Oral Cancer, Socioeconomic Status, and Access to Health Care in BC

Most clustered

Least clusters

Our team decided that a 10 000 metre bandwidth provided an adequate distance for the Kernel Density tool to identify significant clustering of cases outside Metro Vancouver. The area of a cluster is approximately the size of an urban centre. Since most of the treatment centres considered in this project are concentrated in Metro Vancouver, the size of clusters outside Metro Vancouver has little effect on the distance analysis as patients in these areas would have to travel very lengthy distances to reach their treatment facility anyway.

Trial and Error Results

Cases within Metro Vancouver

Most clustered

Least clustered

A much shorter distance was needed to identify significant clustering of cases within Metro Vancouver. This was primarily due to the abundant number of cases existing within a small area. Our team decided that a 300 metre bandwidth was an appropriate distance. A smaller distance results in clusters which lack significance. A larger distance results in clusters which neglect local geographic characteristics.

The kernel density function was used to identify areas of high oral cancer cases. Kernel density was chosen over k-means clustering or hotspot analysis because the kernel density function is able to determine where there is an increased likelihood of oral cancer cases based on spatial dependency in the area around the defined cluster. The function involves placing a symmetrical surface over each of the points and evaluating the distance from the point to a reference location based on a mathematical function. It then sums all the values for the surfaces that reference the location, and the procedure is repeated several times. The resulting raster image displays density estimates for the distribution of points and highlights significant clustering.

The kernel density function requires a defined bandwidth and cell size. To calculate the appropriate bandwidth, our team attempted to utilize the Mutli-Distance Spatial Cluster Analysis (Ripleys K-Function) tool found in ArcMap. Unfortunately, the time require by ArcMap to run this tool was not feasible within the timeline of our project. We therefore relied on intuition and trial and error to acquire the bandwidth appropriate for our project.

In our trials with the Kernel Density tool in ArcMap, it was evident that we could not analyze the injuries within Metro Vancouver and those in the rest of the province under the same variables. The number of injuries within the boundaries of Metro Vancouver are significant higher than the rest of the province due to the disparity in population concentration. The definition of a "cluster" within Metro Vancouver must be defined differently than that outside Metro Vancouver. As a result we separated the injury cases of Metro Vancouver from the injury cases in the rest of the province.

The clusters outside Metro Vancouver identified in this project were the top quantile of a kernel density analysis with a 10000-metre bandwidth.

The clusters within Metro Vancouver identified in this project were the top quantile of a kernel density analysis with a 30-metre bandwidth.