Diagnostic Test Interpretation and Reasoning under Uncertainty pp. 37-62
Authors: (Matt T. Bianchi, Brian Alexander, Neurology Department, Sleep Division, Massachusetts General Hospital, Wang Ambulatory Center, Boston, Massachusetts, and others)
Abstract: Decisions of when to order and how to interpret diagnostic tests form the cornerstone of medical care. However, interpreting a patient‘s test results in the context of their clinical presentation is not always straightforward. The objective rigor of diagnostic tests may be perceived as providing superior information, compared to the clinical evaluation (history and physical findings). Prior studies demonstrate that clinicians at all levels of experience and fields of expertise may not consistently incorporate two critical points of uncertainty in diagnostic decision making: imperfect test accuracy and the pre-test probability of the disease being investigated. This review highlights some pitfalls in diagnostic reasoning in this regard, and emphasizes strategies to incorporate clinical and diagnostic information, including uncertainty therein, to optimize diagnostic reasoning. The Bayesian approach to diagnostic testing, which incorporates information about the test as well as the patient being tested, is more commonly discussed than implemented in routine patient care. Uncertainty about estimating pre-test probability is an important obstacle in the implementation of Bayesian test interpretation. The statistical consequences (and greatest chance for mis-interpretation) are most evident when test results oppose clinical judgment. A simple graphical approach is presented to address the sometimes counter-intuitive relationships between the otherwise familiar concepts of sensitivity, specificity, likelihood ratio, pre-test probability, and predictive value. Additionally, a framework for incorporating uncertainty in pre-test probability demonstrates how (and under what conditions) unexpected results increase diagnostic uncertainty. A simple nomogram can help determine whether a physician, when faced with an unexpected test result, should place more trust in the test result versus their own clinical judgment about disease probability. Clinical examples are provided throughout to highlight practical applications.