayed observation of growth restriction (i.e., the observation in the widespread activity period may be the cumulative effect of previous periods). Moreover, in the vaccinated group, the birth weight distribution in the pre-influenza period was different from the distribution in the period of widespread activity (see Text S1
). However, the difference in mean birth weights (in the vaccinated group) between these two periods was not statistically significant (p
0.74). Since the ostensible increase in birth weight in the widespread activity period compared to the pre-influenza period in the vaccinated group cannot easily be explained by vaccine action, this difference—although non-significant—may suggest confounding vis-à-vis the birth weight outcome.
This study has a few limitations and strengths. Although we assessed and adjusted for many covariates, like any observational study, there is a possibility of residual confounding and selection bias. Moreover, data on influenza infection during pregnancy were not included in the PRAMS dataset. Although the primary explanation of the effects of influenza immunization in pregnancy on birth outcomes is through prevention of infection, having influenza infection data would have provided additional support for our findings. Another issue is that the information regarding maternal influenza immunization was based on recall and could be susceptible to information bias. However, the vaccination rates in our study are similar to the rates computed by other authors for Georgia, and to the United States national level coverage estimated by the National Health Interview Survey 
The PRAMS dataset does not contain information regarding the precise trimester of vaccination. Therefore, the effect of vaccination in a specific trimester could not be evaluated. Moreover, it is possible that mothers of premature infants had less time to receive influenza vaccine than mothers of term infants (i.e., reverse causality). On the other hand, since this was a population-based study with a sampling strategy aimed at producing representative estimates, the temporal distribution of influenza vaccination in pregnancy would be similar to that of the general population, hence adding to the generalizability of our findings. The results of this study, nevertheless, need to be replicated in other populations as it is plausible that the impact of vaccines on birth outcomes would vary with the underlying influenza epidemiology and demographic characteristics.