Recruitment in Multicentre Trials: Prediction and Adjustment Vladimir V. Anisimov 1 , Darryl Downing 2 Valerii V. Fedorov 2 1 GlaxoSmithKline, New Frontiers Science Park (South),Third Avenue, Harlow, Essex, CM19 5AW, United Kingdom. Vladimir.V.Anisimov@gsk.com 2 GlaxoSmithKline, 1250 So Collegeville Rd, PO Box 5089, Collegeville, PA 19426-0989, U.S.A. Valeri.V.Fedorov@gsk.com Summary. There are a few sources of uncertainty/variability associated with the patient recruitment in multicentre clinical trials: uncertainties in prior information, stochasticity in patient arrival and centre initiation processes. Methods of statisti- cal modeling, prediction and adaptive adjustment of recruitment are proposed to address these issues. The procedures for constructing an optimal recruitment design accounting for time and costs constraints are briefly discussed. Key words: patient recruitment, optimal design, multicentre trial, adaptive adjustment. 1 Introduction The recruitment time (time required to complete patient recruitment) is one of the key decision variables at the design stage of clinical trials. Existing tech- niques of recruitment planning are mainly deterministic and do not account for various uncertainties in input information and stochastic fluctuations of the recruitment process. We consider multicentre trials and propose a recruitment model, where the patients enter different centres according to Poisson processes with time- constant rates. This assumption seems to be well accepted; cf. Senn (1997, 1998); Anisimov et al. (2003). We suggest to view these rates as a sample from a gamma distributed population. Other mixing distribution can be used in a similar setting, but a gamma distribution is a conjugate to a Poisson distribution and is a natural candidate for describing prior uncertainties when the Bayesian approach is used. The analysis of a number of completed trials has shown (Anisimov and Fedorov, 2005) that a Poisson-gamma model is in a good agreement with existing data. The model allows to develop the methods for predicting the number of re- cruited patients and the recruitment time together with confidence/credibilty