bs_bs_banner
Journal of Evaluation in Clinical Practice ISSN 1365-2753
Managing uncertainty in diagnostic practice Ashley Graham Kennedy PhD1 1
Assistant Professor of Philosophy, Honors College, Assistant Professor of Clinical Biomedical Science (Secondary), College of Medicine, Florida Atlantic University, Jupiter, FL, USA
Keywords clinical guidelines, diagnosis, epistemology Correspondence Dr Ashley Graham Kennedy Honors College/College of Medicine Florida Atlantic University 5353 Parkside Drive Jupiter, FL 33458 USA E-mail: [email protected]
Abstract In spite of the large medical literature devoted to a discussion of the topic, uncertainty, particularly in diagnostic practice, is sometimes mishandled by clinicians. This, I will argue, can in turn lead to medical and ethical problems in patient care. Via the analysis of two recent case studies, I will propose an explanation for why this kind of uncertainty is sometimes not appropriately dealt with during the process of differential diagnosis in the clinic. Furthermore, I will also give recommendations for improving diagnostic practice by highlighting some representative elements of diagnostic uncertainty and reminding clinicians to be aware of the tendency to overlook this uncertainty in routine clinical cases.
Accepted for publication: 21 December 2014 doi:10.1111/jep.12328
Introduction
Case 1
Much has been written over the years on the concept of medical uncertainty [1–4] and it is now widely accepted (if it was ever in question) by both philosophers and clinicians that uncertainty is pervasive in medical testing methods, clinical diagnostic reasoning, treatment protocols and prognostic evaluations. Furthermore, there is also an extensive and informative medical literature on the topic of bias and uncertainty and how to manage it in clinical practice [5]. However, in spite of this, medical uncertainty, particularly in routine diagnostic practice, is sometimes missed or mishandled by clinicians. In what follows, I will attempt both to explain why this is the case and to bring to light some of the problems that this mishandling of medical uncertainty causes for clinicians and patients. Furthermore, I will also gesture towards a solution to this problem, by suggesting ways in which we might better train physicians to recognize and handle diagnostic uncertainty, particularly in ordinary cases. My analysis will be conducted via the examination of two representative, routine cases, both of which I recently observed. The first case shows how a mishandling of diagnostic uncertainty has the potential to lead to premature closure of a medical case, and thus to potentially cause both medical and ethical harm. The second case serves to highlight the dangers of dismissing diagnostic uncertainty, particularly in routine cases. In cases such as these two, both diagnostic accuracy and doctor–patient interaction could be improved by careful consideration and communication of the medical uncertainty in the process of differential diagnosis.
To begin our analysis, consider the following case study: In May 2013, a 67-year-old man was sent by ambulance from his assisted living home to the emergency department of a large research hospital with the chief complaint of diarrhoea and abdominal pain. Abdominal pain is notoriously non-specific and can indicate any number of conditions of varying degrees of severity; thus, it requires a careful work-up when it presents in the emergency setting. This particular case was rather complicated as the man had a history of prostate cancer. Because of this, he had developed incontinence, which required self-catheterization several times per day. Furthermore, the self-catheterization frequently in turn led to urinary tract infection (UTI). More complicated still, this man had also recently developed a Clostridium difficile infection secondary to antibiotic treatment for a UTI because of self-catheterization. It was unclear from the patient’s history whether or not the treatment for this condition was successful as he reported that he continued to have chronic diarrhoea and stomach pain (which could have been indicative of a persistent C. difficile infection) even after finishing the prescribed course of antibiotic treatment. Furthermore, it was not clear when the diarrhoea had initially started, as the patient’s hospital record indicated that it had been ongoing for several years. Because of this, the man’s symptoms could not be definitely attributed to C. difficile and yet that diagnosis could not be absolutely ruled out either. After the man’s history was taken, the attending physician ordered a urinalysis and a computed tomography (CT) of the abdomen to
Journal of Evaluation in Clinical Practice (2015) © 2015 John Wiley & Sons, Ltd.
1
Managing uncertainty in diagnostic practice
be performed. The urinalysis showed seven white blood cells (normally not present in healthy urine), indicating that a UTI was a possibility, and the CT showed several metastatic masses, indicating that cancer could also be the cause of the man’s symptoms. Hospital records showed that the same masses were visible on a CT scan that had been taken on a previous hospital admission 3 months earlier, indicating that there had not been significant recent growth of the masses. Given these historical- and test-based findings, the man was given a differential diagnosis of UTI versus infectious diarrhoea versus cancer versus gastritis. The man did not have fever or diarrhoea while in the hospital, so the diagnosis of infectious diarrhoea (caused by C. difficile or a virus or some other organism) was difficult to rule in or out, as a stool analysis could not be performed and no clear signs of infection were present.
