Power calculation methods to detect covariates effect when combining observed and simulated data.
Anubha Gupta, Shan Pan, Stefano Zamuner
Clinical Pharmacology Modeling and Simulation, GSK, Stevenage, UK
Objectives: Monte Carlo Mapped Power (MCMP) has been developed as an alternative rapid method [1] with respect to stochastic simulation and re-estimation (SSE) method for power calculation using non-linear mixed effect models based on likelihood ratio test. The aim of this work is to compare MCMP and SSE methods to identify covariates effect when combining existing data with simulated data. A specific example based on population PK model for a monoclonal antibody is presented.
Methods: Routine to implement MCMP [1] and SSE methods are available in PsN [2] for power calculation of a planned study. We borrowed the same principle to calculate the power to detect a covariate effect when combining existing rich PK data from healthy volunteers and simulated sparse PK data from a patient population.
A 2-compartment PK model for monoclonal antibody with first-order absorption and elimination was previously developed using a rich dataset in healthy volunteers (n=62). Sparse data (4 samples from each of 100 subjects) in patient population (n=1000) was simulated assuming significant effect of covariate on bioavailability and was combined with observed data. Higher between–subject variability in clearance was also considered for the patient population. The combined dataset was re-estimated in NONMEM® 7.3 using reduced and full PK model.
To avoid differences in the individual objective function (iOFV) for the real subjects all the parameters values, other than effect of covariate and residual error, were fixed to that estimated from observed data. An R script was developed to calculate the power for a given sample size N based on the change in iOFV between full and reduced model [1]. The results were compared with SSE method still using a combined dataset with existing and simulated data.
Results: The number of subjects required to detect a 27% difference in bioavailability with 90% power were 8 when assuming the same between-subject variability in clearance (26%). A higher (50%) between subject variability in the patient population, increased required sample size to 15. These results were comparable between the extended MCMP method developed in R and SSE method.
Conclusions: The method was successfully implemented to estimate the minimum number of subjects required to detect the difference in bioavailability for a monoclonal antibody using observed PK data and sparse PK study design in the population of interest.
References:
[1] Vong C1, Bergstrand M, Nyberg J, Karlsson MO. Rapid sample size calculations for a defined likelihood ratio test-based power in mixed-effects models. AAPS J. 2012 Jun;14(2):176-86. doi: 10.1208/s12248-012-9327-8. Epub 2012 Feb 17.
[2] Lindbom L, Ribbing J, Jonsson EN. PerlspeaksNONMEM (PsN) -a Perl module for NONMEM related programming. Comput Methods Programs Biomed. 75(2):8594
