The term differential diagnosis is generally thought of as both a noun and verb by clinicians. The noun form of differential diagnosis is the list of all possible conditions that could explain the collection of signs, symptoms, and test results observed in a particular patient at a particular point in time. This list of conditions is organized from most likely (high on the list) to least likely (low on the list). The verb form of differential diagnosis is the medical decision-making process whereby this list is continually updated by eliminating conditions that are considered to be ruled out and adding conditions that may not have been previously considered based on the acquisition of new information. Conditions that remain on the list are also moved up and down in priority based on a continual reanalysis of their likelihood. The goal of diagnostic investigation and problem solving is the elimination of all conditions from the differential diagnosis until a single unifying diagnosis remains.
The number of conditions contained in a differential diagnosis is referred to as its breadth. Generally, the smaller the number of signs, symptoms, and test results available for consideration, the broader the differential diagnosis. A clinician always has the least amount of information available at the time of the patient's initial presentation, and so it behooves him or her to “cast a broad net” by adding many conditions to the list even if they are only remote possibilities. An initial broad differential can be winnowed down later using additional information obtained during the course of further diagnostic investigation. The choice of which conditions to add occurs by pattern recognition, whereby clinicians recognize patterns of signs and symptoms present in disease states that they have seen before. The process of recognizing these patterns, identifying the condition, and adding it to the differential diagnosis is referred to as hypothesis generation.
Accurate pattern recognition and hypothesis generation are the foundation of accurate differential diagnosis, because if a condition does not make it onto the list of differential diagnoses, it can never be confirmed or refuted via further investigation. Research has shown that the source of the improved accuracy of expert diagnosticians is not better acquisition of the signs and symptoms that provide the data set for hypothesis generation, nor is it the number of hypotheses generated from a given data set, but instead, it is the generation of more accurate hypotheses compared with novice diagnosticians. The source of this improved accuracy has been the topic of much debate. However, a rule of thumb that is used frequently by clinicians to describe hypothesis generation is “common things are common.” This seemingly obvious adage means that a given sign or symptom is more likely to be an uncommon manifestation of a common disease than a common manifestation of an uncommon disease. In other words, one should focus on generating hypotheses that are epidemiologically most likely even if they do not seem to fit the pattern perfectly. Recall or availability bias is a type of cognitive error in this process wherein the clinician has a distorted sense of the prevalence of a particular condition based on his or her own personal experience rather than that reported in the scientific literature.
While the amount of data available to the diagnostician is the principal determinant of the breadth of an initial differential diagnosis, the particular characteristics and the quality of the data being considered can also have a profound effect. Certain findings are considered pathognomonic for particular diseases, meaning that the finding is so specific and sensitive that a patient should be considered to have the condition until proven otherwise. An example would be the presence of Kaiser-Fleischer rings in the eyes of patients with Wilson's disease. This single observation on the physical exam would eliminate nearly all other conditions from consideration. Similarly, the quality of the data also has dramatic effects on the breadth of the differential. Demented, mentally ill, or malingering patients may supply a wealth of historical details; however, the reliability of this information would remain suspect, and it might add little value despite its abundance. In these situations, the differential would remain broad despite obtaining a relatively large amount of data.
Once a broad differential has been established based on initial data gathering and hypothesis generation, the list is narrowed by either confirming (ruling in) a single diagnosis or eliminating (ruling out) conditions one by one until a single diagnosis remains. Usually, both approaches are used simultaneously. The order in which conditions in the differential are investigated depends on (a) the urgent or emergent nature of diagnoses on the list, (b) the logistical expediency of obtaining a definitive answer for a particular diagnosis, and (c) the particular cognitive preferences of the diagnostician. Diagnoses that threaten loss of life or function are always investigated first even if they are low on the differential. Dissecting thoracic aortic aneurysm is a relatively rare cause of chest pain and is often near the bottom of the differential. However, it is investigated rapidly, as the consequences of a delayed diagnosis would be devastating. Once all the life-threatening diagnoses have been eliminated from the differential, diagnostic investigation can proceed at a more leisurely pace. If a condition can be excluded simply and easily, it is often pursued next. These are the so-called low-hanging fruit of the diagnostic process, and an example would be excluding a diagnosis of anemia with a simple complete blood count. In general, ruling in is a quicker way to narrow the differential than ruling out because one need only be correct once in the former approach and one needs to be correct N − 1 times (N being the number of conditions in the differential) in the latter.
