dfpk: an R package for a practical implementation of PK measurements in dose-finding studies
Artemis Toumazi (1), Sarah Zohar (1), Frederike Lentz (2), Corinne Alberti (3), Tim Friede (4), Nigel Stallard (5), Emmanuelle Comets (6,7), Moreno Ursino (1)
(1) INSERM, UMRS 1138, team 22, CRC, University Paris 5, University Paris 6, Paris, France, (2) Federal Institute for Drugs and Medical Devices, Bonn, Germany, (3) INSERM, UMR 1123, Hôpital Robert-Debré, APHP, University Paris 7, Paris, France, (4) Department of Medical Statistics, University Medical Center Göttingen, Göttingen, Germany, (5) Statistics and Epidemiology, Division of Health Sciences, Warwick Medical School, The University of Warwick, UK, (6) INSERM, CIC 1414, University Rennes-1, Rennes, France, (7) INSERM, IAME, UMR 1137, University Paris Diderot, Paris, France.
Objectives: Dose-finding, aiming at finding the maximum tolerated dose (MTD), and pharmacokinetics (PK) studies are the first in human studies in the development process of a new treatment. In the literature, to date only few attempts have been made to combine PK and dose-finding and no software implementation is available. Our objective was to implement the five PK-based dose-finding methods developed in [1] in an R package, called dfpk [2].
Methods: By default, AUC is used as PK measure of exposure, but it can be replaced with other PK measures such as Cmax. AUC is treated as a covariate for the probability of toxicity for PKCOV method, as dependent variable in linear regression versus dose for PKLIM method, and in both ways, in two separated regression models, for PKLOGIT, PKTOX and PKPOP methods. All methods were developed in a sequential Bayesian setting: Bayesian parameter estimation is carried out using the rstan package. All available data are used to suggest the dose of the next cohort with a constraint regarding the probability of toxicity. The ggplot2 package is used to create summary plots of toxicities or concentration curves.
Results: dfpk provides, for each method, a function (nextDose) to suggest the dose to give to the next cohort, and a function to run trial simulations (nsim). nextDose requires the method name, the binary toxicity outcomes and the PK measurements for each patient, the panel of doses, the toxicity threshold and the parameter for the prior distributions. The output includes the recommended dose and the estimated probability of toxicity at each dose. nsim requires also the cohort size, the sample size for trial, the number of trials and simulated datasets. The scenario function generates at each dose the toxicity value related to AUC with an underlying one PK compartimental model with linear absorption. It is included as an example. Similar dataframes can be generated directly by the user and passed to nsim.
The online help provides an example of using dfpk for the case-study described in [1].
Conclusions: The developed user-friendly R package dfpk supports the design of innovative dose-finding studies using PK information. It can be downloaded freely from the CRAN repository.
References:
[1] Ursino M, Zohar S, Lentz F, Friede T, Stallard N, Comets E. Dose-finding methods using pharmacokinetics in small populations. Biometrical Journal, 2017; in press
[2] https://CRAN.R-project.org/package=dfpk
Acknowledgments: This research has received funding from the European Union's Framework Programme for research, technological development and demonstration under grant agreement no 602144. This work was part of the InSPiRe project but does not necessarily represent the view of all InSPiRe partners.
