Shared dosimetry error in epidemiological dose-response analyses.

Area of Science:

• Epidemiology
• Radiation Dosimetry
• Biostatistics

Background:

• Sophisticated radiation dose reconstruction systems are crucial for large-scale epidemiological studies.
• These systems often provide multiple dose realizations to represent dosimetry uncertainty.
• Handling this complex dosimetry data in analyses is a significant challenge.

Purpose of the Study:

• To derive expected scores and the information matrix for the linear excess relative risk (ERR) model, accounting for complex dosimetry systems.
• To address the bias introduced by ignoring dosimetry errors when using mean doses in epidemiological analyses.
• To propose methods for adjusting the information matrix to provide unbiased estimates and valid statistical tests.

Main Methods:

• Derivation of expected scores and information matrix for the linear ERR model.
• Utilizing multiple dose realizations from dosimetry systems (e.g., Mayak Worker Dosimetry System 2013).
• Developing bias adjustment methods for the information matrix in radiation epidemiology.

Main Results:

• Treating mean doses yields asymptotically unbiased estimates and valid null hypothesis tests.
• Ignoring dosimetry errors biases the information matrix and standard errors when the excess relative risk slope (β) is non-zero.
• The proposed adjustment method corrects information matrix bias using multiple dose realizations.

Conclusions:

• The developed methods allow for valid statistical inference in radiation epidemiology despite complex dosimetry uncertainty.
• Accurate adjustment of the information matrix is essential for reliable risk estimation in studies with dosimetry errors.
• These methods are applicable to large cohort studies, including the Mayak Worker Cohort and U.S. Atomic Veterans Study.