Ethnic disparities in acquiring 2009 pandemic H1N1 influenza: a case–control study

We conducted a test-negative case–control study, which bases case status on laboratory confirmation of pH1N1 and controls on the same test being negative [18]. Our population was a clinic-based sample, consisting of individuals presenting to clinics for medical care. We collected information on symptoms at the time of testing to restrict our sample to those presenting with influenza-like illness, defined as an acute onset of respiratory illness with fever, cough and with one or more of the following: sore throat, arthralgia, myalgia, or prostration which is likely due to influenza. Ontario residents of all ages who had nasopharyngeal swabs for pH1N1 by real-time reverse-transcriptase polymerase chain reaction (RT-PCR) were eligible for inclusion. We identified study participants as having confirmed pH1N1 through testing at the Public Health Ontario (PHO) Laboratory or Mount Sinai Hospital/University Health Network. The PHO laboratory conducted 100% of the testing in the first months of the pandemic. Later on, a few other laboratories in major centres, the largest being Mount Sinai Hospital/University Health Network, conducted testing as well.

We included individuals with a test date between April 13, 2009 and July 20, 2009 (pandemic wave 1). The study was approved by the Research Ethics Board of University of Toronto, Toronto, Ontario, Canada.

Trained interviewers from the Institute for Social Research at York University, Toronto conducted standardized telephone interviews between August and September 2009 to obtain information on demographics, health status, symptoms, and treatment, including hospitalization. For children 14 and under, interviews were completed by a parental proxy.

Individuals were asked to self-identify with an ethnicity/ethnicities. Ethnicity was based on the question, “People in Canada come from many different ethnic and racial backgrounds. I would like you to identify which ethnicity best describes you. You may choose more than one category.” Ethnicities were then read aloud to the participants. The ethnic groupings were based on the categories similar to Statistics Canada definitions: White, East/Southeast Asian, South Asian, Black, West Asian/Arab, Latin American, and Other. Individuals who identified with multiple ethnic groups were classified as Multi-Racial person. West Asian/Arab, Latin American and Multi-racial were combined with Other because of small sample sizes for these ethnic groups. Aboriginals could not be included in this analysis because of lack of data (<1% of the sample).

Conceptual models of the association between ethnicity and acquisition of pH1N1 were developed a priori to guide the model building. This enabled the consideration of potential confounders and effect modifiers. The variables considered for adults were age, gender, current smoking status, number of visits to their regular primary care doctor in the prior 12 months (low, ≤1; moderate, 2–6; high, >6), completion of post-secondary education, immigrant status (non-immigrant, 0–5 years in Canada, >5 years in Canada), whether there were children in the household, household density (the ratio of the number of individuals in a household to the number of sleeping rooms), previous seasonal influenza vaccination, residence in Toronto (the capital and largest metropolitan city in Ontario), healthcare worker status, and testing date before or after June 11th, 2009 (the date when clinicians in Ontario were instructed to switch from submitting samples from all patients presenting with influenza-like illness to only submitting samples from patients at high risk of complications or for clinical management of hospitalized cases) [19]. In addition to the individual level variables collected through the survey, we examined the impact of deprivation captured through the Deprivation Index (DI) developed by Institut national de santé publique du Québec (INSPQ) [6, 20, 21]. The DI includes social deprivation, material deprivation, total deprivation, employment rate, proportion of the population with less than high school education, and prevalence with low income, where individuals were classified as living in an area with a high prevalence of low income if they resided in dissemination areas that were in the top quintile for low income of the population study. We also considered if individuals were tested in Toronto since this represents the major urban centre of our study population as well as contains the highest proportion of visible minorities and other ethnic groups. Underlying medical conditions were analyzed as a binary variable (≥1 underlying medical conditions, which included diabetes, cancer, heart disease, asthma, chronic obstructive pulmonary disease (COPD), immune disorder, liver disease, kidney or renal disease, splenectomy, sickle cell anemia, seizure disorder, leukemia, cystic fibrosis, cerebral palsy), and obesity was considered separately. In children, all the same variables were considered, with underlying medical conditions restricted to those commonly observed in children (≥ 1 of either asthma or immune disorder).

We first tested the interaction between ethnicity and being a child (< 18 years of age) in order to determine if appropriate to pool adults and children. Given the significant interaction and the different confounders, we then had to stratify all analysis according <18 years of age versus ≥ 18 years of age. Descriptive statistics (means and proportions) were calculated for all variables. In order to develop the multivariable model, the Hosmer and Lemeshow forward model building strategy was used [22]. Briefly, odds ratios were obtained from univariate logistic regression to identify risk factors for acquisition of pH1N1. Variables were then entered into the model based on their effect size and p-value. Variables that had p-values <0.05 were included in the final model, as well as variables that had a significant (>10%) effect on any of the ethnic group’s OR. Age and gender were forced into the multivariate model, as these were deemed important variables to consider for the association. Gender was a priori identified as an important effect modifier and therefore the interaction was tested. At the end of this iterative model building process, a verification step was performed by adding any variable not selected for the original multivariate model one at a time to identify variables that, by themselves, were not influential but may have made an important contribution in the presence of other variables.

In addition, the following sensitivity analyses were performed: including individuals who were tested because of recent travel to Mexico; including individuals who did not present with symptoms at testing; no restrictions implemented (no exclusions); stratifying by presence of risk factors; stratifying by testing prior to June 11th, 2009; and according to Toronto residence.