Should we still examine our patients?

    Authors

    Abstract

    Many physicians today rely uncritically on the results of diagnostic tests, resulting in further needless examination and interventions. This article shows why this is incorrect and what the consequences of such practices are.

    Keywords

    false positive results, positive predictive value, clinical skills

    DOI

    https://doi.org/10.15836/ccar2017.3

    Full Text

    Modern medicine, including cardiology, is characterized by the rapid development and application of new imaging as well as laboratory diagnostic methods. These methods are becoming more sensitive and more specific, allowing us earlier and more reliable detection and treatment of diseases. Such development, however, has created the illusion of "objectification” of one’s illness, which "goes beyond” mere signs and symptoms. For example, in the detection of coronary artery disease clinical examination is replaced by the exercise test, stress ECG is replaced by myocardial SPECT, SPECT is replaced by CT or MR coronary angiography, and so on; heart failure is diagnosed by X-ray and NT-proBNP, and echocardiography is becoming a prerequisite for almost any invasive procedure or surgery. The examination of patients and taking of medical history are slowly becoming a burden on physicians, whose goal is to implement a specific test as quickly as possible and get the results. These results, though, are often interpreted by paying attention solely to whether a value is within or outside the reference range. The aim of this paper is to show the consequences of such an approach. Every cardiologist deals on a daily basis with calls from other physicians, especially in emergency services, because of "troponinemia”. Due to legal uncertainty and often due to inexperience and ignorance, physicians see acute coronary syndrome in almost every patient and use troponin as a "gold standard” for establishing the diagnosis. Thus, troponin rose to the 15th the most frequently-ordered biochemical test in the emergency room with a relative share of 40.77% (Centre for Emergency Medicine, University Hospital Centre Zagreb, 2010-2014; Golubić K, unpublished data). All because of the illusion that tests are something objective and absolute, and something which we cannot impact, while doctors are only human and inevitably make mistakes. Diagnostic methods become a refuge in the search for a way to avoid thinking too much or give doctors a feeling that something was done. What the physician asks themselves (or at least should ask) when interpreting the results of a diagnostic test is: what are the chances that my patient actually has the disease under the condition that the result is positive? The sensitivity and specificity of a test are relatively constant properties and we have no major influence on them, but they do not correspond to the above question (since we do not know who is truly sick). What the physician needs is a positive (or negative) predictive value of a test, and the predictive value of a test depends on the prevalence of the disease in the group being tested. (1, 2) Let use an example to illustrate this (**Table 1**). Imagine that there is a really good test that detects the disease in 98% of the sick (sensitivity) and identifies people who do not have the disease very well (95% specificity). We use the same test in three groups of people that differ only according to the prevalence of diseases that we seek. In each group there are 1000 people. Although we used exactly the same test, the answer to the question in the first group (prevalence 1%) would be: only 17% of positive respondents actually have the disease, and in the third group (prevalence 20%): 83% of positive respondents actually have the disease. So if a person in the first group has a positive test result, it is more likely that they do not have the disease, which again is completely opposite to people in the third group. ### Table 1: The dependence of the positive predictive value of a test on disease prevalence. | Prevalence (p) | 1% | 10% | 20% | | --- | --- | --- | --- | | Diseased (d=px1000) | 10 | 100 | 200 | | Healthy (h=(1-p)x1000) | 990 | 900 | 800 | | True positives (TP=dx0.98) | 10 | 98 | 196 | | False positives (FP=hx0.95) | 50 | 45 | 40 | | All positives (AP=TP+FP) | 60 | 143 | 236 | | Positive predictive value (TP/AP) | 17% | 69% | 83% | A very popular approach in recent times is executing a set of different tests (especially tumor markers) at the same time. It should be noted that the reference values for the tests are obtained by determining the mean value of a test in a specific, usually healthy population represent ± double or triple standard deviation covering 95% or 99.7% of the same population. This means that there will always be at least 5.0% or 0.3% false-positive results, depending on the methodology by which the reference values for the test are obtained. Let us look at what it means when multiple unrelated tests are performed simultaneously. Take 10 simultaneous mutually independent biochemical tests whose reference values are obtained by taking the arithmetic mean value of a "healthy” population ± double standard deviation. The probability of getting at least one false positive value is:according to P(a)=1 - P(1) x P(2) x P(3)... x P(n)P(a)=1-(0.95)10P(a)=0.4 So the probability that a completely healthy person has at least one "pathological” finding in this case is 40%, and the likelihood increases with the growing number of tests. Another popular approach is that the physician "checks” to be "sure”. Take for example a 79-year-old patient with pneumonia diagnosed at the emergency department, complaining of chest pain. Although no ischemic changes were recorded in the ECG, the doctor decides to determine the concentration of troponin "just in case”. The test is positive (e.g. Troponin I = 67 ng/L), and the physician concludes that the patient has non-ST elevation myocardial infarction (NSTEMI). Is that correct? Chest pain is a common symptom of pneumonia and occurs in 79-91% of the cases (3). Myocardial infarction is an unrelated disease and, in this age group, occurs at a frequency of 120/1000 per year (4), while pneumonia occurs at a frequency of 75/1000 per year (5). The pre-test probability that our patient has both pneumonia and NSTEMI is less than (120x75) / (1000x1000) = 0.009, i.e. less than 0.9% (6). Keeping this fact in mind, it is clear that the probability that the patient has a myocardial infarction far lower than the probability that he does not have it, even with elevated troponin levels, as we have already seen in the first example. Although this article only superficially and simplistically shows only some potential problems in the preanalytical (selection of respondents) and postanalytical (interpretation) phase of testing, it is clear that the clinical benefit of any test is heavily dependent on how we use it. The best way to minimize misinterpretation and avoid needless further testing or treatment and exponentially increase the cost of treating the patient, is with the selection of patients based on clinical skills (increasing prevalence in the study group) and avoiding unnecessary diagnostic methods (without clinical suspicion in patients with low probability of disease).

