Companies in the healthcare industry utilize their large amounts of data in combination with natural language processing (NLP) techniques to enhance personalized care, identify high risk patients or automate tedious paperwork. Continuous research in the field leads to new promising applications, including exciting solutions to address COVID-19 and other major challenges.
We had the chance to talk to some leading NLP experts in the healthcare industry and discuss with them the role of NLP in the healthcare sector and the main obstacles when implementing NLP techniques.
Make sure to register for the Healthcare NLP Summit on April 6-7 to learn more about NLP best practices and real-world case studies in the healthcare industry as well as the latest open source libraries, models & transformers you can use today.
According to your opinion: what’s the most promising use of NLP in healthcare?
Doctor-assisting use-cases
“It’s important to remember that our aim is to assist doctors, not to replace them. Therefore, I’m more excited about uses that save doctors time and make important information easily accessible for them, than about complex NLP algorithms making precise diagnoses. Such doctors-assisting use-cases can be medical text summarization, machine-assisted documentation and coding, and on-demand information retrieval. The main principle is to reduce the amount of information and paperwork a doctor needs to deal with, instead of increasing it.
Another promising use of NLP is Population Health, identifying individuals with high risk of developing chronic conditions based on long-term medical records. Preventing people from getting sick is the best way to save time for doctors.” —Rachel Wities, NLP Data Scientist at Zebra Medical Vision
Predicting health conditions
“At the moment, healthcare is diagnostic. We treat a problem a patient is having by monitoring existing conditions. However, what I am really excited about is the move to predictive analytics. By applying NLP to the vast medical records that is currently available, I believe we can build a predictive analytics solution that would identify conditions before they occur.” —Mukesh Mithrakumar, Sr. Machine Learning Engineer at IQVIA
Improve patient care quality
“There are numerous areas where NLP can play a critical role in driving significant value in quality experience delivered to consumers and patients. Starting from the critical area of patient safety, NLP can assist in identifying safety data and potentially severe adverse events, flag them for immediate human review and report it to regulatory authority.
Global Medical Information teams can also use these technologies to mine inquiries and product quality complaints coming from consumers, patients and physicians, to serve them with right contents or potentially identify side effects, while they are also playing a key role in helping healthcare companies follow EMA regulations on relative efficacy analysis of products.
On top of that, NLP and text analytics algorithms can also transform the traditional methodologies and procedures used to evaluate and improve patient/consumer care quality. Evaluating healthcare professional performance and measuring gaps is a crucial task for insurance companies making the switch to value-based reimbursement.” —Prathamesh Karmalkar, Principal Data Scientist at Merck
Understand complex clinical notes
“Clinical notes contain a wealth of information that has frequently gone untapped. It provides a clear reasoning of how a clinical expert or caregiver thinks to derive a treatment plan or modify one. Automated learning of sophisticated clinical reasoning from notes is the breakthrough opportunity for NLP in healthcare.
Today, incorporation of NLP data into AI models is still in early stages, where we are focused on making use of entities extracted from notes and make use of temporal or causal relations.
True AI for healthcare is not just about building accurate classification or predictive models, but enabling explainable, clinical reasoning. If you read a clinical notes document, you can see a clear explanation for why a patient needs a surgery, or why he/she may be a suicide risk. We cannot recommend a surgery to a patient’s family or suggest that someone’s loved ones are at a risk of suicide because a model’s score went up from 0.6 to 0.85.” —Sutanay Choudhury, Chief Scientist at PNNL
How can COVID-19 challenges be addressed using NLP techniques?
