Advertisement

Applying calculations to clinical medicine requires nuance: author's reply

      I thank Prof Stovitz for his letter titled “Applying calculations to clinical medicine requires nuance” [
      • Stovitz S.D.
      Applying calculations to clinical medicine requires nuance: response to Uy.
      ].
      He raised a number of important points in response to the Key Concepts paper on “Estimating pretest probability” [
      • Uy E.J.B.
      Key concepts in clinical epidemiology: Estimating pre-test probability.
      ]:

      1. Diagnostic and treatment decisions in clinical medicine are more nuanced, and there are rarely absolutes. From a shared decision-making perspective, a clinician does not proceed to “definitive treatment.” Rather, clinicians offer patients treatment options, explaining that the treatments may or may not help.

      It is true that diagnostic and treatment decisions in medicine are nuanced.
      In the context of diagnostic tests, these nuances are captured in the clinician's subjective assessments of where to place the diagnostic and therapeutic thresholds for a disease diagnosis under consideration. When deciding on these thresholds, clinicians take into account patient preferences, the severity of the disease, and the invasiveness, side-effects, and cost of both the diagnostic test and the treatment. Shared decision-making in diagnosis is therefore involved not just in treatment decisions but also the approach to diagnostic testing [
      • Uy E.J.B.
      • Ooi S.B.
      • Dans A.L.
      • Dans L.F.
      • Silvestre M.A.A.
      Evaluation of Articles on Diagnosis. In: Painless evidence-based medicine, 2nd edition.
      ].
      Once set, however, these decision thresholds become the objective basis for deciding whether or not the post-test probability from a round of testing would result in (1) definitively ruling out the disease (i.e., post-test probability has crossed the diagnostic threshold), (2) committing to the diagnosis and proceeding to implement treatment (i.e., post-test probability has crossed the therapeutic threshold), or (3) doing further tests for the disease under consideration (i.e., post-test probability is still within the testing range).

      2. Starting the calculation with probability (risk) rather than odds may be inconsequential when prevalence is low but would not work at higher prevalence rates

      The example provided in Section 5 of the paper is indeed a simplified example.
      The post-test probability given a pretest probability of 10% and a likelihood ratio of 5.5 would formally be computed as follows:
      Tabled 1
      Convert Pretest Probability to Pretest Odds0.1/(1.0-0.1) = 0.111
      • Pretest probability/(1-pretest probability)
      Compute Post-test Odds0.11 × 5.5 = 0.611
      • Pretest odds × Likelihood Ratio
      Convert Post-test Odds to Post-test Probability0.61/(1.0 + 0.61) = 0.379
      • Post-test Odds/(1 + Post-test Odds)
      As pointed out, computing directly from pretest probability (without going through the above steps with odds) will overestimate post-test probability when prevalence is higher.
      Tabled 1
      Prevalence/PreTest Probability
      5%10%15%
      Computed Post-Test Probability (Starting computation from pretest probability)27.5%55.0%82.5%
      Computed Post-Test Probability (Starting computation from pretest odds)22.4%37.9%49.3%

      3. The sum of the probabilities across all causes should be around one, although it can be greater than one if people have more than one diagnosis causing their symptoms

      It is true that when a single disease is causing a patient's symptoms, the sum of probabilities across the different diagnoses under consideration should equal 100% [
      • Richardson W.S.
      • Wilson M.C.
      The process of diagnosis.
      ]. In the real world, clinicians cannot pursue all potential diagnoses. Instead, they prioritize which diagnoses to test for depending on which is more likely (i.e., which one has a higher pretest probability), more serious if left undiagnosed and untreated, or more responsive if treatment is offered [
      • Richardson W.S.
      • Wilson M.C.
      • Guyatt G.H.
      • Cook D.J.
      • Nishikawa J.
      Users’ guides to the medical literature: XV. How to use an article about disease probability for differential diagnosis. Evidence-Based Medicine Working Group.
      ]. The ability to estimate pretest probability is therefore an important starting point for prioritizing which differential diagnoses to test for, and subsequently, the type of diagnostic test to use [
      • Uy E.J.B.
      Key concepts in clinical epidemiology: Estimating pre-test probability.
      ].
      I thank Prof Stovitz for raising these points.
      For additional resources on the process of diagnosis, decision thresholds, and computing post-test probability, readers can view Section 8 of the Key Concepts paper on “Estimating pretest probability” [
      • Uy E.J.B.
      Key concepts in clinical epidemiology: Estimating pre-test probability.
      ].

      References

        • Stovitz S.D.
        Applying calculations to clinical medicine requires nuance: response to Uy.
        J Clin Epidemiol. 2022; https://doi.org/10.1016/j.jclinepi.2022.08.008
        • Uy E.J.B.
        Key concepts in clinical epidemiology: Estimating pre-test probability.
        J Clin Epidemiol. 2022; 144: 198-202
        • Uy E.J.B.
        • Ooi S.B.
        • Dans A.L.
        • Dans L.F.
        • Silvestre M.A.A.
        Evaluation of Articles on Diagnosis. In: Painless evidence-based medicine, 2nd edition.
        (Available at:)
        • Richardson W.S.
        • Wilson M.C.
        The process of diagnosis.
        in: Users’ guides to the medical literature A manual for evidence-based clinical practice. 3rd ed. McGraw-Hill Education Medical, New York2015
        • Richardson W.S.
        • Wilson M.C.
        • Guyatt G.H.
        • Cook D.J.
        • Nishikawa J.
        Users’ guides to the medical literature: XV. How to use an article about disease probability for differential diagnosis. Evidence-Based Medicine Working Group.
        JAMA. 1999; 281: 1214-1219

      Linked Article