Redefining the Gold Standard for Cervical Dysplasia Diagnosis in Risk Factor Analysis.
Emily E King, Michelle Berlin, Tomi Mori, Robert Krum, Terry K Morgan. OHSU, Portland, OR; Kaiser-Permanente Northwest, Portland, OR
Background: Diagnosing cervical dysplasia by conventional hematoxylin & eosin (H&E)-based methods is limited by poor interobserver reproducibility and significant inaccuracy, especially in cases of moderate dysplasia (CIN2). An unrecognized consequence of this problem may be effects on cervical dysplasia risk factor analysis. New molecular tools such as immunostaining for p16 appear to improve diagnostic agreement and predictive value; therefore, we hypothesize that combining p16 and H&E-based diagnoses will significantly change understanding of the relationship between risk factors and cervical dysplasia.
Design: Routine H&E, p16, and Ki67 stained slides were prepared from 252 random colposcopic biopsies obtained from Kaiser Northwest, which had at least five years of clinical follow-up (including negative serial Pap tests or surgical excision). Two expert gynecologic pathologists independently reviewed each case blinded to all clinical data. Diagnoses were reported based on H&E only, H&E and p16, and H&E, p16 and Ki67 after sufficient "wash out periods." Kappa statistic and test accuracy were calculated for each method relative to the gold standard based on clinical follow-up. Odds of at least CIN2 were determined in multivariate logistic regression models using a constant panel of risk factors (eg, age, family income, education level) while varying the diagnostic method.
Results: H&E plus p16-based diagnoses of CIN2 were more reproducible and accurate than H&E only, consistent with recent reports (see table). Ki67 staining was not justified. Low family income was only significantly associated with at least CIN2 if using p16 assisted diagnoses (p=0.02 vs. p=0.61). There was a trend toward significant change in the odds of at least CIN2 if the patient had low family income in regression analysis only if using p16 assisted diagnoses (p=0.057).
|H&E and p16||0.48||95%||94%||100%|