Design of experiments for bioengineers in clinical settings

Abstract

This chapter is an introduction for the design of experiments involving human beings. This is a very common scenario in bioengineer practice. The usual problem we are facing is understanding the effects of a new development in the target population. The development could be a new device, a new algorithm, or a piece of software. Facing this problem implies taking a number of decision such as the hypotheses of different factors that can influence the outcome, the analysis of the data to achieve the conclusions, the number of subjects participating in the experiment and the number and definition of the set of measurements to be applied to each subject. This approach is slightly different from a clinical trial. The approach presented in this paper is focussed on understanding the effects of an intervention in a small number of subjects. A clinical trial usually aims at quantifying the effects in the general population. The view we will adopt will be a frequentist approach (opposite to a Bayesian approach) (Fienberg 2005). This is especially important for aspects related with the analysis and almost negligible in the aspects that are related with the experimental design. Besides, for all the examples, we will take into consideration a general linear model, but the conclusions could be extended to a generalized linear model. Again, the impact of our restriction is more important in the part related with the analysis than with the design of the experiment. Although there are many pieces of software devoted to the design of experiments, the examples shown in this chapter have been made using R (R Development Core Team 2012). There are many reasons for this decision: R provides the tools required for advanced statistics, it is widely distributed and it is free. All the examples, have been made using 3 packages: car (Fox and Weisberg (2011)) and phia (de Rosario-Martinez (2012)) for analysis and AlgDesign (Wheeler (2011)) for the design of the experiments.