How Pharmacometrics can improve the power of pharmacogenetic studies

Aims: Pharmacogenetics (PG) studies the relationship between genetic variation and drug response with the aim of providing more precision to the development of personalized medicine. These last years, PG studies have for example identified the abacavir hypersensitivity syndrome association with the HLA-B*5701, HLA-DR7, and HLA-DQ3 variants (1) and the hepatitis C treatment response association with the rs12979860 variant, upstream of the IL28 gene (2). However the impact of PG studies in terms of treatment recommendations is still limited due to low power of studies and lack of precision in the estimation of the association effect size (3). Tessier et al. have shown how using nonlinear mixed effect modeling compared to non-compartmental analysis improved the power of PG tests in pharmacokinetic studies (4). The present work further develops how pharmacometrics can enhance the classic statistical approach in PG.
Methods: Data sets were simulated under the null and several alternative hypotheses. Genetic polymorphisms were simulated approximating the design of the DMET chip (5) with about 1200 genetic polymorphisms across the whole genome. Pharmacokinetic (PK) profiles were simulated using a two-compartment model. Both rich and sparse designs were investigated.
Genetic association was investigated through : i) a classic stepwise linear regression and two types of penalized regressions on the empirical Bayes estimates (EBEs) of the individual PK parameters with the hlasso program (6), ii) a classic stepwise covariate selection with saemix (7), iii) an integrated approach with a penalized regression embedded in the population parameters estimation step using saemix and hlasso and iv) Bayesian inference through informative prior distributions on the genetic effect sizes with Stan (8) and variable selection with JAGs (9).

Results: On EBEs all approaches show similar power, but penalized regression can be much less computationally demanding. Penalized regression should be preferred over stepwise procedures for PK analyses with a large panel of genetic covariates. In all scenarios, the integrated approach detected fewer false positives. A PK phase II study with 300 subjects lacks the power to detect genetic effects on PK using genetic arrays. The integrated approach can simultaneously analyse phase II and clinical routine data and identify when genetic variants affect multiple PK parameters.

Conclusion: Pharmacometrics allows statisticians to analyse together clinical data from multiple studies with different sampling designs. This is crucial as increasing the sample size is necessary to achieve sufficient power to detect realistic and clinically relevant PG effects.

1. Mallal S. et al. Association between presence of HLA-B*5701, HLA-DR7, and HLA-DQ3 and hypersensitivity to HIV-1 reverse-transcriptase inhibitor abacavir. Lancet. 2002;359(9308):727-32.
2. Ge et al. Genetic variation in IL28B predicts hepatitis C treatment-induced viral clearance. Nature. 2009;461(7262):399-401.
3. Holmes et al. Fulfilling the promise of personalized medicine? Systematic review and field synopsis of pharmacogenetic studies. PloS One. 2009;4(12):e7960
4. Tessier A et al. Comparison of Nonlinear Mixed Effects Models and Noncompartmental Approaches in Detecting Pharmacogenetic Covariates. AAPS J. 2015;17(3):597-608
5. Daly TM et al. Multiplex assay for comprehensive genotyping of genes involved in drug metabolism, excretion, and transport. Clin Chem. 2007; 53:1222–1230.
6. Hoggart CJ et al. Simultaneous analysis of all SNPs in genome-wide and re-sequencing association studies. PLoS Genet 2008; 4:e1000130.
7. Comets E et al. SAEMIX, an R version of the SAEM algorithm. Athens, Greece: Population Approach Group in Europe; 2011.
8. Stan Development Team. 2016. The Stan C++ Library, Version 2.9.0.
9. Plummer M. 2003. JAGs: A program for analysis of Bayesian graphical models using Gibbs sampling.