Clinical Decision-Making

How can the Results Apply to my Patient?

Participants in clinical trials represent a selected group of patients. In clinical practice, most patients do not fulfill the inclusion and exclusion criteria of the trials, and the question of whether the results of a trial can be applied to clinical practice is common. Biologic characteristics including gender, the presence of comorbidities, race, age, and underlying pathology of the disease may influence the efficacy and safety of most treatments. When applying the results of a given trial to an individual patient, clinicians should assess if these characteristics were or were not taken into account in the trials or if the patient is belonging to one specific subgroup participating—or not—in the trial. For example, in many malignancies, response to targeted therapies depends on the presence of specific mutations in oncogenes or cell surface receptors. In pulmonary embolism, the increase in the risk of major bleeding related to thrombolytic treatment is found in trials including patients aged >65 years, but not in the trials including younger patients (9).

DIAGNOSTIC ISSUES

General Principles of Diagnostic Testing

In some clinical circumstances, experienced clinicians recognize the disease without need for additional confirmation; for example, this is the case for many dermatologic diseases. Conversely, in many clinical situations, the diagnosis is more challenging and a probabilistic approach is needed. In this approach, clinicians generate a list of possible diagnoses, estimate the probability associated with each potential diagnosis, and then order diagnostic tests which rule in or rule out the different diagnoses. This is, for example, a common clinical scenario for chest pain, dyspnea, anemia, weight loss, or abdominal pain. Considering all possible causes to be equally likely results in unnecessary testing, because in most cases, each possible cause needs a specific test to be confirmed or ruled out. The clinician, instead, should consider a concise list of plausible target conditions; this process leaves the clinician with a working hypothesis considered to be the best possible explanation for the patient’s symptoms. The pretest probability of the disease should then be estimated. This can be done implicitly or by using explicit diagnostic clinical decision rules (CDRs) which have been generated and validated for some diseases; this is the case, for example, for deep vein thrombosis and pulmonary embolism (10,11). Once clinical probability is estimated, clinicians perform a diagnostic test allowing them to assess the posttest probability; this probability rarely equal to zero—which means that the diagnosis is absolutely ruled-out—or to one, which means that the diagnosis is absolutely certain. However, instead of targeting these two extreme values for posttest probability, the clinician should consider reaching a value termed “treatment threshold,” the probability above which treatment is recommended, or a value termed “test threshold,” the probability below which stopping the diagnostic process is safe. These two thresholds vary from one disease to another as a function of the severity of the disease, the side effects and efficacy of treatment, and the complications of further diagnostic tests. When the posttest probability lies between these two threshold values, further testing should be implemented.

What are the Results?

The purpose of a diagnostic test is to change the likelihood the patient has the target diagnosis. The likelihood ratio (LR) for a given test result allows one to assess the posttest probability as a function of the pretest (clinical) probability; the larger the difference, the better the diagnostic performance of the test. The nomogram described by Fagan helps to assess the posttest probability according to the pretest probability and the LR of the test results in a simple manner (12). The LR is the ratio of the true-positive rate and false-positive rate (LR for a positive test or positive LR) or the ratio between false-negative rate and true-negative rate (LR for a negative test or negative LR) (Table 6.1). As a general rule, LRs greater than 10 or less than 0.1 generate large and conclusive changes between pretest and posttest probabilities.

Are the Results Valid?

Diagnostic studies compare the experimental test to a reference standard and provide the classical two-by-two table showing true-positive, true-negative, false-positive, and false-negative tests results (Table 6.2). These studies should fulfill a series of criteria to provide unbiased results. First, the patients should represent the full spectrum of those with the clinical problem. Patients should include those with mild, moderate, and severe disease and should be recruited from all relevant clinical settings—emergency departments, hospital wards, and practitioner’s practice, if appropriate. Patients should be recruited consecutively, including nights and weekends if appropriate and, finally, the definition of the clinical problem must be clear. Second, the classification of patients among those with disease confirmed and those with disease excluded should be as accurate as possible. All patients should undergo the reference diagnostic examination, the diagnostic criteria should be described, the experimental test should not be included in the reference diagnostic strategy, and, finally, the experimental test and the reference standard should be interpreted independently.

In some circumstances, the diagnostic performance of the reference standard is not optimal and includes false-positive and false-negative results. In these cases, randomized comparisons between tests or diagnostic strategies, including different tests, can be performed using clinical endpoints related to the underdiagnosis or overdiagnosis of the suspected clinical condition (13).