Statistics and Its Interface

Volume 8 (2015)

Number 2

Special Issue on Modern Bayesian Statistics (Part II)

Guest Editor: Ming-Hui Chen (University of Connecticut)

A Bayesian phase I/II clinical trial design in the presence of informative dropouts

Pages: 217 – 226

DOI: http://dx.doi.org/10.4310/SII.2015.v8.n2.a9

Authors

Beibei Guo (Department of Experimental Statistics, Louisiana State University, Baton Rouge, La., U.S.A.)

Yong Zang (Department of Mathematical Sciences, Florida Atlantic University, Boca Raton, Fl., U.S.A.)

Ying Yuan (Department of Biostatistics, M.D. Anderson Cancer Center, University of Texas, Houston, Tx., U.S.A.)

Abstract

A phase I/II trial design utilizes both toxicity and efficacy outcomes to make the decision of dose assignment for patients. Because assessing the efficacy endpoint often requires a relatively long follow-up time, phase I/II trials are more susceptible to the missing data problem caused by informative dropouts that are correlated with treatment efficacy and toxicity. In addition, patient outcomes may not be scored quickly enough to apply decision rules that choose treatments or doses for newly accrued patients. To address these issues, we propose a Bayesian phase I/II design that jointly models efficacy, toxicity, and dropout as time-to-event data. Correlations among the three time-to-event outcomes are taken into account by a shared frailty. This joint model strategy accounts for the informative dropouts and has an additional advantage of accommodating a high accrual rate without suspending patient enrollment when toxicity or efficacy outcomes require a long follow-up. Under the Bayesian paradigm, we continuously update the posterior estimate of the model and assign incoming patients to the most desirable dose based on an efficacy-toxicity trade-off utility. Simulation studies show that the proposed design has good operating characteristics with a high probability of selecting the target dose and assigning the most patients to the target dose.

Keywords

Bayesian adaptive design, missing data, nonignorable dropout, dose finding, trade-off

Full Text (PDF format)