- Email: [email protected]
Sidetracks in the diagnosis of gastrointestinal disease
THE AMERICAN JOURNAL OF GASTROENTEROLOGY © 2000 by Am. Coll. of Gastroenterology Published by Elsevier Science Inc.
Vol. 95, No. 11, 2000 ISSN 0002-9270/00/$20.00 PII S0002-9270(00)02044-X
Sidetracks in the Diagnosis of Gastrointestinal Disease Amnon Sonnenberg, M.D., M.Sc. Department of Veterans Affairs Medical Center, and University of New Mexico, Albuquerque, New Mexico
ABSTRACT Patients usually visit a physician for a particular reason, which medicine has termed the “chief complaint.” As patients often have more than a single complaint, it is common for physicians to lose focus, get distracted, and venture down a path other than the one toward resolution of the chief complaint—in other words, to get sidetracked. The aim of the present review is to describe the nature of diagnostic sidetracking, why it occurs, mechanisms in action, and means to prevent it. During a regular diagnostic process, the physician’s suspicions about a diagnosis advance from broad and general to narrow and specific concepts. Each step in the diagnostic process is potentially driven by its own hypothesis, and each individual hypothesis becomes verified or falsified by pointed questions during the history or by medical tests. If a diagnostic process were to focus on the correct diagnosis that eventually explained the chief complaint, the intersections among the consecutively refined diagnostic concepts would converge toward one specific set, with the final diagnosis located inside the nonzero intersection of all consecutive diagnostic concepts. In a faulty diagnostic process, consecutive diagnostic concepts fail to converge, and the final diagnosis and chief complaint do not intersect. In this regard, the physician has pursued a false path and has become sidetracked. The effects of sidetracking range from minimal to disastrous. Sidetracking can delay diagnosis, delay treatment, and waste resources investigating or treating irrelevant medical problems. To avoid sidetracking, physicians should focus on the main problem, avoid getting involved with complementary solutions to secondary medical problems, and verify repetitively during the diagnostic process the relationship between each current working hypothesis and the patient’s major medical problem. (Am J Gastroenterol 2000;95:3039 –3043. © 2000 by Am. Coll. of Gastroenterology)
INTRODUCTION Although patients usually present to their physician with one chief complaint, they may frequently harbor more than one single medical problem. At times, it becomes easy for managing physicians to loose sight of the main medical problem and to find themselves pursuing issues unrelated to the patient’s chief complaint. For instance, a patient sees his
or her physician for symptoms of increasing weakness and fatigue. The complete blood count reveals an iron-deficient anemia, and a subsequent test for fecal occult blood test turns out positive. Questioning the patient about GI symptoms seems the next logical step in the workup. The patient reports symptoms suggestive of gastroesophageal reflux disease (GERD), and a subsequent upper GI endoscopy confirms the diagnosis of erosive reflux esophagitis with single erosions. In the majority of cases, single esophageal erosions cannot explain the presence of anemia. A knowledgeable physician would dismiss the diagnosis as an incidental finding and continue the search for the actual source of GI bleeding or other etiologies of anemia. However, in the clinical routine, one frequently finds patients diagnosed with and treated for disorders that are only loosely connected with their main medical problem. The incidental findings of hemorrhoids in the workup of iron-deficient anemia, of cholecystolithiasis in patients with irritable bowel syndrome, or of hepatitis C infection in the general workup of patients with fatigue constitute other typical examples of correct but irrelevant diagnoses with respect to patients’ chief complaints. The aim of the present article is to describe the nature of diagnostic sidetracking and the mechanisms that contribute to its occurrence, and to suggest means for its prevention.
THE REGULAR DIAGNOSTIC PROCESS Various analytical and cognitive models have been developed to describe the process of diagnostic reasoning (1– 8). For the purpose of the present analysis, the standard Venn diagram from probability textbooks will suffice to illustrate the main features of a regular and sidetracked diagnostic process (Fig. 1) (9, 10). During a regular diagnostic process, the physician’s concepts about the disease advance from being rather broad and general to becoming increasingly more narrow and specific. Each step in the diagnostic process is potentially driven by its own hypothesis. The individual hypotheses become verified or falsified by one or several tests. The tests could comprise a laboratory test, a diagnostic procedure, or, in most instances, simply a pointed question to the patient. As the workup progresses, the diagnostic concepts become more refined, until further refinement is not possible or further refinement bears no thera-
3040
Sonnenberg
AJG – Vol. 95, No. 11, 2000
Figure 1. Venn diagrams of (A) the regular diagnostic process and (B) the sidetracked diagnostic process. The diagnostic process advances from chief complaints of weakness through anemia, GI bleeding, and GI symptoms, to gastric ulcer or gastroesophageal reflux disease. In each step the diagnostician tries to narrow the previous diagnostic concept by overlaying it with a more refined new concept. Whereas in the regular process the sets tend to converge on the intersection of all diagnoses in sequence (marked by an “x”), in a sidetracked process no convergence is achieved.
