Clinical Informatics Board Practice Exam

A key idea tested here is how Bayes’ theorem combines evidence from multiple findings. When multiple observations are used to infer disease status, a common simplifying assumption is that each finding is conditionally independent given whether the disease is present or not. In medical research, however, findings are often not independent: they can share the same underlying pathology, be caused by related physiological processes, or be subject to correlated measurement errors. This means the joint likelihood of observing all findings given the disease cannot be written as the simple product of the individual likelihoods. If dependencies are ignored, the posterior probability of disease can be biased, either overstating or understating the true evidence because the correlations aren’t properly accounted for. To handle this, analysts use models that capture dependencies among findings, such as Bayesian networks or multivariate joint distributions, which provide more accurate and calibrated posteriors. The other statements aren’t inherent limitations of Bayes: data bias is a general data-quality issue, independence is not always true, and medical data do not always come from randomized trials.