Reducing Misclassification Bias in Cervical Dysplasia Risk Factor Analysis With p16-Based Diagnoses

    loading  Checking for direct PDF access through Ovid



Conventional hematoxylin and eosin (HE)–based diagnoses have been the reference standard for cervical cancer risk factor analyses. However, this HE-based method is known to have modest interobserver reproducibility and only moderate predictive value. In contrast, more recent immunohistochemical-based diagnoses using the neoplastic marker p16 are known to improve diagnostic accuracy. Our objective was to test whether p16-based diagnoses would significantly affect high-grade dysplasia (cervical intraepithelial neoplasia 2+) risk factor analysis compared with the current reference standard (HE).

Materials and Methods

Retrospective cohort of 500 index cases were randomly selected from a series of more than 5,000 cervical biopsies performed at Kaiser Permanente Northwest from 1997 to 2003 after a patient’s first abnormal cervical Pap smear (positive for atypical squamous cells of undetermined significance). Subjects were subsequently excluded if they did not have at least 5 years of clinical follow-up, including cervical biopsies, or 3 reproducibly negative Pap smears. This yielded 358 cases for risk factor analysis. The index biopsies and all follow-up biopsies were immunostained for p16 and the proliferation marker Ki-67, which were then independently reviewed by 2 pathologists blinded to clinical outcomes. Data were analyzed by χ2 test and logistic regression modeling.


We observed clinically significant diagnostic errors in 22% of index biopsies. Improved accuracy using p16 strengthened the risk estimate of low family income for cervical intraepithelial neoplasia 2+ (odds ratio = 1.71, 95% confidence interval = 1.09–2.63) compared with HE-based diagnoses (odds ratio = 1.12, 95% confidence interval = 0.72–1.72). The addition of Ki-67 staining did not significantly influence these results.


p16-based diagnoses may affect the power of risk factor analysis, especially when using small cohorts.

Related Topics

    loading  Loading Related Articles