We propose a simulation-based approach to decision theoretic Bayesian optimal design. The underlying probability model is a population pharmacokinetic model which allows for correlated responses (drug concentrations) and patient-to-patient heterogeneity. We consider the problem of choosing sampling times for the anticancer agent paclitaxel, using criteria related to total area under the curve, time above a critical threshold and sampling cost.
Keywords: Clinical trial; Limited sampling strategies; Longitudinal data; Markov chain Monte Carlo; Optimal design; Population model; Random-effects regression.
To appear in Applied Statistics, 2001
The manuscript is available in PostScript format.