Sample Size and Statistical Analyses

The aim of the study was to demonstrate that the trough level of CYP3A4*22 carriers with a reduced dose is non-inferior to the trough level of non-carriers with a standard dose. The trough level was assumed to follow a log-normal distribution. With a non-inferiority margin of − 20% in relative difference (RD) (i.e., a difference of − 0.2231 on log-transformed data) and an average coefficient of variation of 40% (translated to a standard deviation [SD] of approximately 0.40 on the log-scale), at least 198 patients were required for this study of whom at least 33 had to be a CYP3A4*22 carrier (power of 90% and one-sided alpha of 5%) [26]. Only study drugs with at least two CYP3A4*22 carriers per drug were included in the final analysis. Inclusion of patients stopped when all these three criteria were met (i.e., at least 33 CYP3A4*22 carriers, at least two CYP3A4 carriers per drug, at least 165 CYP3A4 wildtype patients).

The two-stage individual patient data (IPD) meta-analysis was applied considering that several drugs were included. Each study drug was seen as a separate study within this analysis. Hereto, the ‘ipdmetan’ package in Stata was used [27]. Non-inferiority of the trough level was evaluated by looking at whether the lower bound of the two-sided 90% confidence interval (CI) for difference between carriers and non-carriers was higher than the non-inferiority boundary of −0.2231 on the log-scale (this boundary is equivalent to a − 20% relative difference and a geometric mean ratio of 0.80).

Non-inferiority regarding drug exposure was also analyzed for the two study drugs with the most CYP3A4*22 carriers separately (i.e., pazopanib and imatinib). Pharmacokinetic data were log normalized before performing an independent sample t-test. Geometric mean ratio and corresponding 90% CI, interpreted as the RD in percentage, were calculated by exponentiation of the mean difference and 90% CI of this difference derived from the independent sample t-test. Exposure was regarded as non-inferior if the lower boundary of the CI of the geometric mean ratio was higher than 0.8 (i.e., equivalent to a −20% relative difference). For these compounds, the incidence of patients with an exposure above TDM targets proposed by Verheijen et al were compared between the CYP3A4*22 carriers and the wildtype patients by the Fisher’s exact test [28].

For the Hardy-Weinberg equilibrium, the distribution of genotypes was tested by a chi-squared test. Severe or medically significant toxicities (i.e., Grade ≥ 3 based on Common Terminology Criteria for Adverse Events [CTCAE] version 4.03) were described with numbers of patients and percentages for CYP3A4*22 carriers and wildtype patients separately. Incidence of dose modifications at the end of study period was described using descriptive statistics.