Optimizing Health Policies with Bayesian Networks
For diagnostic tests used by clinicians, sensitivity and specificity are known and well-understood quantities. Reasoning with these measures in practice, however, is much less straightforward. This challenge manifests itself in many examples of the "base rate fallacy." Fortunately, Bayes' Rule can perfectly resolve any questions of this kind. As a result, general treatment policies can be established as a function of test results.
It becomes more challenging when additional uncertainties enter the picture, such as the base rate of a disease not being known. In the example we discuss in this webinar, the prevalence of malaria varies across different geographies and cannot be established due to the absence of local epidemiological data. Additionally, the malaria test in our example has low specificity, which makes it difficult to rule out the disease. Our objective is to develop a Bayesian network model for establishing an optimal general treatment guideline despite these uncertainties. Furthermore, our Bayesian network model will allow us to evaluate under what hypothetical conditions such a policy would need to change and what variables would be most sensitive in this regard.