It was hypothesized that age, density of health care workers in the community, and H1N1 influenza-level would be significant predictors of azithromycin use. Specifically, it was expected that communities with a high percentage of under 5 year olds would have positive correlations with azithromycin use as would communities with a high percentage of over 65 year olds. The final multivariable linear regression model did not support this hypothesis and the variable of communities with a high percentage of over 65 year olds did not remain in the final model. While communities with a high percentage of under 5 year olds remained in the model as a significant variable, the model indicated that communities with a high percentage of under 5 year olds had a negative correlation with azithromycin use as compared to the referent of communities with a low percentage of under 5 year olds. This is in agreement with the general findings of Akkerman et al. (2004) and Marra et al. (2006) that relative antibiotic use among children is not higher than in other age groups. However, this is in disagreement with the specific findings of Marra et al. (2006) concerning azithromycin. The hypothesized increased use of azithromycin in under 5 year olds was anticipated as a result of Marra et al. (2006) noting an increase in the prescription of this antibiotic in children. Thus the negative correlation found in the current study is surprising. Based on the causal diagram it is possible that the relationship between age and azithromycin use is being impacted by flu-level as an intervening variable and thus this consideration should be accounted for when interpreting results. Low income was not hypothesized as a significant predictor for azithromycin use, however, as a categorical variable it was a significant explanatory variable across all categories for the final model. Compared to the referent category of ≤13.8% low income in the community, communities with between 13.8-15.1% low income had a positive correlation with azithromycin use. The contrasting effect was noted as percentage of low income increased with negative correlations between azithromycin use and communities with 15.1-17.2% and >17.2% low income. This relationship may be associated with accessibility issues. Communities with a higher percentage of low income households may have reduced infrastructure such as public transit and pharmacies leading to a decrease in antibiotic use. It is also possible based on the causal diagram that high vaccination rates, flu-level, and density of health care workers were acting as intervening variables thus impacting the true relationship between income and azithromycin use. Low income and percentage of under 5 year olds interacted significantly in the model with all comparisons showing a positive correlation between a high percentage of under 5 year olds and increasing percentage of low income households in the community. While this interaction was not specifically hypothesized, the a priori causal diagram did highlight the potential for these two variables to interact in some fashion. A quadratic relationship between azithromycin use and density of health care workers in a community was hypothesized such that increased use was anticipated in communities with a low density of health care workers and in communities with a high density of health care workers. This hypothesis was incorrect as density of health care workers was modeled as a continuous variable and met linearity assumptions. There was a positive correlation between density of health care workers and azithromycin use indicating increased use in more serviced communities. This is in opposition to the findings of Cadieux et al. (2007), where antibiotic use was increased in high volume practices. However, it is possible that while these communities had several health care workers, patients may be concentrated among a few practices. This would be in agreement with the findings of Cadieux et al. (2007) and this possibility should be analyzed in future studies to further determine physician-availability and patient-load interactions. The relationship may also be impacted by flu-level and vaccination rates as intervening variables thus interpretation of results should take into account this potential pathway. H1N1 influenza-levels and antibiotic use was a significant predictor of azithromycin use as hypothesized. Azithromycin consumption was expected to increase accordingly with increased H1N1 influenza-levels. This was conjectured as potentially being caused by improper prescription of antibiotics for these viral infections. As compared to high levels of H1N1, communities with low and moderate flu-levels had a negative correlation to azithromycin use. This is in agreement with predictions of Goossens et al. (2005) who noted a peak in antibiotic prescription during the winter months and anticipated this result was due to flu-levels generally being higher at that time of year. After examining significance in the multivariate model, testing for confounding, and reviewing f-test results, it was determined that neither High % Under 65 nor High Community H1N1 Vaccination Rate had a significant impact on the model nor any of the other independent variables in the model. In order to create the most parsimonious model possible, these variables were confidently eliminated. Therefore, communities with High % Over 65 and High Community H1N1 Vaccination Rates would not be expected to have different consumption rates of azithromycin than those without High % Over 65 and High Community H1N1 Vaccination Rates. Interaction terms that were tested in the multivariable model were selected based on significant potential interactions that involved at least one of the predictors of interest. The interaction terms that were assessed included Low Income (categorized) * Health Care Workers per 10000, Low Income (categorized) * High % Under 5, and Low Income (categorized) * High % over 65. The potential interaction between Health Care Workers per 10000 * High % Under 5 was excluded albeit being significant and containing a predictor of interest because there is no biological or otherwise plausible explanation for a direct interaction occurring between these individual variables. After the completion of an F-test for the three remaining interaction terms, the only significant result came from the interaction between Low Income (categorized) * High % Under 5. A contrast was also conducted to ensure that there was a significant difference between the model with the interaction between High % Under 5 * Low Income (categorized) present and the model without (see Table 5). The difference between these models was significant, further supporting our inclusion of this interaction term in the model. There is plausibility for this interaction to occur; Antibiotic prescription has been found to be higher in low income children than children in higher income brackets (Lautenbach et al. 2003), and also higher in the 0-4 age bracket in comparison to older age brackets (Marra et al. 2006). Based on these results, it was the only interaction term that was left in the final multivariate model. Interesting, since Low Income % was categorized, the pattern of interaction coefficients at three categories of Low Income % can be assessed. The coefficients for the interaction between High % Under 5 and the three income categories from 2nd to 4th were 0.18, 0.28, and 0.23. Therefore, generally, as the percentage of households with low income increased, the Defined Daily Doses of azithromycin per 10000 also increased, peaking at the % Low Income bracket between 15.08-17.2%. This pattern further supports with findings by Lautenbach et al. (2003), suggesting that antibiotic prescription is higher in low income children than children in higher income brackets. The final assumptions of the linear regression to be evaluated are homoscedasticity of the residuals and normality of the residual distribution. The residuals were found to be homoscedastic, and the distribution of the residuals did not appear to deviate from normality. The Shapiro-Wilks test conducted on the residuals was deemed significant (P=0.001), however this result can be attributed to the large sample size used in the study. As more data points are included in the test, the chances of a significant test result increase. Very small deviations from normality are detected despite largely being inconsequential, and the result is a significant test result. Therefore, we based our assessment on normality on a graphical assessment and deemed the residuals normally distributed. Outliers identified in the residual analysis included Community IDs 300 and 301. The removal of these points did not significantly affect the magnitude or direction of the coefficients in the model. There are no apparent biological implausibilities or abnormal values in measurements conducted within these communities, and given the comprehensive analysis of potential predictors for increased Defined Daily Doses of azithromycin per 10 000, it is unlikely that an important covariate is missing from the model.