Informatively empty clusters with application to multigenerational studies.

Area of Science:

• Biostatistics
• Epidemiology
• Public Health

Background:

• Multigenerational studies are crucial for public health, assessing exposures across generations.
• These studies face challenges with informative cluster size, where family size affects outcomes.
• The impact of 'empty clusters' (mothers with no children) on analysis is unknown.

Purpose of the Study:

• Evaluate standard methods for informative cluster size with zero cluster sizes.
• Assess the bias introduced by unacknowledged empty clusters in multigenerational analyses.
• Propose and validate a novel joint marginalized approach to handle empty clusters.

Main Methods:

• Evaluated standard methods for informative cluster size, allowing for zero cluster sizes.
• Compared joint models of outcome and size, estimating equation approaches, and a proposed joint marginalized model.
• Utilized simulation studies and analyzed data from the Nurses Health Study on diethylstilbestrol exposure and ADHD risk.

Main Results:

• Standard methods yield biased estimates when informative cluster size includes empty clusters.
• Joint models require explicit inclusion of empty clusters for valid inference.
• Estimating equation approaches inherently omit empty clusters, leading to biased marginal effect estimates.

Conclusions:

• The proposed joint marginalized approach effectively incorporates empty clusters.
• This method offers more intuitive population-averaged effect interpretations than existing methods.
• Accurate analysis of multigenerational studies requires explicit consideration of all cluster sizes, including zero.