Development of Method for Estimating Diplotype Associations with Linear or Binomial Outcomes

Purpose We developed a simulation method to account for phase uncertainty in associations between outcomes and diplotypes imputed from genotypes. Methods Participants were 582 black and 402 white premenopausal women (age 35–49 years) randomly selected from an urban health plan and enrolled in the NIEHS Uterine Fibroid Study. Thirty IGF-I (insulin-like growth factor-I) and 15 IGFBP-3 (IGF binding protein-3) single nucleotide polymorphisms (SNPs) were genotyped based on previous research, functional significance, or haplotype-tagging properties. Diplotypes were estimated using PHASE version 2.1, which reports posterior probabilities reflecting uncertainty of individual diplotype assignments. Our approach uses information on all possible diplotypes from women with data for at least 50% of the SNPs in each gene. Using SAS V9.1, we simulated the study population separately by race 100,000 times and 1) randomly assigned diplotypes to each woman according to her posterior probability of having each possible diplotype; 2) used linear regression to model associations between diplotypes and plasma IGF-I and IGFBP-3 levels; 3) used binary regression to estimate prevalence differences for uterine fibroids; and 4) drew a random variable from a normal distribution to simulate ordinary sampling error. Simulation results are summarized by the mean and the 2.5th and 97.5th percentiles of the simulated distributions. Results Preliminary results for IGFBP-3 confirm the feasibility of the method, which allowed us to retain about 10% of observations that ordinarily would be excluded because of low posterior probabilities. Conclusion This approach uses complete diplotype information, reduces the number of excluded observations, and can be used to estimate associations with both binary and continuous outcomes. It can be applied to any exposure for which potential misclassification (probability of exposure) can be estimated for each individual.