The accuracy of colposcopic biopsy: analyses from the placebo arm of the Gardasil clinical trials

Abstract

We evaluated the overall agreement between colposcopically directed biopsies and the definitive excisional specimens within the context of three clinical trials. A total of 737 women aged 16-45 who had a cervical biopsy taken within 6 months before their definitive therapy were included. Per-protocol, colposcopists were to also obtain a representative cervical biopsy immediately before definitive therapy. Using adjudicated histological diagnoses, the initial biopsies and the same day biopsies were correlated with the surgically excised specimens. The overall agreement between the biopsies taken within 6 months of definitive therapy, and the definitive therapy diagnoses was 42% (weighted kappa = 0.34) (95% CI: 0.29-0.39). The overall underestimation of cervical intraepithelial neoplasia grade 2/3 or adenocarcinoma in situ (CIN2-3/AIS) and CIN3/AIS was 26 and 42%, respectively. When allowing for one degree of variance in the correlation, the overall agreement was 92% for CIN2-3/AIS. The overall agreement between the same day biopsy and definitive therapy specimen was 56% (weighted kappa = 0.41) (95% CI: 0.36-0.47), and the underestimation of CIN2-3/AIS was 57%. There were significant associations in the agreement between biopsies and excisional specimen diagnoses when patients were stratified by age, number of biopsies, lesion size, presence of human papillomavirus (HPV)16/18 and region. Of 178 diagnostic endocervical curettages performed, 14 (7.9%) found any HPV disease. Colposcopic accuracy improved when CIN2 and CIN3/AIS were grouped as a single predictive measure of high-grade disease. Colposcopy functioned well when allowed a one-degree difference between the biopsy and the surgical histologic interpretations, as done in clinical practice. Taking more than one biopsy improved colposcopic accuracy and could improve patient management.