Case analysis This case is representative of many others in the following respect: none of the presenting evidence (including the history and the laboratory and other findings) was such that it allowed for definitively ruling out any of the diagnoses in the initial differential list. In other words, the diagnosis in this case was epistemically uncertain: it was not clear whether the man’s symptoms were due to infectious diarrhoea, UTI, cancer or something else. This case also makes a good study because the diagnostic uncertainty it displays is relatively easy to spot. It might seem surprising then that this diagnostic uncertainty was not recognized and dealt with appropriately by the team that saw the man 3 months prior to the admission that I observed and have described earlier. At that time, as hospital record indicated, the man reported the same symptoms of diarrhoea and abdominal pain that he again reported in May at his second admission. An abdominal CT and a stool culture were performed during his first admission and at that time the stool culture was positive for C. difficile. The man was informed of this result and was given a prescription for oral antibiotic treatment for this infectious condition. The hospital record indicated that the official diagnosis for the man’s complaints was ‘infectious diarrhea due to C. difficile’. Significantly, neither the man nor his gastroenterologist was informed of the results of the CT scan. On the man’s subsequent visit to the emergency department 3 months later, the new attending physician immediately noticed the uncertainty of the case and instead of diagnosing the man with recurrent or undertreated C. difficile, reopened the diagnostic differential and re-examined all of the available evidence. Then, he spent a long time deliberating over what to do, for in medical practice, some sort of action – even if it is merely a communicating between clinician and patient – must always take place even when the epistemic situation is uncertain. In this case, the physician had to decide whether to admit the man to the hospital or to send him back to the assisted living home, and what, if any, treatment the man should be given while in the emergency department. After deliberation, the physician decided not to treat the man for a suspected UTI given the history of C. difficile that had occurred subsequent to previous antibiotic treatment for that condition. Instead, the physician sent the urine out for culture. Thus, given that the immediate risk to the man from any of the diseases in the differential was low, the attending physician decided that he could safely suspend judgment on the case while conducting further 2
A.G. Kennedy
investigation and sent the man home with instructions to follow-up with his gastroenterologist the next week. Finally, the physician also indicated on the man’s discharge instructions that he should be treated for a UTI if bacteria were cultured from the urine sample. I would like to highlight here that suspending judgment, as this physician did, on a diagnostically uncertain case is often the best course of action for several reasons. First, making diagnoses on the basis of insufficient evidence, rather than suspending judgment, is likely to lead to missed or misdiagnosis [6], whereas suspending judgment on a case can allow, as it did in this one, for continued investigation into the cause of the illness in question. In other words, suspending judgment does not have to result in paralysis or inaction. Instead, there is the possibility for action even when judgment is suspended and one form that this can take, aside from that of promoting further investigation, is that of accepting a working diagnosis. Accepting a working diagnosis is an ‘evaluative response’ that allows the clinician to act without requiring belief in any of the initial diagnostic hypotheses [7]. This means, for example, that when a clinician accepts a diagnosis as a basis for action, he or she does not need to believe that the diagnosis is correct. He or she can, instead, simply use the diagnosis on a trial basis. One way that this is often performed in practice is with the use of an ‘empirical trial’. An empirical trial involves giving the patient a treatment for one of the conditions in the differential to see how the patient responds. If the patient responds to the treatment, this is evidence that the patient has the condition in question. Thus, ‘observing a patient’s response (or lack of response) to empiric therapy can provide valuable diagnostic information’ [8]. This means that, at least in some cases, a clinician can arrive at a definitive diagnosis by accepting a tentative one. In fact, many diagnoses are, at least initially, ‘tentative’. There are very few cases in which it is appropriate to call a diagnosis ‘certain’. This is because a diagnosis is a hypothesis [9], the evidence for which can vary greatly in degree. Thus, the cognitive attitude that a clinician should take towards many (and possibly even most) diagnoses is one of acceptance for the purposes of further action, rather than of belief1 that the diagnosis is correct, definitive or true. Second, suspending judgment on a case both requires and allows for an honest communication of the diagnostic uncertainty to the patient and thereby promotes a respectful consideration of the patient as an active participant in the health care team. There are several reasons why this kind of epistemic honesty is beneficial in clinical practice. First, it encourages clinicians to take the patient’s reported symptoms seriously and to refrain from being dismissive in difficult-to-diagnose cases. Second, it also encourages physicians to keep the diagnostic differential open to revision and to thereby avoid the dangers of the overconfidence bias. Berner and Graber [11] (p. S2), who have written on this bias, argue that, ‘physicians in general underappreciate that their diagnoses are wrong’. One of the dangers inherent with an overconfi1
Some, such as Cohen [10], have argued that belief is involuntary. If this is the case, then a physician could not choose whether or not to believe that a specific diagnosis was true or correct. However, he or she could still choose to keep the diagnosis open for re-evaluation. This is what I am suggesting that should be performed in cases in which the evidence is indeterminate. That is, even if a physician involuntarily forms a belief about the diagnosis, he or she can still keep the diagnosis open to investigation, revision or further evaluation.
© 2015 John Wiley & Sons, Ltd.
A.G. Kennedy
dence bias is that it can lead to a breakdown in clinical reasoning. Berner and Graber wrote that an overconfident physician might have problems gathering data, such as failing to elicit complete and accurate information from the patient, failure to recognize the significance of data, such as misinterpreting test results, or most commonly, failure to synthesize or ‘put it all together.’ (p. S8). Another danger with an overconfidence bias is that it tends to lead to definitive action in the face of uncertainty. In fact, the greater the epistemic uncertainty of the diagnostic situation, the more likely clinicians are to exercise overconfidence in the clinic [5,12]. Thus, what comes naturally to many clinicians when they are faced with uncertainty is exactly the opposite of what is called for by best diagnostic practice. Instead of definitive action that closes off further investigation, best diagnostic practice, in many cases, calls for the clinician to either adopt a working diagnosis or leave the investigation open while further evidence is being gathered. Interestingly, Jay Katz, a physician at Yale (see Groopman [5], p. 152–153), has argued that clinicians are willing to admit that medicine is an uncertain practice in theoretical discussion, but that they do not acknowledge the uncertainty inherent in what they do in actual practice. I have found this to be true in my own experience. In discussions with clinicians about medical uncertainty and how to handle it, many tend to think it is ‘obvious’ that (1) medicine is uncertain and (2) that this uncertainty should be acknowledged and handled appropriately. However, there is a tendency among these same clinicians to deny that there is uncertainty in any of their diagnostic cases, including the ones examined in this paper. I think this shows a need for explicit training in the recognition and management of uncertainty during medical school, residency and beyond. In this particular case that we have examined, once the physician had decided to suspend judgment until more information could be gathered, he then faced the difficult question of how to communicate his findings and recommendations to his patient. It was clear from the initial conversation between the two, which I witnessed, that the man was hoping for a solution to, or at least a definitive explanation for, his complaints. The patient had expressed frustration at his ongoing symptoms and multiple emergency room (ER) visits and clearly was looking for an answer as to the cause of his continuing suffering. The most difficult part of the attending physician’s management of this case was that he could not provide what the man wanted: he did not have a definitive answer to give the man concerning the cause of his long-standing symptoms. Of course the physician could have chosen to act as if he did have a definitive answer, but instead, he consciously decided (after much deliberation!) to be honest in his conversation with the patient and to tell the man that he simply did not know what was causing his continuing abdominal pain and diarrhoea. After his initial shock at hearing the words, ‘I don’t know’, the patient quickly expressed his appreciation of the physician’s honesty and was eager to hear his recommendations and plan of action. One could certainly imagine, however, a case in which a patient might not appreciate honesty concerning the uncertainty of his or her diagnosis. Patients are individuals and because of this, it should be acknowledged that uncertainty can be very uncomfortable for some and that therefore it should be communicated thoughtfully and carefully, and, at times, delicately. But it should
© 2015 John Wiley & Sons, Ltd.
