Term: Modeling and Estimation of Rates
Last Updated: 2006-02-15
To estimate and compare rates of events, the count of events for each indicator was modeled using a Poisson or negative binomial distribution, depending on which distribution provided the best model fit. Relative risks were estimated for each region and for each sex within each region. Parameters included in the model consisted of region, sex, age, a region by sex interaction, and if age was modeled as a continuous variable, then both the linear and quadratic terms were included. The reference groups for region and sex were Manitoba and male/female combined sex, respectively. If age was modeled as a categorical variable, then the oldest age group was used as a reference group. To estimate relative risks of rates rather than events, the log of the population count in each region by sex by age stratum was included in the model as an offset. Contrasts were calculated from the parameter estimates of the model to calculate relative risks for each region as well as for each sex within each region. These contrasts also compared the relative risk for each region (or for each sex within each region) to the overall provincial relative risk. The values obtained from the contrasts were actually a linear combination of the natural logarithm of the parameter estimates, so an exponential transformation was necessary to obtain a tangible relative risk of events. Finally, the estimated rates were calculated by multiplying the Manitoba overall crude rate by the appropriate relative risk estimate.