Analysis of binary responses with outcome-specific misclassification probability in genome-wide association studies

Area of Science:

• Genetics
• Biostatistics

Background:

• Binary response traits in human, animal, and plant studies frequently exhibit errors.
• Diagnostic tests with varying sensitivity and specificity lead to differential misclassification between cases and controls, causing false positives and negatives.
• Misclassification in binary responses reduces the statistical power of genome-wide association studies (GWAS).

Purpose of the Study:

• To investigate a threshold model designed to accommodate varying diagnostic error rates between cases and controls.
• To assess the impact of ignoring misclassification on the analysis of binary traits in GWAS.
• To evaluate the performance of a proposed method in identifying true genetic effects and correcting misclassified data.

Main Methods:

• A simulation study was conducted using binary case-control data sets with 2000 individuals each.
• Varying effects for influential single nucleotide polymorphisms (SNPs) and differential diagnostic error rates for cases and controls were simulated.
• A threshold model accommodating varying misclassification was applied and compared to analyses ignoring misclassification.

Main Results:

• Ignoring misclassification led to biased estimates of true SNP effects and inflated estimates for non-influential markers.
• The proposed method demonstrated a substantial reduction in bias and an increase in accuracy, ranging from 12% to 32%.
• The method successfully identified influential SNPs missed in analyses of noisy data and accurately flagged truly misclassified records.

Conclusions:

• The proposed threshold model effectively addresses varying diagnostic errors in binary traits.
• This approach significantly improves the accuracy and statistical power of genome-wide association studies by correcting for misclassification.
• The model's robustness across different simulation parameters highlights its utility in genetic association studies with noisy binary data.