A manifesto for the future of ICU trials

Abstract

1 Think Bayesian Bayesian analysis is an alternate statistical paradigm that answers the question “what is the probability of treatment effect” in contrast to the traditional frequentist approach, which answers the question “what is the probability of these data, assuming no treatment effect?” Under the Bayesian framework, trial information is not biased by “looking at” the data, and the results can be continuously re-estimated and updated as additional information (i.e., patient outcomes) is added to the dataset [6]. [...]the benefit of therapeutics (“signal”) targeting those mechanisms will also vary. To increase the probability of demonstrating the benefit of therapy in treatment-responsive subgroups (where such benefit actually exists)—to find the signal in the noise—heterogeneity in treatment response needs to be characterized as much as possible before and during Phase III trials [12, 13]. Embedding trials within existing data repositories (e.g., clinical registries, electronic health records) to “find” patients, randomly assign treatments, collect data, and ascertain outcomes can increase efficiency and reduce costs.