You are here


08 - 06 - 2018

Drug safety: the predictivity of animal studies specified by Big data

The results of a Big Data study that looked at more than 3,000 drugs over a 70-year period specifies the predictivity for humans of safety studies performed on animals.


Animal model safety studies are an important step in drug development.


For the first time, authors have used Big Data extraction and exploration methods to collect and analyze all reported clinical and preclinical events for drugs marketed in Europe and the US for 70 years.




This study addressed both the statistical significance of a result and the predictive power of the model. The main conclusion is that the level of predictivity of an effect observed in the animal depends on the animal species and the effect.


The predictivity of the effects observed in many animal models is confirmed (such as cardiac arrhythmias or digestive disorders) and less predictive models have been identified. Different tables and graphs present these results.

Many side effects observed in animals are predictive of the effects in humans, while some other effects observed in animals are never seen in humans.

The predictivity of the absence of observed effects in animal models is limited.




The methodological difficulties related to this type of study are discussed. In particular, the data used to carry out the study can be a source of significant bias. This is an observation that gives the full value of this study carried out using a very large amount of data. Other biases are discussed such as differences in the words used to describe the same effect.


Even though regulations and references impose strong constraints on the choice of models, it is up to the researcher to set his requirements on a case-by-case basis.