Misclassification of sex: Assessing an automated sex edit and the example of male breast cancer

Background: Breast cancer (BC) incidence is increasing at a statistically significant faster rate among Florida males compared to US males. Epidemiological studies of male BC in Florida could advance etiologic knowledge, but these data must first be evaluated for validity. Large central cancer registries routinely utilize sex-specific automated edits and other computerized processes to ensure data integrity. However, <15% of cancers are sex-specific. Therefore, the effectiveness of a newly-developed algorithm for identifying miscoded sex, based on name and decade of birth, was investigated. Methods: The algorithm was tested against four cancer sites plus a dataset of BC patients reported as male but identified female after follow-up. The impact of misclassification of sex on cancer rates in Florida was evaluated. Results: For tested sites, the algorithm agreed with reported sex on 68% of cases, 31% could not be assessed, and 0.5% were flagged as incorrect. Results differed by subpopulation. For example, the percents of unassessable cases were higher for Hispanics and nonwhites compared to whites. And a dramatic 21% of male BC cases were identified as incorrect sex. When testing against the dataset of “false male” BC cases, the algorithm correctly identified 81% as female, 19% were unassessed, and none were incorrectly identified as male. Correcting the data resulted in a significant decline in male BC rates. Implications: Clinically significant differences in Florida male BC rates compared to US are unlikely. Although less sensitive for non-whites/ Hispanics, applying the algorithm as an edit improves quality of cancer registry sex data.