During the study period, 4,275 persons were tested for influenza. Of them, 879 (20.6%) were positive for influenza A(H1N1)pdm09 virus, 3,391 (79.3%) were negative for all influenza viruses, and 5 who were positive for influenza A but negative for the pandemic virus were excluded from study. We also excluded 35 persons (8 case-patients, 27 controls) who did not usually reside in Manitoba and 185 infants (26 case-patients, 159 controls) who were <6 months of age. A total of 726 hospitalized test-negative controls and 14 case-patients who were hospitalized during the study period but not around the time of testing were also excluded. Thus, there was a total of 3,310 study participants: 205 hospitalized case-patients, 626 community case-patients, and 2,479 community test-negative controls.

Consistent with previous reports from Manitoba (19,27), we found that the influenza A(H1N1)pdm09 virus–positive case-patients during the first pandemic wave were younger and more socioeconomically disadvantaged than controls (Table 2). Probably because they were younger, community case-patients had fewer prior hospitalizations and physician visits and were less likely than controls to have had a diagnosed chronic or immunocompromising medical condition. Consistent with the literature (20,27), we also found that younger children, pregnant women, residents of northern Manitoba, socioeconomically disadvantaged persons, and persons with chronic diseases were more likely to be hospitalized for infection with the pandemic virus.

About 17% of the community case-patients and 23% of the community controls received TIV during the 2008–09 influenza season (Table 3). The crude OR for the association of TIV receipt with subsequent infection with influenza A(H1N1)pdm09 virus was 0.7 (95% CI 0.6–0.9), corresponding to a vaccine effectiveness estimate of 30%. Adjusting for age, sex, region of residence, SEFI, and week of specimen collection (model B) resulted in an OR of 1.1 (95% CI 0.8–1.4). Additional adjustment for all other measured confounders (model C) did not appreciably change the OR estimates (OR 1.0, 95% CI 0.7–1.4).

In analyses limited by small numbers, study participants who received the seasonal 2007–08 and the 2008–09 TIV had a 40% increased risk for influenza A(H1N1)pdm09 virus infection compared with those who received neither vaccine. In general, there was a trend of increasing risk for influenza A(H1N1)pdm09 virus detection with the receipt of more TIVs over the preceding 5 years (Table 3). However, annual receipt of TIV over the preceding 5 years was inversely associated with the risk for pandemic virus detection. In addition, having ever received PPV23 was not associated with increased risk for influenza A(H1N1)pdm09 detection (OR 0.9, 95% CI 0.6–1.5).

There was no evidence that the association between receipt of the 2008–09 TIV and influenza A(H1N1)pdm09 detection varied by the sex, age, or place of residence of study participants; by the presence of chronic medical conditions in participants; or by the epidemic phase (Table 4). For example, the OR was 1.2 (95% CI 0.6–2.4) for participants who were >50 years of age, and the OR was 0.9 (95% CI 0.6–1.3) for younger participants (p_interaction = 0.31). The OR was 0.9 (95% CI 0.6–1.3) among those who did not have a chronic disease and 1.3 (95% CI 0.7–2.3) for those who did (p_interaction = 0.33).

Among participants with laboratory-confirmed influenza A(H1N1)pdm09 virus infection, the receipt of the 2008–09 TIV was associated with a statistically nonsignificant reduction in the risk for hospitalization (OR 0.6, 95% CI 0.3–1.3) (Table 5). On the other hand, having ever received PPV23 was associated with a statistically nonsignificant increase in risk for hospitalization. However, these analyses were limited by small numbers, which resulted in wide CIs and, likely, unstable estimates.

With 626 community case-patients and 2,479 community controls and by using a 2-sided 5% significance level, our study had ≈80% power to detect an OR as small as 1.3 (30% increase in risk), assuming 19% of controls received the 2008–09 TIV (28). With 205 hospitalized case-patients and 640 community case-patients, the study power was considerably lower for the hospitalization analysis: the smallest detectable OR with 80% power was 1.7.

We found no evidence that receipt of the 2008–09 TIV increased or decreased the risk for laboratory-confirmed influenza A(H1N1)pdm09 virus infections during the first wave of the pandemic in Manitoba. In analyses limited by small numbers, the 2008–09 TIV was associated with a statistically nonsignificant reduction in the risk for hospitalization.

These results are consistent with those in the bulk of the literature. Several studies using different designs (cohort as well as test-negative and conventional case–control studies) from Australia (4,5), England (6), Spain (7,8), and the United States (9–11) found that the 2008–09 TIV neither increased nor decreased the risk for influenza A(H1N1)pdm09 virus infection during the first wave of the pandemic.

