Technology is advancing at an astonishing rate, and artificial intelligence techniques that were “nice to haves” just a few years ago are now considered core competencies for healthcare organizations to keep pace. Natural language processing, or NLP, is a great example. The sheer volume of data in healthcare today, and the fact that most of it is unstructured, means that the only pragmatic option for utilizing and harnessing this data is to adopt innovative technologies, such as NLP.
As the focus shifts from niche solutions for specific problems to more transformative NLP capabilities across the enterprise. The challenges organizations face are the need to blend flexibility with robustness in order to meet the demand of processing huge volumes of data across domains and use cases.
Linguamatics’ 20 years of applying our NLP technology across diverse areas of healthcare, has positioned us well to deliver on all these fronts: flexibility, robustness and domain expertise. Here, I will discuss a few examples detailing how our NLP Data Factory is being applied to feed predictive models, improve adverse event coding and understand the voice of the customer.
Feeding predictive models in insurance
Processing large volumes of data is something that Linguamatics customers have been doing for some time now. For example, one customer surfaced real world insights from 34 million medical records to support a large device company in assessing the performance of cardiac devices. However, recent transformations in the Linguamatics pipeline have meant that this scale of processing can now be automated to create in-stream processing of millions of documents per hour. This is possible through parallel processing managed by an orchestration engine within the NLP Data Factory.
In one example, a large insurer is using our NLP daily to process hundreds of features within their unstructured data at a rate of 8 million documents per hour. The results of this approach are game changing in that a single NLP tool is bringing benefit to multiple teams across the organization, allowing them to extract information from claims to feed downstream underwriting algorithms and predictive models. This capability to reap wholesale benefits without the need for multiple vendors or tools is something that more organizations are seeking.
Meeting pharma’s high-demand safety case processing needs
On the journey from drug discovery through development to post-market surveillance and patient care, the amount of textual information is astronomical, and patients and regulators expect zero gaps in intelligence. That means nuggets of insight within everything from explanatory talks to scientific literature, social media, adverse event reports, and more need to be captured and assessed. When you consider the sheer volume of unstructured text out there applicable to safety, it is actually astonishing that we ever navigated adverse event reporting manually.
A challenge that other NLP offerings have encountered is that they are omni-industry solutions and not tailored to the needs of healthcare. Anyone who has worked in our industry knows that the natural language of drug development and healthcare delivery is a far cry from the natural language of the outside world. Tapping into IQVIA’s deep healthcare domain expertise has allowed us to embed the right terminologies deeply into our NLP Data Factory, which was built to tackle broad, high throughput areas like safety case reporting with phenomenal speed and accuracy.
In one customer example, we used the NLP Data Factory to auto-code adverse events into a standard MedDRA format, reducing their manual coding time by 50 percent while improving quality at the same time. At the onset of the project, the client was finding that 70 percent of their incoming adverse events required manual review. Now, with the right NLP queries in place, the NLP Data Factory is used across the customer’s enterprise to double their auto-coding capacity, while reducing manual coding time by 50% and improving coding consistency compared to manual review.
Capturing the voice of the customer on topics and trends
In the healthcare industry’s pursuit of therapies that meet the needs of patients, the importance of the customer voice cannot be overstated. As the saying goes, successful companies don’t find customers for their product, they find products for their customers. In one particular use case, we worked with a top pharma company who wanted to track trends based on product-related questions across 50 different categories. These categories ranged from simple questions like adverse events or demographic information to more complex topics like contraindication, whether patients are switching medications, off-label use, or an uptick in dose-related questions. They also wanted to identify novel requests and topics coming out of customer verbatims.
From similar collaborations with other customers, we knew that the best way to capture the voice of the customer on standard categories and novel topics concurrently is using a hybrid approach to pair NLP with machine learning. First, we set up an NLP Data Factory workflow to tag and enrich the incoming verbatims with those categories that were already well known. Then we took that same information and used it to identify the right attributes for machine learning classifiers and built an algorithm to cluster the verbatims based on similarity. The customer now uses the resulting system across 20 different product brands and diverse therapeutic areas to better understand what topics people are talking about, how they are changing over time, and signals on novel topics.
So who are the winners?
The launch of the NLP Data Factory has been a massive differentiator, allowing our customers to apply our award-winning NLP to huge volumes of data. It has a flexible deployment model and easily scales with our customers' needs, integrates with diverse sources of unstructured data, and allows you to start extracting value from text on day one. As more technological advances surface, healthcare organizations will continue to gain greater efficiencies and further reduce operational and development costs. Most importantly, patients will increasingly enjoy outcomes and therapies tailored to their true lived experiences as captured in the rich textual data NLP allows us to explore. That’s a win we are all trying to achieve.
