Nursing Informatics and Computer Aided Education (NICE) forms a three-year project, funded by the European Commission under the Phare Tempus programme. In the final NICE year, the main objectives have been achieved. They consist in the creation of a new short cycle degree courses in Nursing Informatics, and the design of computer managed instruction (CMI) tools for nurses. One of these tools is based on the concept of decision trees and automatic learning, in order to enhance the learning process in nursing education. Before the ITIS-ITAB'99 audience, Professor Peter Kokol from the University of Maribor in Slovenia, explained the exact functioning of decision trees and their usefulness in learning how to make correct diagnoses.
Institutions and university colleges in five different countries are involved in the NICE project. The partners are situated in Slovenia, Austria, France, Greece, and Italy. They all are convinced that much of the work that is currently performed by nurses will be done by computer in the future. As a result, nurses will have to be properly educated in Information Technology. This has inspired the NICE partners to the introduction of a new curriculum, and to the development of computer aided tools. The decision tree in fact is an intelligent system, which provides power tools for thinking, as defined by Randy Davis from MIT, Massachusetts.
The nurse's task is mainly based on data gathering and decision taking with information used as input and generated as output. Decision trees allow us to extract knowledge from data, in order to make decisions in cases where it is very difficult to efficiently apply explicit human knowledge. The algorithm for learning a decision tree is trivial. The tool is built hierarchically and used for forecasting to which unique class a training object belongs. The tree not only offers the decision in a previously unseen case but also the explanation of it, which is of vital importance in medical applications. Thus, it deals with the discovery of hidden knowledge, unexpected patterns and new rules.
To produce or induct a tree, Professor Kokol starts with an empty tree and a set of training objects. Each single object is defined by a series of attributes and a class label for category or outcome. Additionally, the tree consists of nodes and edges or links. The node may be internal with splits or external. The external node is a so-called leaf and is labelled with outcomes. In turn, edges are marked with different outcomes of test, which are performed in the source node. The recursive induction process runs in four steps. For all of the training objects with the same outcome, a leaf is generated with exactly that outcome before going one level up in recursion.
Using a heuristic evaluation function, a nurse can find the best attribute for the creation of an internal node with split for the selected attribute. In this way, the training set is split into subsets. For every subset, the nurse again has to check the outcomes, in order to create a new leaf, and so on. As such, the knowledge and the decision making are represented in a fairly simple 2D hierarchical model. The decision tree can be used by nurses to support their decisions in new situations and by students to test their ability in diagnostic decision making. Trainees and nurses can even be encouraged to construct their own decision tree for the new cases they meet.
At present, the team of Professor Kokol has already completed the induction of a decision tree for fifteen types of medical diagnosis. The use of a decision tree is very popular among the students, but the teaching staff as yet, is very reserved about the innovative educational tool. For more news on the latest project developments, we invite you to check out the NICE Web site. We also like to refer to the VMW article EU course teaches Slovenian nurses computer aided health care.