IfADo--Leibniz Research Centre for Working Environment and Human Factors, Ardeystrasse 67, Dortmund, Germany.
Toxicol Lett. 2011 Oct 10;206(2):144-51. doi: 10.1016/j.toxlet.2011.07.003. Epub 2011 Jul 7.
Meta-analyses of individual participant data (IPD) provide important contributions to toxicological risk assessments. However, comparability of individual data cannot be taken for granted when information from different studies has to be summarized. By means of statistical standardization approaches the comparability of data might be increased. An analysis of individual data on the neurobehavioral impact of manganese (Mn) exemplifies challenges and effects of a multilevel statistical procedure. Confounding from individual-level and study-level covariates was shown by analyses of variance, but could be reduced by linear regressions and z-normalization using data of the respective control groups. Fixed models that were used to estimate the impact of the neurotoxic exposure, provided evidence that the employed procedures, especially the z-normalization, effectively reduced variance that was unrelated to the neurotoxic exposure. Even after this statistical treatment the fixed effect models revealed differences among studies that did not seem to be exhaustively explicable by concentration differences obvious from the Mn biomarker at hand. IPD studies using confounded endpoints as effects markers can be reasonably summarized when appropriate statistical operations are employed. For the data at hand the proposed normalization allowed new insights into exposure-effect relationships, in general it appears appropriate to investigate the effect of the independent variable more closely.
个体参与者数据（IPD）的荟萃分析为毒理学风险评估做出了重要贡献。然而，当必须总结来自不同研究的数据时，不能想当然地认为个体数据具有可比性。通过统计标准化方法，可以提高数据的可比性。对锰（Mn）的神经行为影响的个体数据进行分析，举例说明了多层次统计程序的挑战和效果。方差分析表明了个体水平和研究水平协变量的混杂，但可以通过线性回归和使用各自对照组的数据进行 z 归一化来减少。用于估计神经毒性暴露影响的固定模型提供了证据，表明所采用的程序，特别是 z 归一化，有效地减少了与神经毒性暴露无关的方差。即使经过这种统计处理，固定效应模型也揭示了研究之间的差异，这些差异似乎无法完全用手头的 Mn 生物标志物明显的浓度差异来解释。当使用适当的统计操作时，可以合理地总结使用混杂终点作为效应标志物的 IPD 研究。对于手头的数据，建议的归一化允许对暴露-效应关系有新的认识，通常，更仔细地研究自变量的效果是合适的。
