Objective: The purpose of this study was to estimate the optimal performance of cervicography. We compared an arbitrated cervigram classification with an arbitrated referent diagnosis of cervical neoplasia.
Study design: From an initial group of 8460 women, a stratified sample of cervigrams from 3645 women and histologic information from 414 women underwent arbitration. Interobserver agreement was assessed for cervicography and the referent diagnosis. Sensitivity, specificity, and predictive values were estimated for initial and arbitrated cervicography results, compared with the initial and arbitrated referent diagnoses.
Results: For the detection of arbitrated high-grade lesions or cancer, arbitrated cervicography yielded an overall sensitivity of 63.9% and a specificity of 93.7%. Significantly higher sensitivity was associated with younger age and age-related visual characteristics.
Conclusion: Optimization of the cervigram classification improved performance over a single interpretation in this population but suggested the limits of static visual screening.