Screening Experiments for Developing Adaptive Treatment Strategies

Many disorders require multi-component treatments given sequentially in time to achieve and maintain good outcomes. The development of these adaptive treatment strategies is challenging due to the large number of components and sequencing of components. We propose the use of screening experiments in which scientists screen out inactive components. Unfortunately when the goal is to construct adaptive treatment strategies the classical design and analysis of screening experiments can not be directly imported. This is because the classical definitions of effects (say via an ANOVA decomposition) do not represent useful causal effects. We define effects that are both causal and for which we are able to utilize traditional screening (e.g., factorial and fractional factorial) designs. The definition of the causal effects means that the aliasing of effects associated with a fractional factorial does not immediately follow from the defining words, however, somewhat surprisingly, simple modifications can be used to infer aliasing from the defining words.