Improving estimation and prediction in linear regression incorporating external information from an established...

Area of Science:

• Biostatistics
• Statistical modeling
• Biomarker research

Background:

• Utilizing historical linear regression data (coefficients and standard errors) from large studies.
• Integrating this summary information to enhance inference in expanded models with new variables.

Purpose of the Study:

• To improve statistical inference in a new dataset by incorporating historical regression data.
• To develop and evaluate methods for combining existing knowledge with new biomarker information.

Main Methods:

• Formulating historical data as nonlinear constraints on the parameter space.
• Proposing and comparing frequentist and Bayesian solutions.
• Employing a Bayesian transformation approach for approximate Bayesian inference.

Main Results:

• Historical information on E(Y|X) significantly improves estimation efficiency.
• Incorporating historical data enhances the predictive power of regression models (E(Y|X,B)).
• The Bayesian transformation method provides an effective computational approach.

Conclusions:

• Historical regression data can substantially improve statistical models with new biomarkers.
• The proposed methods offer a robust framework for data integration in statistical inference.
• This methodology has practical applications in refining existing prediction models, such as for bone lead levels.