Inclusion bias affects common variant discovery and replication in a health-system linked biobank

Area of Science:

• Genomics
• Biobanking
• Precision Medicine

Background:

• Electronic health record (EHR)-linked biobanks are crucial for precision medicine research.
• Opt-in consent models used by most biobanks may introduce participation and recruitment biases.
• The impact of these biases on genetic association studies is not well understood.

Purpose of the Study:

• To investigate sources of bias in EHR-linked biobanks.
• To evaluate the impact of participation bias on genetic analyses.
• To assess methods for mitigating bias in biobank research.

Main Methods:

• Utilized the UCLA ATLAS Community Health Initiative as a case study.
• Analyzed factors associated with biobank participation using EHR data.
• Applied inverse probability weighting to adjust for enrollment probabilities.
• Evaluated effects on genome-wide association studies (GWAS) and polygenic score phenome-wide association studies (PGS-PheWAS).

Main Results:

• Numerous factors, including healthcare utilization and sociodemographics, significantly influence biobank participation.
• EHR data effectively distinguishes biobank participants from the general healthcare population (AUROC=0.85, AUPRC=0.82).
• Bias adjustment using inverse probability weighting increased replication of known GWAS variants by 54% and affected PGS-PheWAS results.

Conclusions:

• Participation and recruitment bias can significantly affect genetic analyses in EHR-linked biobanks.
• Ad hoc analyses within healthcare systems are essential for identifying and potentially mitigating confounding factors.
• Addressing bias is critical for robust genetic discoveries and risk assessment in precision medicine.