0313 An assessment of the impact of misclassification error on an estimate of effect for occupational exposure to TCDD-contaminated chemicals and ischaemic heart disease0313 An assessment of the impact of misclassification error on an estimate of effect for occupational exposure to TCDD-contaminated chemicals and ischaemic heart disease
通过蒙特卡洛模拟校正暴露和疾病错误分类，发现校正后的比值比中位数高于观察值，提示传统分析可能低估了TCDD暴露与缺血性心脏病的关联。
<h3>Objectives</h3> To quantify the effect of exposure and disease misclassification on the odds ratio (OR) for ischaemic heart disease (IHD) mortality in a retrospective cohort study of trichlorophenol workers exposed to 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD). <h3>Method</h3> Exposure and disease classification parameters were defined using specificity and sensitivity values either reported in the peer-reviewed literature or calculated from available published data. The distribution shapes for each parameter were constructed based on expert judgment and were varied to address the lack of published information on the parameter distributions. Probabilistic uncertainty analysis, which uses Monte Carlo simulation techniques, was then used to sample each parameter distribution, calculating ORs corrected for outcome and exposure misclassification. <h3>Results</h3> Correction for exposure and disease misclassification produced lognormal probability distributions of ORs for IHD with a wider range of possible values than calculated in the traditional epidemiological analysis. The median OR<sub>corrected</sub> was larger than the OR<sub>observed </sub>of 3.05 for all seven scenarios evaluated and ranged from 5.34 to 11.86. Between 70% and 99% of the simulation trials yielded corrected ORs greater than the OR<sub>observed</sub>. <h3>Conclusions</h3> The application of uncertainty analysis to a mortality study of workers occupationally exposed to TCDD-contaminated chemicals provides valuable insight into the magnitude and direction of misclassification error and the impact on an estimate of effect. Further refinement of the parameter distributions and adjustment of the OR<sub>observed</sub> for other study limitations will be necessary to determine whether a true causal relationship between exposure and disease exists or if the effect observed in these workers is an artefact of systematic error.