Data analysis
As a measure of exposure, (i ) C D-1 ratio of atomoxetine was defined as the main outcome measure of interest, as well as (ii ) unadjusted serum concentration, (iii ) daily dose, and (iv ) risk of having undetectable serum concentration of atomoxetine. For outcome measures (i-iii ), linear mixed model analyses (using random intercept and the restricted maximum likelihood model) were used to allow for inclusion of multiple samples per patient with age, sex and blood sampling time as covariate, if not otherwise described. The C D-1 ratios and absolute concentrations of atomoxetine were ln-transformed prior to analysis in order to ensure normal distribution of the data. Furthermore, several sensitivity analyses were performed to estimate the robustness when estimating the effects of CYP2D6/2C19 genotype subgroups, including limiting inclusion of serum samples solely to those withdrawn 4-8 hours post dose. To calculate the risk of having undetectable serum samples, a mixed logistic regression model allowing inclusion of all serum samples per patients was used with risk of having undetectable serum samples as reference. The other proportions (i.e., male sex, samples with available sampling time, samples with dose, undetectable serum samples, patients with at least one undetectable serum sample andCYP2D6/2C19 genotype subgroups) were compared using Fisher’s exact test.
All statistical analyses were performed in SPSS, version 25.0 (IBM SPSS Statistics, Armonk, NY, USA). GraphPad version 4 was used for graphical presentations (GraphPad Software, San Diego, CA). In figure 1, linear mixed model was used as described above but with no covariates and with Sidak-adjusted p-values. For the other analyses, the comparison-wise alpha was set at 0.05 and no multiplicity adjustments were made. The estimated means are given with either standard error (SE) or 95% confidence intervals (CI).