Saddlepoint approximations to score test statistics in logistic regression for analyzing genome-wide association...

Area of Science:

• Genetics
• Biostatistics
• Statistical genetics

Background:

• Normal approximation of the score test statistic in logistic regression can be inaccurate, particularly with imbalanced response data and low minor allele counts.
• Accurate statistical testing is crucial for genome-wide association studies (GWAS) to identify genetic variants associated with diseases.

Purpose of the Study:

• To evaluate the accuracy of saddlepoint approximations for tail probabilities of the score test statistic in logistic regression for GWAS.
• To compare the performance of double saddlepoint methods against single saddlepoint procedures for calculating p-values.

Main Methods:

• Investigated saddlepoint approximations for score test statistic tail probabilities in logistic regression.
• Utilized exact results for a simple logistic regression model and simulations for models with nuisance parameters.
• Compared double saddlepoint methods (for two-sided and mid-p-values) with a single saddlepoint method.

Main Results:

• Saddlepoint approximations significantly improve the accuracy of tail probability calculations compared to normal approximations.
• Improved accuracy was observed even for extreme tail probabilities, which are critical in GWAS.
• The study compared different saddlepoint computational methods on simulated and real-world UK Biobank data.

Conclusions:

• Saddlepoint approximations offer a substantial improvement in accuracy for score test statistics in logistic regression for GWAS.
• These methods are particularly beneficial when dealing with imbalanced data or rare variants.
• The findings support the use of saddlepoint methods for more reliable genetic association studies.