Overcoming attenuation bias in regressions using polygenic indices

Area of Science:

• Genetics
• Biostatistics
• Quantitative genetics

Background:

• Measurement error in polygenic indices (PGIs) attenuates effect estimates in regression analyses.
• Addressing this bias is crucial for accurate genetic association studies.

Approach:

• Compares two methods: Obviously Related Instrumental Variables (ORIV) and PGI Repository Correction (PGI-RC).
• Utilizes simulations to evaluate performance under various conditions (sample size, assortative mating).
• Empirically validates findings using UK Biobank sibling data to predict educational attainment and height.

Key Points:

• PGI-RC generally shows slightly better performance than ORIV, except for very small sample sizes (N < 1000) or significant assortative mating.
• Within-family analyses favor ORIV due to the unavailability of PGI-RC correction factors.
• Empirical application in UK Biobank demonstrated ORIV increases PGI effect estimates compared to meta-analysis-based PGIs.

Conclusions:

• ORIV provides a robust method for estimating genetic effects within families, offering tighter lower bounds for direct genetic effects.
• Both ORIV and PGI-RC offer improvements over standard methods for handling measurement error in PGIs.
• The choice between ORIV and PGI-RC depends on sample characteristics and the specific research question.