Statistical power considerations in genotype-based recall randomized controlled trials

Abstract: Validation of gene × environment interaction results from epidemiological studies in randomized clinical trials (RCTs) is usually hampered by insufficient statistical power. Genotype-based recall (GBR) refers to the design of studies in which subgroups of the population are selected based on their genotypes. In this thesis we test if a GBR approach of recruiting two groups of participants with distinct genetic profiles can yield sufficiently higher statistical power with reduced sample sizes, leading to a reduction of experimental costs, compared with conventional RCTs of comparable sample size. To this end, we modelled GBR sampling, with participants chosen from the two extremes of a genetic risk score distribution, and compare this with conventional ”random” sampling. We performed power calculations using simulations in the R program using assumptions from the Diabetes Prevention Program. We have calculated the required sample sizes to reach sufficient statistical power when analyzing the interaction between a genetic factor and Intensive Lifestyle Intervention in a linear regression model, with 1-year small low-density lipoprotein (LDL) particles level as the outcome. Similarly, the statistical power for the interaction effect between a genetic factor and metformin treatment was simulated in a Cox proportional hazards regression model, with time to developing type 2 diabetes as the outcome. Statistical power for interactions in various scenarios including different effect sizes, allele frequencies, initial sampling frames and error rates were also examined for both types of regression model. Results from almost all simulations confirm that GBR is more powerful than conventional when the object is to detect gene × environment interactions in both the linear regression and Cox regression models. Last but not least, an online application for statistical power calculations has also been developed by this author and made available to the community.