RESUMO
OBJECTIVE: To review the published diagnostic accuracy figures for the performance of colposcopy and to assess how the various forms of bias might explain the very wide range of reported values and the impact they have on quality assurance of cervical screening. METHODS: Publications were only selected where they contained sufficient raw data to enable diagnostic accuracy statistics to be calculated for the detection of cervical intraepithelial neoplasia grade 2+ (CIN2+), as determined by punch biopsy. In addition, both the colposcopic impression at the time of examination and the disease threshold used to determine the need for biopsy must have been reported. RESULTS: Large differences in diagnostic accuracy figures were found when the output of colposcopy was defined either, on the basis that the colposcopist thought there was CIN2+ present or, that the colposcopist considered there to be some disease present and so took a biopsy to confirm this. Weighted mean sensitivity was 68.5% (95% CI 59.9-77.1) for the first method but 95.7% (95% CI 93.4-98.0) for the second method. Weighted mean specificity was 75.9% (95% CI 69.3-82.5) for the first method but 34.2% (95% CI 27.0-41.4) for the second method. Weighted mean PPV was 68.9% (95% CI 64.2-73.6) for the first method but 54.3% (95% CI 46.5-62.1) for the second method. CONCLUSION: The main reason for the wide range of published diagnostic accuracy figures, arises from the use of two different methods of assessing the output of colposcopy. Colposcopic Impression is appropriate when assessing the performance of a colposcopist at the time of examination, but the taking of a biopsy to confirm that Disease is Present should be used when assessing patient management. Accurate assessment of both outcomes is fundamental to any quality assurance programme.