Evaluation Of An Osteoporosis Biomarker Model For Simulation Using Posterior Predictive Check Technique
P. Girard, L. Claret, W. Ebling and S.T. Forgue, M. Heathman, A. Ghosh, C.T. Benson
Pharsight Corp, Mountain View CA and Eli Lilly & Co, Indianapolis IN.
Purpose Osteocalcin (OC) is a biomarker of drug effect on osteoporosis progression. A decrease in OC has been shown to be correlated with a decrease in fractures risks in postmenopausal osteoporotic women. This makes this biomarker interesting for the purpose of clinical trial simulation in earlier phase development, when the length of trials does not allow observation of the clinical endpoint. A posterior predictive check (PPC) technique was used to qualify a model of OC as a prerequisite for trial simulation.
Methods Data were combined across 1,456 postmenopausal women who received placebo (calcium or Ca+vitamin D) or the SERM raloxifene (30, 60 or 150 mg) in 4 parallel-group trials. Serum OC and covariates measured at 0 to 12 months were fitted to an integrated indirect response model:
OC(t)= Ki/Ko (1- f(D) (1-e-Ko t )).
Here, Ki and Ko are 0- and 1-order rate constants, and f(D)= Emax· D/(D50+D) if dose D>0 and f(D)=PL if D=0. Parameters and between-patient variability in Ki, Emax and PL were estimated with NONMEM. Observed OC values were compared to values simulated with a degenerate posterior distribution or a multivariate normal one, with means set to their point estimates and covariance matrix set to the precision matrix.
Results The best model included effects of urinary type 1 collagen on input rate Ki and osteopenia status on maximal SERM effect Emax. Common goodness-of-fit plots suggested an excellent fit; however, both PPC techniques revealed misspecification that could potentially undermine trial simulations. The specific problem was corrected by constraining placebo to have a beneficial effect.
Conclusions PPC accounts for all sources of randomness and is a powerful technique to qualify any population pharmacodynamic model intended to support clinical trial simulation.