Good bad and the ugly of in vivo research Professor Malcolm MacLeod, Chair in Neurology and Translational Neuroscience, University of Edinburgh and Clinical Lead for Neurology at NHS Forth Valley Although large numbers of novel treatment strategies for diseases such as stroke are developed in laboratories each year and improve outcome in animal models, very few prove effective in patients. At present, no systematic or evidence-based criteria inform the decision to venture from preclinical testing into clinical development. Animal experiments testing the efficacy of potential treatments are central to drug development. The outputs from these preclinical studies need to be reliable and valid to improve human health. Systematic analyses of preclinical data have revealed substantial overestimates of treatment effects because of compromised internal and external validity. In addition, animal studies are in general too small to detect reliably the effects they purport to observe; and groups may use models which are highly sensitive to the phenomena being studied at the expense of the generalisability of their conclusions. Most animal studies are therefore underpowered to detect potentially important treatment effects under tightly defined conditions in healthy young animals, let alone to detect smaller but nonetheless potentially important effects in less tightly controlled conditions in older animals with relevant comorbidities. It is therefore likely that for many stroke drugs tested in animals, true biological efficacy in those models is substantially lower than that suggested by the published literature, and in some cases agent itself may be inert. The recognised impact of selection bias (eliminated by randomisation), of performance bias (eliminated by blinded conduct of the experiment) and of measurement bias (eliminated by blinded outcome assessment) have resulted in clear standards for good clinical practice in clinical trials, and to the development of ethical, regulatory, funding, reporting and methodological standards to ensure the validity of their conclusions . At present clear standards such as these do not exist for most preclinical research, and the reporting of measures to avoid bias is uncommon.