Unlocking the power of unstructured data in healthcare: How NLP is transforming patient care

• NLP is transforming healthcare by converting unstructured data from EHRs into actionable insights, enhancing patient care beyond traditional methods
• The impact of NLP on healthcare is multi-faceted, with the potential to improve hospital operations, drive medical research, and enhance quality of care
• Successful NLP integration requires addressing diverse hospital infrastructures and ensuring data privacy and quality, with partnerships enhancing scalability and impact

In today’s digital age, electronic health records (EHRs) have become the backbone of modern healthcare systems. They are often the primary source for a patient’s medical history, past and present conditions, and ongoing treatments. However, many challenges still remain with regard to EHRs, including the disruption of clinicians’ workflows, time-consuming data entry, and the negative impact they can have on the doctor-patient relationship. ¹

Very often healthcare professionals (HCPs) avoid ticking boxes and filling in short answers in their medical notes, preferring to write in the free text portion of medical exam forms. This is because in healthcare, context, possibility, negation, and patient input are pivotal, and can only be ‘coded’ in what is called natural language (in opposition to ‘formal language’ as in those ‘spoken’ by databases). This makes it difficult to interpret and exploit the existing data with legacy technologies, creating a big dependency on manual reviews. Fortunately, all that is changing.

Jorge Tello, engineer, founder, and CEO of the Spanish healthcare technology company Savana, explains how clinical natural language processing (NLP) can be used to address this issue, creating new opportunities to improve patient care and potentially revolutionize healthcare delivery.

HT: What are some challenges with healthcare data today, and how does Savana bridge these gaps?

Jorge Tello: Healthcare data is incredibly intricate and complex. A lot of valuable information is embedded in unstructured free-text fields, like doctors’ notes and patient narratives. One of the big challenges is that traditional electronic medical record (EMR) systems tend to focus on structured data, such as lab test results. Unfortunately, this means they miss out on the nuanced and contextual details that free text captures.

To bridge these gaps, we use advanced text analysis to extract information from free-text fields, capturing the full complexity and context of clinical encounters. By integrating both structured and unstructured data with a quality-driven mindset, we provide a more comprehensive view, ultimately improving patient care. So, in a way, we transform those narrative and complex elements of clinical data into actionable insights, addressing the shortcomings of traditional structured data approaches.

HT: Considering the limitations of traditional EMR systems that focus mainly on structured data, how effective are international classification of diseases (ICD) codes in capturing the full complexity and nuances of patient information?

Jorge Tello:  There are thousands of  ICD codes, but they often feel like administrative simplifications of complex diseases. When you see the ICD code assigned to your condition, it can sometimes appear that it doesn’t fully capture the specifics of your disease. ICD codes are useful for administrative, logistic, and budgeting decisions, but they lack the context needed for clinical decisions, which require more detailed information.

For example, the evolution of a disease, the sequence of events, and how symptoms change over time can’t be effectively captured just through codes. These require detailed narratives—a doctor’s thought process and the surrounding context of the patient’s condition, which aren’t reflected in ICD codes.

After recognizing this problem, we developed a knowledge model that includes these detailed report descriptions. We don’t force the doctor to write in a structured way. Instead, we let doctors document information naturally, as they know what’s important for each case. Our technology then organizes this information into the right “boxes” of knowledge afterward.

This is where NLP comes in. We created and trained over 30 natural language processing models per language for six of the most spoken Western languages. This allows us to fill in these knowledge boxes accurately when we encounter a text. Importantly, we also measure the error in our system to ensure clinical decisions are reliable. We do it through a self-developed methodology that we published, where doctors themselves create gold standards against which results are compared.² Unlike large language models, our results are consistent and reproducible.

This approach helps us address the shortcomings of ICD codes and traditional EMR systems, allowing for more accurate and context-rich clinical information. So, that’s how we tackle the problem.

HT: What are the main use cases of NLP for the healthcare industry and how is it applied in the field? What are its main benefits in terms of patient care and care delivery efficiency?

Jorge Tello: In healthcare, NLP technologies can read and analyze free text in medical records, extracting unstructured data and turning them into valuable insights.

From our perspective, we see three areas where NLP is greatly benefiting healthcare and health systems.

The first relates to health system management, where NLP can provide analytics and insights on how healthcare operations are running within and across hospitals. We can measure how things are done and which results are obtained. This data-driven approach allows healthcare managers and administrators to make informed decisions about resource allocation and process improvements.

The second is the impact it can have on medical research. By analyzing large volumes of patient data, we can compare treatment outcomes at a population level. A real-world example we have is where our technology was used to determine the most effective anticoagulant for cancer patients, leading to published findings that informed clinical guidelines.3 This provides medical societies with new insights into how better care might be delivered for specific conditions.  Furthermore, another important aspect is the identification and management of rare diseases. Many patients suffering from rare conditions are often misdiagnosed due to a lack of specialist knowledge. NLP can help specialists proactively search for symptoms and signs that may indicate a rare disease, ensuring faster and more accurate diagnoses.

The last and perhaps most important is how NLP can serve direct patient care since many individuals spend years visiting multiple doctors to get a correct diagnosis. Our technology can provide dynamic care suggestions based on a patient’s history and current clinical guidelines. We can for instance alert doctors to important patient information they might have overlooked, such as comorbidities or overdue routine tests, thus driving more proactive and personalized care. By leveraging NLP, we can streamline this process, ensuring that patients are directed to the right specialist sooner. This not only reduces the emotional and financial burden on patients but also leads to more effective treatment plans.

HT: What are the most significant challenges you see for integrating technologies like yours into healthcare systems, and what recommendations do you have to address these challenges?

Jorge Tello:  Implementing NLP in healthcare systems comes with several challenges. One of the biggest hurdles we face is the sheer diversity in hospital infrastructures.

Every hospital has its unique set of systems and processes, which means a one-size-fits-all approach just doesn’t work. We have to look at each hospital individually and come up with flexible integration strategies – whether that’s direct integration with more advanced hospitals or batch processing of information in other cases. While this requires an initial investment, the long-term benefits are well worth it.

Then there’s the all-important issue of data privacy and security. Medical records often contain sensitive personal information, and keeping that data confidential is crucial. At Savana, we have developed a module that anonymizes data locally before processing it. This means personal identifiers never leave the hospital, supporting both legal requirements and cultural concerns about data privacy. The same approach has to be taken for data quality management, with a special focus on clinical plausibility checks.

Another significant challenge is the hesitation around adopting cloud-based solutions, even though they’re often more secure than local systems. Hospitals need to have a balance between local and cloud operations to take advantage of the computational power required for NLP. The initial costs might seem high, but the long-term benefits – like proactive patient care, fewer diagnostic delays, and better resource allocation – far outweigh the expense. Proactive management of patient care not only leads to better health outcomes but also makes healthcare more affordable over time.

At Savana, we’re strong advocates for transforming healthcare into a more proactive care system. By working closely with hospitals and health authorities, we can show just how beneficial NLP can be, creating a win-win situation for everyone involved.

HT: Finally, how important are partnerships with larger companies for startups like yours?

Jorge Tello: Partnerships with large companies are incredibly important for startups like ours.

At first, we were a bit hesitant, worrying about potential conflicts of interest that could limit our opportunities. However, we’ve found that working with large companies that have a strong digital culture and significant industry penetration offers significant advantages.

These collaborations help us widen our network, boost our capabilities, and connect with more stakeholders in the healthcare space. Plus, the resources and expertise they offer are invaluable. These partnerships have been game-changers, allowing us to scale our solutions more effectively and make a bigger impact in patient care than we could on our own.