Knowing the accuracy of clinical tests in practice is useful to any clinician hoping to take an evidence-based approach to their practice, and the work of Astin and colleagues1 provides a useful summary on the performance of clinical tests used to diagnose colorectal cancer in primary care. However, I believe there are two shortcomings to their analysis.
In part of their analysis, the authors chose to aggregate the positive predictive values (PPV) of the tests. While it may be tempting to aggregate the PPVs of primary studies when they only report on the positive test results, this is potentially misleading. The problem is that it tells you nothing about the accuracy of the test or whether the test adds anything to the diagnostic process. Since the PPV depends on both the prior probability and the likelihood ratio of the test, a high PPV could be the result of high prior probability for colorectal cancer. For instance, the reported pooled estimate of the PPV for rectal bleeding was 8.1%. Feasibly, this could result from a prior probability of 8.1% for colorectal cancer and a likelihood ratio of 1, in which case rectal bleeding as a diagnostic test for colorectal cancer is useless and clinicians should avoid using this test. Alternatively, with a prior probability of 0.81%, a PPV of 8.1% would mean the test has a positive likelihood ratio of around 11, making it a very good test for clinicians to use, and one that clearly adds to the diagnostic process.
The second shortcoming relates to whether the sensitivity and specificity should be aggregated using univariate methods when there is potential for the two to be associated, not least due to a changing diagnostic test threshold. For this reason the Cochrane Diagnostic Test Accuracy Working Group recommend a bivariate approach when aggregating diagnostic test data,2 and it would have been interesting to see whether taking this more rigorous line had a material effect on the summary results.
- © British Journal of General Practice, January 2011
