Premium
Evaluating measurement error in estimates of worker exposure assessed in parallel by personal and biological monitoring
Author(s) -
Symanski Elaine,
Greeson Nicole M.H.,
Chan Wenyaw
Publication year - 2007
Publication title -
american journal of industrial medicine
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.7
H-Index - 104
eISSN - 1097-0274
pISSN - 0271-3586
DOI - 10.1002/ajim.20422
Subject(s) - statistics , observational error , medicine , correlation , standard error , exposure assessment , multiple exposure , correlation coefficient , random error , econometrics , mathematics , computer science , geometry , computer vision
Background While studies indicate that the attenuating effects of imperfectly measured exposure can be substantial, they have not had the requisite data to compare methods of assessing exposure for the same individuals monitored over common time periods. Methods We examined measurement error in multiple exposure measures collected in parallel on 32 groups of workers. Random‐effects models were applied under both compound symmetric and exponential correlation structures. Estimates of the within‐ and between‐worker variances were used to contrast the attenuation bias in an exposure‐response relationship that would be expected using an individual‐based exposure assessment for different exposure measures on the basis of the intra‐class correlation coefficient (ICC). Results ICC estimates ranged widely, indicative of a great deal of measurement error in some exposure measures while others contained very little. There was generally less attenuation in the biomarker data as compared to measurements obtained by personal sampling and, among biomarkers, for those with longer half‐lives. The interval ICC estimates were oftentimes wide, suggesting a fair amount of imprecision in the point estimates. Ignoring serial correlation tended to over estimate the ICC values. Conclusions Although personal sampling results were typically characterized by more intra‐individual variability than inter‐individual variability when compared to biological measurements, both types of data provided examples of exposure measures fraught with error. Our results also indicated substantial imprecision in the estimates of exposure measurement error, suggesting that greater emphasis needs to be given to studies that collect sufficient data to better characterize the attenuating effects of an error‐prone exposure measure. Am. J. Ind. Med. 50:112–121, 2007. © 2006 Wiley‐Liss, Inc.