Taking the Pain out of Literature Searches: From Manual to Automation

Aug 30, 2016

By Sharad Prakash
signal detection in drug safety pharmacovigilanceThe premise is simple: Manual efforts, in any part of managing adverse events (AEs), increase risk. But, life sciences organizations are facing increased manual workloads to manage the deluge of information coming from multiple sources. And, this is taking a toll on pharmacovigilance departments as they struggle to meet demands around risk minimization. As I discussed in my last blog, life sciences companies now have to deal with a digitized landscape, with adverse events (AEs) coming from multiple sources – from the more traditional avenues to literature databases and even social media.

The issue of literature searches is a vast one and worth considering in more depth. In September last year, the European Medicines Agency (EMA) assumed responsibility for managing literature searches for some substances in selected medical literature. However, companies still must continue to search literature databases to meet regulatory requirements, including for the Food and Drug Administration (FDA).

Most companies have processes in place as part of their SOPs to address the complex issue of risk management and literature searches. Typically, companies have a group of people whose job it is to screen databases, including the literature databases, for potential signals. They will look through titles, citations, abstracts, keywords, mentions of the product name or product class in major literature databases — MedLine, PubMed and Embase. If they come across a citation, they will go through a process of triaging and validating those citations and passing the information onto a medical reviewer for further action.

That reviewer will determine whether the AE is new or has previously been observed and followed up on. If it’s new, pharmacovigilance will go through the usual cycle of further verifying those signals or associations, citing the new source and attaching and archiving that source along with the medical evaluation and other follow up actions.

All these activities are very manual – from going through different databases to manually getting a list of citations, reviewing each separately, annotating an item, adding comments, passing on to the reviewer, who in turn adds his or her comments.  And, doing this across a broader spectrum of information sources, and at different speeds, is putting undue pressure on the review process that is causing failures and gaps.

Struggle for a Clear View of Safety Signals

Given that each of these databases are scanned separately outside of the company’s own database, it’s a real struggle to get a complete view of potential safety signals and coordinate activities. The question companies grapple with is, how do they manage the life cycle of a signal they come across, not just in literature databases but in other sources – social media, public healthcare databases, and so on? How do they document all the information they have assessed?

The answer lies in automation, which would allow reviewers to define their searches and specify frequency – such as weekly or monthly – and be sent a list of citations to their inbox. This is made possible by text analytics, visualization (Fig 1) , natural language processing and artificial intelligence, which extracts semantic concepts and relationship information from the literature body and then searches for causality and hidden associations related to potential adverse drug reactions (ADRs). By looking at the textual information – whether in a citation, abstract or the entire literature – an automated solution built on artificial intelligence can identify signals for reviewers rather than have them manually search for those signals.

Figure 1:benefit risk management


This is undoubtedly the direction regulators would like to see the industry move. In fact, during the DIA meeting in June, representatives from the FDA talked about the challenge of going through each case in their system to find potential ADRs. They recognize the value of automation not only for their system, but for literature databases to remove the need to search through thousands of pages of literature. While the ultimate decision of what to do with those signals is up to the reviewer, the use of smart algorithms to automate the search process will vastly simplify, standardize and improve the literature search process.

Learn more about managing literature searches through automation by contacting us at sales@arisglobal.com