Information extraction from previous research
“With a fast-spreading coronavirus across the globe, there is an extreme need for solutions to break the chain of infections. There is significant research and literature around similar challenges during previous epidemics, which may not be specific for the current situation but still very valuable. This might provide us with the right approaches, which will help us fight this battle. We, as AI & NLP researchers, should leverage this research, ideas, reports or any data to find close to accurate and quickly actionable insights to control the spread via medical or non-pharma interventions. With this, we should plan to build an engine which can help community members to find the right literature using the methods of NLP, Deep Learning & Search.” —Prathamesh Karmalkar, Principal Data Scientist at Merck
Rachel Wities, NLP Data Scientist at Zebra Medical Vision, also highlights the value of NLP to extract information from new research and further highlights the ability of NLP to simply the search results: “One main challenge of COVID-19, both for physicians and for the general population, is the increasing load of new information. Physicians struggle to stay up-to-date with the new research, which seems to be growing exponentially, while laymen struggle to understand the medical terminology in their Google search results, and can’t tell which information sources are reliable.
NLP summarization, text simplification and information extraction techniques can be used to help doctors and the general population separate the wheat from the chaff, and easily access relevant and reliable information.”
What is the biggest challenge you see in using NLP applications in healthcare?
Biased datasets
“The biggest challenge in using NLP in any application is ensuring equitable and respectful treatment of people. Recent advances in large language models, which are mostly trained on huge datasets of internet text, have shown immense promise in enabling powerful text-based analysis and interactions. But, they have also shown an ability to reflect the biased and harmful language that can be found in their training data. Healthcare is a field where disparate treatment of people can lead to serious harm, and so much care must be taken in deploying NLP solutions.” —James Wexler, Staff Software Engineer at Google
Gaining the doctor’s trust
“The biggest challenge, as always, is in the human element. The key to success in NLP healthcare applications is gaining the trust and cooperation of the clinical personnel who are supposed to use it. Much has been said about the importance of explainability, but explainability is just one part of it — using the correct precision/recall threshold can be just as important for gaining the doctor’s trust, and neither can be successful without a solid understanding of the doctor’s workflow and his needs. This mission is even more complicated given the fact that even the doctors themselves often can’t predict their needs in different situations, which makes flexibility and interactivity very important in such applications.” —Rachel Wities, NLP Data Scientist at Zebra Medical Vision
Thin margin of error
“Applying NLP techniques to healthcare comes with the same risk as self-driving cars, overpromising a revolutionary technology with a thin margin of error. Nobody dies if a BERT model makes mistakes in recommending my email subject, or the next word to type. But the margin of error is thin or would garner more headlines if an AI model recommends a surgery when it should not or qualifies someone as a suicide risk when he or she is not.
NLP-driven healthcare has not yet reached the technological maturity level of self-driving cars but we can look at them and learn how a few unfortunate incidents or embarrassing mistakes (which from a technical perspective is not embarrassing, we can explain) can cause a setback for a field in terms of funding, attraction talent, driving product adoption etc.” —Sutanay Choudhury, Chief Scientist at PNNL
Exponential growth in unstructured data
“Due to complex regulations and compliances, the healthcare and life sciences industries have been slow in adoption of text analytics and NLP. Industries are facing significant challenges in analyzing the data due to its unstructured textual nature. Because of massive digital disruption across the globe, there is a sharp rise in the generation of naturally written forms of electronic data. This explosive growth of unstructured clinical data, medical data, regulatory data and healthcare data has prioritized the use of innovative technologies of NLP and text analytics. The key challenges in adoption of NLP are exponential growth in unstructured data in the form of texts from various business teams within the organization. This ever-growing data remains untouched and therefore companies are unable to identify actionable insights and recommendations that can generate significant value for consumers and patients across the globe.” —Prathamesh Karmalkar, Principal Data Scientist at Merck
Data interoperability
"A key barrier to the utility of NLP is that of interoperability between data sources. Interoperability between data sources allows greater ease in communicating necessary patient data and, in turn, would improve patient safety through improved accuracy of clinical documentation relied on for providing a patient’s medical care." —Joseph Plasek, Postdoctoral Research Fellow at Mass General Brigham
Are you interested in learning more about NLP use cases and challenges in the healthcare industry? Register now for the free Healthcare NLP Summit coming up virtually from April 6-7.
