By Chris Funk, Ph.D., Senior Medical Informaticist, Wolters Kluwer, Health Language
Twitter: @wkhealth
Today’s healthcare leaders would be hard-pressed to design forward-thinking IT strategies without considering advances in artificial intelligence (AI). A term that is taking healthcare by storm, AI represents a wide range of technologies that draw from existing data to learn, reason and adapt—ultimately helping users and their organizations answer mission-critical questions and improve decision-making.
In healthcare, two sub-groups of AI, natural language processing (NLP) and machine learning, are already delivering value across a variety of use cases and producing notable return on investment by extracting critical information from free text—the unstructured documentation that comprises as much as 80% of the patient record and is largely missing from present-day analytics strategies.
Different from traditional natural language processing techniques, Clinical Natural Language Processing (cNLP), is tuned to understand healthcare’s specialized, unique language, and is able to address vital tasks such as identification of medical terms, interpretation of attributes (e.g. negation, uncertainty and severity) and mapping terms within a clinical document to industry standards. Having a lexicon of how physicians document common jargon, abbreviations, misspellings is important to capture the complete clinical view of the patient so it can be integrated with all other data sources.
Because unstructured documentation elements can serve as the basis for determining medical necessity, informing Hierarchical Conditions Categories (HCC), and improving quality measure reporting, healthcare organizations must have access to these valuable insights to improve performance, quality of care and reimbursement.
cNLP: The Value Proposition
Much of reporting and analytics today is built around analyzing, patient cohorts, which are groups of patients sharing specific characteristics. For example, the following patient characteristics may be used to define a diabetic patient cohort: elevated HbA1C lab value; taking medications such as Metformin; symptoms such as increased thirst and hunger.
There are many reasons for defining patient cohorts such as quality measures reporting and population health initiatives. However, the success of these efforts relies on the healthcare organization’s ability to accurately and completely identify all patients who meet the pre-defined criteria of the cohort. The following use cases demonstrate areas where cNLP can add notable value to patient cohort strategies:
- Risk Stratification: cNLP can help care managers identify high-risk patients by accelerating patient chart review and extracting pre-identified information from free text fields within the electronic health record (EHR). Even with complex, well-documented diagnoses such as diabetes, key indicators, like eye and foot exams, are often missed. While this data is valuable for determining the severity of a diabetic patient’s health, documentation of these exam orders does not always show up in structured, reportable EHR templates.
- Quality Measure Reporting: Reimbursement, reputation and the bottom-line are tied to the accuracy of quality measure reporting. For example, payer organizations must report quality measures for the Healthcare Effectiveness Data and Information Set (HEDIS) program, which impacts publicly-available Centers for Medicare and Medicaid Services Star Ratings and reimbursement. Some quality measures include inclusion and exclusion criteria for conditions such as, acute bronchitis. Yet, documentation demonstrating the secondary diagnosis needed to demonstrate this criterion is often found in free text fields notated by the clinician, at the point-of-care.
- Medical Necessity Review: cNLP can extract key data required by payers to justify coverage of prescribed therapies. For example, if a physician wants to prescribe an MRI, elements supporting the necessity of this test may reside in free text.
- Predictive Analytics: When cNLP is applied to extract and normalize clinical insights to industry standards, healthcare organizations can more accurately empower predictive analytics and machine learning models. For example, care teams can combine structured and unstructured data, real time labs to predict sepsis in real time.
Forward Thinking Technology
Today’s providers and payers are dedicating specialized resources, often clinicians, nurses and informaticists, to do the work of cNLP. These high-paid, skilled professionals manually comb through patient medical records looking for clinical insights and data required to identify gaps-in-care or documentation. This process is tedious and time consuming—reviewing a single patient record can take anywhere from 30 minutes to three hours depending on the information being collected and can consist of up to 700+ pages.
Fortunately, there is a better way. cNLP technologies enable significant workflow efficiencies and reduce the potential for human error by automatically extracting and codifying data contained in unstructured text. Advanced cNLP solutions can aggregate and normalize key data, including problem lists, diagnoses, medications, allergies or labs, so that it can be semantically shared or ingested into a data repository.
In addition, comprehensive solutions ensure that the context of clinical information extracted is properly identified to provide deeper insights. For example, cNLP can help identify which HCC or HEDIS measure is appropriate.
The future of cNLP is promising as the industry endeavors to capture the wealth of information that currently resides in unstructured documentation. As healthcare organizations continue to mature their analytics and AI initiatives for next-generation healthcare, cNLP must become a priority in order to succeed in a competitive marketplace.