Blog
What do top pharma want from their AI vendors?
Jane Reed, Director NLP, Safety & Regulatory
Oct 19, 2023

There is a huge amount of discussion going on, around the value – and potential pitfalls – of artificial intelligence, particularly since the arrival of chatGPT and other large language models. From my perspective, the concerns that we are hearing tend to be around the pros and cons of established tried and trusted technology to solve key challenges in drug development, compared to the potential of new innovations. And our experience is that the best approach is using a blend of technologies; bringing the right expertise to the table to know when to apply the most appropriate AI/ML method. And balancing cost, accuracy and convenience is important.

I always enjoy listening to what our clients have to say on the matter. Recently, I was chatting to an attendee in one networking session at our NLP Summit. This person works at a top 10 pharma that have been getting value from IQVIA’s Natural Language Processing technology for many years (almost since the founding days of Linguamatics), and they talked about the benefits of working with IQVIA for their NLP needs, compared to some of the up-and-coming opensource tools.

It was an interesting discussion and so I thought I’d share some of their key points. Maybe these resonate with you as you look for new solutions:

  • Reliability of software
    IQVIA NLP have been providing software and services for 20+ years, and from the customer viewpoint that means that software quality is assured; it has been through the rigorous test cycle that is expected for commercial off-the-shelf software. The client noted that they are focused on drug development not software development, and it’s best to play to their strengths. Opensource tools are great for experimentation but not necessarily the best for robust work or production workflows.
  • Compliance around content
    Access to key public domain data sources is critical, but “public domain” is often mis-understood; it does not mean that data and documents on public websites can be ingested and provided with AI/ML augmentation. This demands additional licencing, which again is something that experienced vendors can provide.
  • Auditability
    The customer commented that it’s critical to be able to understand how data is generated, for milestone investment decisions, defining go/no go criteria. This is straightforward with white-box rules-based NLP, but less transparent for black-box machine learning technologies.
  • Usability
    This customer has been using our interactive extraction GUI (previously known as i2e), for several years, and finds it very user-friendly for non-coders. In addition, we discussed issues around the compute time needed to run a query – this is significantly lower for most rule-based methods compared to that needed for deep learning models.
  • Named entity recognition and normalization to standard ontologies
    Being able to find all the key entities in life science is important. And this customer commented that it was just as important to be able to normalize these to up-to-date standard ontologies, such as MeSH, NCI Thesaurus, MedDRA. It’s a considerable burden to maintain and update these, and a significant benefit to have this provided by a trusted vendor. These ontologies, and the ability to standardize to the latest version, are hugely valuable to enable reusability and interoperability of text-mined data with other data sets.

The speakers at our two Summit days evidenced that researchers, scientists and clinicians in healthcare and pharma are aware of the benefits of different NLP tools and know when to use the best tool for the task. For example, Chad Konchak from NorthShore University Health System talked about how they are leveraging analytics capabilities and a new AI initiative around natural language processing to aid in health equity; with the mission of reducing healthcare disparities in the care they provide. Jenny Uyei & Yuting Kuang (IQVIA) talked about using a combination of rules-based NLP in combination with machine learning for literature search and evidence synthesis projects such as synthesising relevant real world data on COVID-19 vaccine effectiveness, antiviral effectiveness, and long COVID. And Emilie Louvet (AstraZeneca) talked about using rules-based NLP and deep learning (e.g. BERT) to find relevant clinical trial endpoints from literature, to assist clinical trial study design.

If you’d like to learn more about IQVIA NLP and how we can help you find the best NLP tool for the task, please contact us, or listen to the talks from the Summit here.

Related solutions

Contact Us