Rheumatology Advance Access originally published online on April 4, 2006
Rheumatology 2006 45(6):779-780; doi:10.1093/rheumatology/kel083
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
LETTER TO THE EDITOR |
Positive and negative predictive values from published studies can be misleading for decision-making in clinical practice
Charité Campus Benjamin Franklin, Medizinische Klinik I, Rheumatologie, Hindenburgdamm 30, 12200 Berlin, Germany
Correspondence to: M. Rudwaleit. E-mail: martin.rudwaleit{at}charite.de
SIR, There are few laboratory tests which may be of help in diagnosing ankylosing spondylitis (AS) [1]. With interest we read the article by Duftner and co-workers on the presence and mean concentrations of antibodies cross-reacting with a purified 28 kDa Drosophila antigen in patients with AS, patients with other inflammatory conditions and healthy controls [2]. The authors concluded that a positive test result has a high predictive value, thus offering an additional tool for further diagnostic decision-making. Although this antibody test may in fact have some additional diagnostic value in AS and other spondylarthropathies, we strongly disagree with the conclusion that this test is of diagnostic relevance because of a high positive predictive value (PPV), as is suggested by the title, the abstract and throughout the discussion. We disagree because the PPV is highly dependent on the test setting, in particular on the prevalence of the disease (pretest probability) in the test setting, and therefore is subject to change. A high PPV in one setting can drop dramatically in another setting, although both the sensitivity and the specificity of the test do not change!
Using a cut-off level of 
75 U/ml, the authors found 30.7% of 371 AS patients to be positive for this antibody cross-reacting with the 28 kDa antigen, whereas only 3 of 37 (8.1%) healthy controls were positive, resulting in a PPV of 97.4%. The PPV for a cut-off level of 60 U/ml was 95.7% (sensitivity 36.1%, specificity 83.8%).
The high PPV reported by the authors mainly resulted from the high prevalence of the disease in their study population (90.9%) who underwent the test. If instead of 37 controls, 370 healthy controls had been included, for example, the PPV for a cut-off level 75 U/ml would have dropped from 97.4 to 79.1%, applying the same sensitivity (30.7%) and specificity (91.9%) to the test.
The authors suggest using this test as a diagnostic tool in patients with suspected AS. If this test is being applied to all patients with chronic back pain in primary care, no more than 5% of such patients can be expected to have AS or early AS according to Underwood and Dawes [3]. If we consider that the 371 AS patients investigated by the authors constitute the 5% of patients with chronic back pain who have AS, a total of 7049 patients with non-inflammatory (mechanical) back pain would had been tested for this antibody as well (constituting 95% of all patients with chronic back pain; total population, 7420). Assuming that for a cut-off level of 75 U/ml the frequency of a positive test result (8.1%) is the same in patients with non-inflammatory back pain as in healthy controls, 571 of 7049 patients with non-inflammatory back pain can be expected to be positive, whereas the percentage of AS patients with a positive test result would again be 30.7% (114 out of 371 positive). In this setting of 371 AS patients and 7049 controls (5% prevalence of AS), the PPV would be only 16.6%, and the PPV would be 10.5% if a cut-off level 60 U/ml had been chosen.
Such low PPVs may tell the reader that this test is not useful at all. Moreover, the resulting negative predictive values (NPV) of 96.2 and 96.1%, respectively, in this setting would suggest that in clinical practice a negative test result might be more helpful in ruling out AS than would a positive test result in ruling in AS. However, ruling out the disease by a negative test result is not appropriate either, as correctly discussed by the authors, because 64–70% of AS patients are negative for this antibody!
The authors are rather vague as to the question of which population of patients should be tested—all patients with chronic back pain or only patients with inflammatory back pain? The latter suggestion can only be found in the key messages at the end of the paper and is not discussed elsewhere in the paper. If only patients with inflammatory back pain (IBP) were tested (sensitivity and specificity of IBP in AS are both around 75%), the PPV calculated by us was 91.4% and the NPV was 30.6%. However, 93 of 371 AS patients (25%) would not have undergone testing for this antibody because around 25% of AS patients had not fulfilled criteria for IBP [1, 4, 5].
The examples above illustrate the heavy dependence of positive and negative predictive values on the prevalence of the disease within the population tested, i.e. on the number of subjects tested in each group (patients and controls), although the sensitivity and specificity of the test that is applied stay absolutely the same. Thus, a high PPV, such as 95%, by no means implies that this test is helpful in daily practice. Following this line of argument, PPVs and NPVs are of highly limited value and are best avoided since they can strongly mislead the clinician reader.
Instead, the likelihood ratio (LR) is a much better description of the value of any diagnostic test [6, 7]. The LR is much more robust since the LR does not depend on the number of patients or controls tested, and thus does not depend on the prevalence of the disease. Using a cut-off level of 75 U/ml, the antibody test cross-reacting with the 28 kDa Drosophila antigen had a positive (LR +) of 3.8, which is comparable to the diagnostic value of the presence of IBP [1, 5], enthesitis of the heel, or peripheral arthritis [1, 4]. The LR+ of 2.2 for a positive test result using a lower cut-off level of 60 U/ml implies a relatively small diagnostic gain [8], which, in the case of suspected AS, is comparable to the diagnostic gain of an elevated ESR or CRP [1].
We agree with the authors that the test for antibodies cross-reacting with 28 kDa Drosophila antigen may be of additional diagnostic value in suspected AS, assuming a LR+ of 3.8 if the high cut-off level of >75 U/ml is chosen. Pending results from future studies on sensitivity in early disease and on specificity, this antibody may expand the diagnostic laboratory armamentarium in early AS, which is currently limited to ESR/CRP (LR+ 2.5) and HLA-B27 (LR+ 9.0) [1].
The authors have declared no conflicts of interest.
References
- Rudwaleit M, van der Heijde D, Khan MA, Braun J, Sieper J. How to diagnose axial spondyloarthritis early. Ann Rheum Dis 2004;63:535–43.
[Abstract/Free Full Text] - Duftner C, Dejaco C, Klauser A, Falkenbach A, Lakomek HJ, Schirmer M. High positive predictive value of specific antibodies cross-reacting with a 28-kDa Drosophila antigen for diagnosis of ankylosing spondylitis. Rheumatology 2006;45:38–42.
[Abstract/Free Full Text] - Underwood MR, Dawes P. Inflammatory back pain in primary care. Br J Rheumatol 1995;34:1074–7.
[Abstract/Free Full Text] - Rudwaleit M, Khan MA, Sieper J. The challenge of diagnosis and classification in early ankylosing spondylitis: do we need new criteria? Arthritis Rheum 2005;52:1000–8.[CrossRef][Web of Science][Medline]
- Rudwaleit M, Metter A, Listing J, Sieper J, Braun J. Inflammatory back pain in ankylosing spondylitis – a reassessment of the clinical history for screening and for diagnosis. Arthritis Rheum 2006;54:569–78.[CrossRef][Web of Science][Medline]
- Sackett DL, Haynes RB, Guyatt GH, Tugwell P. Clinical epidemiology: a basic science for clinical medicine, 2nd edn. Boston: Little, Brown, 1991;69–152.
- Grimes DA, Schulz KF. Refining clinical diagnosis with likelihood ratios. Lancet 2005;365:1500–5.[CrossRef][Web of Science][Medline]
- Jaeschke R, Guyatt GH, Sackett DL. Users’ guides to the medical literature. III. How to use an article about a diagnostic test. B. What are the results and will they help me in caring for my patients? The Evidence-Based Medicine Working Group. JAMA 1994;271:703–7.
[Abstract/Free Full Text]
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||