Methods and design
A cross-sectional cohort design using administrative data was used for this study.
Calgary Zone Public Health maintains a database to track individual as well as community immunization rates. Nurses and clerks are responsible for entering vaccination data into the database. Initial database entries are derived from live births in the zone and new entries are made when individuals move to the zone and contact Public Health themselves or register their child in school. Clerks also verify that all children enrolled in area schools are included in the database by cross-referencing with school registration data at each immunization time point. A data analyst within Alberta Health Services maintains the database, cleans and verifies data on a continuous basis. This database was used for analysis.
Individual data for all grade 5 (ages 9–11) and grade 9 (ages 13–15) girls for school years 2009–2010 and 2010–2011 was available in the database, as well as 2008–2009 for grade 5 girls. Grade 9 girls did not receive the vaccine in 2008–2009. The database captured 100% of students registered in area schools. Data for girls who were home-schooled and registered with a board of education outside of the City of Calgary were not available. Girls attending Hutterite colony schools (religion-based, rural communities that are isolated from mainstream communities) and those who were home-schooled, but included in the database, were excluded from the study for consistency and the challenge in identifying a vaccination delivery system. Together these girls comprised less than 0.5% of the total sample.
Data included: postal code, school system’s religious affiliation (public [non-denominational], Catholic, and private), grade, vaccination delivery model (“in-school”, “community”), Hepatitis B vaccination status (none, incomplete, complete), and number of HPV vaccine doses received. HPV vaccination status was considered complete if three doses were received within one year; receipt of one or two doses was considered incomplete.
The Pampalon material deprivation index was used as a proxy measure to identify socioeconomic status (SES) for each student based on the geographic location of the student’s residence [
3]. Each student’s postal code was linked to the corresponding dissemination area (the smallest standard geographic area for which census data are available and is comprised of 400 to 700 individuals). Statistics Canada 2006 census data on income, education, and employment were used to calculate deprivation factor scores for each dissemination area and the Alberta population was divided into quintiles based on the scores for each area [
3]. Each student in our study, based on their residence’s dissemination area, was assigned a deprivation category of 1 (least deprived) through 5 (most deprived) to correspond with highest through lowest SES.
All analyses were performed in Stata S/E Version 12 [
4]. Descriptive statistics were conducted with categorical variables expressed as frequencies and percentages, with differences in the distributions examined using chi-square tests. A multivariable logistic regression model was developed using an interactive approach [
5] where the outcome variable was complete vaccination (3 doses) versus none/incomplete vaccination (0, 1, or 2 doses). Five variables were considered for entry into the model. The three primary predictor variables were delivery system, school type, and SES. Delivery model and school type were subsequently collapsed into one variable (see below). Hepatitis B vaccination status and grade (as a proxy for age) were also considered because Smith et al. found vaccination history to be the strongest predictor of initiation of HPV vaccination in Ontario, and Reiter et al. found that the daughter’s age was the strongest predictor of parents’ intent to vaccinate [
6,
7]. For estimation of coefficients and their standard errors, clustering by dissemination area was included to account for correlation between girls living in the same area.
To examine the possibility of interaction and confounding of school type and service delivery model on the relationship between SES and vaccination a new combined variable was created, resulting in the following five school types: public - in-school, private - in-school, private - community, Catholic - in-school, Catholic - community. In addition, due to small cell sizes and similarity of adjacent vaccination rates, some categories of SES were collapsed for logistic regression analysis. The new categories of SES included: High (highest SES category), Medium (second and third categories), and Low (fourth and fifth SES categories).
The possibility of interaction between the two primary variables (delivery model/school type and SES) was considered before inclusion of other variables. The significance of interaction terms was assessed using the Likelihood Ratio Statistic in the model, not accounting for clustering, since an equivalent statistic is not available for clustered data. Interaction terms were considered significant and included in the model if p < 0.05. The remaining predictor variables were eligible for inclusion in the multivariable model if they were significant at p < 0.1 in an individual logistic regression model or if there was evidence of confounding of the primary relationship under observation. Variables were retained in the multivariable model if p < 0.05.
Since this study was cross-sectional in design, it was possible to calculate the estimated proportion vaccinated, along with a 95% confidence interval, for various combinations of the coefficients to aid with interpretation of the final model. The probabilities were generated using the ‘margins’ command in Stata 12, accounting for grade and Hepatitis B vaccination status in these estimates and displayed using a profile plot.
This study was reviewed by the Chair of the University of Calgary Conjoint Health Research Ethics Board and deemed to contain no ethical matters that precluded its conduct.