AI in Healthcare: Practical Applications of Clinical NLP
While there’s been a lot of buzz about the promise of artificial intelligence (AI) in healthcare, health leaders are recognizing that AI holds the potential to diagnose and treat disease, improve processes, and better manage underlying operational, financial, and patient health data through which they can innovate and maximize value.
Specifically, within healthcare, Natural Language Processing (NLP), a specific branch of AI, has quickly proven value by enabling healthcare organizations the ability to efficiently leverage unstructured data, which represents nearly 80% of all healthcare data.
By applying NLP technology, healthcare organizations can optimize manual workflows of reviewing unstructured text within patient medical records in order to uncover clinical insights to empower high value uses cases.
In the recent webinar “Applying AI in Healthcare: Practical Applications of Clinical NLP to Drive Value in Your Organization” (watch now on demand), hosted by Health Data Management, three experts from Wolters Kluwer discussed how hospitals and health plans can take advantage of using clinically tuned NLP technology to improve patient outcomes and enhance performance.
Before jumping right into the topic of AI, Brian Diaz, Sr. Director of Strategy of Health Language at Wolters Kluwer, kicked off the webinar by discussing the foundational concepts of healthcare data, and why it is essential to maximize the use of all healthcare data types in order to get the maximum value when deploying AI technology. Brian walked the audience through a graphic that illustrated the different sources of healthcare data.
Leverage All Data for Maximum Value
First is claims data which offers information on the services provided to a patient. This data type is structured and is represented in industry standardized terminologies such as ICD-10, CPT®, UB-04®, HCPCS etc. Because of the structure of the data, this information is the most widely used and shared. However, claims data lacks the clinical detail on a patient’s healthcare interaction and experience.
Second is clinical data which represents near-real time information about the patient’s encounter with the healthcare provider. While this data offers the most insight on a patient’s healthcare journey, the data itself can be represented in structured, semi-structured, or even unstructured data types.
Labs and drugs can be coded in standard terminologies such as LOINC® and RxNorm within the electronic health record (EHR). Yet the medical history, pathology, and diagnostic reports can be captured in PDF images and free text observation fields also within the EHR. Due to the variability of these data sources, clinical insights are often overlooked because of the manual efforts required to uncover them.
The third source of data that Brian mentions are emerging data types such as telehealth, genomics, and social determinants of health (SDoH). These data sources are new and typically unstructured.
There have been regulations introduced throughout the years to standardize how data is represented to promote interoperability. However, it is often a challenge to manage and leverage the variability of healthcare data.
Luckily AI can help us accelerate the transformation of healthcare data - from information to knowledge and ultimately to wisdom, so we can improve patient outcomes and succeed under value-based care.
NLP Extracts Valuable Insights from Unstructured Text
At this point of the webinar Dr. Chris Funk, Senior Medical Informaticist at Health Language, explained what NLP is and how it is being used within the healthcare industry. The ultimate objective of NLP is to read, decipher, understand, and make sense of the languages in a matter that is valuable for a specific use case.
While NLP has been around since the 1950’s, its introduction into the healthcare industry is relatively new due to the complexity and variation of the clinical domain. Traditional NLP technique rely on lexical and syntactic features – all present in the average newspaper article or research paper; however, because of healthcare’s sublanguage which is comprised of incorrect formatting and sentences, incorrect grammar, clinical jargon, and variety of acronyms, etc., it is naturally more difficult for computers to process and comprehend. With that being said, it is essential that traditional NLP be clinically trained before being successfully deployed within the healthcare environment.
Today, NLP is primarily being used to identify and extract relevant clinical information from unstructured documents, and then is resolving the results to an industry recognized standard, so that it can be understood by a computer.
Once a healthcare organization decides to explore NLP technology, the next step is to understand what kind of NLP technique is appropriate in order to deliver results for your specific use case. Chris outlines the four most common techniques used to train NLP models: dictionary based, rules based or expert systems, machine learning, and deep learning. Chris makes the point that it is important to understand that results and output may vary due to the selected technique’s generalizability of the algorithms and the customization and interpretability of the results. The takeaway here is one size does not fit all. However, for optimal results an ensemble of techniques may need to be deployed in order to completely solve a client use case.
Importance of Clinical Expertise
While the technology is very important, what is perhaps the most important component of any solution is the level of clinical expertise. Any NLP solution must have the intelligence to understand the clinical relevance in order to extract appropriate information for a specific use case. This clinical expertise includes a thorough understanding of:
- Healthcare terminologies - the language of healthcare
- Mappings – to understand the complexity of clinical conditions
- Value sets – used to define population cohorts, decision support rules, or clinical quality measures
- Rules and Guidelines – of how to appropriately act on the data uncovered, such as treatment rules or reimbursement and billing procedures.
To summarize, Chris explains that the end goal is to choose a solution that can uncover the most clinically relevant insights previously trapped in unstructured data types which can deliver value for an organizations specific use case.
Practical Applications of AI: Three Use Cases
Next the presenters review three distinct use cases where clinical NLP has been deployed to deliver significant value. I encourage you to watch the webinar On Demand to hear directly from the experts on these three use cases.
- Hear from Chris Funk, Ph.D., Sr. Medical Informaticist, on how clinical NLP is being used to automate the interpretation of pathology reports to increase accuracy and speed of detecting Adenomas within pathology (colonoscopy) reports, resulting in not only timely detection and treatment for the patient, but also improved quality measure reporting.
- Hear from Dr. Stephen Claypool, MD., Medical Director of Surveillance, about how clinical NLP is powering early and accurate Sepsis detection intelligence built into real-time clinical surveillance solutions at the point-of-care.
- Hear from Brian Diaz, Sr. Director of Strategy, about how clinical NLP can accelerate the chart review process to find supporting evidence for accurate risk adjustment coding, resulting in optimized workflows and maximized reimbursement.
To close out the webinar, Brian concluded by reinforcing that no matter the use case, the value cNLP brings is clear in that it can automate labor-intensive processes, optimize existing workflows, and increase the accuracy of the data used to inform initiatives that influence patient care and reimbursement.
When considering a vendor to partner with, healthcare leaders should make sure they are well versed in all elements of helping you form a foundation of high-quality data. Your partner should be able to help you establish a single source of truth for all reference data, be able to connect the dots between claims and clinical data by mapping to standard terminologies, and must be able to extract and codify clinical insights found in unstructured text to provide a more complete view of patient health.
If you missed the live webinar, watch it on demand by clicking here.
The Health Language Clinical Natural Language Processing (cNLP) Solution optimizes manual medical record review, automates the review of unstructured data, extracts clinically relevant data, and codifies extracted data to industry standards. Watch our two-minute video or speak to an expert to learn more about to the Health Language cNLP Solution.
CPT® is a registered trademark of the American Medical Association (AMA).
LOINC® is a registered trademark of Regenstrief Institute, Inc.
UB-04® is a registered trademark of the American Hospital Association (AHA).