Page - 32 - in Applied Interdisciplinary Theory in Health Informatics - Knowledge Base for Practitioners

Figure 4. Entropy as a function of tumour thickness t (in mm) as a test for 10-year fatality from malignant melanoma. 5. Using information theory to prioritise laboratory tests We are beginning to build up a number of approaches to using techniques for assisting with the choice of diagnostic tests in a clinical setting. First, we used relative entropy to discriminate between two candidate tests. Then we looked at the variation of entropy with the outcome of some diagnostic test (for a binary outcome) to identify the range of outcomes for which the test is informative. It should be emphasised that these are not the only tools available, but published work to date has argued for their value as additional tools that may aid in decision making. Lee and Maslove [6] made the case that an information theoretic approach could have particular value in the case of identifying redundancy in tests in an intensive care unit (ICU). Parsimony in the case of blood tests is particularly important in the ICU context as repeated bloodwork can: Cause anaemia and increase the need for blood transfusions; Cause patient discomfort; Disrupt sleep; Lead to delirium. The challenge, then, is to identify which blood tests are the most informative at a system wide level. A key issue here is that there may be some level of redundancy between laboratory tests; that is, especially over time, some tests may add little information over previously conducted tests. In cases where we can identify that there is a high degree of mutual information between tests (either through the same test being repeated too frequently, of for two different tests having too strong a dependency on 0 0.2 0.4 0.6 0.8 1 1.2 0 2 4 6 8 10 12 14 Tumour thickness t P.Krause / InformationTheoryandMedicalDecisionMaking32