Diagnostic tests: Asking the right questions


You have picked up the next chart and have drawn your differential diagnosis based on the patient's demographic, chief complaint, and vital signs.

Pattern Recognition vs Probabilistic Diagnostic Reasoning2:
Pattern Recognition
See it and recognize disorder
⇓
Compare post-test probability with threshold (usually pattern recognition implies near 100% and so above threshold)
Probabilistic Diagnostic Reasoning
Clinical assessment generates pretest probability
⇓
New information generates post-test probability (may be iterative)
⇓
Compare post-test probability with threshold

You can approach diagnosing diseases by using either of these two patterns or a combination of both. Keep in mind these two methods can also complement each other. Pattern recognition is more of an intuitive approach, often referred to as System 1 thinking. It is much faster, heuristic. Probabilistic diagnostic reasoning may be referred as System 2 thinking, meaning it’s slower, but more analytical and systematic.


After you draw your differential diagnosis and before seeing the patient, you have a pretest probability of the diagnosis that ails the patient. After you obtain a focused history and physical exam, you have gathered information that will help you draw a more accurate post-test probability and narrow your differential diagnosis. The essence of being a doctor does not lie on the tests, therapies, signs, or symptoms, but on how you use them. Every sign and every symptom represents a diagnostic test.

 Understanding statistical terms helps us interpret diagnostic test results.

 Sensitivity:
  • SnOUT: If a test has a high sensitivity and the elicited test is negative, you have essentially ruled OUT the disease.  
  • How accurately the test picks up patients WITH disease
Specificity:
  • SpIN: If a test has a high specificity and the test is positive, you have essentially ruled IN the disease. 
  • How accurately the test picks up patients WITHOUT disease 
As you can see there are limitations when applying sensitivity and specificity on patients since we don't know if the have the disease or not. Predictive values have their limitations as well, these change with prevalence. Therefore, a test used to detect disease in a population with high prevalence would not be adequate in a population with low prevalence. Keep in mind that tests should not be used to replace your clinical judgement.

    Prevalence:
    • This is your pretest probability, used before you go into the room to see the patient. 
    Likelihood Ratio (LR):
    • Compares results of patients with disease vs patients without disease
    • More accurate than sensitivity and specificity
    • Helps you derive the post-test probability
    • Takes sensitivity and specificity into account simultaneously 
    • When LR >1 it means the probability of disease increases 
    • When LR <1 it means the probability of disease decreases 
    • When LR = 1, the probability of disease is unchanged 
    • Before ordering a test, eliciting a symptom, or finding a sign, ask yourself: How will the absence or presence of this factor change my post-test probability? 
    • Use the Fagan Nomogram to determine post-test probability.

    Does the post-test probability drawn after your assessment change your threshold?
    When the post-test probability falls between test threshold and treatment threshold, further investigation needs to be done. If it lies above the testing threshold it is encouraged to treat, but if it falls below the testing threshold, it is encouraged to pursue a different diagnosis.


    Diagnostic and Test Threshold 7



    There are three aspects that determine test and treatment thresholds:2
    1. Properties of the test
    2. The disease prognosis
    3. The nature of the treatment

    Changing testing thresholds
    If
    Then
    • Test: safe and less expensive
    • Disease: worse prognosis
    • Treatment: safe and effective
    The LOWER the test threshold
    • Test: not as safe, more expensive
    • Disease: benign
    • Treatment: not safe or effective
    The HIGHER the test threshold


    Changing treatment thresholds 
    If
    Then
    • Test: safe and less expensive
    • Disease: benign prognosis
    • Treatment: expensive and major adverse effects
    The HIGHER the treatment threshold
    The characteristics require a higher diagnostic certainty so that we cause less harm from this treatment.
    • Test: less safe, more invasive than the treatment
    • Disease: worse prognosis
    • Treatment: safer than the test
    The LOWER the treatment threshold
    The treatment may be more preferable than the test.


    Four Lessons of Diagnostic Testing as per David Newman 4
    1. All tests are imperfect 
    2. Context trumps results 
    3. All tests have a threshold 
    4. Likelihood ratios have it all
    These four axioms are important to keep in mind when navigating the diagnostic and therapeutic process in medicine. If used appropriately, these four axioms can help us avoid harming patients.



    References:
    1. The Rational Physical Examination: Systematic reviews of the diagnostic properties of the history and the physical examination.

    2. Users' Guides to the Medical Literature:     A Manual for Evidence-Based Clinical Practice, 2nd Edition

    3. NNT Website by David Newman that looks into the number-needed-to-treat of different therapies and also the website contains a scale where likelihood ratios can be manipulated to calculate post-test probabilities.

    4. SMART EM: by David Newman podcast on diagnostic tests

    5. A universal model of diagnostic reasoning.     Croskerry, P; Academic Medicine; 2009 Aug;84(8):1022-8.

    6. A life at risk: a website with lots of LR on signs, symptoms, tests

    7. Hayden SR, Brown MD. Likelihood ratio: A powerful tool for incorporating the results of a diagnostic test into clinical decisionmaking. Ann Emerg Med. 1999. May;33(5):575-80.

    8.     Anthony K. Akobeng. Understanding diagnostic tests 2: likelihood ratios, pre- and post-test probabilities and their use in clinical practice. Acta Paediatrica; 2007 Apr;96(4):487-91. Epub 2007 Feb 14.

    9. Paucis Verbis from Academic Life in Emergency Medicine: multiple cards with likelihood ratios, pre- and post-test probabilities

    10. mdcalc: Bayesian, sensitivities, specificities, probabilities: if you know the prevalence, sensitivity, and specificity of a disease mdcalc has a calculator to obtain the likelihood ratios, and the predictive values.

    11. Center for Evidence Based Medicine. Interactive Nomogram January 2009