Automation of literature search and review

automation of literature search and review

Over the years, pharmacovigilance (PV) processes have relied on PV professionals manually sifting through large volumes of data to identify, assess, and report Adverse Event (AE) information. However, this approach is increasingly facing challenges as the workload continues to grow within fixed time constraints. Addressing these challenges can involve labour-intensive methods or leveraging technology. Technology-enabled processes offer sustainability over time and potential savings in both time and costs.

Within PV, scientific and medical literature review is particularly amenable to the introduction of natural language understanding technology. This is due to the structured nature of the data and the relatively low ratio of AE information to the volume of journal articles reviewed. To expedite and ensure accuracy in identifying articles containing reportable AE information, a simple three-step approach has been developed utilising a technology solution. An initial test of the technology solution showed that this new approach could pick up all AE terms with no false-negative results and go on to deliver further benefits.1

Literature review: A starting point for technology-enabled approaches

In searching for a “beach head” in PV for technology to demonstrate its capabilities, literature review of scientific and medical publications is a good starting point. A PV literature report review involves monitoring a set of journals at least once a week as mandated by regulation to see if any reportable AEs can be identified for a particular drug.

This area of PV lends itself to the involvement of technology for several reasons. First, the articles are written by professionals using standard medical terminology. Second, the articles exist in text format and are therefore relatively easy to search for. Third, the ratio of AEs found to volume of articles reviewed is very low. These factors make it possible for technology-enabled approaches to show significant benefits in the review of literature reports.3

Does technology play a role in PV literature review?

The recent advances in machine learning and natural-language processing require this consensus to be revisited, particularly in scientific and medical literature review. Machine learning is increasing the impact of technology, particularly in work processes that were previously considered complex.

The current process of literature review involves manual screening of articles in a set of journals to select those of interest for further investigation. The selected articles are then reviewed by a trained PV professional for AE information that meets the reporting criteria. Finally, the articles with reportable AEs are then forwarded to the PV case-processing and case-reporting team. This process is very time consuming, with a low percentage of AEs found for each review cycle.

By embedding technology (natural-language processing) within the literature-review process, the time taken to do the work can be significantly reduced, saving the expert intervention to focus on those parts of the process on which it will have the most productive impact.

How does automation work for literature search and review?

There are three simple steps to carrying out a literature review of scientific and medical journals using existing technology.

  • First, the search string needs to be established by ensuring that all the relevant AE search terms are uploaded, as well as all the journals that are regularly searched. This involves loading the MedDRA dictionary, names of medicines, commonly used AE terms, and journal articles onto the technology solution. The uploaded information can always be updated to include new search terms or journals.
  • Second, a search schedule must be set to run and identify any articles that contain the names of drugs (branded and generic) for which the Marketing Authorization Holder (MAH) has PV responsibility. For articles containing references to the specific drugs, another search must be conducted to identify those that contain predetermined AE search terms.
  • The third step involves analysing the output of search. For articles that have the names of drugs and at least one identified AE, a context analysis can be performed to better understand the nuance around the identified AE term; for example, the word “headache” (possibly negative) may have been identified, but the context analysis may reveal that the full statement was, “the patient did not experience any headache” (most likely positive).

The articles from the process that refer to reportable AEs can then be reviewed by PV professionals, with priority given to articles with negative AE contexts. Outputs with no drug names and no AEs identified (usually making up the bulk of journal articles reviewed) will require no further action. After a search cycle, a quick review can be done to see if any adjustments need to be made to the search infrastructure (additional AE terms, better context analysis, etc.), so the next run produces even better results.

Effective medical literature monitoring workflows with AI

Incorporating technology solutions into the existing PV literature review process has led to the creation of a new proposed process for carrying out literature-report reviews. This new process shows that significant savings in manual hours can be accrued by employing technology-enabled approaches.

One AI tool is a scientific literature monitoring platform used for active safety surveillance which is simple to use, fully web-enabled, and powered by AI. For any CRO, the monitoring of scientific literature is a high-volume activity requiring immense time from specialists each day. This has been tackled by developing an AI tool with end-to-end screening solution addressing the real requirements.

Achieving process efficiency offers the following benefits:

  • Increases productivity: More efficient processes would help increase output with the same resources.
  • Reduces errors: Efficient processes leverage automation technologies, which will reduce and even eliminate common errors.
  • Manual tracker maintenance elimination: Unerring approach reduced the time involved in entering the data manually and potential loss of data.
  • Audit/inspection readiness: All documents/sources maintained within literature tool, no room for errors in documentation.
  • Improves customer satisfaction: With leaner processes, employees can focus on the activities that increase customer satisfaction.
  • Lowers operating costs: Reduced lead times and eliminated instances of waste would ultimately lower overall operating costs.
  • Improves employee engagement: When processes are optimised, employees can focus on more productive work.
  • Maximises resource utilisation: Understanding how a process works helps in forecasting and utilising the resources to their full potential.4

AI productivity gains can save professionals time and effort. They can also lead to results of higher quality/data accuracy. By focusing time on results that matter, more articles/abstracts can be screened, leading to strategies that more comprehensively search the literature.

Comparing the current labour-intensive approach to literature review with the proposed technology-enabled approach, the introduction of technology can yield significant savings in manual hours and redirect the same efforts towards aspects that necessitate the attention of highly trained PV professionals. All of this can be achieved while maintaining the accuracy of the PV process.


  1. World Health Organization. Essential medicines and health products: pharmacovigilance. 2016. pharmvigi/en/
  2. Mittmann N, Knowles SR, Gomez M, et al. Evaluation of the extent of under-reporting of serious adverse drug reactions: the case of toxic epidermal necrolysis. Drug Saf 2004;27:477–87. doi:10.2165/00002018-200427070-00004.
  3. Enejo, B. Aisabokhae, E. Pharmacovigilance literature review in the age of precision medicine.
  4. Iain J. Marshall & Byron C. Wallace Toward systematic review automation: a practical guide to using machine learning tools in research synthesis

About the authors

V S N Raju DatlaV S N Raju Datla is associate director of PV at Navitas Life Sciences and has 13+ years of experience. His expertise includes processing/reviewing, handling, and managing end-to-end PV services (clinical and post-marketing ICSR, literature monitoring, aggregate reports, signal management, and RMP).

Shabana Banu ShaikShabana Banu Shaik is a senior drug safety associate in PV operations at Navitas Life Sciences. She is a Pharm D graduate and has 6+ years of experience in literature monitoring and end-to-end ICSR processing. Shaik is fascinated by streamlining and automating global literature monitoring.

Dr Pushpa BasavanapalliDr Pushpa Basavanapalli is director of PV at Navitas Life sciences, with 15 years of PV experience. She is responsible for managing end-to-end PV services and known for her ability to establish strong relationships with sponsors.