peutic consequence. In the example from above, “anemia” represents a hypothesis to explain weakness and fatigue. Similarly, “GI bleeding” represents a subsequent hypothesis to explain “anemia.” The two hypotheses are verified by the complete blood count and the fecal occult blood test, respectively. The diagnostic process would stop, for instance, if upper GI endoscopy revealed a gastric ulcer. Figure 1A represents the regular diagnostic process, in which a well chosen sequence of hypotheses and corresponding tests helps to narrow down the initial broad diagnostic concept to a final diagnosis. There is always an appreciable overlap between various symptoms and diagnoses. For instance, there is only a partial match between weakness and anemia, as only a fraction of all weakness relates to anemia and a large proportion of all anemia occurs without symptoms of weakness or fatigue. Similarly, the partial overlap between anemia and GI bleeding stems from the fact that only a fraction of all anemia is caused by GI bleeding and only a fraction of all GI bleeding causes anemia. Finally, although some GI bleeding may be associated with GI symptoms, the majority of GI symptoms occur unaccompanied by bleeding. In case of epigastric pain, the workup would probably continue with an upper GI endoscopy that would then reveal a gastric ulcer. Despite the imperfect match between the primary complaint and its subsequent refinements, the sequence of increasingly refined diagnoses is well capable of honing down to the final diagnosis.
DIAGNOSTIC SIDETRACKS Figure 1B illustrates what happens if the diagnostic process becomes sidetracked. Again, the same example of a patient with the initial complaints of weakness and fatigue is used. The sole difference between the first and second scheme concerns the final transition from GI symptoms to gastric ulcer or erosive esophagitis. Although the final diagnosis of a bleeding gastric ulcer falls partly inside the area covered by weakness and anemia, erosive esophagitis clearly lies
outside these two initial diagnostic concepts. The diagnosis of GERD is made based on a correct and logical chain of inferences, and each diagnostic step of refinement can be justified perfectly well. There is nothing wrong about asking a patient with GI bleeding about potential symptoms suggestive of GI disease, and it is prudent to pursue any such symptoms with upper GI endoscopy. An endoscopic verification of the final diagnosis lends additional credence to the veracity of the diagnosis. Only if one looks back to check the relationship between the initial problem and the final diagnosis does it become obvious that one has lost focus and entered a diagnostic sidetrack. In dealing with diagnostic sidetracks, it is important first to stress the difference between an incorrect and an irrelevant diagnosis, as sidetracking often leads to correct but— with respect to the chief complaints— unimportant diagnoses. At times it may prove difficult to convince diagnosticians that they have truly lost track. A physician may point out the correctness of his/her diagnosis and provide impeccable justification for each diagnostic step leading up to the final diagnosis. Second, one needs to appreciate the difference between a verification of individual links and an overall verification of the entire diagnostic chain (Fig. 2). The connection between each two diagnostic steps may have been established by a well founded hypothesis and its subsequent verification by a well established diagnostic test. Nevertheless, a chain of individually valid and verified connections between each two consecutive diagnostic links does not guarantee the overall validity of the chain, nor does it guarantee the validity of the connection between any two nonadjacent links within the diagnostic chain. In contrast to the absolute certainty associated with a purely mathematical or logical chain of arguments, the connections in a diagnostic chain are based on probability values only. This leads to a partial degeneracy of the diagnostic chain; and, for each refinement in the diagnostic process, its validity with respect to the starting point needs to be re-established. Third, there is frequent confusion between the verification of a diagnosis and that of the diagnostic process itself. The transition
AJG – November, 2000
Sidetracks in Diagnosis of GI Disease
3041
Figure 2. A sidetracked diagnostic process depicted as a chain of consecutive hypotheses leading to increasingly refined diagnostic concepts about the chief complaint. The various tests serve to verify each individual diagnostic concept, but not the overall logical consistency of the chain. To avoid becoming sidetracked, the diagnostician needs to verify each new diagnosis in relation to all previous diagnostic findings.
between consecutively more refined diagnoses is driven by a sequence of hypotheses about potential causes for the chief complaint. Once an hypothesis has been phrased in terms of a diagnostic concept, this concept becomes verified or falsified by a medical test. By checking whether a suspected diagnosis is present or absent, the tests help to verify or falsify an individual diagnostic concept and its underlying hypothesis. The test rarely, if ever, checks on the overall appropriateness or validity of the diagnosis itself. In other words, by confirming the presence or absence of a particular diagnosis, the test does not confirm its general relevance within the diagnostic chain.
A PROBABILISTIC APPROACH TO SIDETRACKING For the diagnostic process to focus on one diagnosis that will eventually explain the initial complaints, the intersections among the consecutively refined diagnoses must converge toward one specific set. The final diagnosis will be located inside the nonzero intersection of all preceding diagnostic concepts (Fig. 1). In probability theory the symbol 艚 is used to denote the intersection between two sets. The following abbreviations are used: CC ⫽ chief complaints of weakness and fatigue, A ⫽ anemia, GIB ⫽ GI bleeding, GIS ⫽ GI symptoms, and GU ⫽ gastric ulcer. According to the principles of probability theory (9 –11), the probability P of the intersection CC艚A艚GIB艚GIS艚GU equals: P(CC艚A艚GIB艚GIS艚GU) ⫽ P(CC) P(A兩CC) P(GIB兩CC艚A) P(GIS兩CC艚A艚GIB) P(GU兩CC艚A艚GIB艚GIS).
