Philip E. Castle*, Mark Schiffman, Cosette M. Wheeler, Nicolas Wentzensen, and Patti E. Gravitt
Initially submitted June 1, 2009; accepted for publication August 5, 2009.
Correspondence to Dr. Philip E. Castle, Division of Cancer Epidemiology and Genetics, National Cancer Institute, 6120 Executive Boulevard, Room 5004, MSC 7234, Bethesda, MD 20892-7234 (e-mail: firstname.lastname@example.org).
Misclassification of exposure and surrogate endpoints of disease can obscure causal relations. Using data from the Atypical Squamous Cells of Undetermined Significance/Low-Grade Squamous Intraepithelial Lesion Triage Study (ALTS, 1997–2001), the authors explored the impact of exposure (human papillomavirus (HPV) detection) and endpoint (histologic cervical precancer) classification on their mutual association. Women referred into this study with an atypical squamous cells of undetermined significance Papanicolaou test with satisfactory results for all 4 HPV tests were included in this analysis (n ¼ 3,215; 92.2%). HPV testing results were related to differentdefinitions of cervical precancer, based on paired, worst 2-year histologic diagnoses, by calculating clinical sensitivity, specificity, and odds ratios. The authors found that HPV test sensitivity increased and specificity decreased with increasing certainty of cervical precancer, with HPV testing having the highest sensitivity (92%–98%) and lowest specificity (46%–54%) for consensus cervical intraepithelial neoplasia grade 3 (CIN 3). The overall accuracy of each HPV test, as measured by odds ratios, was greatest for consensus CIN-3 diagnoses, from 2- to 4-fold greater than for a less stringent precancer definition of any diagnosis of CIN 2 or more severe. In summary, there was convergence of greater certainty of carcinogenic HPV with greater certainty of a precancerous diagnosis, such that all 4 HPV tests almost always tested positive in women most likely to have cervical precancer. Finding increasingly strong associations when both test and diagnostic misclassification are reduced is a useful sign of ‘‘true association’’ in molecular epidemiology.