Final PK model
A one compartment PK model with first order absorption/conversion and elimination with proportional residual error model provided the best fit to the observed serum oxypurinol-time data. Using a two-compartment PK model or other residual error models provided similar fits, so the simpler model was retained. Covariance between BSV forCL/fm and V/fm was tested but this resulted in similar BSV estimates; therefore, the covariance was not included.
Model development steps for the oxypurinol PK model are summarized inSupplemental Table 2 . The final model included FFM onCL/fm and V/fm allometric scaled using the theoretical value (0.75 forCL/fm and 1 for V/fm ), renal function using estimated CrCL, and SLC22A12rs505802C>T. Using TBW as a covariate onCL/fm improved the model fit but failed to improve the fit when used as covariate on V/fm . On the other hand, using either AJBW or FFM as a covariate on bothCL/fm and V/fm improved the fit. The selection of FFM as a covariate instead of AJBW was based on previous findings that FFM was also found to be a significant covariate.27,29 HMG-CoA reductase inhibitors and drugs that decrease SU reduced CL/fm by about 48% and 30%, respectively, but the effect was not statistically significant in the SCM step (Supplemental Table 2 ). In addition toSLC22A12 rs505802C>T as a covariate onCL/fm , CARMIL1 rs742132A>G and PDZK1 rs12129861G>A were found to be significant during the forward selection step but were excluded during backward elimination step. The combined absorption and formation rate constant (Kfm ) and its BSV were fixed to initial estimates (which is similar to the value, 0.92 h-1 reported in the literature47) due to insufficient data to support the parameter estimates and high shrinkage.48
The BSV in CL/fm decreased from 42.8% to 28.3%, and V/fm decreased from 40.7% to 32.4% after including significant covariates. The results of the base and final (including covariate) PK models are summarized in Table 2 , and the final estimates for CL/fm andV/fm for oxypurinol are given by:
\begin{equation} CL/f_{m}\ (L/h)=1.05\ (L/h)\times\left(\frac{standardized\ CrCL\ (mL/min)}{100\ (mL/min)}\right)^{0.45}\times\left(\frac{\text{FFM\ }\left(\text{kg}\right)}{70\ \left(\text{kg}\right)}\right)^{0.75}\times\ (\mathbf{1}\text{\ for\ }SLC22A12\ rs505802\ CC,\ \mathbf{1.32}\text{\ for\ CT\ and\ }\mathbf{1.64}\text{\ for\ \ TT})\ \nonumber \\ \end{equation}\begin{equation} V/f_{m}\ (L)=59.3\ (L)\times{(\frac{FFM\ (kg)}{70\ (kg)})}^{1}\ \nonumber \\ \end{equation}
To estimate the renal and non-renal CL/fm of oxypurinol, a PK model with both serum and urine oxypurinol data was fitted. The renal CL/fm was 0.77 L/h (77%) and non-renal CL/fm was 0.23 L/h (23%) (Supplemental Table 3 ). Similar to the PK model with serum oxypurinol, the estimated CrCL and SLC22A12rs505802C>T were found to be significant with renalCL/fm . NSAIDs, CARMIL1rs742132A>G, and SLC2A9 rs1014290C>T were found to be significant in the forward selection step but not in the backward elimination step on renal CL/fm . No covariates were found to be significant with non-renalCL/fm .