Managing uncertainty in diagnostic practice
always be communicated and never hidden. When communicating diagnostic uncertainty to a patient, a clinician should at all times be aware of how her patient is receiving the news. Some patients, for example, will want detailed information and others will not. The key to effective communication of uncertainty in diagnosis is to carefully consider the needs of the individual patient and to always couple the news of the uncertainty with an assurance that the clinician will continue to investigate the case until a more definitive diagnosis can be made. The man’s prior visit to the ER had proceeded quite differently. Even though the presenting evidence at that time also rendered the case diagnostically uncertain (he was found to have both a C. difficile infection and to have metastatic masses in the abdomen), the team that initially managed the patient’s case settled on a definitive diagnosis of infectious diarrhoea as the cause of the man’s symptoms, without communicating the uncertainty of the situation to the patient. What this meant is that the patient’s metastatic masses were completely ignored and neither he nor his regular physician was notified of this finding. The reasons for the mishandling of the diagnostic uncertainty in this case are not clear. It is entirely possible that the clinicians were over worked, over tired or just too busy on that given day, leading them to favour quick diagnostic decisions over careful consideration of the full differential list.2 Whatever the reason for the mishandling, the failure to appropriately consider the diagnostic uncertainty in this case was serious, even bordering on malpractice, and put both the patient (for health reasons) and the hospital (for legal reasons) at risk of significant harm. Because the diagnosis of C. difficile was definitively recorded in the patient’s chart, no further investigation into the cause of his abdominal pain and diarrhoea was conducted by the physicians that the man saw in the interim between his two hospital visits. In other words, the case was prematurely closed, as a consequence of the lack of recognition and acknowledgment of the uncertainty in the initial diagnosis. Ultimately, this cost the man many months of suffering as well as multiple ER and clinic visits. It might have even shortened his life, as what was missed in this case was cancer. There are many lessons that can be learned from an examination of this case. The first is that settling too quickly on a single diagnosis is not in the best medical interest of the patient even when the patient (and/or the clinician) is clearly hoping for a definitive answer to or solution for the presenting symptoms. Instead, a willingness to suspend judgment in the face of an uncertain diagnostic situation, and thus to keep open the scientific investigation into the cause of a patient’s complaints, is preferable to too quickly settling on a single diagnosis in the face of underdetermining evidence. Diagnostic uncertainty is never comfortable for patients or clinicians and this is one reason that it sometimes mishandled or brushed aside. There are other reasons as well. During the man’s first ER admission, the attending physician might have reasoned that, given the man’s age, the metastatic masses found on the 2 Another possibility is that the clinicians were not able to correctly interpret the probabilities associated with the likelihood of each of the conditions in the differential. A discussion of probabilities and their interpretations is relevant and important for clinical practice. However, it is beyond the scope of this paper and I therefore will leave it to a subsequent one.
3
Managing uncertainty in diagnostic practice
abdominal CT were unlikely to be of immediate harm and that informing the man of their presence would not be of any therapeutic benefit. Thus, the uncertainty in this case might have been ignored because the other conditions in the diagnostic differential were not considered to be of appreciable risk to the patient. There is a pervasive presumption among clinicians that if the highest, or most immediate, risk diagnosis in the differential is treated, then the other possible diagnoses can be eliminated or ignored [13]. There are good, practical, reasons for treating high risk and/or immediate dangers first. Certainly, treating the man’s C. difficile infection, which was an immediate risk, was appropriate. But the error in this case was that the uncertainty in the diagnosis was not acknowledged, nor was it communicated, and thus the cancer that was found on the man’s CT scan was never investigated. Instead, only the C. difficile diagnosis was communicated to the man and recorded in the hospital record.
Case 2 In June 2013, a 29-year-old man was admitted to the emergency department of another major hospital with the chief complaint of chest pain and shortness of breath. He denied any history of allergies or asthma. Upon examination, he was found to have a productive cough, which produced green sputum, and significant difficulty with breathing. The family members that had driven him to the hospital made a specific request that the man be given antibiotic therapy for his condition. However, because chest imaging was negative for pneumonia, the attending physician refused this request and instead made a diagnosis of viral upper respiratory infection.3 He then treated the patient with albuterol. Fifteen minutes after the albuterol treatment, the patient reported easier breathing but continued to look quite ill. The physician then explained to the man that he had a bad cold, that he definitely did not need antibiotics and that his condition would improve on its own.
Case analysis I would like to suggest that this case was handled inappropriately because the diagnostic uncertainty of the situation was completely ignored. The man displayed symptoms (difficulty breathing and productive cough with green sputum) that are consistent with both viral and bacterial respiratory tract infection. What this means is that the diagnosis of a viral infection in this case, and in the many like it, is inherently uncertain because the evidence is underdetermining. There is no way to simply look at a patient and determine by physical examination alone, whether or not he 3 Many clinicians are reticent to prescribe antibiotic therapy without confirmatory signs such as bacterial culture or X-ray showing infection because of the growing problem of antibiotic resistance. This problem is indeed urgent, multifaceted and widespread: the Center for Disease Control (CDC) estimates that in the United States alone, an estimated 23 000 deaths each year are associated with drug-resistant bacterial infections. Clinicians have an important role to play in a potential solution to this problem, by prescribing antibiotics to patients only when needed and not simply whenever they are requested. But the question of when they are needed is not always easy to answer, and in attempting to answer it, clinicians need to take care not to be dismissive of their patients in the process, which is what happened in this case, and what, unfortunately, happens in many cases like it.