(1)
The probability of the chief complaints is P(CC) ⫽ 1, as this is what led the patient to see the physician in the first place. P(AⱍCC) represents the conditional probability of anemia, given that the patient complains about weakness and fatigue. Similarly, P(GIBⱍCC艚A) represents the conditional probability of GI bleeding in a patient with both complaints of
weakness and anemia. P(GISⱍCC艚A艚GIB) represents the conditional probability of GI symptoms in a patient who complains about weakness and who also has anemia and GI bleeding. In principle, each of the conditional probabilities from equation [1] could be calculated using Bayes’ formula (10 – 13). For instance: P(A兩CC) ⫽
P(CC兩A) 䡠 P(A) ¯ ) 䡠 P(A ¯) P(CC兩A) 䡠 P(A) ⫹ P(CC兩A
(2)
¯ represent anemia and nonanemia, respecwhere A and A tively. P(CCⱍA) could be interpreted loosely as the sensitivity of weakness and fatigue for the presence of anemia, ¯ ) would similarly denote its specificity. whereas 1 ⫺ P(CCⱍA For the last term of equation [1], Bayes’ formula yields: P(GU兩CC艚A艚GIB艚GIS) ⫽ P(CC兩GU) 䡠 P(A兩GU) 䡠 P(GIB兩GU) 䡠 P(GIS兩GU) 䡠 P(GU) ¯U ¯ ) 䡠 P(A兩G ¯U ¯ ) 䡠 P(GIB兩G ¯U ¯ ) 䡠 P(GIS兩G ¯U ¯ ) 䡠 P(G ¯U ¯) nominator ⫹ P(CC兩G
(3) ¯U ¯ represent gastric ulcer and no gastric where GU and G ulcer, respectively. Although the formula may look daunting, its nominator contains nothing more but the a priori probability of gastric ulcer, multiplied by the sensitivity of each previous symptom with respect to gastric ulcer. For the diagnostic process to converge on a nonzero intersection of all intermittent diagnoses, P(CC艚A艚GIB艚 GIS艚GU) ⬎ 0; that is, each term of equation [1] must be greater than zero, and each nominator of Bayes’ formulas, as exemplified by equations [2] and [3], needs to be greater than zero. This latter requirement is met only if the individual terms P(CCⱍGU) ⬎ 0, P(AⱍGU) ⬎ 0, . . . , P(GISⱍGU) ⬎ 0. In other words, for gastric ulcer to represent a valid refinement of the diagnostic process, it needs to intersect with each of the previous diagnostic concepts, including the initial chief complaint (CC), and each of the previous diagnostic concepts needs to carry some sensitivity with respect to the occurrence of gastric ulcer. It is easy to see now why GERD fails as an explanation
3042
Sonnenberg
of the initial chief complaints and how it leads to a sidetrack rather than to a resolution of the chief complaints. Although there is a nonzero intersection between GERD and its immediate diagnostic precursor GIS, P(GISⱍGERD) ⬎ 0, and possibly a even small nonzero intersection between GERD and GIB, P(GIBⱍGERD) ⬎ 0, GERD does not intersect with weakness and fatigue. It would be extremely unlikely to have weakness as a sensitive sign of GERD, and its probability approximates zero: P(CCⱍGERD). Because clinical practice deals with probabilities, it is rarely (if ever) possible to rule out or rule in an association with absolute certainty. For instance, Cameron erosions have been reported be associated with iron deficiency anemia and, in a rare instance, GERD may be the sole explanation for weakness (14). A prudent clinician, however, would continue to pursue other diagnoses, such as colorectal cancer or other potential sources of GI bleeding.
ROOTS AND CONSEQUENCES OF SIDETRACKING It seems occasionally as if the particularly diligent and knowledgeable physicians are more prone to fall into the trap of sidetracking. Because they know more, they discover more subtle signs, search for more associations among symptoms and test results, and establish more diagnoses. Part of the problem also relates to the type of sweeping diagnostic tools that are now available for the workup of patients. A general laboratory panel may indicate impaired organ function that was not suspected in the first place. Similarly, CT scan of the abdomen and chest, abdominal ultrasound, or upper GI endoscopy are frequently used as “fishing instruments” for vaguely phrased diagnostic hypotheses. As such broad screening tools are prone to reveal incidental findings, they often provide the clinician with answers to questions not asked. Getting sidetracked prolongs the diagnostic process. In hospitalized patients, it lengthens the hospital stay and increases hospitalization costs. Because the pursuit of an irrelevant diagnosis delays the management of the true diagnosis, the patient is exposed to an increased risk from his or her yet undiagnosed and untreated primary condition. The consequences of sidetracking, however, are not limited to the finding of one irrelevant diagnosis, as physicians frequently may spend appreciable amounts of time and resources in verifying and treating the incidental diagnosis. The diagnostic sidetrack may branch out, result in yet another set of secondary diagnoses, and introduce a variety of complications associated with the use of the unnecessary medical procedures and therapies that lead the managing physician further and further astray from the primary focus of the patient–physician encounter. A lengthier and more complicated management is more prone to complications and failure (15). For instance, a patient with mitral valve prolapse is admitted to the hospital for cardiac catheterization. During the routine physical examination on admission, a digital examination reveals traces of blood in the stool. To rule out
AJG – Vol. 95, No. 11, 2000
colorectal cancer, a colonoscopy is performed under antibiotic prophylaxis. The colonoscopy is unremarkable, but subsequently the patient develops pseudomembranous colitis from the use of antibiotics. In a second example, a patient with a typical presentation and laboratory panel suggestive of alcoholic hepatitis undergoes an abdominal CT scan to rule out other potential causes of jaundice. No gallstones are seen, but the common bile duct of 9 mm seems dilated nevertheless. A subsequent ERCP confirms the normal appearance of the pancreaticobiliary ducts, but the patient develops pancreatitis as a complication of the endoscopic procedure. Examples like these abound in clinical routine (16). The phenomenon of sidetracking is by no means confined to gastroenterology and occurs in all areas of medicine. It may involve gastroenterology more often than other subspecialties, however, because nonspecific abdominal symptoms are frequent in the general population and because the possibility of GI bleeding is located at the crossroads of many differential diagnoses. The diagnostic and therapeutic armamentarium of gastroenterologists involves relatively invasive procedures that render sidetracks into the realm of gastroenterology especially expensive. The development of new diagnostic tests and therapies for hepatitis C virus, as well Helicobacter pylori, has left many physicians uncertain as to how to pursue their incidental discoveries.
