No Association between 2008–09 Influenza Vaccine and Influenza A(H1N1)pdm09 Virus Infection, Manitoba, Canada, 2009

Receipt of seasonal inactivated trivalent vaccine neither increased nor decreased the risk for pandemic influenza virus infection.

T he nature of the relationship between receipt of the 2008-09 seasonal inactivated trivalent infl uenza vaccine (TIV) and the risk for infection with the pandemic (H1N1) 2009 virus strain, hereafter referred to as A(H1N1) pdm09, remains unclear. A case-control study in Canada that used data from a network of sentinel physicians monitoring infl uenza vaccine effectiveness in the provinces of Alberta, British Colombia, Ontario, and Quebec found an increased risk for infl uenza A(H1N1)pdm09 virus infection among persons who received the 2008-09 TIV (odds ratio [OR] 1.7, 95% CI 1.0-2.7); an increased risk for severe illness was not detected (1). In addition, 3 other studies conducted by the same team found a 1.4-to 2.5fold increased risk for infection (laboratory confi rmed) with infl uenza A(H1N1)pdm09 virus among persons who received the 2008-09 TIV (1). The 3 studies were 1) a testnegative case-control study in Ontario; 2) a household transmission cohort study in Quebec; and 3) a conventional case-control study using population controls in Quebec.
The results of these studies in Canada were not confi rmed by studies conducted elsewhere. In fact, several studies using different designs found that TIV partially prevented or had no effect on infections with the pandemic strain (2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15)(16). It has been suggested that the fi nding in Canada of an increased risk for infl uenza A(H1N1)pdm09 virus infection among persons who received the 2008-09 TIV might be unique to Canada; the increased risk might be related to the use of the domestically manufactured vaccine (1) or to the timing of the pandemic in relation to the most recent infl uenza season and the types of circulating infl uenza strains during that season (17,18). At the time of the pandemic, the Canadian province of Manitoba was not part of the Canadian vaccine effectiveness monitoring network. However, the availability of a provincewide, population-based immunization registry and laboratorybased infl uenza surveillance system provided a unique opportunity to investigate these issues in Manitoba.
In the fi rst wave of the pandemic (May-August 2009), Manitoba was more severely affected than any other Canadian province, accounting for 50% of hospital intensive care unit admissions attributable to the virus in Canada (19,20). TIVs used in Manitoba during the 2008-09 infl uenza season were identical to those used elsewhere in Canada; they included 15 μg hemagglutinin each of A/ Brisbane/59/2007 (H1N1)-like virus, A/Brisbane/10/2007 (H3N2)-like virus, and B/Florida/4/2006-like virus. These were the 3 strains recommended that year by the World Health Organization for infl uenza vaccines in the Northern and Southern Hemispheres (21). In Manitoba, as in other provinces, ≈75% of the administered seasonal infl uenza vaccine doses were manufactured domestically (Fluviral; GlaxoSmithKline, Mississauga, Ontario, Canada); imported vaccines, predominantly Vaxigrip (Sanofi Pasteur Ltd, Toronto, Ontario, Canada), comprised the remaining 25%. The live attenuated infl uenza vaccine was not available in Canada during the 2008-09 season.
To investigate whether use of TIV in Manitoba was associated with infl uenza A(H1N1)pdm09 virus infection during the fi rst wave of the pandemic, we conducted a population-based case-control study using data from Cadham Provincial Laboratory (CPL) and the Manitoba Immunization Monitoring System. The test-negative casecontrol design used in this study is similar to the design of the Ontario study (1).

Data Sources
This study was conducted using de-identifi ed records obtained by linking the CPL database, the Manitoba Immunization Monitoring System, and other Manitoba Health (MH) administrative databases after securing the approval of the Research Ethics Board of the University of Manitoba and the Health Information Privacy Committee of MH. MH provides publicly funded health insurance coverage to 99% of the 1.2 million residents of Manitoba. The coverage includes laboratory testing and hospital and outpatient physician services, including immunization and laboratory services. Eligibility for coverage is not based on age or income. For administrative purposes, MH maintains several centralized electronic databases that can be linked by using a unique health services number.