4
A.G. Kennedy
or she has a viral or bacterial respiratory tract infection. It is possible, however, to hone in on the probability of bacterial infection by taking a careful patient history. It is well known that in the general population viral upper respiratory infections are more common than bacterial ones. However, this statistic does not apply to all individual cases. In patients with a history of recurrent bacterial respiratory infection, that condition is more common than the former. However, the patient was not informed of this diagnostic uncertainty, nor was he ever questioned about a history of bacterial respiratory infections. Instead, he was told that his infection was certainly viral. That made his diagnosis both epistemically and ethically problematic. The diagnosis was epistemically problematic because the evidence that was gathered was insufficient to support it. Furthermore, it was ethically problematic because it led the physician to be dismissive of the patient, by not communicating the epistemic situation to him. As we have seen, given the presenting evidence in this case, it was not possible to make a definitive determination between the two possible diagnoses (bacterial vs. viral respiratory infection). However, the attending physician never even entertained the hypothesis that the patient’s infection was bacterial, even though the evidence was consistent with this. If he had entertained this possibility, or perhaps had used it as a working diagnosis, the physician might have cultured the patient’s sputum or considered an empirical trial of antibiotic therapy, thereby keeping the investigation into the cause of the patient’s symptoms open. Even if the diagnosis of viral infection turned out to be correct, making this diagnosis, without either considering, or informing the patient of, the possibility that the illness was bacterial instead of viral was harmful to the patient for two reasons. First, it was harmful because it prematurely closed the investigation into the cause of the patient’s illness, possibly resulting in misdiagnosis. Second, it was harmful because it was dismissive of the patient. Telling the patient that he had a cold and did not need antibiotics might have been true. But it might have been false. In this case, the risk of long-term physical harm to the patient because of misdiagnosis was low because he was young and in relatively good health. If he did have a bacterial infection, and not a viral one, this infection could likely be safely treated at a later time, if symptoms persisted. While delaying the proper treatment of the infection would have caused some extra suffering (and perhaps missed days of work) for the patient, it is unlikely that it would have posed a serious or long-term risk to his health. But this is not always the case: in some situations, misdiagnosis can be serious or even fatal. Thus, even in cases in which the risk to the patient is perceived by the clinician to be low or not immediate, diagnostic uncertainty should not be ignored.
Teaching clinicians to recognize diagnostic uncertainty I have presented two examples of what many clinicians would consider ‘ordinary’ diagnostic cases. I picked these cases precisely because they do exhibit key elements of the ordinary – in both the presenting symptoms and in the inherent uncertainty in the differential diagnosis, and in the second case, in the way that it was handled – and are therefore representative of the kinds of diagnostic cases that many clinicians routinely encounter in practice. But my hope is that the way in which these ordinary, routine, diagnos-
© 2015 John Wiley & Sons, Ltd.
A.G. Kennedy
tic cases are handled in the clinical setting will begin to change. An effective management of diagnostic uncertainty begins with an ability to recognize and acknowledge it in the clinical setting. And, as Katz’s work has shown, clinicians are not always able, or willing, to do this in practice, even though they readily do so in theoretical discussion. Furthermore, even when diagnostic uncertainty is recognized, this alone is not enough for effective diagnostic practice: it must also be communicated. Unfortunately, in many cases, diagnostic uncertainty is not communicated and patient health outcomes suffer for it. So what can be done about this problem of recognizing and communicating uncertainty so that diagnostic accuracy and patient health outcomes will improve? The answer to this question is that clinicians must be taught to acknowledge (rather than dismiss) and to communicate to their patients, any uncertainty in differential diagnosis, especially in those cases that are perceived to be ‘ordinary’ or routine. If this is performed in routine cases, it will become automatic in even the most complex situations. In the second case study that I presented earlier, it is likely that the epistemic uncertainty was missed for one of two reasons. It might have been missed because it was eclipsed by a concern for antibiotic stewardship. In other words, the physician in that case might have thought that the risk (to the public) of prescribing an antibiotic to the coughing patient was higher than withholding it and so rushed to a determination of viral illness so that an antibiotic would not be prescribed. Certainly, as I have already noted, the problem of antibiotic resistance is real and worsening. However, the existence of this problem does not eliminate the epistemic uncertainty in such diagnostic cases, nor does it legitimize the negative ethical implications of ignoring it. Clinicians should be reminded of this. Even when immediate or grave physical risk to the patient will be low if some of the diagnoses in the differential4 are ignored, an honest evaluation of, and communication to, the patient of the epistemic situation should still always take place. If this is put into practice in low-risk cases, it will become automatic
Managing uncertainty in diagnostic practice
in high-risk cases, and in all instances it will serve only to enhance diagnostic accuracy.
References 1. Fox, R. C. (1957) Training for uncertainty. In The Student-Physician: Introductory Studies in the Sociology of Medical Education (eds R. K. Merton, G. Reader & P. L. Kendall), pp. 207–241. Cambridge: Harvard University Press. 2. Atkinson, P. (1984) Training for certainty. Social Science and Medicine, 19 (9), 949–956. 3. Upshur, R. E. (2000) Seven characteristics of medical evidence. Journal of Evaluation in Clinical Practice, 6 (2), 93–97. 4. Djulbegovic, B., Hozo, I. & Greenland, S. (2011) Uncertainty in clinical medicine. In Philosophy of Medicine (ed. F. Gifford), pp. 299–356. Oxford: North Holland. 5. Groopman, J. (2008) How Doctors Think. New York: Houghton Mifflin. 6. Kennedy, A. G. (2013) Differential diagnosis and the suspension of judgment. The Journal of Medicine and Philosophy, 38 (5), 487–500. 7. Elliott, K. C. & Willmes, D. (2013) Cognitive attitudes and values in science. Philosophy of Science, 80 (5), 807–817. 8. Brett, A. S. & Powell, C. (2011) Approach to laboratory testing and imaging in aging. In Case-Based Geriatrics: A Global Approach (eds V. A. Hirth, D. Wieland & M. Dever-Bumba), pp. 14–21. New York: McGraw-Hill. 9. Simon, J. R. (2011) How to make real, constructive, progress in medicine. Journal of Evaluation in Clinical Practice, 17 (5), 847–851. 10. Cohen, J. (1992) An Essay on Belief and Acceptance. New York: Oxford University Press. 11. Berner, E. S. & Graber, M. L. (2008) Overconfidence as a cause of diagnostic error in medicine. The American Journal of Medicine, 121 (5 Suppl.), S2–S23. 12. Croskerry, P. & Norman, G. (2008) Overconfidence in clinical decision making. The American Journal of Medicine, 121 (5 Suppl.), S24–S29. 13. Than, M. P. & Flaws, D. F. (2009) Communicating diagnostic uncertainties to patients: the problems of explaining unclear diagnoses and risk. Evidence-based Medicine, 14 (3), 66–67.
4
Of course it is never the case that all possible causes will be put into the differential list. The clinician must decide which possibilities to rule in or out by considering risk, probability and treatability.
© 2015 John Wiley & Sons, Ltd.
