| Adverse drug reactions (ADRs) pose a serious public health burden, yet the role of genetics in drug response remains incompletely characterized. Thiazide diuretics, commonly used anti-hypertensives, may cause QT interval (QT) prolongation, a major drug development barrier that increases risk for highly fatal and difficult to predict ventricular arrhythmias. We thus examined whether common SNPs modified the association between thiazide use (17% mean prevalence) and QT or its component parts (QRS interval, JT interval) by performing ancestry-specific, trans-ethnic, and cross-phenotype genome-wide analyses of European (66%), African American (15%), and Hispanic (19%) populations (N-78,199). Analyses leveraged longitudinal data, incorporated corrected standard errors to account for underestimation of interaction estimate variances, and evaluated evidence for pathway enrichment. Although no loci achieved genome-side significance (P<5x10-8), we found suggestive evidence (P<5x10-6) for SNPs modifying the thiazide-QT association at 22 loci, including biologically plausible ion transport loci (e.g. NELL1, KCNQ3).;Given our highly plausible, but only suggestive findings and our observational cohort setting, we next examined the influence of prevalent user bias and exposure misclassification on pharmacogenomics studies conducted in observational settings. Specifically, we simulated three study designs (longitudinal, cross-sectional, new user), two control groups (whole cohort, active comparator), and two scenarios (extreme or modest drug effects) to enable comparison of 12 settings. For each setting, we simulated N=120,000 participants, conducted 10,000 iterations, applied an alpha=5x10-8, and introduced varying degrees of prevalent user bias and drug exposure misclassification. When large drug effects (>10 ms change in QT) or exposure misclassification were present, drug-SNP interaction estimates were biased (bias range: 0.02--3.4 ms) across settings. Under no settings did power to detect the drug-SNP interaction estimate exceed 80% for effects less than 2 ms; detection of drug effects below 2 ms required a longitudinal design with at least 150,000 participants. Results from this dissertation suggest that despite leveraging longitudinal data in 78,199 participants, our study was likely underpowered to detect modest or clinically significant pharmacogenomics effects on QT. Future pharmacogenomics efforts will require even larger sample sizes and innovative methods to enable prevention of ADRs in the large and increasingly at-risk population exposed to medications. |
