The Use of Joint Models in Analysis of Aggregate Endpoints in RERF Cohort Studies

Area of Science:

• Radiation epidemiology
• Biostatistics
• Cancer research

Background:

• Radiation risk estimation commonly uses Poisson regression on aggregate endpoints.
• Current methods require a surrogate organ dose, which may be problematic for diverse endpoints.
• Aggregate endpoints combine various organ-specific events, necessitating robust analytical approaches.

Purpose of the Study:

• To compare the inference from standard aggregate endpoint analysis with joint analysis methods.
• To evaluate the impact of using surrogate versus organ-specific radiation doses.
• To address potential biases in radiation risk estimation for aggregate endpoints.

Main Methods:

• Utilized maximum likelihood theory to compare analytical approaches.
• Performed a re-analysis of the Radiation Effects Research Foundation (RERF) Life Span Study all cancer mortality data.
• Investigated the use of single surrogate organ doses versus organ-specific doses within joint analysis frameworks.

Main Results:

• Standard aggregate endpoint analysis and joint analysis with a single surrogate dose yield nearly identical inference for radiation dose response, background, and effect modification.
• Using organ-specific doses in joint analysis can introduce bias in intercept parameters, particularly when sex-specific outcomes are included and dose estimates are sex-undefined.
• A simple correction can mitigate bias when using organ-specific doses with sex-specific outcomes.

Conclusions:

• For aggregate endpoints, joint analysis with a single surrogate dose is an appropriate and effective method for radiation risk estimation.
• Joint analysis with organ-specific doses may be interpretable for aggregate endpoints only if endpoints are analyzed separately for dose-response estimation.
• Inference about dose response remains complex and requires careful consideration for both aggregate and organ-specific analyses.