1578 Canadian Family Physician • Le Médecin de famille canadiendVOL 5: NOVEMBER • NOVEMBRE 2005
Researchers’ Page
What constitutes clinical evidence?
A dynamic approach to clinical diagnosis
Akbar Soltani, MD Alireza Moayyeri, MD
T
wo commonly accepted parameters describing the diagnostic power of various clinical tests are sensitivity and specifi city, and likelihood ratios.1 Evidence-based authorities have repeatedly highli- ghted the advantages of likelihood ratios in clinical decision making.2,3 A less emphasized attribute of likelihood ratios, however, is their role in a dynamic approach to clinical diagnosis.Sensitivity and specifi city
Sensitivity (the proportion of patients with a disease who have positive test results)1 and spe- cificity (the proportion of patients without the disease who have negative test results) are fami- liar to family physicians.1,2 If testing is intended to establish or rule out a particular diagnosis, these measures can help physicians greatly. A negative test result for a highly sensitive question will rule out the diagnosis, and obtaining a positive test result from a highly specifi c test will establish the diagnosis.
Unfortunately, some clinicians seek clinical evi- dence to replace their uncertainty with complete certainty.4 Typically, evidence for such physicians is something (a test result) that entirely proves or disproves a particular diagnosis. This approach, however, seems rarely, if ever, applicable in family medicine. Uncertainty is an ineradicable part of cli- nical decision making.5 In reality, the total number
of clinical tests with 100% sensitivity or specifi city is very low, and their application in routine clinical practice poses a variety of problems.6
Diagnoses are frequently proved or disproved based on sensitivity or specifi city of a single test.
With this outlook, using tests with sensitivity or specifi city nearing 100% becomes logical. Tests with lower sensitivity and specifi city can appear to be less eff ective. Moreover, getting a positive test result from a sensitive test (or a negative result from a specifi c one) does not change the pretest position of a clinician; this is why these parame- ters are considered static. We need a tool that can facilitate a dynamic diagnosis to assemble the information derived from a constellation of clini- cal tests.
Likelihood ratios
Likelihood ratios (LRs) are measures that express the relative likelihood that a given test result would be expected in a patient with (as opposed to one without) a condition.7 For ins- tance, in the scenario below, a history of can- cer is 14.7 times more likely to be found in a patient with a spinal neoplasm.8 Likelihood ratios for positive test results are derived from the equation sensitivity/1-specificity; L Rs for negative test results are derived from the equa- tion 1-sensitivity/specificity.1,2 For instance, LRs
College Collège
VOL 5: NOVEMBER • NOVEMBRE 2005dCanadian Family Physician • Le Médecin de famille canadien 1579 for a positive history of cancer are calculated as
0.31/1-0.98 = 14.7. Physicians can multiply their pretest estimate of presence of the disease by this measure to reach a posttest estimate.7
Evidence-based approaches convey the presence of uncertainty in every clinical decision in terms of “probabilistic reasoning.”2,3 Using this approach, we cannot (and really do not need to) completely prove or disprove a diagnosis. Rather, we usually intend to refi ne our estimate of the probability of a disease in order to decrease our uncertainty and refi ne our position regarding a predefi ned diagnos- tic threshold.
It is essential to consider results of each test in clinical practice within the context of informa- tion obtained from other tests. Likelihood ratios are dynamic because they can be easily used in a sequence of tests.1,3 Using LRs, clinicians can reestimate the probability of a disease according to the integration of history, physical examination, and laboratory results.
Clinical scenario
You are consulted by a 62-year-old man who has had severe back pain for 3 months. As he has a history of cancer, he is worried about recurrence.
His weight has remained stable, and results of x- ray examination of the lumbar and thoracic spine were normal. If you know that the prevalence of cancer in patients with low back pain is about 0.7%, how would you approach this patient? Would you recommend magnetic resonance imaging?
(Diagnostic properties of characteristics and tests are summarized in Table 1.8)
Resolution of clinical scenario
Using the sensitivity and specifi city of the given tests, diagnosis of cancer could not be proved or disproved. Using positive and negative LRs, however, gives a combined LR of 37.1 (by multi- plying 2.7, 14.7, 0.9, 2.6, and 0.4). Th is number can change the position of the pretest estimate from 0.7% to about 20%; that cannot be ignored. Th us, the patient requires further evaluation, including magnetic resonance imaging.
Summary
The evidence-based diagnostic approach appre- ciates physicians’ inability to eliminate uncertain- ties in clinical medicine. Likelihood ratios can be a dynamic tool for refi ning probability in clinical settings. Family physicians can use these parame- ters for better estimation of the strength of clinical evidence.
Dr Soltani is an Assistant Professor in Endocrinology and Dr Moayyeri is a Research Fellow on the
Evidence-Based Medicine Working Team of the Endocrinology and Metabolism Research Center in the Shariati Hospital at the Tehran University of Medical Sciences in Iran.
References
1. Sackett DL, Haynes RB, Guyatt GH, Tugwell P. Clinical epidemiology: a basic science for cli- nical medicine. 2nd ed. Boston, Mass: Little, Brown, and Co; 1991.
2. Sackett DL, Straus SE, Richardson WS, Rosenberg W, Haynes RB. Evidence based medicine:
how to practice and teach EBM. 2nd ed. Edinburgh, UK: Churchill Livingstone; 2000.
3. Guyatt G, Rennie D. Users’ guides to the medical literature: a manual for evidence-based practice. Chicago, Ill: American Medical Association Press; 2002.
4. Laupacis A, Sekar N, Stiell IG. Clinical prediction rules. A review and suggested modifi ca- tions of methodological standards. JAMA 1997;277:488-94.
5. Djulbegovic B. Acknowledgment of uncertainty: a fundamental means to ensure scientifi c and ethical validity in clinical research. Curr Oncol Rep 2001;3:389-95.
6. Hawkins DM, Garrett JA, Stephenson B. Some issues in resolution of diagnostic tests using an imperfect gold standard. Stat Med 2001;20:1987-2001.
7. Deeks JJ, Altman DG. Diagnostic tests; 4: likelihood ratios. BMJ 2004;329:168-9.
8. Jarvik JG, Dayo RA. Diagnostic evaluation of low back pain with emphasis on imaging. Ann Intern Med 2002;137:586-97.
Table 1. Characteristics of patients and properties of various tests for diagnosing cancer
PATIENT CHARACTERISTICS AND
PROPERTIES OF TESTS SENSITIVITY (%) SPECIFICITY (%)
POSITIVE LIKELIHOOD
RATIO
NEGATIVE LIKELIHOOD
RATIO
Age > 50 y 77 71.0 2.7 0.32
History of cancer 31 98.0 14.7 0.70
Unexplained weight loss 15 94.0 2.7 0.90
Duration of pain > 1 mo 50 81.0 2.6 0.62
Plain radiography 60 99.5 120 0.40
Magnetic resonance imaging 93 97.0 31 0.07
Values are adapted from Jarvik and Dayo.8
