In this issue of Academic Emergency Medicine, Viau et al.1 assess the yield of computed tomography (CT) of the head among patients presenting with syncope. The systematic review included a total of 17 studies (15 retrospective chart reviews and two prospective) that included adult emergency department (ED) patients presenting with the chief complaint of syncope or those admitted to the hospital for syncope. Approximately half of these ED patients underwent head CT with a 1.2 and 3.8% yield of serious intracranial conditions identified in admitted patients and ED patients, respectively. Serious intracranial conditions were selected based on clinical relevance, previous literature, and consensus among the coauthors.1 While this systematic review is not directly a diagnostic accuracy trial, it gives us an opportunity to discuss how to appraise the evidence pertaining to diagnostic tests within the context of clinical efficiencies in an era of value-based purchasing. Over the past 20 years there has been a shift in the understanding and approach to diagnostic testing in medicine. 1. Tests rarely produce yes and no answers. Better understanding of diagnostic test interpretation in the clinical setting has resulted in dismissing the notion of yes or no answers (i.e., rule in or rule out). Diagnostic tests, which include history, physical exam, labs, and imaging, typically change the probability of the disease or condition and the concept of test and treatment thresholds originated to assist in managing these posttest probabilities. If a test result increases the probability of the condition to a level beyond the treatment threshold, the probability is high enough to indicate that the benefits of treatment outweigh the harms. If the diagnostic test result reduces the probability of the disease below the test threshold (i.e., acceptable miss rate), the presence of the disease is unlikely enough that the clinician should consider other diagnoses to explain the patient's symptoms. In cases where the posttest probability of the condition falls between these thresholds, additional testing will typically be necessary. These thresholds vary based on factors such as the severity of the medical condition and the potential benefits and harms associated with treatment.2 For example, the treatment threshold for uncomplicated urinary tract infection in a healthy young woman3 would be quite different from the treatment threshold for acute pulmonary embolism.4 In fact, statistical models can produce test and treatment thresholds for most medical conditions based on the operating characteristics of the tests as well as benefits and harms of treatment.5 For example, the test (synovial fluid white blood cell analysis) and treatment thresholds for septic arthritis in adults have been reported to be approximately 5 and 39%, respectively.6 This means that if the clinical probability of septic arthritis is below 5%, further testing might not be helpful (or even harmful). On the other hand, if the probability of septic arthritis is above 39%, treatment should be started as further diagnostic testing is not likely to increase the yield and could potentially harm the patient.6 2. Using likelihood ratios to measure probability of the disease. Likelihood ratios are the tools that are used to manage probability of the disease in diagnostic testing.7 This initial probability of a disease is usually determined based on the prevalence of the disease or from previous studies and is called pretest probability. The pretest probability is determined before any medical history is obtained, any physical examination is performed, or any test is ordered. Each additional pertinent positive or negative information or test will change the pretest probability and produce a new probability called posttest or posterior probability. The change from one probability to another is possible because of likelihood ratios. The likelihood ratio for a positive test increases the probability of the disease and the likelihood ratio for a negative test decreases the probability of the disease. The extent by which the probabilities are changed depends on the numerical value of the likelihood ratios.3, 7 As a rule of thumb, positive likelihood ratios more than 10 significantly increase the disease probability and negative likelihood ratios less than 0.1 significantly reduce it.7 3. Focusing on what matters. Another shift in paradigm pertaining to diagnostic testing is measuring the value of a diagnostic test by its impact on patient-centered outcomes. Nowadays patients, clinicians, and payers require more than analytical or technical characteristics and accuracy from a diagnostic test. They would also expect the test to lead to health benefits.8 Highly accurate tests might not change the outcome of the patients. For example, N-terminal-pro brain natriuretic peptide (NT-proBNP) and brain natriuretic peptide (BNP) are very sensitive diagnostic tests in identifying acute decompensated congestive heart failure. However, the knowledge of these values rarely affect the patient-centered outcomes of patients with this condition.9 Coronary angiography, a highly accurate diagnostic test in diagnosing coronary artery stenosis, might not be beneficial or could even be harmful (number needed to harm of 50 for major hemorrhage, myocardial infarction, or stroke because of the procedure) in patients with stable coronary artery disease.10 This is probably the most important change in approach to diagnostic testing as clinicians are more mindful of the concept of overdiagnosis and the harms associated with it.11 Simple “routine” tests might pose significant inconvenience and harm to patients. For example, performing routine preadmission chest x-rays or routine chest x-rays in patients with chest pain could result in accidental findings such as benign nodules. These findings automatically put the patients on a path that requires multiple doctor visits and potentially significant radiation exposure (follow-up CT scans of chest). The psychological harms associated with these findings and the prolonged stress until the nodules are deemed benign should be emphasized more. 4. Diagnostic yield. In addition to accuracy, the value of a diagnostic test lies in the probability that it provides the information needed to establish a definitive diagnosis. The difference between diagnostic accuracy and diagnostic yield is that the latter varies in each clinical scenario.12 The diagnostic yield of CT scan of the head in patients with a focal neurologic deficit and for diagnosing stroke might be quite different from that of CT scan of the head for syncope for identifying a possible etiology. Diagnostic yield is generally higher when the probability of the disease is higher, and the test is used in a proper clinical context.12 This is impacted by spectrum bias: when patients are at the milder end of the disease spectrum, many tests are not definitive. Overuse of the test, whether because of a “just-in-case” strategy or defensive medical practice, significantly reduces the diagnostic yield. Routine ordering of coagulation profile (PT and PTT) in patients who are not on anticoagulants or have no evidence of liver failure is an example of a low-yield test.13 Blindly ordering tests with low diagnostic yield in any clinical scenario has implications for the patients, physicians, and also the laboratory. Overuse of tests and “overdiagnosing” may make the patients’ management more confusing and problematic. This practice may produce unexplained, abnormal results that generate additional testing that could be harmful to the patient. However, these harms are rarely measured as a quality indicator. In addition to overburdening the laboratories and radiology suites, overtesting certainly contributes to already inflated health care costs. 