Estimating correlations from epidemiological data in the presence of measurement error.