Most of the reviewed articles also failed to ensure against test review bias. This bias occurs when a lack of suitable blinding of the reviewers who interpret the evaluated test or the reference standard results prevents an independent assessment. Test review bias is especially serious in investigations wherein the reference standard requires subjective interpretation and prior knowledge of the results of the new diagnostic test would influence the reference standard results. Our study showed that only 37% of the study articles incorporated appropriate blinding in their study designs. Retrospective cohort studies are especially prone to test review bias because investigators design the study after the results of the evaluated diagnostic test and the reference standard have already been interpreted add comment buy claritin online. It would be unlikely, therefore, that test results in a retrospective study would be independently evaluated. In our review of study articles, we found that 27% utilized a retrospective cohort design, which would be expected to inflate the measured diagnostic accuracies of the tests.
Workup bias (or ‘Verification” bias) appeared to be a potential problem for many of the study articles. This bias occurs when patients with some results of the new diagnostic test are not equally likely to be examined by the reference standard. Workup bias can affect investigations wherein some study patients with negative results of the test do not undergo the reference standard evaluation and are assumed to be free of disease. This form of bias inflates the measured sensitivity and specificity of the new diagnostic test. Workup bias can also result when patients with negative test results do not undergo evaluation by the reference standard and are excluded from further analysis, which increases sensitivity but decreases specificity. Articles were given credit in our review for fulfilling the workup bias standard if they (1) compensated for workup bias by submitting patients, who were not examined by the reference standard, to long-term follow-up to limit the false-positive rate, or (2) provided post hoc mathematical corrections. Despite these allowances, only 63% of studies were designed appropriately to limit the effects of workup bias.