Validating the use of census data on education as a measure of socioeconomic status in an occupational cohort

Purpose Adjusting for smoking status or a reliable surrogate [such as socioeconomic status (SES)] is critically important in occupational epidemiology studies when any smoking-related cancer or cardiovascular disease is an outcome of interest. Sometimes, however, data on smoking patterns or individual-level smoking surrogates such as job title, education, pay scale or other measures of SES are not readily available in occupational cohorts. Methods and materials To obtain a surrogate measure for missing smoking or individual-level SES data, we demonstrate a method used to obtain and geocode residential address histories which were then linked to area-level SES measures from the United States Census in three test samples and then in a full cohort of workers from the Mound nuclear weapons facility in Dayton, Ohio, USA. The mean educational attainment of the Census Block Group was used to derive a categorical estimate of educational attainment which was compared to self-reported (SR) education available from Mound worker histories using Kappa statistics. Lung cancer mortality patterns between area-derived (AD) and SR education were investigated using Standardized Mortality Ratios (SMR) and Cox Proportional Hazards models with stratification or adjustment by either SR or AD education. Results Home address histories were obtained from linkages of individual worker data to online resources. In the test cohorts, mean educational attainment was the Census Block Group measure found to have the largest magnitude association with individual-level SES measures. Among 7251 Mound workers, 5685 (78.4%) had at least one residential address match (mean 4.9 addresses) identified. The SR and AD educational attainment measures were highly correlated (weighted Kappa coefficient 0.10, p < .0001). SMR patterns by SR and AD educational attainment were similar, with steadily decreasing mortality with increased educational attainment by either measure. Cox models for lung cancer using AD education produced similar results as those using SR education as an adjustment factor. Conclusion When individual-level SES indicators are not available for statistical adjustment, area-level SES measures can serve as a reliable surrogate when investigating outcomes that are affected by lifestyle factors such as smoking.