0411 Separating within- and between-group exposure effects in a panel study on pesticide use and early biological effects in the Corn Farmers study
该研究通过分组均值中心化方法，分离了农民个体内和组间（暴露组与对照组）的农药暴露效应，发现组间差异主导了效应估计，为面板数据分析提供了方法学参考。
<h3>Objectives</h3> We aimed to estimate the effect of pesticides on selected early biological effects among farmers, allowing for different effects of within-person and between-group (unexposed controls versus farmers) changes over time. Using a group-level estimate of exposure is a well-known approach to reduce impact of measurement error on estimated exposure-response relations. With only few exposure groups this results in an ecological study design, with potential for “aggregation” bias. By group-mean centering of individually assigned exposures it is possible to separately estimate within-individual and between-group exposure effects <h3>Method</h3> Pesticide exposure information, blood and urine were collected throughout a growing season from male corn farmers (n = 30), and non-farming controls (n = 10). We used a hierarchical mixed model to relate change in cumulative exposure to atrazine and 2.4-D from before to during the spraying season to plasma levels of 22 immuno-modulatory cytokines and compared this to a more conventional model using individual exposures only. <h3>Results</h3> Model fit for group-mean centred models averaged better than for non-centred models (lower AICs for 20/22 models for atrazine and 18/22 models for 2.4-D). Estimates for between-group differences in exposure were very similar to those from conventional (non-centred) models, while standard errors for estimates based on within-individual differences were relatively large (5x those for estimates based on between-group differences). <h3>Conclusions</h3> Group-mean centering of exposures allowed us to estimate and contrast exposure-response relations based on differences between groups and within individuals. Comparison with a more conventional approach, ignoring the clustering of individuals, showed that effect estimates were dominated by differences between groups.