Importance of epidemiology and biostatistics in deciding clinical strategies for using diagnostic tests: a simplified approach using examples from coronary artery disease
RE Patterson
and
SF Horowitz
Department of Medicine (Cardiology), Emory University School of Medicine, Carlyle Fraser Heart Center, Crawford Long Hospital, Emory University, Atlanta, Georgia 30365.
The explosion of costly new medical diagnostic technologies demands a common sense approach to help physicians decide appropriate indications and strategies for use of these tests. This simple, nonmathematical review focuses on the assessment of coronary artery disease, but the approach can be generalized to other medical problems. This clinical approach to diagnostic testing strategies is based on seven sequential questions: 1. What is the clinical probability that this patient has a specific disease characteristic based on clinical data? 2. What is the overall objective for management of this patient based on the overall status of the patient? 3. Most importantly, what specific questions need to be answered about the patient's condition before the physician can recommend the most appropriate management (e.g., whether the patient has coronary disease, whether an anatomic lesion is functionally significant, whether a myocardial region is reversibly ischemic or irreversibly infarcted, whether a particular therapy has had good or bad effects or what is the patient's prognosis)? The key point is for the physician to formulate a specific clinical question about the patient before the test. 4. The physician must then ask how well does the test answer the particular clinical question about the patient. Here the physician needs to understand the sensitivity and specificity of the test, especially because they are influenced by various clinical biases. 5. Next, the physician must ask how to interpret the reliability of a positive or negative test result in the individual patient. This requires understanding predictive value and predictive error of a given result and how they are influenced by the clinical data as described by Bayes' theorem. 6. Next, the physician must ask what further tests or therapies will be recommended for the patient. The physician can estimate in advance how different test results would alter management plans and he can then allow this estimate to help determine indications for the test. There is some controversy concerning whether to use Bayes' theorem or multivariate analysis to estimate the final probability of a disease characteristic. 7. Finally, in this era of quality assurance, professional review and cost containment, it behooves each physician to ask whether the data provided by the particular tests were worth the cost, inconvenience and risk for that particular patient.
