Using Natural Language Processing to Extract Critical Data in Digestible Format

Natural Language Processing (NLP) technologies allow machines to use their algorithmic power to break down the complexity of human language. At Dovel, we apply NLP to complex data sets to display it in a concise and digestible format. Our Innovation and Technology Group (ITG) recently created a new application, Vaccine Adverse Event Surveillance Portal. This powerful dashboard, hosted in our Discover platform,  uses NLP to integrate complex sets of data ingested from the Vaccine Adverse Event Reporting System (VAERS), a system for aggregating adverse event reports from doctors, nurses, volunteers, and other medical professionals, as well as data from DailyMed.gov which contains the official labels for vaccines, drugs and other therapeutics.

The use of NLP is critical in filtering this data as there is little standardization in the language used to report adverse events. For example, a doctor might diagnose a patient as having a “fever,” but a nurse might diagnose the patient with “pyrexia,” which are different ways to refer to the same condition. Similarly, the difference between “symptoms” (cough, muscle pain, congestion) and a “diagnosis” (respiratory tract infection) is sometimes blurred and adds another layer of complexity in how medical data is reported. Our solutions combine NLP tools with a strong semantic foundation that utilizes terminologies such as MedDRA, SNOMED CT, etc. to harmonize and link medical language. This allows us to dig into medical data to get an accurate picture of the reported data and avoid data duplication, data gaps, and ambiguous results.

Applying NLP not only helps to digest the complex data, but it also allows us to visualize the data for easier consumption and action. We did this by using the Tableau Data Visualization Tool. The resulting dashboard provides a user-friendly interface to explore adverse events reported against the top 20 most common vaccines in the United States with a few clicks. Recently we have included the results of the COVID19 vaccines adverse events reported to VAERS. Information can easily be organized based on questions like, “What were the most commonly reported adverse events for the flu vaccine since 2017?” Data can also be quickly sorted to see if there are a significant number of reports for adverse events that are unexpected (i.e., that were not encountered during vaccine development) or if a certain part of the country experience more reported incidents of adverse events than the rest of the country. We have further added demographic information such as gender, age, and location information.

The most powerful use case for this portal is the ability for organizations like the FDA to identify opportunities for post-market analysis.   By identifying the adverse events (AEs) that are not present in the label (i.e., unlabeled AEs), but have a higher percentage of reports, the portal provides a powerful visualization tool that manufacturers can use to update their current labels and provide better protection to the consumers.

Dovel has long been involved in tracking adverse events in our nation’s food and health products. Dovel helps create and maintain data standards that are used as the semantic building blocks of data exchange and aggregation. Our subject matter experts work with RxNorm, UMLS Metathesaurus, SNOMED CT, Medical Subject Headings (MeSH) as well as hundreds of other terminologies that are being used as data standards. We developed a solution that allows hospitals to access, submit, interact with, and consume confidential and protected hemovigilance data to ensure the safety of the nation’s blood supply. Our team also led the modernization of the database for all food and drug related issues and recalls, tracking and monitoring incidents from first report to closure of the issue. The Vaccine Adverse Event Surveillance solution is currently tracking around 20 vaccines including COVID-19 vaccines.

Visit us at Discover to see this dashboard in action.