A cautionary note on using secondary phenotypes in neuroimaging genetic studies

Area of Science:

• Genetics
• Neuroimaging
• Biostatistics

Background:

• Genome-wide association studies (GWASs) often employ case-control designs, which can lead to biased sampling.
• Standard logistic regression for primary disease traits is robust to this bias, but linear regression for secondary phenotypes may yield biased results.

Purpose of the Study:

• To investigate the validity of standard linear regression for secondary phenotypes using Alzheimer's Disease Neuroimaging Initiative (ADNI) data, despite known biases in case-control sampling.
• To assess whether published studies using ADNI secondary phenotypes have been affected by potential analytical problems.

Main Methods:

• Analysis of real ADNI data.
• Simulation studies to model the impact of biased sampling on secondary phenotype association testing.
• Illustration of specialized methods for valid secondary phenotype analysis.

Main Results:

• Standard linear regression analysis of secondary phenotypes in the ADNI dataset demonstrated generally valid results, with only minor biases and slightly inflated Type I errors.
• The findings contrast with the expected theoretical biases in genetic epidemiology for secondary phenotypes from case-control data.

Conclusions:

• Standard linear regression for secondary phenotypes appears valid for the ADNI dataset, contrary to general expectations.
• Caution is advised when applying these methods to other datasets, and specialized methods should be considered.