Methods. As part of a larger clinical trial, 1850 patients were recruited into a diagnostic group or a screening group depending on their history of abnormal Papanicolaou smear results. For all patients, colposcopic examination was performed and biopsies were obtained from abnormal and normal colposcopic sites. The gold standard of test result accuracy was the histologic report of biopsies.

Results. The diagnostic group had a 29.0% prevalence of high-grade squamous intraepithelial lesions (HGSIL) or cancer, and the screening group had a 2.2% prevalence of HGSIL or cancer. Using a disease threshold of HGSIL or worse, we found that in the diagnostic group, colposcopy had a sensitivity of 0.982 and a specificity of 0.451 when the test threshold was LGSIL and a sensitivity of 0.714 and a specificity of 0.813 when the test threshold was HGSIL. Using the same disease threshold, in the screening group, colposcopy had a sensitivity of 0.286 and a specificity of 0.877 when the test threshold was LGSIL and a sensitivity of 0.191 and a specificity of 0.961 when the test threshold was HGSIL. The areas under the receiver operating characteristic (ROC) curves for the diagnostic and screening groups at the disease threshold of HGSIL or worse were 0.821 (95% CI 0.79, 0.85) and 0.587 (95% CI 0.56, 0.62), respectively. Changing the disease threshold to LGSIL or worse demonstrated similar tradeoffs of sensitivity and specificity. However, the overall performance of colposcopy using a LGSIL or worse disease threshold demonstrated inferior performance, with area under the ROC curve of 0.776 (95% CI 0.74, 0.91) for diagnostic patients and 0.577 (95% CI 0.55, 0.61) for screening patients.

Conclusions. We demonstrated that colposcopy performs well in the diagnostic setting and poorly in the screening setting. Colposcopy should not be used for the screening of cervical intraepithelial neoplasia.