Researchers from the UPM's School of Computing and another researcher from Holland’s Radboud University Nijmegen have published an article in the Decision Support Systems journal proposing a technique for improving decision making in medical environments.
The research focused on patients with the disease known as non-Hodgkin lymphoma of the stomach, which accounts for 5% of all gastric neoplasms. This disease is believed to be caused by a chronic Helicobacter pylori infection.
The proper diagnosis and medical treatment of this type of cancer depends on the use of uncertain information. To further their knowledge of this disease, physicians have to resort to data tables. Data tables give a broad overview of a set of factors within a group of patients for pattern extraction. These patterns are helpful for preventing, diagnosing and treating this lymphoma.
These decision-making processes, which affect the health and lives of many people, are complicated: the size of the tables used is massive, which is an obstacle for analysing and interpreting their contents.
What the research team proposed in the article was to introduce what are known as KBM2L Lists to support clinical decision making in patients suffering from this disease. These lists are dynamic knowledge representations that are capable of representing the analysed system knowledge in a summarized and understandable form.
As they explain in their article, this proposal is an attractive alternative to today’s clinical decision-making methods, as the system is able to extract and present patterns from an influence diagram for physicians to interpret. Influence diagrams are a way of representing decision-making problems as they are perceived by the decision maker, in this case the physician.
In a preliminary evaluation of this technique, the researchers generated KBM2L lists specific to non-Hodgkin lymphoma of the stomach that one of the authors, both a computer scientist and experience physician, interpreted. The KBM2L lists could be applied to other diseases.
The authors of this research are Concha Bielza and Juan Fernández del Pozo, from the Decision Analysis and Statistics Group at the UPM’s School of Computing, and Peter F. Lucas, from Radboud University Nijmegen’s Information Science and Computing Institute in Holland.