MEANS OF PREVENTION The remedies against getting sidetracked in the routine diagnostic process seem rather straightforward. Many aspects of the diagnostic workup resemble the line of thought characteristic of a scientific investigation. Both processes depend crucially on a logical sequence of hypotheses that need to be verified or falsified (17). In most instances, one should not pursue diagnoses (or hypotheses) unrelated the primary problem. Few diagnostic tests— be they interview questions, laboratory tests, or endoscopic procedures— have a perfect sensitivity or specificity to allow one to rule in or out a particular diagnosis completely. The imperfection inherent in clinical practice forces the clinician to use diagnostic thresholds and to abandon hypotheses that fall below a threshold and seem improbable. It is therefore unrealistic to anticipate that we can truly falsify diagnostic hypotheses on a regular basis. Because the diagnostic process is driven by concerns about medical urgency, expediency, and economics, in general it should be focused on finding a positive diagnosis rather than excluding a negative one. During a diagnostic process, the intactness of each individual relationship between the evolving hypotheses and the chief complaint needs to be checked repetitively. If possible, lesser incidental findings on the diagnostic path should be ignored or reserved for a solution at a later time. One should not provide complementary solutions to secondary medical problems but, rather, focus on the main problem, and follow the adage “if it’s not broken don’t fix it.” One must not fall
AJG – November, 2000
in love with one’s own diagnostic acumen or pursue findings primarily because they are uncommon, interesting, or would lead to an elegant solution. Patients rightfully care only about outcome—not statistics, elegance, or becoming an interesting case.
ACKNOWLEDGMENT Supported by a grant from the Centers for Disease Control and Prevention. Reprint requests and correspondence: Amnon Sonnenberg, M.D., M.Sc., Gastroenterology Section, Department of Veterans Affairs Medical Center 111-F, 1501 San Pedro Drive SE, Albuquerque, NM 87108. Received Mar. 27, 2000; accepted Jul. 5, 2000.
REFERENCES 1. Feinstein AR. An analysis of diagnostic reasoning. I. The domains and disorders of clinical macrobiology. Yale J Biol Med 1973;46:212–32. 2. Feinstein AR. An analysis of diagnostic reasoning. II. The strategy of intermediate decisions. Yale J Biol Med 1973;46: 264 – 83. 3. Feinstein AR. An analysis of diagnostic reasoning. III. The construction of clinical algorithms. Yale J Biol Med 1974;47:5–32. 4. Szolovits P, Patil RS, Schwartz WB. Artificial intelligence in medical diagnosis. Ann Intern Med 1988;108:80 –7.
Sidetracks in Diagnosis of GI Disease
3043
5. Elstein AS, Shulman LS, Sprafka SA. Medical problem solving. Cambridge, MA: Harvard University Press, 1988. 6. Moskowitz AJ, Kuipers BJ, Kassirer JP. Dealing with uncertainty, risks, and tradeoffs in clinical decisions. A cognitive science approach. Ann Intern Med 1988;108:435– 49. 7. Schmidt HG, Norman GR, Boshuizen HPA. A cognitive perspective on medical expertise: Theory and implications. Acad Med 1990;65:611–21. 8. Bo¨ckenholt U, Weber EU. Toward a theory of hypothesis generation in diagnostic decision making. Invest Radiol 1993; 28:76 – 80. 9. Drake AW. Fundamentals of applied probability theory. New York: McGraw-Hill, 1967:1–39. 10. Parzen E. Modern probability theory and its application. New York: John Wiley, 1960:1–147. 11. Pearl J. Probabilistic reasoning in intelligent systems—Networks of plausible inference. San Francisco: Morgan Kaufmann, 1988:29 –75. 12. Weinstein MC, Fineberg HV, Elstein AS, et al., eds. Clinical decision analysis. Philadelphia: WB Saunders, 1980. 13. Pauker SG, Kassirer JP. Decision analysis. N Engl J Med 1987;316:250 – 8. 14. Bornstein JD, Brazer SR. Cameron erosions. Gastrointest Endosc 1999;49:514. 15. Sonnenberg A, Inadomi JM, Bauerfeind P. Reliability block diagrams to model disease management. Med Decis Making 1999;19:180 –5. 16. Sonnenberg A, Townsend WF. Preoperative testing for fecal occult blood: A questionable practice. Am J Gastroenterol 1992;87:1410 –7. 17. Sonnenberg A. How to tell the truth and avoid the urge to lie: Planning of a career in medical research. Hepatology 1995; 22:974 – 82.