5
Journal of Evaluation in Clinical Practice ISSN 1365-2753
Managing uncertainty in diagnostic practice Ashley Graham Kennedy PhD1 1
Assistant Professor of Philosophy, Honors College, Assistant Professor of Clinical Biomedical Science (Secondary), College of Medicine, Florida Atlantic University, Jupiter, FL, USA
Keywords clinical guidelines, diagnosis, epistemology Correspondence Dr Ashley Graham Kennedy Honors College/College of Medicine Florida Atlantic University 5353 Parkside Drive Jupiter, FL 33458 USA E-mail: [email protected]
Abstract In spite of the large medical literature devoted to a discussion of the topic, uncertainty, particularly in diagnostic practice, is sometimes mishandled by clinicians. This, I will argue, can in turn lead to medical and ethical problems in patient care. Via the analysis of two recent case studies, I will propose an explanation for why this kind of uncertainty is sometimes not appropriately dealt with during the process of differential diagnosis in the clinic. Furthermore, I will also give recommendations for improving diagnostic practice by highlighting some representative elements of diagnostic uncertainty and reminding clinicians to be aware of the tendency to overlook this uncertainty in routine clinical cases.
Accepted for publication: 21 December 2014 doi:10.1111/jep.12328
Introduction
Case 1
Much has been written over the years on the concept of medical uncertainty [1–4] and it is now widely accepted (if it was ever in question) by both philosophers and clinicians that uncertainty is pervasive in medical testing methods, clinical diagnostic reasoning, treatment protocols and prognostic evaluations. Furthermore, there is also an extensive and informative medical literature on the topic of bias and uncertainty and how to manage it in clinical practice [5]. However, in spite of this, medical uncertainty, particularly in routine diagnostic practice, is sometimes missed or mishandled by clinicians. In what follows, I will attempt both to explain why this is the case and to bring to light some of the problems that this mishandling of medical uncertainty causes for clinicians and patients. Furthermore, I will also gesture towards a solution to this problem, by suggesting ways in which we might better train physicians to recognize and handle diagnostic uncertainty, particularly in ordinary cases. My analysis will be conducted via the examination of two representative, routine cases, both of which I recently observed. The first case shows how a mishandling of diagnostic uncertainty has the potential to lead to premature closure of a medical case, and thus to potentially cause both medical and ethical harm. The second case serves to highlight the dangers of dismissing diagnostic uncertainty, particularly in routine cases. In cases such as these two, both diagnostic accuracy and doctor–patient interaction could be improved by careful consideration and communication of the medical uncertainty in the process of differential diagnosis.
To begin our analysis, consider the following case study: In May 2013, a 67-year-old man was sent by ambulance from his assisted living home to the emergency department of a large research hospital with the chief complaint of diarrhoea and abdominal pain. Abdominal pain is notoriously non-specific and can indicate any number of conditions of varying degrees of severity; thus, it requires a careful work-up when it presents in the emergency setting. This particular case was rather complicated as the man had a history of prostate cancer. Because of this, he had developed incontinence, which required self-catheterization several times per day. Furthermore, the self-catheterization frequently in turn led to urinary tract infection (UTI). More complicated still, this man had also recently developed a Clostridium difficile infection secondary to antibiotic treatment for a UTI because of self-catheterization. It was unclear from the patient’s history whether or not the treatment for this condition was successful as he reported that he continued to have chronic diarrhoea and stomach pain (which could have been indicative of a persistent C. difficile infection) even after finishing the prescribed course of antibiotic treatment. Furthermore, it was not clear when the diarrhoea had initially started, as the patient’s hospital record indicated that it had been ongoing for several years. Because of this, the man’s symptoms could not be definitely attributed to C. difficile and yet that diagnosis could not be absolutely ruled out either. After the man’s history was taken, the attending physician ordered a urinalysis and a computed tomography (CT) of the abdomen to
Journal of Evaluation in Clinical Practice (2015) © 2015 John Wiley & Sons, Ltd.
1
Managing uncertainty in diagnostic practice
be performed. The urinalysis showed seven white blood cells (normally not present in healthy urine), indicating that a UTI was a possibility, and the CT showed several metastatic masses, indicating that cancer could also be the cause of the man’s symptoms. Hospital records showed that the same masses were visible on a CT scan that had been taken on a previous hospital admission 3 months earlier, indicating that there had not been significant recent growth of the masses. Given these historical- and test-based findings, the man was given a differential diagnosis of UTI versus infectious diarrhoea versus cancer versus gastritis. The man did not have fever or diarrhoea while in the hospital, so the diagnosis of infectious diarrhoea (caused by C. difficile or a virus or some other organism) was difficult to rule in or out, as a stool analysis could not be performed and no clear signs of infection were present.
