Generalized case-control sampling under generalized linear models.

Area of Science:

• Biostatistics
• Epidemiology
• Statistical Modeling

Background:

• Standard case-control studies use outcome-dependent sampling (ODS) for binary responses.
• Generalized case-control (GCC) studies extend ODS to nonbinary outcomes, posing analytical challenges.
• Existing GCC analyses often rely on parametric generalized linear models (GLMs), which are sensitive to model misspecification.

Purpose of the Study:

• To develop a novel, unifying, and robust approach for analyzing GCC study data.
• To extend semiparametric GLM methodology to accommodate ODS in nonbinary response settings.
• To provide a more flexible and reliable alternative to existing parametric GLM-based GCC analyses.

Main Methods:

• Developed a semiparametric extension of the generalized linear model (GLM) for GCC data.
• Employed a conditional likelihood approach to address biased sampling inherent in ODS designs.
• Investigated estimation and inference procedures under the conditional likelihood framework, including misspecification scenarios.

Main Results:

• The proposed semiparametric GLM approach demonstrates superior robustness to model misspecification compared to parametric methods.
• Extensive simulation studies confirmed the flexibility and validity of the new methodology.
• Application to the Asset and Health Dynamics Among the Oldest Old study showcased practical utility.

Conclusions:

• The novel semiparametric GLM provides a versatile and robust solution for analyzing GCC data with various response distributions and sampling schemes.
• This approach offers improved statistical validity and flexibility for researchers utilizing ODS in epidemiological and health studies.
• The methodology addresses key limitations of existing parametric approaches, enhancing the reliability of findings from complex study designs.