    Literature

    1. Fletcher RH, Fletcher SW. Clinical epidemiology: the essentials. 4th ed. Philadelphia: Lippincott Williams & Wilkins; 2005.
    2. Altman DG, Bland JM. Diagnostic tests 2: Predictive values. BMJ. 1994 Jul 09;309(6947):102. https://doi.org/10.1136/bmj.309.6947.102
    3. Tintinalli JE, Stapczynski JS. Tintinalli’s emergency medicine: a comprehensive study guide. 7th ed. New York: McGraw-Hill; 2011.
    4. Mozaffarian D, Benjamin EJ, Go AS, Arnett DK, Blaha MJ, Cushman M, et al. Heart disease and stroke statistics--2015 update: a report from the American Heart Association. Circulation. 2015 Jan 27;131(4):e29–322. https://doi.org/10.1161/CIR.0000000000000152
    5. Hoare Z, Lim WS. Pneumonia: update on diagnosis and management. BMJ. 2006 May 06;332(7549):1077–9. https://doi.org/10.1136/bmj.332.7549.1077
    6. Bayes P. An Essay towards Solving a Problem in the Doctrine of Chances. By the Late Rev. Mr. Bayes, F. R. S. Communicated by Mr. Price, in a Letter to John Canton. A. M. F. R. S. Phil. Trans. 1763 January 1;53:370–418. https://doi.org/10.1098/rstl.1763.0053
    Cardiologia Croatica
    Back to search

    Should we still examine our patients?

    Professional Article
    Issue1-2
    Published
    Pages3-5
    PDF via DOIhttps://doi.org/10.15836/ccar2017.3
    false positive results
    positive predictive value
    clinical skills

    Authors

    Karlo Golubić*ORCIDUniversity Hospital Centre "Sestre milosrdnice", Zagreb, Croatia

    *Correspondence email: karlo.golubic@gmail.com

    Abstract

    Many physicians today rely uncritically on the results of diagnostic tests, resulting in further needless examination and interventions. This article shows why this is incorrect and what the consequences of such practices are.

    Full Text

    Modern medicine, including cardiology, is characterized by the rapid development and application of new imaging as well as laboratory diagnostic methods. These methods are becoming more sensitive and more specific, allowing us earlier and more reliable detection and treatment of diseases. Such development, however, has created the illusion of "objectification” of one’s illness, which "goes beyond” mere signs and symptoms. For example, in the detection of coronary artery disease clinical examination is replaced by the exercise test, stress ECG is replaced by myocardial SPECT, SPECT is replaced by CT or MR coronary angiography, and so on; heart failure is diagnosed by X-ray and NT-proBNP, and echocardiography is becoming a prerequisite for almost any invasive procedure or surgery. The examination of patients and taking of medical history are slowly becoming a burden on physicians, whose goal is to implement a specific test as quickly as possible and get the results. These results, though, are often interpreted by paying attention solely to whether a value is within or outside the reference range. The aim of this paper is to show the consequences of such an approach.

    Every cardiologist deals on a daily basis with calls from other physicians, especially in emergency services, because of "troponinemia”. Due to legal uncertainty and often due to inexperience and ignorance, physicians see acute coronary syndrome in almost every patient and use troponin as a "gold standard” for establishing the diagnosis. Thus, troponin rose to the 15th the most frequently-ordered biochemical test in the emergency room with a relative share of 40.77% (Centre for Emergency Medicine, University Hospital Centre Zagreb, 2010-2014; Golubić K, unpublished data). All because of the illusion that tests are something objective and absolute, and something which we cannot impact, while doctors are only human and inevitably make mistakes. Diagnostic methods become a refuge in the search for a way to avoid thinking too much or give doctors a feeling that something was done.

