Dose escalation studies: a comparison between NONMEM and a novel Bayesian tool
A. Russu (1), G. De Nicolao (1), I. Poggesi (2), M. Neve (2), L. Iavarone (2), R. Gomeni (2)
(1) Dipartimento di Informatica e Sistemistica, Università di Pavia, Italy; (2) GlaxoSmithKline, CPK Modelling & Simulation, Verona, Italy
Objectives: Evaluate modeling approaches and decision making criteria in dose escalating Phase I studies. In these studies, subjects receive increasing dose levels and, at each escalation, decision is made on whether next doses are to be administered based on some safety/tolerability constraints. In recent years, there has been a growing interest in Bayesian methods applied to such experimental settings [1, 2]. A comparison was therefore carried out between a NONMEM estimation procedure and two Bayesian approaches implemented within the R/WinBUGS environment. The different approaches were compared in terms of parameter estimates, predictive performance on new and existing subjects, and reliability of prediction limits.
Methods: Ten phase I dose escalation studies and 20 simulated datasets were analyzed. The dose-response relationship was modeled as a linear model using log-transformed doses and exposure metrics. In NONMEM VI (FOCE interaction method), additive intersubject variability and residual error were used. The two Bayesian methods, namely Empirical Bayes (EB) and MCMC were implemented within a user-friendly software tool based on R and WinBUGS. Estimated parameters, predictive root mean square error (RMSE) and % data within 90% prediction limits were used as comparison criteria.
Results: In all cases, it was found that all the considered methods provided comparable values of parameter estimates and predictive RMSE. However, the prediction limits provided by NONMEM turned out to be overly optimistic. Bayesian procedures provided either realistic or only slightly conservative intervals.
Conclusions: Satisfactory estimates and reliable prediction intervals, suitable for individual risk assessment, were obtained using both Bayesian approaches. The advantage of using the suboptimal EB scheme is two-fold: it helps finding relevant hyperparameter regions and allows a reciprocal cross-check with MCMC. NONMEM, though providing realistic point estimates, tends to systematically underestimate prediction intervals, possibly due to neglecting fixed effects uncertainty. Moreover, individual confidence intervals are not easily obtained in a rigorous way. Both aspects may impact on the quality of risk assessment during dose-escalation.
 Whitehead, J., Zhou, Y., Patterson, S., Webber, D. and Francis, S. (2001). Easy-to-implement Bayesian methods for dose escalation studies in healthy volunteers. Biostatistics 2, 47-61
 D.A. Berry, P.Müller, A.P. Grieve, M. Smith, T.Parke, R. Blazek, N. Mitchard, M. Krams. Adaptive Bayesian designs for dose-ranging drug trials. in C.Gatsonis, R.E. Kass, B. Carlin, A. Carriquiry, A. Gelman, I. Verdinelli, M. West (eds.), Case Studies in Bayesian Statistics V. New York: Springer-Verlag; 2001: 99-181