"Does Replication help with Experimental Biases in Clinical Trials?" Prof David Teira (UNED, Madrid)

During the last decade, a replication crisis has been detected in many experimental fields, and, in particular, in drug testing in clinical trials. Experimental outcomes published in top journals do not stand the test of reproduction. A widespread interpretation of this crisis puts the blame on the experimenters’ financial biases. Clinical trials are regulatory experiments in which a treatment may gain or not market access: the financial stakes for the sponsor of the development of the treatment are high. Therefore, the sponsor may put direct or indirect pressure on the experimenter to obtain a positive outcome. Often, once this pressure is relaxed, in further replications of the trial, the original positive outcome vanishes. The implicit assumption in this interpretation is that, once we correct for the sponsor biases, trials will become more replicable than they actually are. We want to contest this interpretation of the replication crisis with an analysis of the concept of experimental bias in clinical trials. We will focus on the biases that may flaw the design and conduct of the test. Our basic claim is that replication in experiments is only valuable once the experimenters have agreed on a standardized intervention and a list of debiasing controls to be implemented in the trial. Replicability mainly helps us in controlling for unintended deviations from the protocol, once the relevant debiasing procedures have been implemented. But the major problems with trials lie elsewhere: either in improperly debiased tests or in trials with clinically irrelevant variables. Against a widespread intuition, we will defend that the outcomes in these latter trials are perfectly replicable. If we want better trials, fostering replicability (good as it may be) is perhaps not helpful in itself.