Study Population
All Manitoba residents >6 months of age who had respiratory specimens submitted to CPL, the province's only public health laboratory, for infl uenza testing during April 27-August 21, 2009, were included in this analysis. During the pandemic, guidelines for testing patients seeking care for infl uenza-like illness were issued by MH; anecdotal evidence indicated that, to a great extent, physicians followed the guidelines. Patients were tested in hospital and ambulatory care settings; the specimens obtained were predominantly nasopharyngeal and nasal swab samples. For the duration of the fi rst pandemic wave, all infl uenza testing in Manitoba was completed at CPL by using a real-time duplex reverse transcription PCR (RT-PCR) developed by the National Microbiology Laboratory (22). We obtained information about infl uenza testing from CPL's electronic database.

Study Design
Consistent with the design of the Ontario test-negative case-control study (1), we identifi ed 3 nonoverlapping study groups: 1) the hospitalized cases group comprised persons in the study population (as defi ned above) who had real-time duplex RT-PCR test results positive for infl uenza A(H1N1)pdm09 virus and who had been admitted to a hospital in Manitoba around the time of testing (within ± 1 week of collection of their fi rst infl uenza A(H1N1)pdm09 virus-positive specimen); 2) the community cases group comprised A(H1N1)pdm09-positive persons who had not been hospitalized during April-August 31, 2009; 3) the community controls group comprised persons who were not hospitalized during this period and who tested negative for infl uenza A and B. Identifi cation of these 3 groups enabled us to assess whether use of TIV was associated with the detection of infl uenza A(H1N1)pdm09 virus infection (by contrasting the odds of TIV use among community casepatients and community controls). Identifi cation of the 3 groups also enabled us to assess whether use of TIV was associated with increased risk for hospitalization as an indication of the severity of disease (by contrasting the odds of TIV use among hospitalized and community casepatients). Information about hospitalization status was obtained from the Hospital Separation database.

Determination of Vaccination Status
For all study participants, information about receipt of TIV and polyvalent pneumococcal polysaccharide (PPV23) vaccine during or before the 2009-10 infl uenza season was obtained from the Manitoba Immunization Monitoring System, the population-based provincewide registry that has recorded virtually all vaccinations administered to Manitobans since 1988. In addition to details about patients, the database stores information about the date of vaccination and the type and dose, but not the brand name, of the vaccine administered. The recorded vaccination information is considered highly complete and accurate (23).

Information about Potential Confounders
Study participants were assigned to a neighborhood of residence based on their postal code as recorded in the MH Population Registry. Information about socioeconomic status was obtained by using the postal code of residence and a previously validated area-based Socioeconomic Factor Index (SEFI) (24).
Information about coexisting diseases and propensity to seek health care (measured as the number of hospital and family physician visits in the previous 12 months) was obtained from the Hospital Separation and Physician Claims databases. Since 1971, these databases have recorded information about most hospital admissions and outpatient physician visits, respectively. Previously validated algorithms were used to identify various chronic diseases and other indications for vaccination (25) ( Table 1). Immunosuppression was defi ned as having a diagnosis of cancer, AIDS, or another immunodefi ciency disorder or as receiving prescriptions for immunosuppressive drugs. Information about the use of immunosuppressant and antimicrobial drugs and neuraminidase inhibitors was obtained from the Drug Program Information Network, the comprehensive database of all out-of-hospital prescriptions dispensed in Manitoba. Pregnancy status was determined from the databases mentioned above by using disease and tariff codes for different conditions and procedures indicative of ongoing pregnancy or the completion of pregnancy (26) ( Table 1).

