Reliability of retrospectively assessed categorical menstrual cycle length data

Prospective diaries have been shown to be more reliable than retrospective questionnaires for the assessment of menstrual cycle length collected as continuous data. We investigated the reliability of retrospectively collected categorical cycle length data.

In 2005, 67 women from Illinois and Vermont completed a menstrual diary and retrospective questionnaire. Based on commonly used cutpoints for normal cycle length, we created 2-, 3- and 4-category cycle length variables, calculated percent agreement, Cohen's kappa and prevalence index and investigated factors associated with accurate reporting.

Overall agreement between diary and questionnaire responses was highest for the 2-category cycle length variable, <=35 vs. > 35 days (93%), and the 3-category variable, <25, 25-35 and > 35 days (82%). While Cohen's kappa fell below 0.6, its magnitude was limited by a high prevalence of "normal"" cycle length. Women with regular cycles were slightly more likely to have concordant responses.

In epidemiologic studies where cycle length categories are of interest, the limitations of prospective diaries in terms of cost, compliance, attrition and power must be carefully weighed against the potentially decreased reliability of questionnaire data.