Vol. 95, No. 11, 2000 ISSN 0002-9270/00/$20.00 PII S0002-9270(00)02044-X
Sidetracks in the Diagnosis of Gastrointestinal Disease Amnon Sonnenberg, M.D., M.Sc. Department of Veterans Affairs Medical Center, and University of New Mexico, Albuquerque, New Mexico
ABSTRACT Patients usually visit a physician for a particular reason, which medicine has termed the “chief complaint.” As patients often have more than a single complaint, it is common for physicians to lose focus, get distracted, and venture down a path other than the one toward resolution of the chief complaint—in other words, to get sidetracked. The aim of the present review is to describe the nature of diagnostic sidetracking, why it occurs, mechanisms in action, and means to prevent it. During a regular diagnostic process, the physician’s suspicions about a diagnosis advance from broad and general to narrow and specific concepts. Each step in the diagnostic process is potentially driven by its own hypothesis, and each individual hypothesis becomes verified or falsified by pointed questions during the history or by medical tests. If a diagnostic process were to focus on the correct diagnosis that eventually explained the chief complaint, the intersections among the consecutively refined diagnostic concepts would converge toward one specific set, with the final diagnosis located inside the nonzero intersection of all consecutive diagnostic concepts. In a faulty diagnostic process, consecutive diagnostic concepts fail to converge, and the final diagnosis and chief complaint do not intersect. In this regard, the physician has pursued a false path and has become sidetracked. The effects of sidetracking range from minimal to disastrous. Sidetracking can delay diagnosis, delay treatment, and waste resources investigating or treating irrelevant medical problems. To avoid sidetracking, physicians should focus on the main problem, avoid getting involved with complementary solutions to secondary medical problems, and verify repetitively during the diagnostic process the relationship between each current working hypothesis and the patient’s major medical problem. (Am J Gastroenterol 2000;95:3039 –3043. © 2000 by Am. Coll. of Gastroenterology)
INTRODUCTION Although patients usually present to their physician with one chief complaint, they may frequently harbor more than one single medical problem. At times, it becomes easy for managing physicians to loose sight of the main medical problem and to find themselves pursuing issues unrelated to the patient’s chief complaint. For instance, a patient sees his
or her physician for symptoms of increasing weakness and fatigue. The complete blood count reveals an iron-deficient anemia, and a subsequent test for fecal occult blood test turns out positive. Questioning the patient about GI symptoms seems the next logical step in the workup. The patient reports symptoms suggestive of gastroesophageal reflux disease (GERD), and a subsequent upper GI endoscopy confirms the diagnosis of erosive reflux esophagitis with single erosions. In the majority of cases, single esophageal erosions cannot explain the presence of anemia. A knowledgeable physician would dismiss the diagnosis as an incidental finding and continue the search for the actual source of GI bleeding or other etiologies of anemia. However, in the clinical routine, one frequently finds patients diagnosed with and treated for disorders that are only loosely connected with their main medical problem. The incidental findings of hemorrhoids in the workup of iron-deficient anemia, of cholecystolithiasis in patients with irritable bowel syndrome, or of hepatitis C infection in the general workup of patients with fatigue constitute other typical examples of correct but irrelevant diagnoses with respect to patients’ chief complaints. The aim of the present article is to describe the nature of diagnostic sidetracking and the mechanisms that contribute to its occurrence, and to suggest means for its prevention.
THE REGULAR DIAGNOSTIC PROCESS Various analytical and cognitive models have been developed to describe the process of diagnostic reasoning (1– 8). For the purpose of the present analysis, the standard Venn diagram from probability textbooks will suffice to illustrate the main features of a regular and sidetracked diagnostic process (Fig. 1) (9, 10). During a regular diagnostic process, the physician’s concepts about the disease advance from being rather broad and general to becoming increasingly more narrow and specific. Each step in the diagnostic process is potentially driven by its own hypothesis. The individual hypotheses become verified or falsified by one or several tests. The tests could comprise a laboratory test, a diagnostic procedure, or, in most instances, simply a pointed question to the patient. As the workup progresses, the diagnostic concepts become more refined, until further refinement is not possible or further refinement bears no thera-
3040
Sonnenberg
AJG – Vol. 95, No. 11, 2000
Figure 1. Venn diagrams of (A) the regular diagnostic process and (B) the sidetracked diagnostic process. The diagnostic process advances from chief complaints of weakness through anemia, GI bleeding, and GI symptoms, to gastric ulcer or gastroesophageal reflux disease. In each step the diagnostician tries to narrow the previous diagnostic concept by overlaying it with a more refined new concept. Whereas in the regular process the sets tend to converge on the intersection of all diagnoses in sequence (marked by an “x”), in a sidetracked process no convergence is achieved.
