Login / Signup

Pharmacogenetic Dose Modeling Based on CYP2C19 Allelic Phenotypes.

Julia Carolin StinglJason RadermacherJustyna WozniakRoberto Viviani
Published in: Pharmaceutics (2022)
Pharmacogenetic variability in drug metabolism leads to patient vulnerability to side effects and to therapeutic failure. Our purpose was to introduce a systematic statistical methodology to estimate quantitative dose adjustments based on pharmacokinetic differences in pharmacogenetic subgroups, addressing the concerns of sparse data, incomplete information on phenotypic groups, and heterogeneity of study design. Data on psychotropic drugs metabolized by the cytochrome P450 enzyme CYP2C19 were used as a case study. CYP2C19 activity scores were estimated, while statistically assessing the influence of methodological differences between studies, and used to estimate dose adjustments in genotypic groups. Modeling effects of activity scores in each substance as a population led to prudential predictions of adjustments when few data were available ('shrinkage'). The best results were obtained with the regularized horseshoe, an innovative Bayesian approach to estimate coefficients viewed as a sample from two populations. This approach was compared to modeling the population of substance as normally distributed, to a more traditional "fixed effects" approach, and to dose adjustments based on weighted means, as in current practice. Modeling strategies were able to assess the influence of study parameters and deliver adjustment levels when necessary, extrapolated to all phenotype groups, as well as their level of uncertainty. In addition, the horseshoe reacted sensitively to small study sizes, and provided conservative estimates of required adjustments.
Keyphrases
  • electronic health record
  • big data
  • primary care
  • healthcare
  • climate change
  • case report
  • single cell
  • machine learning
  • artificial intelligence
  • drug induced
  • deep learning
  • adverse drug