Finding relevant information in large volumes of unstructured text using conventional keyword search can be an arduous process.
Pharmacovigilance teams know this well – they are tasked with monitoring the effects of drugs licensed for use. This market, valued at $1 billion in 2015, is predicted to exceed $8 billion by 2024. Literature monitoring is a key component of pharmacovigilance – and a special challenge. Faced with a range of spontaneous reporting systems, time is often wasted on false positives and dead-ends.
Biomedical literature can be a rich source of the signals that pharmacovigilance teams need to do their work. However, scientific journal articles are not designed with the special needs of these teams in mind – leaving potentially valuable information locked in unstructured research narratives and reducing the recall of literature screening approaches.
The underreporting of adverse drug reactions by healthcare professionals and patients is also a recognized issue.
Patients’ narratives of drugs and their side effects on social media represent an additional data source for postmarketing drug safety surveillance. A 2014 study conducted by Epidemico which examined 6.9 million social media posts discovered 4,401 tweets resembling an ADR.
The industry is also faced with the prospect of negative drug reactions that don’t feature in HCP reports, but do appear in literature.
Related Reading: What a Study of 15 Million Articles Can Teach us About Text Mining
Text mining and pharmacovigilance working together
Machine analysis can help assuage these challenges. The process, which uses natural language processing (NLP) techniques to swiftly analyze huge quantities of text, can transform every stage of the drug development journey. This means algorithms can identify potential adverse drug reactions within a data set at scale, reducing false positives.
Text mining tools can also help teams fine-tune their queries and see an improvement in search strategy management. Keyword-based search strategies can often be convoluted, messy, and overly-specific, frequently including every synonym possible such as brand name, substance name, pre-release name, as well as a whole range of adverse reactions. These searches can be difficult to update and maintain. Text mining or a semantically-enriched approach can help simplify those queries, making them more powerful and the results easier to interpret.
Looking to the future
A study by Elsevier looked at how pharmacovigilance teams not currently using text mining would like to incorporate it into their workflows. The results show that needs vary: some want to overcome taxonomy and indexing issues, others want to use it to mine multiple sources.
Whatever the objective, the industry is taking a more data-driven approach to pharmacovigilance. But the journey has only just begun. In the years to come we will see more advanced NLP, algorithms and platforms adding pharmacovigilance-friendly value to data. But right now, text mining means less time spent chasing false positives and less risk of missing vital information, which in turn means better patient care.
Check out Copyright Clearance Center’s text mining solutions here.