Case analysis This case is representative of many others in the following respect: none of the presenting evidence (including the history and the laboratory and other findings) was such that it allowed for definitively ruling out any of the diagnoses in the initial differential list. In other words, the diagnosis in this case was epistemically uncertain: it was not clear whether the man’s symptoms were due to infectious diarrhoea, UTI, cancer or something else. This case also makes a good study because the diagnostic uncertainty it displays is relatively easy to spot. It might seem surprising then that this diagnostic uncertainty was not recognized and dealt with appropriately by the team that saw the man 3 months prior to the admission that I observed and have described earlier. At that time, as hospital record indicated, the man reported the same symptoms of diarrhoea and abdominal pain that he again reported in May at his second admission. An abdominal CT and a stool culture were performed during his first admission and at that time the stool culture was positive for C. difficile. The man was informed of this result and was given a prescription for oral antibiotic treatment for this infectious condition. The hospital record indicated that the official diagnosis for the man’s complaints was ‘infectious diarrhea due to C. difficile’. Significantly, neither the man nor his gastroenterologist was informed of the results of the CT scan. On the man’s subsequent visit to the emergency department 3 months later, the new attending physician immediately noticed the uncertainty of the case and instead of diagnosing the man with recurrent or undertreated C. difficile, reopened the diagnostic differential and re-examined all of the available evidence. Then, he spent a long time deliberating over what to do, for in medical practice, some sort of action – even if it is merely a communicating between clinician and patient – must always take place even when the epistemic situation is uncertain. In this case, the physician had to decide whether to admit the man to the hospital or to send him back to the assisted living home, and what, if any, treatment the man should be given while in the emergency department. After deliberation, the physician decided not to treat the man for a suspected UTI given the history of C. difficile that had occurred subsequent to previous antibiotic treatment for that condition. Instead, the physician sent the urine out for culture. Thus, given that the immediate risk to the man from any of the diseases in the differential was low, the attending physician decided that he could safely suspend judgment on the case while conducting further 2
A.G. Kennedy
investigation and sent the man home with instructions to follow-up with his gastroenterologist the next week. Finally, the physician also indicated on the man’s discharge instructions that he should be treated for a UTI if bacteria were cultured from the urine sample. I would like to highlight here that suspending judgment, as this physician did, on a diagnostically uncertain case is often the best course of action for several reasons. First, making diagnoses on the basis of insufficient evidence, rather than suspending judgment, is likely to lead to missed or misdiagnosis [6], whereas suspending judgment on a case can allow, as it did in this one, for continued investigation into the cause of the illness in question. In other words, suspending judgment does not have to result in paralysis or inaction. Instead, there is the possibility for action even when judgment is suspended and one form that this can take, aside from that of promoting further investigation, is that of accepting a working diagnosis. Accepting a working diagnosis is an ‘evaluative response’ that allows the clinician to act without requiring belief in any of the initial diagnostic hypotheses [7]. This means, for example, that when a clinician accepts a diagnosis as a basis for action, he or she does not need to believe that the diagnosis is correct. He or she can, instead, simply use the diagnosis on a trial basis. One way that this is often performed in practice is with the use of an ‘empirical trial’. An empirical trial involves giving the patient a treatment for one of the conditions in the differential to see how the patient responds. If the patient responds to the treatment, this is evidence that the patient has the condition in question. Thus, ‘observing a patient’s response (or lack of response) to empiric therapy can provide valuable diagnostic information’ [8]. This means that, at least in some cases, a clinician can arrive at a definitive diagnosis by accepting a tentative one. In fact, many diagnoses are, at least initially, ‘tentative’. There are very few cases in which it is appropriate to call a diagnosis ‘certain’. This is because a diagnosis is a hypothesis [9], the evidence for which can vary greatly in degree. Thus, the cognitive attitude that a clinician should take towards many (and possibly even most) diagnoses is one of acceptance for the purposes of further action, rather than of belief1 that the diagnosis is correct, definitive or true. Second, suspending judgment on a case both requires and allows for an honest communication of the diagnostic uncertainty to the patient and thereby promotes a respectful consideration of the patient as an active participant in the health care team. There are several reasons why this kind of epistemic honesty is beneficial in clinical practice. First, it encourages clinicians to take the patient’s reported symptoms seriously and to refrain from being dismissive in difficult-to-diagnose cases. Second, it also encourages physicians to keep the diagnostic differential open to revision and to thereby avoid the dangers of the overconfidence bias. Berner and Graber [11] (p. S2), who have written on this bias, argue that, ‘physicians in general underappreciate that their diagnoses are wrong’. One of the dangers inherent with an overconfi1
Some, such as Cohen [10], have argued that belief is involuntary. If this is the case, then a physician could not choose whether or not to believe that a specific diagnosis was true or correct. However, he or she could still choose to keep the diagnosis open for re-evaluation. This is what I am suggesting that should be performed in cases in which the evidence is indeterminate. That is, even if a physician involuntarily forms a belief about the diagnosis, he or she can still keep the diagnosis open to investigation, revision or further evaluation.
© 2015 John Wiley & Sons, Ltd.
A.G. Kennedy
dence bias is that it can lead to a breakdown in clinical reasoning. Berner and Graber wrote that an overconfident physician might have problems gathering data, such as failing to elicit complete and accurate information from the patient, failure to recognize the significance of data, such as misinterpreting test results, or most commonly, failure to synthesize or ‘put it all together.’ (p. S8). Another danger with an overconfidence bias is that it tends to lead to definitive action in the face of uncertainty. In fact, the greater the epistemic uncertainty of the diagnostic situation, the more likely clinicians are to exercise overconfidence in the clinic [5,12]. Thus, what comes naturally to many clinicians when they are faced with uncertainty is exactly the opposite of what is called for by best diagnostic practice. Instead of definitive action that closes off further investigation, best diagnostic practice, in many cases, calls for the clinician to either adopt a working diagnosis or leave the investigation open while further evidence is being gathered. Interestingly, Jay Katz, a physician at Yale (see Groopman [5], p. 152–153), has argued that clinicians are willing to admit that medicine is an uncertain practice in theoretical discussion, but that they do not acknowledge the uncertainty inherent in what they do in actual practice. I have found this to be true in my own experience. In discussions with clinicians about medical uncertainty and how to handle it, many tend to think it is ‘obvious’ that (1) medicine is uncertain and (2) that this uncertainty should be acknowledged and handled appropriately. However, there is a tendency among these same clinicians to deny that there is uncertainty in any of their diagnostic cases, including the ones examined in this paper. I think this shows a need for explicit training in the recognition and management of uncertainty during medical school, residency and beyond. In this particular case that we have examined, once the physician had decided to suspend judgment until more information could be gathered, he then faced the difficult question of how to communicate his findings and recommendations to his patient. It was clear from the initial conversation between the two, which I witnessed, that the man was hoping for a solution to, or at least a definitive explanation for, his complaints. The patient had expressed frustration at his ongoing symptoms and multiple emergency room (ER) visits and clearly was looking for an answer as to the cause of his continuing suffering. The most difficult part of the attending physician’s management of this case was that he could not provide what the man wanted: he did not have a definitive answer to give the man concerning the cause of his long-standing symptoms. Of course the physician could have chosen to act as if he did have a definitive answer, but instead, he consciously decided (after much deliberation!) to be honest in his conversation with the patient and to tell the man that he simply did not know what was causing his continuing abdominal pain and diarrhoea. After his initial shock at hearing the words, ‘I don’t know’, the patient quickly expressed his appreciation of the physician’s honesty and was eager to hear his recommendations and plan of action. One could certainly imagine, however, a case in which a patient might not appreciate honesty concerning the uncertainty of his or her diagnosis. Patients are individuals and because of this, it should be acknowledged that uncertainty can be very uncomfortable for some and that therefore it should be communicated thoughtfully and carefully, and, at times, delicately. But it should
© 2015 John Wiley & Sons, Ltd.
