Despite this lack of evidence, she does not simply dismiss her initial diagnosis. Although she faces conflicting evidence, she maintains her clinical suspicion of myocardial injury caused by ACS. She orders the tests repeated in two hours and has the patient monitored closely for any signs of worsening condition. Because of persistent pain he receives additional morphine. Meanwhile the clinician considers alternative diagnoses which could mimic the symptoms of ACS.
Commentary: Why does the clinician maintain her initial suspicion? While the negative results are sufficient to make her hold off treatment based on the initial diagnosis, she wants to avoid prematurely turning away from the commonest disease, especially since not treating it could have potentially very negative effects for the health of the patient. The case does not fit the textbook picture of cardiac injury caused by ischemia. However, she knows both from her own experience and epidemiological studies that atypical disease presentations are not uncommon: the pain may be referred to the jaw instead of chest, the T-waves in ECG may not develop early, etc.
Due to this uncertainty, she initiates two lines of action. Firstly, although she does not have a single and simple hypothesis about what could cause the atypical presentation, she decides to repeat the tests. The fact that ACS is the most common cause of chest pain and the life-threatening character of the diagnosis makes it reasonable to repeat the tests. Secondly, her lowered confidence in the ACS diagnosis is reflected in the fact that she decides to start systematically generating alternative diagnoses in order to select a differential diagnosis.
Generation and selection of differential diagnosis: The clinician thinks of alternative diagnoses. She asks herself a number of questions to guide her thinking: “what else could explain acute chest pain?”, “which are the most common causes?”, “what would be the most serious disease to miss?” “what could I test effectively and quickly?”, “which conditions are effectively treatable?” She can think of a wide range of possibilities. For instance, she briefly considers Chagas disease, but decides this is too rare in the U.S.A. to merit immediate consideration. Another possibility would be referred pain e.g. from acute pancreatitis or another abdominal organ. In certain clinical circumstances this can be confirmed by re-examination after delaying an hour or more. In the present case, there is no time to delay, as the pain seems to be continuous even with morphine analgesia. She ultimately decides to focus on four alternative hypotheses in addition to acute coronary syndrome:
  1. Acute pulmonary embolism (a blockage of a lung artery, e.g. by a blood clot).
  2. Acute aortic syndromes, (different kinds of damage to the aorta, the main artery leading blood out of the heart and into the body).
  3. Pericarditis (inflammation to the pericardium, the sac surrounding the heart).
  4. Gastrointestinal reflux disease.
Commentary: At this stage, the clinician starts systematically generating possible explanations. She probably has already thought of alternative possibilities, but she now tries to explicitly elicit and search her memory and clinical experience using a series of interrogatives. The case presents a puzzling picture, so the clinician decides initially to cast her net widely. She asks herself which other diagnosis could explain the symptoms. Her first concern is to make sure she has not forgotten to think of a potentially dangerous condition. However, she needs to quickly limit herself to a short list of actionable diagnoses. Trying to actively consider and rule out all disease candidates is not practically possible, especially when the patient is unstable. Thus, the physician asks herself further questions to limit and focus her search, prioritizing conditions that (i) are common for this type of patient, (ii) would be dangerous if left untreated and (iii) allows of effective testing and treatment.
In this case the clinician chooses to generate hypotheses based on her own experience and background knowledge. Sometimes the generation of hypotheses happens almost automatically: the clinician recognizes a known pattern and immediately recalls the most common and important diagnoses. In more atypical cases, it can be necessary to employ a more structured or directed form of thinking. Other possibilities would have been to use one of the existing artificial intelligence programs designed to assist medical diagnosis (e.g. Isabel, DXplain) or request a colleague to review the data and offer a second opinion. All of these would classify as different strategies for hypothesis generation.
Current computer programs are based on the prevalence and weighting of signs and symptoms linked to diseases. They indicate which conditions are most the common causes of the symptoms entered into the program and red-flag potentially life-threatening diagnoses. The cardinal arrangement of the diagnoses is based on the experience of the writers of the program and the epidemiology of diseases commonly encountered in the indicated age group, sex and geographical region. Using a computer program can be helpful for reminding a clinician of rare but dangerous conditions. Many experienced physicians however consider them of limited usefulness.10
First, they tend to generate a fairly long list, which is not particularly helpful in an emergency situation. Trying to work through an extensive list of possibilities is not a feasible strategy especially when the patient is unstable. It may also subject the patient to unnecessary and potentially harmful over-testing. Second, physicians do not know how the program assigns weights to the different signs and symptoms and to the prevalence of the disease. The computer program is based on geographically common epidemiologic findings in specific diseases and populations, but this population level information cannot be translated directly to the individual case. Experienced physicians will be attuned to the concrete clinical setting (how stable is the patient, what are the urgent problems), details of the case (e.g. medical history, country of origin, foreign travel, use of drugs, smoking) and the patient’s response to therapy (e.g. pain relief and normalization of heart rate, breathing). These facts have to be interpreted and the physician has to judge whether or not the findings are properly perceived and integrated into the diagnostic picture. So, physicians will in any case need to draw on their experience and insight to interpret the results.