Once a diagnosis has been ruled in, the remainder of the diagnoses are assumed to be ruled out based on the principle of parsimony, or Ockham's razor. The principle is attributed to the 14th-century logician William of Ockham and states that “the explanation of any phenomenon should make as few assumptions as possible, eliminating those that make no difference in the observable predictions of the explanatory hypothesis.” Practically, this means that all of the observable signs, symptoms, and test results should be explained by a single diagnosis. If a single condition has crossed the threshold of evidence to be accepted as the unifying diagnosis, then all other diagnoses must be rejected. Even if a diagnosis has been confirmed, the particular cognitive preferences of a diagnostician will still factor into the ongoing investigation. Some diagnosticians may continue to rule out conditions as they prefer to “leave no stone unturned.”
All diagnosticians are subject to bias in the medical decision making involved in narrowing the differential diagnosis. Two common types of cognitive bias are confirmation bias and anchoring bias. Confirmation bias arises when a clinician only performs further testing in an effort to confirm a diagnosis that he or she already believes to be true and does not test other hypotheses that might refute the favored diagnosis. Anchoring bias is similar but distinct in that it results from a failure to add new diagnoses to the differential or adjust the position of old diagnoses based on new information. The clinician becomes anchored to the original differential and is blinded to new possibilities.
“No evidence of disease is not evidence of no disease” is a phrase often used to describe the fact that a clinician's inability to detect a condition at a particular point in time does not mean that it is not present currently or was not present in the recent past. This is especially true for conditions that have waxing-and-waning courses, such as occult gastrointestinal bleeding. Commonly, 80% of upper gastrointestinal bleeding has stopped by the time of presentation to medical attention. Nonbleeding ulcers or varices are often found on esophagogastroduodenoscopy and presumed to be the source, but in a significant number of cases, the source of the bleeding cannot be found because active bleeding is no longer visible at the time of the diagnostic investigation. Failure to find a source of bleeding despite thorough investigation does not mean that gastrointestinal bleeding has been ruled out as a cause of the patient's presenting signs and symptoms, and consequently, it cannot be eliminated from the differential diagnosis.
When a workup is entirely negative and the differential diagnosis still contains more than a single diagnosis, watchful waiting is sometimes employed as a passive diagnostic strategy if the patient's condition is stable. The hope is that the condition causing the presenting symptoms will reactivate and new observations can be made at that time, which will allow the differential diagnosis to be narrowed.
When a workup is negative but the differential is relatively small and/or the patient's condition is deteriorating, a strategy of diagnostic and therapeutic intervention can be employed to confirm a diagnosis. If the therapy is narrowly directed at a particular diagnosis and the patient responds to treatment, the individual diagnosis in question is considered to be ruled in, and further diagnostic workup is unnecessary. When this strategy is employed, it is important that a diagnostic response to treatment be defined clearly and prospectively and that only a single, narrowly directed therapy be used at any one time. If multiple therapies are employed simultaneously, a causal relationship between treatment and disease cannot reliably be inferred, and therefore, a diagnosis cannot be reliably confirmed based on response to treatment. This type of obfuscation of the differential diagnosis often occurs when broad-spectrum antibiotics are used to treat an infection of unclear etiology. The patient may have improvement in fever, white blood cell count, and bacteremia, but the signs and symptoms that would have helped localize the infection have not been allowed to develop.
The process of differential diagnosis is critical to medical decision making, because without an accurate diagnosis, decisions about treatment become extremely difficult. The medical decision making involved in differential diagnosis is complex and subject to the underlying cognitive biases of clinicians. Diagnostic testing is not without the potential to harm patients. Consequently, risk/ benefit decisions must be made to determine whether the additional diagnostic information provided by a test or procedure is warranted. Skilled differential diagnosticians balance these risks with their degree of confidence that the correct single unifying diagnosis has been selected from the list of possibilities generated during the process of differential diagnosis.
Diagnostic Process, Making a Diagnosis, Errors in Clinical Reasoning, Heuristics
Related Credo Articles
The process of differentiating between two or more psychiatric disorders that have similar or overlapping signs or symptoms. This practice...
noun the identification of one disease from a number of other similar diseases by comparing the range of symptoms of each ...
(ca. 1860) : the distinguishing of a disease or condition from others presenting with similar signs and symptoms