Integrating dynamic mixed-effect modelling and penalized regression to explore genetic association with pharmacokinetics

CONTEXT: In a previous work, we have shown that penalized regression approaches can allow many genetic variants to be incorporated into sophisticated pharmacokinetic (PK) models in a way that is both computationally and statistically efficient. The phenotypes were the individual model parameter estimates, obtained a posteriori of the model fit and known to be sensitive to the study design. OBJECTIVE: The aim of this study was to propose an integrated approach in which genetic effect sizes are estimated simultaneously with the PK model parameters, which should improve the estimate precision and reduce sensitivity to study design. METHODS: A total of 200 data sets were simulated under the null and each of the following three alternative scenarios: (i) a phase II study with N=300 participants and n=6 sampling times, wherein six unobserved causal variants affect the drug elimination clearance; (ii) the addition of participants with a residual concentration collected in clinical routine (N=300, n=6 plus N=700, n=1); and (iii) a phase II study (N=300, n=6) in which four unobserved causal variants affect two different model parameters. RESULTS: In all scenarios the integrated approach detected fewer false positives. In scenario (i), true-positive rates were low and the stepwise procedure outperformed the integrated approach. In scenario (ii), approaches performed similarly and rates were higher. In scenario (iii), the integrated approach outperformed the stepwise procedure. CONCLUSION: A PK phase II study with N=300 lacks the power to detect genetic effects on PK using genetic arrays. Our approach can simultaneously analyse phase II and clinical routine data and identify when genetic variants affect multiple PK parameters.