Simulation–Extrapolation for Bias Correction with Exposure Uncertainty in Radiation Risk Analysis Utilizing Grouped Data
Abstract
Summary In observational epidemiological studies, the exposure that is received by an individual often cannot be precisely observed, resulting in measurement error, and a common approach to dealing with measurement error is regression calibration (RC). Use of RC, which requires assumptions about the distribution of unknown error-free (true) variables, leads to concern about the possibility of bias due to misspecification of that distribution. The simulation–extrapolation (SIMEX) method, in contrast, does not require a distributional assumption. However, analyses of large cohorts may be performed by using grouped or person-year data, and application of SIMEX to grouped data is not straightforward, particularly when there is a mixture of classical and Berkson measurement errors. We compared RC and SIMEX with grouped data analyses to assess robustness of the RC method to misspecification of the true dose distribution. We also applied SIMEX assuming mixtures of classical and Berkson errors and compared the results with those obtained by using RC for classical error only. SIMEX had less bias than RC and performed well regardless of the true dose distribution, whereas RC based on a misspecified true dose distribution showed greater bias than when based on the correctly specified true dose distribution.
- Authors:
- Publication Date:
- Sponsoring Org.:
- USDOE
- OSTI Identifier:
- 1958665
- Alternate Identifier(s):
- OSTI ID: 1401313
- Grant/Contract Number:
- HS0000031; DE‐HS0000031
- Resource Type:
- Published Article
- Journal Name:
- Journal of the Royal Statistical Society, Series C: Applied Statistics
- Additional Journal Information:
- Journal Name: Journal of the Royal Statistical Society, Series C: Applied Statistics Journal Volume: 67 Journal Issue: 1; Journal ID: ISSN 0035-9254
- Publisher:
- Oxford University Press
- Country of Publication:
- United Kingdom
- Language:
- English
Citation Formats
Misumi, Munechika, Furukawa, Kyoji, Cologne, John B., and Cullings, Harry M. Simulation–Extrapolation for Bias Correction with Exposure Uncertainty in Radiation Risk Analysis Utilizing Grouped Data. United Kingdom: N. p., 2017.
Web. doi:10.1111/rssc.12225.
Misumi, Munechika, Furukawa, Kyoji, Cologne, John B., & Cullings, Harry M. Simulation–Extrapolation for Bias Correction with Exposure Uncertainty in Radiation Risk Analysis Utilizing Grouped Data. United Kingdom. https://doi.org/10.1111/rssc.12225
Misumi, Munechika, Furukawa, Kyoji, Cologne, John B., and Cullings, Harry M. Wed .
"Simulation–Extrapolation for Bias Correction with Exposure Uncertainty in Radiation Risk Analysis Utilizing Grouped Data". United Kingdom. https://doi.org/10.1111/rssc.12225.
@article{osti_1958665,
title = {Simulation–Extrapolation for Bias Correction with Exposure Uncertainty in Radiation Risk Analysis Utilizing Grouped Data},
author = {Misumi, Munechika and Furukawa, Kyoji and Cologne, John B. and Cullings, Harry M.},
abstractNote = {Summary In observational epidemiological studies, the exposure that is received by an individual often cannot be precisely observed, resulting in measurement error, and a common approach to dealing with measurement error is regression calibration (RC). Use of RC, which requires assumptions about the distribution of unknown error-free (true) variables, leads to concern about the possibility of bias due to misspecification of that distribution. The simulation–extrapolation (SIMEX) method, in contrast, does not require a distributional assumption. However, analyses of large cohorts may be performed by using grouped or person-year data, and application of SIMEX to grouped data is not straightforward, particularly when there is a mixture of classical and Berkson measurement errors. We compared RC and SIMEX with grouped data analyses to assess robustness of the RC method to misspecification of the true dose distribution. We also applied SIMEX assuming mixtures of classical and Berkson errors and compared the results with those obtained by using RC for classical error only. SIMEX had less bias than RC and performed well regardless of the true dose distribution, whereas RC based on a misspecified true dose distribution showed greater bias than when based on the correctly specified true dose distribution.},
doi = {10.1111/rssc.12225},
journal = {Journal of the Royal Statistical Society, Series C: Applied Statistics},
number = 1,
volume = 67,
place = {United Kingdom},
year = {Wed Apr 19 00:00:00 EDT 2017},
month = {Wed Apr 19 00:00:00 EDT 2017}
}
https://doi.org/10.1111/rssc.12225