Reproducibility of reported farming activities and pesticide use among breast cancer cases and controls. A comparison of two modes of data collection.

PURPOSE: Farming is associated with exposure to many potential hazards including pesticides and other agents, but the quality of self-reported data on farm exposures has not been well studied.
METHODS: The reproducibility of self-reported farming history was evaluated among women in a population-based, case-control study of breast cancer in North Carolina. Thirty cases and 31 controls were randomly re-interviewed by telephone an average of 13.8 months after the initial interview. The initial interview was based on a farm-by-farm questionnaire, while the repeat interview was based on a shorter ever/never questionnaire. Agreement was estimated using proportions in exact agreement, kappa (kappa), and intraclass correlation coefficients (ICC).
RESULTS: In general, group prevalences and means were higher on re-interview. Kappa estimates ranged from 0.15 to 0.84 among cases, and 0.26 to 0.87 among controls, with most estimates falling between 0.5 and 0.8. Moderate to almost perfect agreement (kappa) was observed for questions on crop work (0.47-0.70), crop type (0.56-0.82), pesticide application to tobacco (0.77), and farm residence (0.84). ICC estimates for continuous variables showed fair to substantial agreement (0.30 to 0.69 among cases, 0.38 to 0.69 among controls). Older cases, less educated cases, cases who lived on more than one farm, and cases with longer time intervals between interviews gave lower total agreement than similar groups of controls.
CONCLUSIONS: Agreement estimates in this study are similar to those for other types of exposure information typically collected in epidemiologic studies. Nevertheless, a farm-by-farm method of exposure assessment may be preferable to an ever/never determination.