Accumulating evidence indicates high risk of bias in preclinical animal research, questioning the scientific validity and reproducibility of published research findings. Systematic reviews found low rates of reporting of measures against risks of bias in the published literature (e.g., randomization, blinding, sample size calculation) and a correlation between low reporting rates and inflated treatment effects. That most animal research undergoes peer review or ethical review would offer the possibility to detect risks of bias at an earlier stage, before the research has been conducted. For example, in Switzerland, animal experiments are licensed based on a detailed description of the study protocol and a harm–benefit analysis. We therefore screened applications for animal experiments submitted to Swiss authorities (n = 1,277) for the rates at which the use of seven basic measures against bias (allocation concealment, blinding, randomization, sample size calculation, inclusion/exclusion criteria, primary outcome variable, and statistical analysis plan) were described and compared them with the reporting rates of the same measures in a representative sub-sample of publications (n = 50) resulting from studies described in these applications. Measures against bias were described at very low rates, ranging on average from 2.4% for statistical analysis plan to 19% for primary outcome variable in applications for animal experiments, and from 0.0% for sample size calculation to 34% for statistical analysis plan in publications from these experiments. Calculating an internal validity score (IVS) based on the proportion of the seven measures against bias, we found a weak positive correlation between the IVS of applications and that of publications (Spearman’s rho = 0.34, p = 0.014), indicating that the rates of description of these measures in applications partly predict their rates of reporting in publications. These results indicate that the authorities licensing animal experiments are lacking important information about experimental conduct that determines the scientific validity of the findings, which may be critical for the weight attributed to the benefit of the research in the harm–benefit analysis. Similar to manuscripts getting accepted for publication despite poor reporting of measures against bias, applications for animal experiments may often be approved based on implicit confidence rather than explicit evidence of scientific rigor. Our findings shed serious doubt on the current authorization procedure for animal experiments, as well as the peer-review process for scientific publications, which in the long run may undermine the credibility of research. Developing existing authorization procedures that are already in place in many countries towards a preregistration system for animal research is one promising way to reform the system. This would not only benefit the scientific validity of findings from animal experiments but also help to avoid unnecessary harm to animals for inconclusive research.

In animal experiments, animals, husbandry and test procedures are traditionally standardized to maximize test sensitivity and minimize animal use, assuming that this will also guarantee reproducibility. However, by reducing within-experiment variation, standardization may limit inference to the specific experimental conditions. Indeed, we have recently shown in mice that standardization may generate spurious results in behavioral tests, accounting for poor reproducibility, and that this can be avoided by population heterogenization through systematic variation of experimental conditions. Here, we examined whether a simple form of heterogenization effectively improves reproducibility of test results in a multi-laboratory situation. Each of six laboratories independently ordered 64 female mice of two inbred strains (C57BL/6NCrl, DBA/2NCrl) and examined them for strain differences in five commonly used behavioral tests under two different experimental designs. In the standardized design, experimental conditions were standardized as much as possible in each laboratory, while they were systematically varied with respect to the animals' test age and cage enrichment in the heterogenized design. Although heterogenization tended to improve reproducibility by increasing within-experiment variation relative to between-experiment variation, the effect was too weak to account for the large variation between laboratories. However, our findings confirm the potential of systematic heterogenization for improving reproducibility of animal experiments and highlight the need for effective and practicable heterogenization strategies.