Mining Spatial Gradual Patterns: Application to Measurement of Potentially Avoidable Hospitalizations

Gradual patterns aim at automatically extracting co-variations between variables of data sets in the form of “the more/the less” such as “the more experience, the higher salary”. This data mining method has been applied more and more in finding knowledge recently. However, gradual patterns are still not applicable on spatial data while such information have strong presence in many application domains. For instance, in our work we consider the issue of potentially avoidable hospitalizations. Their determinants have been studied to improve the quality, efficiency, and equity of health care delivery. Although the statistical methods such as regression method can find the associations between the increased potentially avoidable hospitalizations with its determinants such as lower density of ambulatory care nurses, there is still a challenge to identify how the geographical areas follow or not the tendencies. Therefore, in this paper, we propose to extend gradual patterns to the management of spatial data. Our work is twofold. First we propose a methodology for extracting gradual patterns at several hierarchical levels. In addition, we introduce a methodology for visualizing this knowledge. For this purpose, we rely on spatial maps for allowing decision makers to easily notice how the areas follow or not the gradual patterns. Our work is applied to the measure of the potentially avoidable hospitalizations to prove its interest.