Mixed Model Association with Family-Biased Case-Control Ascertainment

Area of Science:

• Population Genetics
• Statistical Genetics
• Genetic Epidemiology

Background:

• Standard mixed models are widely used for genetic association studies but can be unreliable with family sampling bias and case-control ascertainment.
• Previous methods like the liability threshold-based mixed model association statistic (LTMLM) addressed ascertainment in unrelated samples.
• Family-biased case-control ascertainment, where related individuals are non-randomly selected, significantly impacts heritability estimates and mixed model association statistics.

Purpose of the Study:

• To introduce a novel family-based association statistic (LT-Fam) robust to family-biased case-control ascertainment.
• To evaluate the calibration and power of LT-Fam compared to existing methods under family sampling bias.
• To demonstrate the importance of accounting for family sampling bias in genetic association studies using related samples.

Main Methods:

• Developed the LT-Fam statistic, computed from posterior mean liabilities (PML) under a liability threshold model.
• LT-Fam incorporates published narrow-sense heritability estimates to ensure correct calibration.
• Simulations and analysis of the CARe cohort (type 2 diabetes) with induced family-biased ascertainment were performed to compare LT-Fam with the Armitage trend test (ATT), standard mixed model association (MLM), and case-control retrospective association test (CARAT).

Main Results:

• LT-Fam demonstrated correct calibration in simulations (average χ² = 1.00-1.02 for null SNPs) and in the CARe cohort.
• In contrast, ATT, MLM, and CARAT showed mis-calibration in both simulations and the CARe cohort under family-biased ascertainment.
• LT-Fam achieved comparable or higher statistical power than other methods in certain scenarios.

Conclusions:

• The proposed LT-Fam statistic provides accurate genetic association analysis in the presence of family sampling bias.
• Standard mixed models and other tested methods are susceptible to mis-calibration when family sampling bias is present.
• Accounting for family sampling bias is crucial for reliable genetic association studies utilizing case-control datasets with related individuals.