By Elizabeth Marshall, MD, MBA
The healthcare industry is beginning to realize the immense promise that natural language processing (NLP) technology holds in mining unstructured data for valuable information that can improve health outcomes and promote efficiency.
While NLP use cases such as prior authorizations and clinical trial matching certainly have potential, the technology’s greatest benefit may be its ability to uncover actionable insights that enhance both the quality and safety of care—insights previously buried in unstructured clinical documentation.
Over the last decade or so, the industry has made enormous progress in digitizing significant amounts of clinical, administrative, and billing data, but this effort has not come without problems. Up to 80% of this data is stored in an unstructured format, meaning that clinical and financial decision-makers often can’t access important information. Effectively, the data in their own systems is hiding in plain sight.
Rather than relying on expensive and time-consuming manual chart reviews, more healthcare organizations (HCO) are turning to NLP to mine unstructured data from the clinical notes and free-text fields of electronic health record (EHR) systems. NLP tools make unstructured data usable by automatically identifying and extracting key concepts from large volumes of clinical documentation. HCOs can then transform this information into structured data to improve the efficiency and precision of quality and safety initiatives.
What’s driving NLP adoption in healthcare?
Several recent industry trends have combined to spark interest in using NLP to unlock the value of the unstructured data that is essentially at HCOs’ fingertips but just out of reach. The first factor is simply the vast amount of healthcare data created since the digitization of health records. This data not only includes information from EHRs, but also patient-reported information from secured communication entered via emails in patient portals.
More data has equated to more issues with data integrity, as vital patient data in EHRs is often incomplete, missing, or spread over several disconnected provider information systems that may not be capable of sharing it. The industry regards data integrity seriously, ranking it as the second most significant technology hazard on a list of top patient safety concerns for HCOs, according to a report from the ECRI Institute.
Additionally, patient safety remains top of mind for most healthcare leaders—and rightfully so, despite progress across the industry in recent years. We still have plenty of room for improvement, as illustrated by this metric from the Centers for Disease Control and Prevention: On any given day, about one in 25 hospital patients have at least one healthcare-associated infection. Further, in 2016, an avoidable injury occurred in nine out of every 100 patient stays as a result of a bad medication reaction, an injury during a procedure, a fall, an infection, or other avoidable factors, according to the Agency for Healthcare Research and Quality.
Another key factor driving healthcare interest in NLP is the industry’s ongoing shift from fee-for-service to value-based care, which has led to increased focus on tracking patient safety, closing care gaps, and improving quality-reporting efforts. To obtain the full reimbursement that they’ve earned in value-based agreements, HCOs must measure, track, and report on their quality activities. Value-based care has also highlighted the importance of addressing social determinants of health (SDoH), which frequently exert substantial influence on health outcomes.
The problems for many HCOs, though, is that vital SDoH information—such as social and economic factors, lifestyle choices, and living conditions—is not always easily accessible for clinical decision-making because it is often trapped in clinical notes. Far too often, clinicians aren’t aware of important SDoH details until after those details have played a harmful role in a patient’s health status.
Finally, advances in artificial intelligence (AI) technologies have made new solutions more accessible to HCOs of varying sizes. For example, AI-based NLP solutions are increasingly sophisticated and don’t necessarily require teams of expensive data scientists.
One ACO’s NLP experience: Care gaps closed and revenues increased
Here’s an example of an NLP use case for an accountable care organization (ACO) that serves Medicare, Medicaid, and commercial health insurance populations through value-based agreements with payers.
These value-based contracts require the ACO to track and report initiatives that illustrate improvements in the delivery of quality care, as well as any related financial savings. Quality reporting on heart failure, for example, involves recording information on the ejection fraction of all patients in the covered population. Then, to achieve quality reporting and improvement goals, the ACO must identify at-risk patients to minimize gaps in care quality and enroll these patients in safety-net programs.
The ACO’s quality reporting initiatives had previously struggled to obtain important patient data—such as the comprehensive identification of patients with various conditions and quality metrics—because the required information was often captured in clinical notes. After studying the problem, the ACO’s leadership found that providers were often including clinical details in narrative form as free text. The issue was particularly acute when structured fields for such information were unavailable or difficult to find, but it was also due to acknowledged challenges associated with clinician burnout.
While narrative-based data entry is necessary to capture clinical complexity and nuance and is frequently preferred to structured data entry, its use also creates data-completeness challenges because the free-text information does not automatically translate into structured information on patient problem lists.
To solve the challenge, the ACO implemented an NLP-based AI solution that enabled it to identify vital unstructured data in patient records, closing information gaps and, subsequently, care gaps. Prior to the ACO’s adoption of the NLP solution, nurses had to manually review 1,000 charts to identify a single care gap. Post-implementation, the ACO has reduced manual review to just six charts for each care gap identified.
In the program’s initial phase, the NLP solution identified 92 instances in which patients were documented in the narrative as having chronic obstructive pulmonary disease or congestive heart failure, but whose conditions were not entered into a structured format. As a result of their identification, the patients became eligible for the ACO’s population-based disease management programs, which in many cases led to higher-quality care after the ACO was able to close chronic disease–related care gaps.
NLP: The key to filling healthcare’s data gaps
As the digitization of healthcare continues, the industry no longer lacks data—it lacks a reliable way of making sense of mountains of disparate data. As information fragmentation continues to exact a toll on the healthcare industry in the form of higher costs and lower-quality care, NLP technology will play an increasingly important role in making that data usable, allowing decision-makers to acquire insights that improve care quality and patient safety.
Elizabeth Marshall is director of clinical analytics at Linguamatics, an IQVIA company. Dr. Marshall is a research physician dedicated to the development and implementation of informatics solutions that improve the effectiveness and quality of patient care.