The business benefits of text mining in life sciences

Teams will need to harness technology strategically and set up processes to ensure that the contents of the dossier match the contents of the IDMP/SPOR dataset for each submission. Under IDMP, Substances Products Organisations and Referentials (SPOR) data services provide the vehicle for implementation of ISO IDMP standards in the regulatory and e-health worlds.

One option is to pull data from documents as an ongoing operational process, but this approach is likely to be very labour-intensive and offers companies very little additional benefit beyond compliance.

The opposite option is to leverage well-structured data to generate content. Structured content-authoring technology, in which documents are assembled automatically from pre-approved content fragments or data sets, would appear to be the optimal long-term option. However, the technology is not yet mature enough to offer a failsafe and simple-to-use solution, allowing dossiers to be created intelligently using approved source data.

A better approach, at least for the time being, is to establish parallel processes to prepare documents and data, keeping both in tight alignment and ensuring this is the case as a quality control requirement until the final submission.

In this context, companies would do well to harness an already proven technology: advanced text mining. This has strong potential, both at a data extraction and quality checking level, and for ongoing data and content maintenance.

The accuracy of such tools has reached around 95% in the context of automated data extraction, meaning that teams can place a lot of trust in it, saving human resources for an oversight role or to home in on more complex use cases.

Text-mining technology uses machine learning and natural language processing to help teams detect patterns or data points in existing documents, such as content around the composition of a drug, any counter-indications, or manufacturing detail. Once identified, it can extract this information and encode it properly using the correct controlled vocabularies and flow it into the company’s RIM system for onward processing.

On top of the technology’s strong track record at doing this accurately, text-mining tools are also very good at detecting whether the original data used in the documents was wrong, flagging this as a potential quality or consistency issue.

In initial data collection use cases, where text mining is already gaining traction, the technology is helping to improve the efficiency of IDMP data extraction from a range of different documents, automatically populating RIM data records directly from those static files. This provides teams with a good foundation for data enrichment, allowing skilled professionals to focus their time on populating additional fields that are now needed.

At an ongoing data maintenance level, advanced text-mining tools support proper data validation and user guidance to ensure that data is and remains complete, consistent and properly encoded, ready for a final review and approval by a human supervisor. This vital validation step ensures that discrepancies are picking up and gaps identified and flagged to the experts overseeing the data quality.

The return-on-investment potential of advanced data-mining tools in both data extraction and data maintenance use cases is impressive, as long as the technology is harnessed appropriately within the context of end-to-end regulatory information management processes.

In a data extraction context, where a set of documents must be read to populate product records, potentially taking someone four hours per record, a text-mining solution can, very conservatively, halve that time. With a potential saving of hundreds of euros per record, companies processing tens of thousands of authorised records per year could see cost savings run into the millions.

For data and content validation – checking the consistency of data and eCTD dossiers – the potential to use smart text mining to compare product records with submission document content is enormous and can generate considerable business value. By at least halving the current error/discrepancy rate via automated content validation, companies could, on average, yield a cost saving of hundreds of euros per submission, multiplied by 10,000 submissions annually.

Text mining can play a vital and direct role in improving efficiency, reducing costs, improving quality and minimising errors. More than that, advanced text mining has a meaningful role as part of a broader, end-to-end RIM capability, aiding planning, editing and formatting throughout, through its ability to validate data across the entire lifecycle.

Once responsible teams are made aware of text-mining solutions, it usually takes only a small proof-of-concept study to showcase the potential and lay to rest any concerns about the technology’s accuracy and efficacy.

Ideally, text-mining technology should be deployed seamlessly as part of a broader RIM project, as part of an IDMP data migration initiative, as companies press on with data cleaning, structuring and importing, ready for the IDMP go-live date.

Whether or not teams are exposed to the technology directly, when assessing how they will accomplish their projects and keep within their allotted timeframes and budgets, text-mining should be very firmly on the radar.