Reproducibility of skin characteristic measurements and reported sun exposure history

Background The aim of the present study is to investigate the reproducibility of information on sun exposure, skin characteristics and sunburn collected through a standardized questionnaire used in the multi-centre South European case-control study on skin cancer, 'Helios'. We also intended to use results from reproducibility analysis for correcting odds ratio (OR) estimates from the original study. Methods We re-interviewed, with the same questionnaire, a sample of 115 cases of basal cell and squamous cell carcinoma of the skin and 119 population controls, 18-26 months apart, in four centres of Italy, Spain and France. The questionnaire included questions on skin characteristics, sunburns and sun exposure histories. We investigated agreement, studying the association between the difference of the two measures and a set of explanatory variables. According to the results of the reproducibility analysis we corrected OR estimates from the original study simultaneously adjusting for random measurement error. Results Hair and eye colour showed high agreement with intra-class correlation coefficients (ICC) of 0.81 and 0.74, respectively. Lifetime sun exposure showed substantial agreement with ICC ranging from 0.68 for time spent doing outdoor work to 0.79 for time spent outdoors during holidays and holidays at the beach. The poorest agreement was found for number of lifetime sunburns (ICC = 0.25), while sunburns during childhood showed a substantial agreement (Cohen's Kappa = 0.67). Lack of reproducibility was mainly associated with subjects' education, while no significant differences were observed between cases and controls. The corrected OR showed a moderate increase with a reinforcement of the effect of sun exposure and skin reaction to sun exposure. Conclusion Overall, there was good reproducibility, particularly in the case of sun exposure histories, between answers given on two different occasions to a questionnaire administered by trained interviewers. However, since measurement error can substantially bias OR toward the null value, it should be taken into account in estimates of the effect of sun exposure on risk of skin cancer.