Objective: To estimate the reproducibility of the assessment of visual characteristics of cervical lesions used to judge lesion grade from colposcopy.
Methods: Digitized cervical images from 862 women enrolled in the ASCUS-LSIL Triage Study were obtained after application of 5% acetic acid. Each image was later assessed online by two randomly assigned evaluators from a pool of 20 experienced colposcopists. Interobserver agreement beyond chance was assessed by kappa statistics.
Results: Of 862 evaluable images with paired assessments, 607 were considered to have an acetowhite lesion by both evaluators, 171 by one, and 84 by neither. Kappa values (95% confidence intervals) for agreement were 0.22 (0.17-0.27) for color, 0.24 (0.18-0.30) for margins, 0.22 (0.16-0.29) for mosaicism, 0.17 (0.11-0.23) for punctation, 0.11 (0.00-0.22) for atypical vessels, and 0.26 (0.22-0.31) for modified Reid Index score.
Conclusion: Characteristics used to assign colposcopic grade are poorly reproducible when used to assess static cervical images from women with borderline cytology results.
Level of evidence: II.