A simulation study to define the number of patients necessary to meet primary pharmacokinetic (pk) aims of phase I oncology trials

Abstract

2579 Introduction: The primary pharmacokinetic parameters defined in oncology phase I trials include estimates of drug clearance and determining the relationship between dose and drug exposure as measured by AUC. The numbers of patients needed for reliable estimates of these parameters, however, has not been defined. We hypothesized that a threshold number of patients existed above which meaningful improvement in such estimates would not occur. Methods: We identified 28 English language adult phase 1 oncology publications (2004-2006) from which individual patient dose and AUC data could be extracted. Analyses of clearance precision were performed in PASS 2000. Simulations were performed by determining a distribution of r2 coefficients for the regression of AUC on dose for each trial, for sample sizes of 2 to n-1, where n was the total patient number studied. Median values for the simulated r2 and a 90% confidence interval (CI) were obtained for every simulation. A stability boundary of 0.1 around the simulated r2 value was calculated. A 90% confidence interval around the observed r2 value was also bootstrapped. An r2 value was considered stable when the simulated r2 CI fell within the stability boundary for all subsequent sample sizes. Results: Analyses of the precision of mean clearance estimates over a range of confidence intervals demonstrated that the incremental gain in clearance precision estimates was marginal beyond the study of 8 to 12 patients. Linearity simulations of AUC vs dose demonstrated the median (range) of patients needed to determine a stable r2 estimate was 7 (4 - 16), with a mean (±SD) of 7.8 ± 3.2. Conclusions: These simulations show that, in general, only 8–12 patients are needed to meet the primary PK objectives of phase I oncology trials. While such modest sample sizes are not sufficient to define pharmacokinetic-dynamic-genetic relationships, such relationships rarely emerge from phase I trials. Our simulations indicate that the number of patients needed to meet PK objectives may be more modest than previously anticipated, and should be readily achievable with voluntary participation in the PK component of oncology phase 1 trials. No significant financial relationships to disclose.