The future of biostatistics: expecting the unexpected.

The paper explores the future of (clinical) biostatistics. In the first part of the paper important trends in biostatistics are reviewed: new applications and more complex data; causal models for observational data; cross-validation-based model building; graphical chain and random effect models; faster computing and new algorithms as Markov chain Monte Carlo; generalized estimating equations and pseudo-likelihood; pooled data sets for meta-analysis and prognostic modelling. In the second part some dreams and nightmares of the author are sketched. Dreams are: implementation of prognostic and diagnostic models in the clinic; an instantaneous numeric integrator; much more and better organized follow-up data; disease mapping in space and time. 'Nightmares', that is, issues which it is hoped will go out of use in the future, are: P-value; rank tests; exact methods; meta-analysis; matched case-control studies.