Dr Faiz Shakeel,Dr Muhammad Fawad ul Hassan, Dr Aemun Javed
Many studies have demonstrated that trained observers are more likely to a radiograph abnormal when it is reviewed in a specially assembled sample with high disease prevalence than when the same film is interpreted as part of a group with lower disease prevalence. At present there are very limited previously described context bias studies. The diagnostic sensitivity in the 2 groups was markedly higher for 3 readers in group A but unchanged in the other 3 readers. The mean±SE of sensitivity was significantly higher for group A than for group. Among the 6 radiologists, diagnostic specificity was higher in group for 3, lower for 2, and unchanged for 1 reader. The mean of specificity was not remarkably different between the groups. The potential effects of con¬ text bias, and other biases that may be recognized in the future, cannot be eliminated by randomization or adjusted with simple post hoc computations, and, therefore, the effects should be contemplated or dealt with when an experiment is designed. To get unbiased estimates of diagnostic test accuracy for new and existing technologies, radiologists and clinicians should recognize that, despite the popularity of such quantitative measures as sensitivity, specificity, and likelihood ratios, their derivation and clinical significance still remain underexplored and sometimes dubious.