RepFail: Understanding Replication Failure in Animal Research – Lack of Scientific Rigor, Low Statistical Power, or Standardization?

Principal Investigator: Hanno Würbel
Project Team: Bernhard Völkl (Senior Scientist)
Janja Novak (Senior research assistant)
Ivana Jaric (Postdoc)
Marianna Rosso (PhD student)
Funded by: Swiss National Science Foundation (SNSF)

Increasing evidence indicates that the scientific validity and reproducibility of preclinical biomedical animal research is at stake. My group has been pivotal in producing substantial evidence that results from single-laboratory animal experiments may generally have poor external validity and thus be poorly reproducible when studies are replicated across laboratories. In particular, we have provided proof-of-concept that rigorous standardization of animal experiments could be an important factor mediating spurious results resulting in poor reproducibility. Based on this, the overall hypothesis underlying this project is that poor reproducibility is caused by a failure to account for biological variation, whereby reproducibility is compromised when biological variation is standardized away instead of being incorporated into experimental design. This hypothesis will be further developed, explored, and tested against competing hypotheses, namely poor scientific rigor in experimental conduct (risk of bias) and small sample sizes (low statistical power). To this end, we will employ systematic reviews and meta-analyses of the published literature, computer simulations based on empirical datasets, and experimental studies using the mouse as model species, including a large multi-laboratory study to formally test the main hypotheses. We predict that scientific rigor and high statistical power are necessary, but not sufficient, to guarantee reproducibility, and that heterogenization of study samples is essential for reproducibility. This work will greatly benefit from cross-fertilization and synergies with the Horizon 2020 IMI2 project EQIPD. The new results will advance our understanding of the sources of poor reproducibility in animal research and help develop experimental designs that account for both biological variation and random error, thereby providing significant new knowledge for more effective research strategies to both basic and preclinical animal research. By improving scientific validity such experimental designs will ultimately enhance the value of animal research, thereby facilitating the ethical justification of animal experiments in the harm-benefit analysis.