peutic consequence. In the example from above, “anemia” represents a hypothesis to explain weakness and fatigue. Similarly, “GI bleeding” represents a subsequent hypothesis to explain “anemia.” The two hypotheses are verified by the complete blood count and the fecal occult blood test, respectively. The diagnostic process would stop, for instance, if upper GI endoscopy revealed a gastric ulcer. Figure 1A represents the regular diagnostic process, in which a well chosen sequence of hypotheses and corresponding tests helps to narrow down the initial broad diagnostic concept to a final diagnosis. There is always an appreciable overlap between various symptoms and diagnoses. For instance, there is only a partial match between weakness and anemia, as only a fraction of all weakness relates to anemia and a large proportion of all anemia occurs without symptoms of weakness or fatigue. Similarly, the partial overlap between anemia and GI bleeding stems from the fact that only a fraction of all anemia is caused by GI bleeding and only a fraction of all GI bleeding causes anemia. Finally, although some GI bleeding may be associated with GI symptoms, the majority of GI symptoms occur unaccompanied by bleeding. In case of epigastric pain, the workup would probably continue with an upper GI endoscopy that would then reveal a gastric ulcer. Despite the imperfect match between the primary complaint and its subsequent refinements, the sequence of increasingly refined diagnoses is well capable of honing down to the final diagnosis.
DIAGNOSTIC SIDETRACKS Figure 1B illustrates what happens if the diagnostic process becomes sidetracked. Again, the same example of a patient with the initial complaints of weakness and fatigue is used. The sole difference between the first and second scheme concerns the final transition from GI symptoms to gastric ulcer or erosive esophagitis. Although the final diagnosis of a bleeding gastric ulcer falls partly inside the area covered by weakness and anemia, erosive esophagitis clearly lies
outside these two initial diagnostic concepts. The diagnosis of GERD is made based on a correct and logical chain of inferences, and each diagnostic step of refinement can be justified perfectly well. There is nothing wrong about asking a patient with GI bleeding about potential symptoms suggestive of GI disease, and it is prudent to pursue any such symptoms with upper GI endoscopy. An endoscopic verification of the final diagnosis lends additional credence to the veracity of the diagnosis. Only if one looks back to check the relationship between the initial problem and the final diagnosis does it become obvious that one has lost focus and entered a diagnostic sidetrack. In dealing with diagnostic sidetracks, it is important first to stress the difference between an incorrect and an irrelevant diagnosis, as sidetracking often leads to correct but— with respect to the chief complaints— unimportant diagnoses. At times it may prove difficult to convince diagnosticians that they have truly lost track. A physician may point out the correctness of his/her diagnosis and provide impeccable justification for each diagnostic step leading up to the final diagnosis. Second, one needs to appreciate the difference between a verification of individual links and an overall verification of the entire diagnostic chain (Fig. 2). The connection between each two diagnostic steps may have been established by a well founded hypothesis and its subsequent verification by a well established diagnostic test. Nevertheless, a chain of individually valid and verified connections between each two consecutive diagnostic links does not guarantee the overall validity of the chain, nor does it guarantee the validity of the connection between any two nonadjacent links within the diagnostic chain. In contrast to the absolute certainty associated with a purely mathematical or logical chain of arguments, the connections in a diagnostic chain are based on probability values only. This leads to a partial degeneracy of the diagnostic chain; and, for each refinement in the diagnostic process, its validity with respect to the starting point needs to be re-established. Third, there is frequent confusion between the verification of a diagnosis and that of the diagnostic process itself. The transition
AJG – November, 2000
Sidetracks in Diagnosis of GI Disease
3041
Figure 2. A sidetracked diagnostic process depicted as a chain of consecutive hypotheses leading to increasingly refined diagnostic concepts about the chief complaint. The various tests serve to verify each individual diagnostic concept, but not the overall logical consistency of the chain. To avoid becoming sidetracked, the diagnostician needs to verify each new diagnosis in relation to all previous diagnostic findings.
between consecutively more refined diagnoses is driven by a sequence of hypotheses about potential causes for the chief complaint. Once an hypothesis has been phrased in terms of a diagnostic concept, this concept becomes verified or falsified by a medical test. By checking whether a suspected diagnosis is present or absent, the tests help to verify or falsify an individual diagnostic concept and its underlying hypothesis. The test rarely, if ever, checks on the overall appropriateness or validity of the diagnosis itself. In other words, by confirming the presence or absence of a particular diagnosis, the test does not confirm its general relevance within the diagnostic chain.
A PROBABILISTIC APPROACH TO SIDETRACKING For the diagnostic process to focus on one diagnosis that will eventually explain the initial complaints, the intersections among the consecutively refined diagnoses must converge toward one specific set. The final diagnosis will be located inside the nonzero intersection of all preceding diagnostic concepts (Fig. 1). In probability theory the symbol 艚 is used to denote the intersection between two sets. The following abbreviations are used: CC ⫽ chief complaints of weakness and fatigue, A ⫽ anemia, GIB ⫽ GI bleeding, GIS ⫽ GI symptoms, and GU ⫽ gastric ulcer. According to the principles of probability theory (9 –11), the probability P of the intersection CC艚A艚GIB艚GIS艚GU equals: P(CC艚A艚GIB艚GIS艚GU) ⫽ P(CC) P(A兩CC) P(GIB兩CC艚A) P(GIS兩CC艚A艚GIB) P(GU兩CC艚A艚GIB艚GIS).
