Until recently, the use of natural language processing (NLP) in healthcare has been primarily limited to research efforts and population health within academic medical centers. However, with the proliferation of unstructured data from electronic medical records, providers are now seeking to harness the potential of their data and considering a variety of use cases for NLP technology.[1] That’s the conclusion of a recent KLAS report entitled “Natural Language Processing: Glimpses into the Future of Unstructured Data Mining.”

The report includes insights from 58 provider organizations and examines the various ways providers are currently leveraging NLP technology, as well as some of use cases poised for wider adoption. Coding and documentation applications represent the broadest use of NLP engines. But it is clear providers have a growing interest in NLP solutions that advance their population health initiatives. An increasingly popular use case, involves applications that use NLP to mine unstructured data within patient populations and include predictive analytics to identify at-risk patient populations.

 A few of the major findings from KLAS’s report are summarized below.

How is NLP being used today?


During his January 2015 State of the Union speech, President Obama announced details of his administration’s Precision Medicine Initiative, which promises to accelerate the development of tools and therapies that are customized to individual patients. Precision medicine focuses on disease treatment and prevention and considers the variability in genes, environment, and lifestyle between individual patients.

Precision medicine takes into account healthcare’s relatively minor role in impacting a patient’s overall health and well-being, compared to the larger roles of genetics, health behaviors, and social and environmental factors. The precision medicine approach thus requires that providers have access to a wealth of patient-specific data. Thanks to advancements in genetic testing and new technologies, such as patient portals and remote monitoring devices, a wide variety of patient data is now readily available. Unfortunately, clinicians may have difficulty extracting data that is clinically relevant because much of the information is stored in an unstructured format.

Consider how a physician would glean information from a paper medical chart prior to EMRs. To understand a patient’s complete health status, the doctor would search through pages and pages of notes - obviously a time-consuming and error-prone task.


Shifting payment models based on quality and value are fueling the demand for insights into the health of populations. This demand requires the analysis of vast amounts of patient data. For example, before healthcare organizations can implement pre-emptive care programs, they must first identify the relative risk of their patient population. This is based on a variety of clinical, financial, and lifestyle factors, including:

  • Problem list of patients, especially chronic conditions
  • Procedures, medications and other hospital data
  • Claims information
  • Risk factors such as tobacco, alcohol and drug use
  • Availability and accessibility of health services and social support.

As illustrated in Figure 1, a healthcare population typically includes a relatively small percentage of the highest-risk patients, though these least healthy patients usually account for the biggest percentage of overall healthcare costs.

 

Figure 1: Level of patient risk associated with population segments and
their cost implications; a relatively small segment of the population
accounts for a disproportionate percentage of healthcare costs


What if physicians could offer patients access to a potentially life-preserving test, but could not easily identify which of their patients were eligible?

That is the exact situation many providers have found themselves in since Medicare announced it would begin covering lung cancer screening for patients meeting a certain set of criteria.

In a decision memo published February, 2015, CMS agreed to make Medicare coverage available for a low dose computed tomography (LDCT) lung cancer screening for eligible patients. Patients who are between ages 55 and 77, asymptomatic, are either a current smoker or have quit within the last 15 years, and, have a tobacco smoking history of at least 30 pack-years can now qualify for an annual preventative screening.

CMS added the coverage after determining there was sufficient evidence that LDCT procedures were cost-effective for high risk populations. A study by the National Lung Cancer Screening Trial, for example, found that 12,000 deaths a year could be avoided if high-risk patients underwent a LDCT scan. Lung cancer is currently the leading cause of cancer-related death among both men and women in the US.


Natural Language Process (NLP) is a powerful tool for uncovering hidden secrets within unstructured text to analyze trends and reveal insights.

In healthcare, 60% of the 1.2 billion clinical documents produced in the US each year reside in unstructured narrative documents that would be largely inaccessible for data mining and quality measurement without NLP tools.

With NLP technology, organizations can unlock rich data to analyze patient populations and ultimately improve patient care.

In recent years, the use of NLP in healthcare has primarily been limited to disease-coding and research applications; however, Linguamatics was interested in discovering new opportunities that leverage NLP to enhance patient care and improve hospital efficiency.
 

Surveying healthcare system CMIOs

To that end, Linguamatics, with the support of the American Medical Informatics Association (AMIA), surveyed healthcare system CMIOs and asked them to share their visions for ways to leverage NLP to enhance patient care and improve hospital efficiency.

The participating CMIOs expressed overwhelming support for using NLP to help preserve the patient narrative and provide the insights required to meet accountable care objectives, including care delivery goals and the pro-active identification of high-risk patients.

They also voiced interest in leveraging NLP for a variety of other applications, including: