Distilling a High Precision Drug Adverse Effect Benchmark Using Wikipedia’s Wisdom of the Crowd

ABSTRACT Background Large datasets of relational medical data, such as the adverse effects of drugs or vaccines, typically attain their large size, by relying on automatic, or semi-automatic, methods for generation. This often comes with a compromise on the precision of the generated data, which can be at least partially alleviated by having experts curate the data. Aim Since having experts review a large dataset can be costly and time consuming, here we suggest using Wikipedia for this task – that is, augment the automatic generation step by an automatic curation step based on the expert knowledge accumulated in Wikipedia. Methods To curate a dataset of adverse drug effects (ADEs), we suggest retrieving the Wikipedia page associated with the drug, and checking whether the ADE appears in the sections describing adverse effects. Drug indications, typically described in the opening paragraph of the page, are similarly filtered out. Results We use the method to curate two large adverse drug effect datasets and show that the obtained datasets have a much higher precision relative to their originating ones. Conclusions Algorithms which aim to infer drug-ADE relations, should at the very least be able to identify the “clear cut” cases. The high-precision benchmark constructed herein may therefore be a valuable resource for the evaluation of such algorithms.