The lack of protective effects against influenza A(H1N1)pdm09 virus is not surprising given the substantial antigenic divergence between the pandemic virus and recently circulating seasonal influenza A (H1N1) viruses among humans (29) and the lack of a cross-reactive antibody response to the pandemic strain in serologic studies of TIVs for humans and animals (12–14,30). However, 2 case–control studies from Mexico have indicated a protective effect (35%–74%), especially against severe infections (2,3). Concerns about possible selection bias and uncontrolled confounding were raised about both studies (31,32), although a reanalysis of the second study that attempted to address these concerns confirmed the initial results (33). Lower levels of seroconversion among TIV-vaccinated compared with unvaccinated persons were observed among nurses in a cohort study in Canada (15) and among military personnel in a cohort study in Singapore (16). However, it is unclear whether the results from these subpopulations are applicable to the general population. In the Singapore study, TIV was not protective against seroconversion among community participants. Similar reservations might be applicable to a US case–control study that reported a protective effect for the 2008–09 TIV among active-duty military service members (34).

On the other hand, increased risk for influenza A(H1N1)pdm09 virus infection with receipt of the 2008–09 TIV was reported for US military beneficiaries who sought care for influenza-like illness at Navy clinics in San Diego County, California, USA, during the first wave of the pandemic (35). However, the positive association with confirmed subtype H1N1 infection was seen only in univariate analyses restricted to active-duty members and was not observed for other study groups. In a small pilot study from Hong Kong, 31% of children who were randomly selected to receive TIV in November 2008 had serologically confirmed influenza A(H1N1)pdm09 virus infection, compared with 12% of the children who received a placebo (30). However, there were no significant differences between the 2 groups in rates of influenza-like illness, acute respiratory symptoms, or PCR-confirmed pandemic infections. Four studies from Canada, including the aforementioned Ontario study, have also reported increased risk with TIV use, especially among younger persons (1). The inconsistency between our results and those of the Ontario study could be due to bias or residual confounding in either study.

Major strengths of our study include its population-based design and relatively large sample size. Because of the availability of accurate automated vaccination records (23), this study was less susceptible to recall bias and to misclassification of exposure status, issues that are common in observational studies in which vaccination information is self-reported. Misclassification of disease status was minimized by use of an accurate diagnostic test (RT-PCR) (22). However, it is well-known that viral RNA is occasionally not detectable by RT-PCR (e.g., because of delay in specimen collection), which means that some case-patients in our study might have been misclassified as controls. It is difficult to predict the direction of resulting bias. If the likelihood of false-negative results was not related to receipt of vaccine, our estimates would generally bias toward the null, masking any associations (36). If false-negative results were more likely among the unvaccinated persons (which could be the case if lack of vaccination and the delay in getting tested are caused by lack of timely access to primary care), our OR estimates could have been biased downwards, potentially masking any harmful effects of vaccination. We did not have information about testing delay, but we used proxies for access to health care (e.g., frequency of physician encounters) to adjust for factors that might be associated with promptness of testing. Stratifying the analysis by quintiles of the number of physician visits in the previous year did not result in any significant differences in the estimated ORs.

To further control for confounding by access to and propensity to seek health care, we employed a test-negative case–control design, in which all participants were tested for influenza. We believe that using test-negative controls was the most practical way to sample controls from the population that gave rise to case-patients in our study (37). Had we sampled from the population at large, some of these controls would have been persons who would have never been tested for influenza if they had it (e.g., because of asymptomatic infection or because of lack of timely access to ambulatory care) and would have never appeared in our database as case-patients. The resulting bias could lead to underestimation of vaccine effectiveness if, as expected, receipt of the vaccine is positively associated with better access to ambulatory care and, therefore, to testing.

The controls in our study appeared to be representative of their respective age groups in the Manitoba population, and in general, they had characteristics similar to those for the control group in the Ontario case–control study (1) (Table 6). For instance, in our study the percentage of controls who received the 2008–09 TIV was ≈8% for participants 12–19 years of age, 14.5% for those 20–34 years of age, 19% for those 35–44 years of age, 28% for those 45–64 years of age, and 57% for those >65 years of age (Table 6). The corresponding percentages for the Manitoba population were 14%, 13%, 18%, 28%, and 67%, respectively (38). In the years leading to the pandemic, influenza vaccination policy in Manitoba was consistent with the recommendations of the Canadian National Advisory Committee on Immunization (21).

Information about several confounders was obtained from administrative databases. The completeness and accuracy of the MH database are well established, and these databases have been used extensively in studies of postmarketing surveillance of various drugs and vaccines (39). However, it is possible that there was a measurement error in some variables, which could result in residual confounding. In addition, the protective effects we observed against hospitalization might be related to confounding by factors that were not measured in this study, e.g., functional capacity (healthy vaccinee bias) (40).

Results from our study in Manitoba corroborate findings from studies outside Canada that the 2008–09 TIV neither increased nor decreased the risk for influenza A(H1N1)pdm09 virus infection. Additional epidemiologic and experimental investigations are needed to clarify the relationship between TIV use and infection with the pandemic strain.