New method for early detection of side-effects of drugs

Using a so-called “data-mining” method, it is possible to automatically find previously unknown side-effects of drugs in the huge WHO database of side-effect reports. This is demonstrated in a doctoral dissertation by Andrew Bate at Umeå University in Sweden.

The use of pharmaceuticals sometimes causes side-effects. By gathering reports about suspected cases of side-effects, it is possible to detect previously unknown ties between a certain drug and a side-effect at an early stage, so-called “signals.” The World Health Organization, WHO, database is the world’s largest collection of such side-effect reports. Today the database, created as early as 1968, contains more than 2.8 million case reports from 70 countries around the world.

To analyze such a huge amount of data to discover hitherto unknown patterns can be likened to searching for a needle in a haystack, and it is virtually impossible to do manually. The dissertation describes the use of a so-called “data-mining” method to automatically find signals at an early stage with the help of computers. The method is called BCPNN (Bayesian Confidence Propagation Neural Network) and employs a special type of statistic, a Bayesian statistic, in combination with so-called neural networks. The dissertation shows how BCPNN has been elaborated and tested for routine use in signal detection, refinement of signals, and finding complex patterns in the WHO database. The utility of the results is directly related to the quality of the data in the database, and the method should be used to find new suspected correlations (between a drug and a reaction) rather than to assess whether the drug in question actually caused the side-effect. That determination still relies on clinical analyses of the case reports as such.