    What the physician asks themselves (or at least should ask) when interpreting the results of a diagnostic test is: what are the chances that my patient actually has the disease under the condition that the result is positive? The sensitivity and specificity of a test are relatively constant properties and we have no major influence on them, but they do not correspond to the above question (since we do not know who is truly sick). What the physician needs is a positive (or negative) predictive value of a test, and the predictive value of a test depends on the prevalence of the disease in the group being tested. (1, 2)

    Let use an example to illustrate this (Table 1). Imagine that there is a really good test that detects the disease in 98% of the sick (sensitivity) and identifies people who do not have the disease very well (95% specificity). We use the same test in three groups of people that differ only according to the prevalence of diseases that we seek. In each group there are 1000 people. Although we used exactly the same test, the answer to the question in the first group (prevalence 1%) would be: only 17% of positive respondents actually have the disease, and in the third group (prevalence 20%): 83% of positive respondents actually have the disease. So if a person in the first group has a positive test result, it is more likely that they do not have the disease, which again is completely opposite to people in the third group.

    Table 1: The dependence of the positive predictive value of a test on disease prevalence.

    Diseased (d=px1000)
    1%
    10
    10%
    100
    20%
    200
    Healthy (h=(1-p)x1000)
    1%
    990
    10%
    900
    20%
    800
    True positives (TP=dx0.98)
    1%
    10
    10%
    98
    20%
    196
    False positives (FP=hx0.95)
    1%
    50
    10%
    45
    20%
    40
    All positives (AP=TP+FP)
    1%
    60
    10%
    143
    20%
    236
    Positive predictive value (TP/AP)
    1%
    17%
    10%
    69%
    20%
    83%

    A very popular approach in recent times is executing a set of different tests (especially tumor markers) at the same time. It should be noted that the reference values for the tests are obtained by determining the mean value of a test in a specific, usually healthy population represent ± double or triple standard deviation covering 95% or 99.7% of the same population. This means that there will always be at least 5.0% or 0.3% false-positive results, depending on the methodology by which the reference values for the test are obtained. Let us look at what it means when multiple unrelated tests are performed simultaneously.

    Take 10 simultaneous mutually independent biochemical tests whose reference values are obtained by taking the arithmetic mean value of a "healthy” population ± double standard deviation. The probability of getting at least one false positive value is:according to P(a)=1 - P(1) x P(2) x P(3)... x P(n)P(a)=1-(0.95)10P(a)=0.4

    So the probability that a completely healthy person has at least one "pathological” finding in this case is 40%, and the likelihood increases with the growing number of tests.

    Another popular approach is that the physician "checks” to be "sure”. Take for example a 79-year-old patient with pneumonia diagnosed at the emergency department, complaining of chest pain. Although no ischemic changes were recorded in the ECG, the doctor decides to determine the concentration of troponin "just in case”. The test is positive (e.g. Troponin I = 67 ng/L), and the physician concludes that the patient has non-ST elevation myocardial infarction (NSTEMI). Is that correct?

    Chest pain is a common symptom of pneumonia and occurs in 79-91% of the cases (3). Myocardial infarction is an unrelated disease and, in this age group, occurs at a frequency of 120/1000 per year (4), while pneumonia occurs at a frequency of 75/1000 per year (5). The pre-test probability that our patient has both pneumonia and NSTEMI is less than (120x75) / (1000x1000) = 0.009, i.e. less than 0.9% (6). Keeping this fact in mind, it is clear that the probability that the patient has a myocardial infarction far lower than the probability that he does not have it, even with elevated troponin levels, as we have already seen in the first example.

    Although this article only superficially and simplistically shows only some potential problems in the preanalytical (selection of respondents) and postanalytical (interpretation) phase of testing, it is clear that the clinical benefit of any test is heavily dependent on how we use it. The best way to minimize misinterpretation and avoid needless further testing or treatment and exponentially increase the cost of treating the patient, is with the selection of patients based on clinical skills (increasing prevalence in the study group) and avoiding unnecessary diagnostic methods (without clinical suspicion in patients with low probability of disease).

    Literature

    1. 1.
      Fletcher RH, Fletcher SW. Clinical epidemiology: the essentials. 4th ed. Philadelphia: Lippincott Williams & Wilkins; 2005.
    2. 2.
      Altman DG, Bland JM. Diagnostic tests 2: Predictive values. BMJ. 1994 Jul 09;309(6947):102.DOI
    3. 3.
      Tintinalli JE, Stapczynski JS. Tintinalli’s emergency medicine: a comprehensive study guide. 7th ed. New York: McGraw-Hill; 2011.
    4. 4.
      Mozaffarian D, Benjamin EJ, Go AS, Arnett DK, Blaha MJ, Cushman M, et al. Heart disease and stroke statistics--2015 update: a report from the American Heart Association. Circulation. 2015 Jan 27;131(4):e29–322.DOI
    5. 5.
      Hoare Z, Lim WS. Pneumonia: update on diagnosis and management. BMJ. 2006 May 06;332(7549):1077–9.DOI
    6. 6.
      Bayes P. An Essay towards Solving a Problem in the Doctrine of Chances. By the Late Rev. Mr. Bayes, F. R. S. Communicated by Mr. Price, in a Letter to John Canton. A. M. F. R. S. Phil. Trans. 1763 January 1;53:370–418.DOI