Social media has been suggested as a possibly valuable data source for signal detection in pharmacovigilance. This study focused on the evaluation of disproportionality analysis in combined Twitter and Facebook data.

A large number of drugs and a breadth of adverse events were considered. Two different reference sets were used to benchmark predictive performance, one based on labelling changes and one based on validated safety signals.

Twitter/Facebook data displayed no predictive value for either of the reference sets, which was contrasted by considerably better performance for the conventional pharmacovigilance data source VigiBase. Therefore, broad-ranging statistical safety signal detection in Twitter and Facebook cannot be recommended.