Effects of bias on the results of diagnostic studies of carpal tunnel syndrome.

PURPOSE: Design weaknesses in studies of diagnostic tests can lead to biases that produce misleading results. The extent to which such biases exist in studies of diagnostic tests for carpal tunnel syndrome (CTS) is unknown, and no studies address whether these biases influence published estimates of the performance of CTS diagnostic tests. The present study investigates these issues. METHODS: We conducted a systematic review of studies of tests used to diagnose CTS in adults. Whether a study had design-related biases was determined using the Quality Assessment of Diagnostic Accuracy Studies instrument, by determining whether a study had a case control design, and by considering each study's level of evidence. We used meta-regression to determine whether any of these factors influenced estimates of the sensitivity, specificity, and diagnostic odds ratio of diagnostic tests. RESULTS: A total of 23 studies of 129 diagnostic test evaluations met our inclusion criteria. Of those, 87% (20 of 23 studies) contained at least 1 bias and, on average, each study contained 2 biases. The most common design-related bias was spectrum bias (61% of studies [14 of 23 studies]), which was perfectly correlated with use of a case control design. Spectrum bias was associated with a statistically significant overestimation of sensitivity, specificity, and the diagnostic odds ratio. Test review bias (which was clearly present in 35% of studies [8 of 23 studies]) was associated with statistically significant overestimation of sensitivity and the diagnostic odds ratio. These results were not influenced by whether the diagnostic test under study was a clinical or an electrodiagnostic test. CONCLUSIONS: Well-designed studies of diagnostics for CTS are uncommon, and shortcomings in the design of studies of clinical and electrodiagnostic tests are associated with overestimates of the performance of these tests. Consequently, most articles on diagnostic tests for CTS are unlikely to report results that are applicable to actual clinical practice.