Meeting Details
Date: 24-06-12
Time: 11:00-12:00
Location: Hybrid: Rm 6E032/34 and via Teams
Presenter: Dr. Mark Little
A new method of adjusting for correlated dose errors, or where there may be substantial curvature in dose response, and comparison with other methods of dose error correction
Mark P Little, Nobuyuki Hamada, Lydia B Zablotska
The problem of allowing for measurement error in exposure when estimating exposure-response relationships has been the subject of much interest in epidemiology. It is well recognized that measurement error can alter substantially the shape of this relationship and hence the derived population risk estimates. In particular, low-dose radiation risks are often extrapolated from groups exposed at much higher levels of dose, and this extrapolation can be adversely affected by errors in assessed exposures. Recently, there has been much attention paid to methods of dealing with shared errors in estimating exposures, which are common in many datasets, and particularly important in occupational and environmental settings.
Various methods of correcting for exposure error have been proposed. Regression calibration (RC) is easy to use, and has been frequently applied. Various full-likelihood methods such as Monte Carlo maximum likelihood (MCML) and Bayesian Markov Chain Monte Carlo (MCMC) have also been used, although these are computationally much more cumbersome. More recently frequentist model averaging (FMA) and 2D Monte Carlo with Bayesian model averaging (2DMC+BMA) methods have been used in simulated and real data where there are substantial inter-individual correlations in dose, and claims made that these outperform methods such as RC.
We test RC, MCML, FMA and 2DMC+BMA against a modification of RC, the extended regression calibration (ERC) method in various sets of simulated datasets (of leukemia after radiation exposure) with varying amounts of Berkson and classical error, both shared and unshared. 2DMC+BMA has poor coverage for the dose coefficients; coverage of RC, MCML and FMA is better, although uniformly too high for FMA and MCML, and best for ERC. Bias in predicted relative risk is generally smallest for ERC, and largest for quasi-2DMC+BMA and FMA methods, with RC and MCML exhibiting bias in predicted risk somewhat intermediate between the other two methods. In summary, the new ERC method leads to the lowest error in risk estimates using exposure estimates with shared errors in comparison to previously proposed methods and is particularly suited to situations with low to moderate amounts of shared and unshared Berkson errors.
Articles for Discussion:
- Little, M. P., Hamada, N., & Zablotska, L. B. (2023). A generalisation of the method of regression calibration. Scientific Reports, 13(1), 15127. https://doi.org/10.1038/s41598-023-42283-y
- Little, M. P., Hamada, N., & Zablotska, L. B. (2024). A generalisation of the method of regression calibration and comparison with Bayesian and frequentist model averaging methods. Scientific Reports, 14(1), 6613. https://doi.org/10.1038/s41598-024-56967-6
- NCRP Commentary No 34 2024 Recommendations on statistical approaches to account for dose uncertainties in radiation epidemiologic risk models.