(1)
The probability of the chief complaints is P(CC) ⫽ 1, as this is what led the patient to see the physician in the first place. P(AⱍCC) represents the conditional probability of anemia, given that the patient complains about weakness and fatigue. Similarly, P(GIBⱍCC艚A) represents the conditional probability of GI bleeding in a patient with both complaints of
weakness and anemia. P(GISⱍCC艚A艚GIB) represents the conditional probability of GI symptoms in a patient who complains about weakness and who also has anemia and GI bleeding. In principle, each of the conditional probabilities from equation [1] could be calculated using Bayes’ formula (10 – 13). For instance: P(A兩CC) ⫽
P(CC兩A) 䡠 P(A) ¯ ) 䡠 P(A ¯) P(CC兩A) 䡠 P(A) ⫹ P(CC兩A
(2)
¯ represent anemia and nonanemia, respecwhere A and A tively. P(CCⱍA) could be interpreted loosely as the sensitivity of weakness and fatigue for the presence of anemia, ¯ ) would similarly denote its specificity. whereas 1 ⫺ P(CCⱍA For the last term of equation [1], Bayes’ formula yields: P(GU兩CC艚A艚GIB艚GIS) ⫽ P(CC兩GU) 䡠 P(A兩GU) 䡠 P(GIB兩GU) 䡠 P(GIS兩GU) 䡠 P(GU) ¯U ¯ ) 䡠 P(A兩G ¯U ¯ ) 䡠 P(GIB兩G ¯U ¯ ) 䡠 P(GIS兩G ¯U ¯ ) 䡠 P(G ¯U ¯) nominator ⫹ P(CC兩G
(3) ¯U ¯ represent gastric ulcer and no gastric where GU and G ulcer, respectively. Although the formula may look daunting, its nominator contains nothing more but the a priori probability of gastric ulcer, multiplied by the sensitivity of each previous symptom with respect to gastric ulcer. For the diagnostic process to converge on a nonzero intersection of all intermittent diagnoses, P(CC艚A艚GIB艚 GIS艚GU) ⬎ 0; that is, each term of equation [1] must be greater than zero, and each nominator of Bayes’ formulas, as exemplified by equations [2] and [3], needs to be greater than zero. This latter requirement is met only if the individual terms P(CCⱍGU) ⬎ 0, P(AⱍGU) ⬎ 0, . . . , P(GISⱍGU) ⬎ 0. In other words, for gastric ulcer to represent a valid refinement of the diagnostic process, it needs to intersect with each of the previous diagnostic concepts, including the initial chief complaint (CC), and each of the previous diagnostic concepts needs to carry some sensitivity with respect to the occurrence of gastric ulcer. It is easy to see now why GERD fails as an explanation
3042
Sonnenberg
of the initial chief complaints and how it leads to a sidetrack rather than to a resolution of the chief complaints. Although there is a nonzero intersection between GERD and its immediate diagnostic precursor GIS, P(GISⱍGERD) ⬎ 0, and possibly a even small nonzero intersection between GERD and GIB, P(GIBⱍGERD) ⬎ 0, GERD does not intersect with weakness and fatigue. It would be extremely unlikely to have weakness as a sensitive sign of GERD, and its probability approximates zero: P(CCⱍGERD). Because clinical practice deals with probabilities, it is rarely (if ever) possible to rule out or rule in an association with absolute certainty. For instance, Cameron erosions have been reported be associated with iron deficiency anemia and, in a rare instance, GERD may be the sole explanation for weakness (14). A prudent clinician, however, would continue to pursue other diagnoses, such as colorectal cancer or other potential sources of GI bleeding.
ROOTS AND CONSEQUENCES OF SIDETRACKING It seems occasionally as if the particularly diligent and knowledgeable physicians are more prone to fall into the trap of sidetracking. Because they know more, they discover more subtle signs, search for more associations among symptoms and test results, and establish more diagnoses. Part of the problem also relates to the type of sweeping diagnostic tools that are now available for the workup of patients. A general laboratory panel may indicate impaired organ function that was not suspected in the first place. Similarly, CT scan of the abdomen and chest, abdominal ultrasound, or upper GI endoscopy are frequently used as “fishing instruments” for vaguely phrased diagnostic hypotheses. As such broad screening tools are prone to reveal incidental findings, they often provide the clinician with answers to questions not asked. Getting sidetracked prolongs the diagnostic process. In hospitalized patients, it lengthens the hospital stay and increases hospitalization costs. Because the pursuit of an irrelevant diagnosis delays the management of the true diagnosis, the patient is exposed to an increased risk from his or her yet undiagnosed and untreated primary condition. The consequences of sidetracking, however, are not limited to the finding of one irrelevant diagnosis, as physicians frequently may spend appreciable amounts of time and resources in verifying and treating the incidental diagnosis. The diagnostic sidetrack may branch out, result in yet another set of secondary diagnoses, and introduce a variety of complications associated with the use of the unnecessary medical procedures and therapies that lead the managing physician further and further astray from the primary focus of the patient–physician encounter. A lengthier and more complicated management is more prone to complications and failure (15). For instance, a patient with mitral valve prolapse is admitted to the hospital for cardiac catheterization. During the routine physical examination on admission, a digital examination reveals traces of blood in the stool. To rule out
AJG – Vol. 95, No. 11, 2000
colorectal cancer, a colonoscopy is performed under antibiotic prophylaxis. The colonoscopy is unremarkable, but subsequently the patient develops pseudomembranous colitis from the use of antibiotics. In a second example, a patient with a typical presentation and laboratory panel suggestive of alcoholic hepatitis undergoes an abdominal CT scan to rule out other potential causes of jaundice. No gallstones are seen, but the common bile duct of 9 mm seems dilated nevertheless. A subsequent ERCP confirms the normal appearance of the pancreaticobiliary ducts, but the patient develops pancreatitis as a complication of the endoscopic procedure. Examples like these abound in clinical routine (16). The phenomenon of sidetracking is by no means confined to gastroenterology and occurs in all areas of medicine. It may involve gastroenterology more often than other subspecialties, however, because nonspecific abdominal symptoms are frequent in the general population and because the possibility of GI bleeding is located at the crossroads of many differential diagnoses. The diagnostic and therapeutic armamentarium of gastroenterologists involves relatively invasive procedures that render sidetracks into the realm of gastroenterology especially expensive. The development of new diagnostic tests and therapies for hepatitis C virus, as well Helicobacter pylori, has left many physicians uncertain as to how to pursue their incidental discoveries.