An important part of this is the ability to recognize the important signs and symptoms, i.e. to recognize that a certain fact, or combination of facts, is important information which needs to be taken into account. For instance, had a patient travelled to sub-Saharan Africa, this might suggest anemia as a result of malaria; sedentary life style would place coronary ischemic heart disease higher in the list of diagnoses; family history of diabetes in mother and grandmother would interpret weight gain, hypertension, and dyslipidemia into the spectrum of metabolic syndromes. Of course, neither a computer nor a human reasoner can take a given piece of evidence into account before it is recognized as evidence. As with hypothesis generation, trying to take into account every single piece of information is not feasible, nor necessarily very efficient. Perspicacious observation is here an intricate part of the reasoning process itself. As Peirce noted, “abductive inference shades into perceptual judgment without any sharp line of demarcation between them” (1932-1958, §5.181). These computer programs are, at best, an aid to, rather than a replacement for, clinical judgement and experience.
To have a colleague review the data, whose difference in experience, training and background knowledge may lead him to think of other possibilities, can also be a way to ensure that important diagnoses are not overlooked. At other times, it grows out of the physician being confused by the picture or even dissatisfaction with her own thought process. Although this strategy for generating hypotheses cannot guarantee to be as exhaustive as a computer program, drawing on the judgement of a colleague has the advantage of being better attuned to the concrete clinical situation, and so being more likely to generate suggestions that are plausible and useful in context.
The experience and training of the clinician, or her colleagues, play a crucial role in several respects in the selection of a differential diagnosis. First, the clinician has to make a wise choice of which strategy for generating new suggestions strikes the right balance between expediency, exhaustiveness and quality in the concrete situations. Although one can discuss general considerations of advantages and disadvantages of different strategies, as above, the choice ultimately has to rely on the judgement of the physician. Second, it is the training, experience and background knowledge of the clinician that allows her to recognize patterns and recall possible diagnoses. Finally, in choosing which diagnostic hypothesis to focus on the clinician needs to judge which diagnoses are most likely in the concrete case and decide how to weigh this, e.g., against the seriousness and urgency of the disease, its testability and its treatability.
Prioritize hypothesis for testing: Cardiac enzymes after two hours are borderline elevated, with c-Troponin is 98.5% of the normal range. The clinician ponders if perhaps the origin of the troponin elevation is from the epicardium or the pleura, rather than the heart muscle, reassessing the hypotheses of enzyme origin and cause of chest pain. She requests a cardiologist consultation. Meanwhile, she is at the bedside. She next considers pulmonary embolism and orders a chest-computed tomogram (CT-scan) with contrast media to search for the embolism (the blockage).
Commentary: The clinician now decides to request a second opinion from a specialist colleague. Meanwhile she prioritizes the pulmonary embolism hypothesis for testing. There are several good reasons for this. First, she currently considers pulmonary embolism one of the most likely diagnoses. Second, pulmonary embolism and acute aortic syndromes are both emergency conditions and would require immediate treatment. Pericarditis is also a very serious condition but less urgent, whereas gastrointestinal reflux disease is not an immediate threat. Third, a CT-scan is a highly reliable way to detect embolism. Fourth, the chest CT-scan might also show a widened mediastinum (the area containing the heart and the pericardium), a possible sign of pericarditis. It would also show the thoracic aorta (the part of aorta situated in the chest-region), a possible site of any aortic syndrome. This last point is an example of how a test can have other epistemic consequences in addition to testing hypotheses directly, in this case by potentially providing clues for future hypothesis generation. In sum, the CT-scan would be a reliable test of the most likely and serious condition while potentially also providing information relevant for the two other most serious conditions.
Further puzzling results: The results of the CT-scan adjusted to an early phase of contrast injection are reported as negative for pulmonary embolism, but the ascending aorta is reported to be prominent, measuring 4.3 cm in diameter (normal ascendingaorta is 3.63 to 3.91 cm ). The patient still complains of chest pain but feels relieved by increased dose of morphine, and breathing is improved somewhat by continuous 100% oxygen therapy.