5. Do no harm. There is an increasing awareness that preventing medical harm must become one of the pillars of modern health care. Similar to treatment strategies, diagnostic tests are associated with harms.11 No test is perfect, so patients require nuanced decisions rather than groupthink conformity. Understanding of this concept has resulted in the proposal of a strategy called “deliberate clinical inertia.”14, 15 This refers to the art of doing nothing as a positive response. Adopting deliberate clinical inertia, i.e., the art of not intervening, as a specific measurable indicator, would be a novel patient-centered quality initiative that could counterbalance the current perspective of overtesting. This overtesting reflects our cultural intolerance of uncertainty that drives imaging requests. Developing skills in dealing with uncertainty, competing diagnostic approaches, harm–benefit trade-offs, societal pressures, and shared decision making is essential in developing deliberate clinical inertia. When these issues are explained to patients in a patient-friendly format, patients often choose the “less is more” option.14, 15 This approach is particularly important for heterogeneous conditions such as syncope. The systematic review by Viau et al.1 likely overreports the proportion having a serious diagnosis as the proportion of syncope cases undergoing head CT was only about half. It is very likely that patients with higher pretest probability of serious intracranial pathology were selected for head CT. This also means that doctors are already making patient-centered decisions without high-quality evidence, using information derived from experience and intuition (which is also a component of evidence-based medicine). Clinical training needs to go beyond robotic rule following to hone expert clinical judgment. Getting information (e.g., CT head) in the absence of judgment can be misleading, as we must have a pretest probability. 6. Using systematic reviews for assessing the accuracy of diagnostic tests. Systematic reviews with or without meta-analysis remain the highest level of evidence for assessing the diagnostic accuracy of diagnostic tests. However, the process is not as straightforward as interventions. Some of the challenges facing such systematic reviews are related to the method of reporting the results. It is not always feasible to distinguish the limitations in methodology of the original trials from limitations of reporting the results. Recently, most journals encourage investigators to follow specific guidelines for reporting their findings such as those recommended by STARD (Standards for Reporting of Diagnostic Accuracy) statement.16 Following these guidelines are a positive move toward standardized reporting of such trials, reducing the confusion when distinguishing methodologic flaws from reporting flaws. Unfortunately, such guidelines are frequently ignored in emergency medicine diagnostic trials.17 Another challenge of systematic reviews addressing diagnostic tests are difficulty in measuring publication bias.18 Currently, there is no registry for trials measuring the accuracy of diagnostic tests and it is not possible to predict which diagnostic test trial gets published and which ones do not. Commonly, there are many differences in the design and quality of diagnostic accuracy studies. These differences (also known as heterogeneity) could affect the interpretation of their results. Heterogeneity may reflect differences between the studies in their definition of a positive test, study design, patient characteristics, place of test in diagnostic pathway, etc.19 Excessive heterogeneity might result in systematic biases and affect the estimates of diagnostic performance. Therefore, a structured appraisal of methodologic quality of studies included in the systematic reviews and meta-analyses is a critical step. The quality assessment is performed with the objective of evaluating the effects of potential sources of bias on estimates of test accuracy. Additionally, quality assessment allows for evaluating the effect of hypothesized clinical sources of heterogeneity on estimates of test accuracy. Unfortunately, when it comes to diagnostic testing accuracy, the use of statistical tests of heterogeneity does not reliably indicate absence of heterogeneity. It is recommended that the authors of systematic reviews assume the presence of heterogeneity and develop models that can identify heterogeneity and account for it.19 While this systematic review fails to produce an exact diagnostic yield for CT scan of the head in patients with syncope due to certain limitations, it produces valuable information that could be used in clinical practice.1 The authors of the systematic review recommend estimating a pretest probability for serious intracranial pathologies for each patient with syncope before ordering the test. This pretest probability is estimated based on patients’ risk factors, possibility of other pathologies mimicking syncope (e.g., seizure), and also preferences. The authors suggest that when this assessment generates a low pretest probability, CT scan of the head should be deferred.1 This systematic review also provides the opportunity to broaden our views of diagnostic testing in light of many recent paradigm shifts. There are many elements beyond diagnostic accuracy that should be considered before a test is ordered or interpreted. When benefits of the test in question are not clear or the possible harms are considerable, adopting a deliberate clinical inertia strategy might be a reasonable alternative to diagnostic testing. The benefits of this strategy should be explained to the patients and their families to ensure respect for the patient's circumstances and preferences, and to incorporate them in the shared clinical decision-making process. In summary, diagnostic strategies require adopting a global view and diligent consideration of patient-centered factors. As an anonymous proverb indicates, “For most diagnoses all that is needed is an ounce of knowledge, an ounce of intelligence, and a pound of thoroughness.”