Correspondence

Clinical Problem-Solving: High on the Differential

N Engl J Med 1998; 338:767-769March 12, 1998

Article

To the Editor:

The commentary in the Clinical Problem-Solving article “High on the Differential” (Nov. 6 issue) 1 contains a misconception that, in my mind, illustrates very nicely why expert systems may sometimes lead to erroneous calculations of the likelihood of a diagnosis in a particular patient.

The underlying disease in the 48-year-old patient — chronic renal failure with several phases of dialysis treatment and three kidney transplantations — clearly has an impact on the pretest probability of coronary artery disease. Although chronic uremia is not a conventional risk factor for coronary artery disease, the estimated prevalence of coronary artery disease in patients who are undergoing dialysis is in the range of 17 to 34 percent.2 The rate of death from cardiovascular causes in patients undergoing dialysis is about 20 times greater than in the age-matched general population.3 The figures are approximately the same for patients who undergo renal transplantation.4 Likewise, the sensitivity and specificity of the thallium stress test may be different for this particular patient population,5 possibly because of specific disorders of the left ventricular wall.6

All this has to be taken into account when one is considering the diagnostic strategy for congestive heart failure or dyspnea on exertion in a patient with chronic renal failure. In applying diagnostic algorithms to patients, one has to be very careful not to overlook the implications of the specific history of an individual patient.

Jan Gossmann, M.D.
Klinikum der J.W. Goethe Universität, 60590 Frankfurt am Main, Germany

6 References

1. 1

   Lurie JD, Sox HC. High on the differential. N Engl J Med 1997;337:1377-1381
   Full Text | Web of Science | Medline

2. 2

   Murphy SW, Parfrey PS. Screening for cardiovascular disease in dialysis patients. Curr Opin Nephrol Hypertens 1996;5:532-540
   CrossRef | Medline

3. 3

   London GM, Drueke TB. Atherosclerosis and arteriosclerosis in chronic renal failure. Kidney Int 1997;51:1678-1695
   CrossRef | Web of Science | Medline

4. 4

   Braun WE. Long-term complications of renal transplantation. Kidney Int 1990;37:1363-1378
   CrossRef | Web of Science | Medline

5. 5

   Marwick TH, Steinmuller DR, Underwood DA, et al. Ineffectiveness of dipyridamole SPECT thallium imaging as a screening technique for coronary artery disease in patients with end-stage renal failure. Transplantation 1990;49:100-103
   CrossRef | Web of Science | Medline

6. 6

   London GM, Fabiani F, Marchais SJ, et al. Uremic cardiomyopathy: an inadequate left ventricular hypertrophy. Kidney Int 1987;31:973-980
   CrossRef | Web of Science | Medline

To the Editor:

Lurie and Sox estimate the pretest probability of coronary artery disease in the patient under discussion to be approximately 0.40. Although this may be true for most patients with evidence of left ventricular diastolic dysfunction in the absence of symptoms or definite signs of coronary artery disease, I believe that this particular patient's pretest probability was in fact much lower. His high-output state inadvertently turned the echocardiography performed at rest into an unstandardized one-stage stress test, with a heart rate of 129 beats per minute, close to the maximal heart rate of 136 beats per minute achieved later during thallium scintigraphy. Since systolic function was normal at echocardiography under the conditions mentioned by the authors, it was a priori highly unlikely that another imaging technique would correctly reveal evidence of clinically significant coronary artery disease. Thallium stress testing therefore should have been omitted from this patient's diagnostic workup.

Ulrich H. Koehler, M.D.
Kardiologische Gemeinschaftspraxis, D-67065 Ludwigshafen, Germany

Author/Editor Response

The authors reply:

To the Editor: We agree that physicians must consider the patient's specific history in applying diagnostic algorithms. However, subjective probability estimates should be grounded in data from well-designed studies to minimize the effect of numerous biases that can plague the diagnostic process.1 Drs. Gossmann and Koehler point out the importance and difficulty of accurately estimating prior probabilities; physicians often need to deal with clinical subtleties unaddressed by published data. The principles of medical decision making can help the clinician use available data more effectively and the researcher perform studies that are more clinically relevant.

One relevant principle is conditional independence. To use the post-test probability from one test result as the pretest probability for a second test, the two tests must measure independent aspects of the disease (i.e., the second test must perform equally well regardless of the results on the first test). Clinical findings are rarely independent, however, and physicians tend to overestimate the probability of disease when multiple redundant features are present.2 In the patient under discussion, end-stage renal disease, hypertension, and diastolic heart failure are interrelated, and it is unclear whether the existence of a 34 percent prevalence of coronary artery disease in patients undergoing dialysis should increase the already high probability estimate of 0.4. Physicians need improved data on the predictive value of clinical findings, both singly and in combination.3

A related issue is the problem of spectrum bias in the evaluation of diagnostic tests. Studies that compare patients with classic or severe disease with normal, healthy volunteers often overestimate the usefulness of a test in practice.4 Although the study of thallium imaging cited by Dr. Gossmann relates specifically to dipyridamole testing and may not be relevant to the patient under discussion,5 the principle that researchers should evaluate test performance across a broad range of disease severity and comorbidity is crucial.

Finally, Dr. Koehler's point is intriguing. How do the operating characteristics of a static echocardiogram in a patient with tachycardia compare with those of a formal stress echocardiogram? Does the context of test interpretation affect test performance? Will the interpretation of an imaging study be different depending on the specific question asked of the reader? Answers to these questions would most likely allow more precise decision making in the future.

Jon D. Lurie, M.D.
Veterans Affairs Medical Center, White River Junction, VT 05009

Harold C. Sox, M.D.
Dartmouth–Hitchcock Medical Center, Lebanon, NH 03756

5 References

1. 1

   Tversky A, Kahneman D. Judgment under uncertainty: heuristics and biases. Science 1974;185:1124-1131
   CrossRef | Web of Science | Medline

2. 2

   Sox HC Jr, Blatt MA, Higgins MC, Marton KI. Medical decision making. Boston: Butterworths, 1988.
   

3. 3

   Simel DL, Rennie D. The clinical examination: an agenda to make it more rational. JAMA 1997;277:572-574
   CrossRef | Web of Science | Medline

4. 4

   Ransohoff DF, Feinstein AR. Problems of spectrum and bias in evaluating the efficacy of diagnostic tests. N Engl J Med 1978;299:926-930
   Full Text | Web of Science | Medline

5. 5

   Marwick TH, Steinmuller DR, Underwood DA, et al. Ineffectiveness of dipyridamole SPECT thallium imaging as a screening technique for coronary artery disease in patients with end-stage renal failure. Transplantation 1990;49:100-103
   CrossRef | Web of Science | Medline