210 Attenuation of exposure effects over time: a simulation study
通过模拟动态队列，评估了易感人群耗竭和测量误差对暴露-反应趋势随时间衰减的影响，发现生物学通路饱和或经典测量误差更可能导致明显衰减。
<h3>Background</h3> Positive exposure-response trends often diminish at higher cumulative exposure, correlated with longer follow-up time. Depletion of susceptibles, increased measurement error at higher cumulative exposure, and saturation of biological pathways, have all been postulated as reasons for attenuation. <h3>Methods</h3> We conducted simulations to evaluate rate ratios over time under different assumptions about susceptibility to exposure effects and measurement error; we evaluated exposure-response trends to determine whether attenuation was evident. We simulated a dynamic cohort in which entry occurred over time; the metric of interest was duration of exposure. We also considered cross-sectional analyses in which follow-up started only after a certain point of time. Simulations considered 10,000 subjects enrolled from 1940–2010 and followed through 2020. Ten simulations were conducted for each scenario and exposure-response parameters averaged. An excess relative risk model was used to generate the relationship between duration of exposure and disease, controlling for age. Measurement error of both classical and Berkson type were simulated, with increasing error with increasing exposure. Cox regression was used to evaluate exposure-response trends. <h3>Results</h3> Under all scenarios considered with less than 100% susceptibility among the exposed, there was evidence of depletion of susceptibles over follow-up time. However, under realistic scenarios considered here, there was only modest evidence of attenuation of a linear exposure-response trend due to depletion of susceptibles. Classical measurement error, but not Berkson error, produced attenuation. Cross-sectional analyses did not dramatically change attenuation patterns. <h3>Conclusions</h3> Marked attenuation of exposure-response trends over follow-up time is more likely due saturation of biological pathways or, perhaps less likely, to classical measurement error - than to either the depletion of susceptibles or Berkson measurement error.