Managing uncertainty in diagnostic practice
always be communicated and never hidden. When communicating diagnostic uncertainty to a patient, a clinician should at all times be aware of how her patient is receiving the news. Some patients, for example, will want detailed information and others will not. The key to effective communication of uncertainty in diagnosis is to carefully consider the needs of the individual patient and to always couple the news of the uncertainty with an assurance that the clinician will continue to investigate the case until a more definitive diagnosis can be made. The man’s prior visit to the ER had proceeded quite differently. Even though the presenting evidence at that time also rendered the case diagnostically uncertain (he was found to have both a C. difficile infection and to have metastatic masses in the abdomen), the team that initially managed the patient’s case settled on a definitive diagnosis of infectious diarrhoea as the cause of the man’s symptoms, without communicating the uncertainty of the situation to the patient. What this meant is that the patient’s metastatic masses were completely ignored and neither he nor his regular physician was notified of this finding. The reasons for the mishandling of the diagnostic uncertainty in this case are not clear. It is entirely possible that the clinicians were over worked, over tired or just too busy on that given day, leading them to favour quick diagnostic decisions over careful consideration of the full differential list.2 Whatever the reason for the mishandling, the failure to appropriately consider the diagnostic uncertainty in this case was serious, even bordering on malpractice, and put both the patient (for health reasons) and the hospital (for legal reasons) at risk of significant harm. Because the diagnosis of C. difficile was definitively recorded in the patient’s chart, no further investigation into the cause of his abdominal pain and diarrhoea was conducted by the physicians that the man saw in the interim between his two hospital visits. In other words, the case was prematurely closed, as a consequence of the lack of recognition and acknowledgment of the uncertainty in the initial diagnosis. Ultimately, this cost the man many months of suffering as well as multiple ER and clinic visits. It might have even shortened his life, as what was missed in this case was cancer. There are many lessons that can be learned from an examination of this case. The first is that settling too quickly on a single diagnosis is not in the best medical interest of the patient even when the patient (and/or the clinician) is clearly hoping for a definitive answer to or solution for the presenting symptoms. Instead, a willingness to suspend judgment in the face of an uncertain diagnostic situation, and thus to keep open the scientific investigation into the cause of a patient’s complaints, is preferable to too quickly settling on a single diagnosis in the face of underdetermining evidence. Diagnostic uncertainty is never comfortable for patients or clinicians and this is one reason that it sometimes mishandled or brushed aside. There are other reasons as well. During the man’s first ER admission, the attending physician might have reasoned that, given the man’s age, the metastatic masses found on the 2 Another possibility is that the clinicians were not able to correctly interpret the probabilities associated with the likelihood of each of the conditions in the differential. A discussion of probabilities and their interpretations is relevant and important for clinical practice. However, it is beyond the scope of this paper and I therefore will leave it to a subsequent one.
3
Managing uncertainty in diagnostic practice
abdominal CT were unlikely to be of immediate harm and that informing the man of their presence would not be of any therapeutic benefit. Thus, the uncertainty in this case might have been ignored because the other conditions in the diagnostic differential were not considered to be of appreciable risk to the patient. There is a pervasive presumption among clinicians that if the highest, or most immediate, risk diagnosis in the differential is treated, then the other possible diagnoses can be eliminated or ignored [13]. There are good, practical, reasons for treating high risk and/or immediate dangers first. Certainly, treating the man’s C. difficile infection, which was an immediate risk, was appropriate. But the error in this case was that the uncertainty in the diagnosis was not acknowledged, nor was it communicated, and thus the cancer that was found on the man’s CT scan was never investigated. Instead, only the C. difficile diagnosis was communicated to the man and recorded in the hospital record.
Case 2 In June 2013, a 29-year-old man was admitted to the emergency department of another major hospital with the chief complaint of chest pain and shortness of breath. He denied any history of allergies or asthma. Upon examination, he was found to have a productive cough, which produced green sputum, and significant difficulty with breathing. The family members that had driven him to the hospital made a specific request that the man be given antibiotic therapy for his condition. However, because chest imaging was negative for pneumonia, the attending physician refused this request and instead made a diagnosis of viral upper respiratory infection.3 He then treated the patient with albuterol. Fifteen minutes after the albuterol treatment, the patient reported easier breathing but continued to look quite ill. The physician then explained to the man that he had a bad cold, that he definitely did not need antibiotics and that his condition would improve on its own.
Case analysis I would like to suggest that this case was handled inappropriately because the diagnostic uncertainty of the situation was completely ignored. The man displayed symptoms (difficulty breathing and productive cough with green sputum) that are consistent with both viral and bacterial respiratory tract infection. What this means is that the diagnosis of a viral infection in this case, and in the many like it, is inherently uncertain because the evidence is underdetermining. There is no way to simply look at a patient and determine by physical examination alone, whether or not he 3 Many clinicians are reticent to prescribe antibiotic therapy without confirmatory signs such as bacterial culture or X-ray showing infection because of the growing problem of antibiotic resistance. This problem is indeed urgent, multifaceted and widespread: the Center for Disease Control (CDC) estimates that in the United States alone, an estimated 23 000 deaths each year are associated with drug-resistant bacterial infections. Clinicians have an important role to play in a potential solution to this problem, by prescribing antibiotics to patients only when needed and not simply whenever they are requested. But the question of when they are needed is not always easy to answer, and in attempting to answer it, clinicians need to take care not to be dismissive of their patients in the process, which is what happened in this case, and what, unfortunately, happens in many cases like it.
