Finding the needle in the PV haystack
Injecting technology solutions into the literature review process in pharmacovigilance (PV)

Introduction

Many life science companies are exploring, with limited success, the application of digital technologies to business operations. One area that has failed in recent years to deliver the promised benefits or required quality is the automation of scientific and medical literature review for reportable adverse event (AE) information.

The proof of the technology

Based on a hypothesis that technology has sufficiently matured to handle intelligent tasks such as literature review, a team of pharmacovigilance (PV) and digital experts convened to evaluate the potential of technology to effectively search literature articles for reportable AE information.

The team conducted a single blind experiment comparing the outputs of a manual review of a number of open-access medical literature with those of a smartly stitched technology solution (built by combining existing solutions such as IBM Watson Natural Language Understanding and PDFMiner library for Python coding language).

To test the hypothesis, a sample of journal articles (in English) from neuroscience publications was selected for the experiment. Fifty percent of the articles were about Attention Deficit Hyperactivity Disorder (ADHD) and covered the ADHD drug "methylphenidate". Of the articles mentioning the drug name, 40 percent contained AE terms associated with methylphenidate, while 30 percent mentioned AEs in a negative context. The main task was to use the technology solution to review all of the selected articles and identify those that contained AE information (including the AE context).

The experiment was conducted using a simple three-step process that had been previously described in the paper titled “Pharmacovigilance literature review in the age of precision medicine”.