Joris Mestdagh, SPF Public Health

Statistical Techniques in Nursing Related Models

Abstract:

In order to finance the nurses in hospitals, we create so-called NRGs or Nursing Related Groups. These NRG’s are clusters of patients which have very similar care. It is a current objective of the SPF Public Health to create such Nursing Related Groups.
In order to do this, we utilize many techniques from unsupervised learning. For example, we will try to visualize data and do a dimension reduction with self organizing maps, principal component analysis, princals, t-SNE, neural networks, recursive feature selection.
On the other hand, we try to obtain interesting clusters of similar patients. For this, we try some clustering techniques such as k-means, agglomerative clustering and DBSCAN.
In this talk, we will illustrate how well these techniques work on the specific data set of nursing care. We review the costs and benefits of each technique both theoretically as applied to this project of nursing related models.