MEANS OF PREVENTION The remedies against getting sidetracked in the routine diagnostic process seem rather straightforward. Many aspects of the diagnostic workup resemble the line of thought characteristic of a scientific investigation. Both processes depend crucially on a logical sequence of hypotheses that need to be verified or falsified (17). In most instances, one should not pursue diagnoses (or hypotheses) unrelated the primary problem. Few diagnostic tests— be they interview questions, laboratory tests, or endoscopic procedures— have a perfect sensitivity or specificity to allow one to rule in or out a particular diagnosis completely. The imperfection inherent in clinical practice forces the clinician to use diagnostic thresholds and to abandon hypotheses that fall below a threshold and seem improbable. It is therefore unrealistic to anticipate that we can truly falsify diagnostic hypotheses on a regular basis. Because the diagnostic process is driven by concerns about medical urgency, expediency, and economics, in general it should be focused on finding a positive diagnosis rather than excluding a negative one. During a diagnostic process, the intactness of each individual relationship between the evolving hypotheses and the chief complaint needs to be checked repetitively. If possible, lesser incidental findings on the diagnostic path should be ignored or reserved for a solution at a later time. One should not provide complementary solutions to secondary medical problems but, rather, focus on the main problem, and follow the adage “if it’s not broken don’t fix it.” One must not fall
AJG – November, 2000
in love with one’s own diagnostic acumen or pursue findings primarily because they are uncommon, interesting, or would lead to an elegant solution. Patients rightfully care only about outcome—not statistics, elegance, or becoming an interesting case.
ACKNOWLEDGMENT Supported by a grant from the Centers for Disease Control and Prevention. Reprint requests and correspondence: Amnon Sonnenberg, M.D., M.Sc., Gastroenterology Section, Department of Veterans Affairs Medical Center 111-F, 1501 San Pedro Drive SE, Albuquerque, NM 87108. Received Mar. 27, 2000; accepted Jul. 5, 2000.
REFERENCES 1. Feinstein AR. An analysis of diagnostic reasoning. I. The domains and disorders of clinical macrobiology. Yale J Biol Med 1973;46:212–32. 2. Feinstein AR. An analysis of diagnostic reasoning. II. The strategy of intermediate decisions. Yale J Biol Med 1973;46: 264 – 83. 3. Feinstein AR. An analysis of diagnostic reasoning. III. The construction of clinical algorithms. Yale J Biol Med 1974;47:5–32. 4. Szolovits P, Patil RS, Schwartz WB. Artificial intelligence in medical diagnosis. Ann Intern Med 1988;108:80 –7.
Sidetracks in Diagnosis of GI Disease
3043
5. Elstein AS, Shulman LS, Sprafka SA. Medical problem solving. Cambridge, MA: Harvard University Press, 1988. 6. Moskowitz AJ, Kuipers BJ, Kassirer JP. Dealing with uncertainty, risks, and tradeoffs in clinical decisions. A cognitive science approach. Ann Intern Med 1988;108:435– 49. 7. Schmidt HG, Norman GR, Boshuizen HPA. A cognitive perspective on medical expertise: Theory and implications. Acad Med 1990;65:611–21. 8. Bo¨ckenholt U, Weber EU. Toward a theory of hypothesis generation in diagnostic decision making. Invest Radiol 1993; 28:76 – 80. 9. Drake AW. Fundamentals of applied probability theory. New York: McGraw-Hill, 1967:1–39. 10. Parzen E. Modern probability theory and its application. New York: John Wiley, 1960:1–147. 11. Pearl J. Probabilistic reasoning in intelligent systems—Networks of plausible inference. San Francisco: Morgan Kaufmann, 1988:29 –75. 12. Weinstein MC, Fineberg HV, Elstein AS, et al., eds. Clinical decision analysis. Philadelphia: WB Saunders, 1980. 13. Pauker SG, Kassirer JP. Decision analysis. N Engl J Med 1987;316:250 – 8. 14. Bornstein JD, Brazer SR. Cameron erosions. Gastrointest Endosc 1999;49:514. 15. Sonnenberg A, Inadomi JM, Bauerfeind P. Reliability block diagrams to model disease management. Med Decis Making 1999;19:180 –5. 16. Sonnenberg A, Townsend WF. Preoperative testing for fecal occult blood: A questionable practice. Am J Gastroenterol 1992;87:1410 –7. 17. Sonnenberg A. How to tell the truth and avoid the urge to lie: Planning of a career in medical research. Hepatology 1995; 22:974 – 82.