4
A.G. Kennedy
or she has a viral or bacterial respiratory tract infection. It is possible, however, to hone in on the probability of bacterial infection by taking a careful patient history. It is well known that in the general population viral upper respiratory infections are more common than bacterial ones. However, this statistic does not apply to all individual cases. In patients with a history of recurrent bacterial respiratory infection, that condition is more common than the former. However, the patient was not informed of this diagnostic uncertainty, nor was he ever questioned about a history of bacterial respiratory infections. Instead, he was told that his infection was certainly viral. That made his diagnosis both epistemically and ethically problematic. The diagnosis was epistemically problematic because the evidence that was gathered was insufficient to support it. Furthermore, it was ethically problematic because it led the physician to be dismissive of the patient, by not communicating the epistemic situation to him. As we have seen, given the presenting evidence in this case, it was not possible to make a definitive determination between the two possible diagnoses (bacterial vs. viral respiratory infection). However, the attending physician never even entertained the hypothesis that the patient’s infection was bacterial, even though the evidence was consistent with this. If he had entertained this possibility, or perhaps had used it as a working diagnosis, the physician might have cultured the patient’s sputum or considered an empirical trial of antibiotic therapy, thereby keeping the investigation into the cause of the patient’s symptoms open. Even if the diagnosis of viral infection turned out to be correct, making this diagnosis, without either considering, or informing the patient of, the possibility that the illness was bacterial instead of viral was harmful to the patient for two reasons. First, it was harmful because it prematurely closed the investigation into the cause of the patient’s illness, possibly resulting in misdiagnosis. Second, it was harmful because it was dismissive of the patient. Telling the patient that he had a cold and did not need antibiotics might have been true. But it might have been false. In this case, the risk of long-term physical harm to the patient because of misdiagnosis was low because he was young and in relatively good health. If he did have a bacterial infection, and not a viral one, this infection could likely be safely treated at a later time, if symptoms persisted. While delaying the proper treatment of the infection would have caused some extra suffering (and perhaps missed days of work) for the patient, it is unlikely that it would have posed a serious or long-term risk to his health. But this is not always the case: in some situations, misdiagnosis can be serious or even fatal. Thus, even in cases in which the risk to the patient is perceived by the clinician to be low or not immediate, diagnostic uncertainty should not be ignored.
Teaching clinicians to recognize diagnostic uncertainty I have presented two examples of what many clinicians would consider ‘ordinary’ diagnostic cases. I picked these cases precisely because they do exhibit key elements of the ordinary – in both the presenting symptoms and in the inherent uncertainty in the differential diagnosis, and in the second case, in the way that it was handled – and are therefore representative of the kinds of diagnostic cases that many clinicians routinely encounter in practice. But my hope is that the way in which these ordinary, routine, diagnos-
© 2015 John Wiley & Sons, Ltd.
A.G. Kennedy
tic cases are handled in the clinical setting will begin to change. An effective management of diagnostic uncertainty begins with an ability to recognize and acknowledge it in the clinical setting. And, as Katz’s work has shown, clinicians are not always able, or willing, to do this in practice, even though they readily do so in theoretical discussion. Furthermore, even when diagnostic uncertainty is recognized, this alone is not enough for effective diagnostic practice: it must also be communicated. Unfortunately, in many cases, diagnostic uncertainty is not communicated and patient health outcomes suffer for it. So what can be done about this problem of recognizing and communicating uncertainty so that diagnostic accuracy and patient health outcomes will improve? The answer to this question is that clinicians must be taught to acknowledge (rather than dismiss) and to communicate to their patients, any uncertainty in differential diagnosis, especially in those cases that are perceived to be ‘ordinary’ or routine. If this is performed in routine cases, it will become automatic in even the most complex situations. In the second case study that I presented earlier, it is likely that the epistemic uncertainty was missed for one of two reasons. It might have been missed because it was eclipsed by a concern for antibiotic stewardship. In other words, the physician in that case might have thought that the risk (to the public) of prescribing an antibiotic to the coughing patient was higher than withholding it and so rushed to a determination of viral illness so that an antibiotic would not be prescribed. Certainly, as I have already noted, the problem of antibiotic resistance is real and worsening. However, the existence of this problem does not eliminate the epistemic uncertainty in such diagnostic cases, nor does it legitimize the negative ethical implications of ignoring it. Clinicians should be reminded of this. Even when immediate or grave physical risk to the patient will be low if some of the diagnoses in the differential4 are ignored, an honest evaluation of, and communication to, the patient of the epistemic situation should still always take place. If this is put into practice in low-risk cases, it will become automatic
Managing uncertainty in diagnostic practice
in high-risk cases, and in all instances it will serve only to enhance diagnostic accuracy.
References 1. Fox, R. C. (1957) Training for uncertainty. In The Student-Physician: Introductory Studies in the Sociology of Medical Education (eds R. K. Merton, G. Reader & P. L. Kendall), pp. 207–241. Cambridge: Harvard University Press. 2. Atkinson, P. (1984) Training for certainty. Social Science and Medicine, 19 (9), 949–956. 3. Upshur, R. E. (2000) Seven characteristics of medical evidence. Journal of Evaluation in Clinical Practice, 6 (2), 93–97. 4. Djulbegovic, B., Hozo, I. & Greenland, S. (2011) Uncertainty in clinical medicine. In Philosophy of Medicine (ed. F. Gifford), pp. 299–356. Oxford: North Holland. 5. Groopman, J. (2008) How Doctors Think. New York: Houghton Mifflin. 6. Kennedy, A. G. (2013) Differential diagnosis and the suspension of judgment. The Journal of Medicine and Philosophy, 38 (5), 487–500. 7. Elliott, K. C. & Willmes, D. (2013) Cognitive attitudes and values in science. Philosophy of Science, 80 (5), 807–817. 8. Brett, A. S. & Powell, C. (2011) Approach to laboratory testing and imaging in aging. In Case-Based Geriatrics: A Global Approach (eds V. A. Hirth, D. Wieland & M. Dever-Bumba), pp. 14–21. New York: McGraw-Hill. 9. Simon, J. R. (2011) How to make real, constructive, progress in medicine. Journal of Evaluation in Clinical Practice, 17 (5), 847–851. 10. Cohen, J. (1992) An Essay on Belief and Acceptance. New York: Oxford University Press. 11. Berner, E. S. & Graber, M. L. (2008) Overconfidence as a cause of diagnostic error in medicine. The American Journal of Medicine, 121 (5 Suppl.), S2–S23. 12. Croskerry, P. & Norman, G. (2008) Overconfidence in clinical decision making. The American Journal of Medicine, 121 (5 Suppl.), S24–S29. 13. Than, M. P. & Flaws, D. F. (2009) Communicating diagnostic uncertainties to patients: the problems of explaining unclear diagnoses and risk. Evidence-based Medicine, 14 (3), 66–67.
4
Of course it is never the case that all possible causes will be put into the differential list. The clinician must decide which possibilities to rule in or out by considering risk, probability and treatability.
© 2015 John Wiley & Sons, Ltd.
5
![[Dealing with diagnostic uncertainty in general practice].](/assets/img/hold.png)