Statistical Analysis
We used unconditional logistic regression models (fi tted to community case-patients and community controls) to estimate odds ratios (ORs) for the association between the receipt of the 2008-09 TIV and subsequent infection with laboratory-confi rmed infl uenza A(H1N1)pdm09 virus while adjusting for confounding. Results are presented for unadjusted models (model A) and for models that were adjusted a priori for age, sex, place of residence, SEFI, and week of specimen collection (to account for changes in infection incidence and laboratory testing practices) (model B). Other potential confounders (Table 1) were included in the fully adjusted models if their inclusion resulted in a >2% change in crude ORs. Using this criterion, we also adjusted the fi nal models (model C) for pregnancy, antiviral drug use, presence of a chronic or immunocompromising medical condition, and number of hospital admissions and family physician visits in the previous 12 months. Model C also included mutual adjustment for the 2007-08 TIV and the PPV23.
These analyses were repeated after stratifi cation by potential confounders and effect modifi ers, such as age group, place of residence, epidemic phase (before and after the peak), and presence of chronic conditions. We also assessed for possible effect modifi cation between the 2008-09 TIV and the 2007-08 TIV and the PPV23. The statistical signifi cance of adding the interaction terms was assessed by using a likelihood ratio test. Similar analyses, contrasting the odds of TIV use among hospitalized casepatients with those among community case-patients, were performed to assess whether use of the 2008-09 TIV was associated with increased risk for hospitalization.

Results
During the study period, 4,275 persons were tested for infl uenza. Of them, 879 (20.6%) were positive for infl uenza A(H1N1)pdm09 virus, 3,391 (79.3%) were negative for all infl uenza viruses, and 5 who were positive for infl uenza 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 testnegative controls.
Consistent with previous reports from Manitoba (19,27), we found that the infl uenza A(H1N1)pdm09 viruspositive case-patients during the fi rst 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 infl uenza season ( Table 3). The crude OR for the association of TIV receipt with subsequent infection with infl uenza 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 infl uenza A(H1N1)pdm09 virus infection compared with those who received neither vaccine. In general, there was a trend of increasing risk for infl uenza 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 infl uenza 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 infl uenza A(H1N1)pdm09  (Table 5). On the other hand, having ever received PPV23 was associated with a statistically nonsignifi cant 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% signifi cance 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 was considerably lower for the hospitalization analysis: the smallest detectable OR with 80% power was 1.7.

Discussion
We found no evidence that receipt of the 2008-09 TIV increased or decreased the risk for laboratory-confi rmed infl uenza A(H1N1)pdm09 virus infections during the fi rst wave of the pandemic in Manitoba. In analyses limited by small numbers, the 2008-09 TIV was associated with a statistically nonsignifi cant 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 casecontrol studies) from Australia (4,5), England (6), Spain (7,8), and the United States (9)(10)(11) found that the 2008-09 TIV neither increased nor decreased the risk for infl uenza A(H1N1)pdm09 virus infection during the fi rst wave of the pandemic.
The lack of protective effects against infl uenza A(H1N1)pdm09 virus is not surprising given the substantial antigenic divergence between the pandemic virus and recently circulating seasonal infl uenza 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)(13)(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 confi rmed 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 infl uenza A(H1N1)pdm09 virus infection with receipt of the 2008-09 TIV was reported for US military benefi ciaries who sought care for infl uenza-like illness at Navy clinics in San Diego County, California, USA, during the fi rst wave of the pandemic (35). However, the positive association with confi rmed 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 confi rmed infl uenza A(H1N1)pdm09 virus infection, compared with 12% of the children who received a placebo (30). However, there were no signifi cant differences between the 2 groups in rates of infl uenza-like illness, acute respiratory symptoms, or PCR-confi rmed 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 populationbased 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 misclassifi cation of exposure status, issues that are common in observational studies in which vaccination information is self-reported. Misclassifi cation 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 misclassifi ed as controls. It is diffi cult 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 falsenegative 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 signifi cant differences in the estimated ORs.
To further control for confounding by access to and propensity to seek health care, we employed a testnegative case-control design, in which all participants  (24), and week of specimen collection, no. hospital admissions and family physician visits in previous 12 mo, pregnancy, presence of a chronic or immunocompromising medical condition, and antiviral drug use. Model also included mutual adjustment for 2007-09 TIV and PPV23.
were tested for infl uenza. We believe that using testnegative 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 infl uenza 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, infl uenza 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 fi ndings from studies outside Canada that the 2008-09 TIV neither increased nor decreased the risk for infl uenza 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.
Dr Mahmud is an assistant professor in the Department of Community Health Sciences, University of Manitoba, and a medical offi cer of health at the Winnipeg Regional Health Authority. His primary research interests include evaluation of vaccine effectiveness and safety and cancer chemoprevention.