Measurement errors in epidemiological studies of human papillomavirus and cervical cancer

Measurement errors have been an important concern in studies of human papillomavirus (HPV) and anogenital cancers. Misclassification of HPV infection status is a possible explanation for incoherent findings in previous epidemiological studies purporting to show an etiological role for HPV in cervical cancer. Even low levels of misclassification of HPV infection can cause severe underestimation of HPV prevalence in field surveys, bias the association between HPV and sexual activity, and impair the ability to control statistically the relation between sexual activity and neoplasia by viral status. The present report focuses on aspects that have not been elaborated in two previous studies (Kaldor, 1989; Franco, 1991). Emphasis is given to effects of measurement errors of HPV infection status in biasing the association with cervical neoplasia under more complex scenarios, e.g., differential misclassification between cases and controls in a case-control study, and non-differential misclassification of both viral infection and cervical neoplasia in cross-sectional cytology surveys and cohort studies. Some simple numeric formulae are given that allow the correction of prevalence rates and epidemiological measures of effect, such as the odds ratio and the relative risk, under the latter conditions of misclassification constraint. These formulae have been used to correct estimates from recent epidemiological studies using hypothetical misclassification scenarios in order to obtain clues on the magnitude of the underlying relationship between HPV and cervical cancer.