81% of individuals went to a healthcare provider at least once in the last year. While that is not surprising, pairing it with another stat tells a very different story.
80% of healthcare data is unstructured today.
Unstructured data serves a critical purpose, but it cannot be pigeonholed. And when billions of patients are concerned, the very nature of unstructured data gives way to inefficient care on a grand scale. According to a survey by HIMSS Media, 56% of healthcare professionals cited unstructured data as a key barrier to optimizing clinical workflows.
Structured data in spreadsheets and relational databases are presented in a way that facilitates easier search and analysis. It is functional. But unstructured data in handwritten prescriptions and pathology reports defy rigid rules and systems to convey a deeper awareness of a patient’s condition. It can be insightful.
The Dichotomy of Structured and Unstructured Data
The importance of both kinds of data depends on how they are managed. Structured data lends itself to the organization and mechanical analysis, a task befitting most current systems. Imagine a spreadsheet sprawling with names, dates, heights, weights, currencies, blood types, diagnostic codes, and many more values. Such data is mandatory for insurance companies in the event of reimbursement claims. Its management is more straightforward and requires little human intervention.
However, structured data has clinical value as it is backed by unstructured data. The problem lies in the inability of typical analytic solutions to analyze this data since it exists in the form of free text and narrative. Here’s a hypothetical situation at a clinic:
“Hello, doc, I’m here for my monthly checkup.”
“Yes, of course, please have a seat. Let’s take a look at your vitals. Your blood pressure is 130/80, a little on the higher side.”
“Is that manageable, doc?”
“Of course! But you shouldn’t take too much stress. How does your schedule look like this week?”
“I’ll be at work for most days of the week. I was planning to take a leave on Thursday, it’s my daughter’s birthday. But there’s some urgent business at work so I wasn’t sure about taking it.”
“I would urge you to take that leave. With your busy schedule, you could do 10 to 15 minutes of meditation every day. Perhaps this stress at work is making you spend less time with your family. Get out of the office early whenever possible. I’m prescribing a diuretic for you as well as a chest X-ray. Take the medicine for a week and pay me a visit once you get the report.”
“Thanks, doc. I’ll try to follow what you said.”
“Please do. Take care.”
A 130/80 level of blood pressure and the prescribed diuretic can fit neatly in a spreadsheet. But the recommendations of meditation and spending more time at home cannot. The forthcoming X-ray report won’t find a place either. Yet it is those bits of information that reveal a lot more about the patient’s current and future state of health than a single blood pressure level or a medication. Both these facts might be specific to the patient but are also generic in the information they convey.
A study in the Journal of the American Medical Informatics Association (JAMIA) states that real-world data in electronic health records (EHRs) provide more accuracy, but only when it is mined by trained algorithms. The study found that with structured EHR data, the average precision and recall were 98.3% and 51.7% respectively. With unstructured data, the corresponding rates were 95.3% and 95.5%.
NLP is the Way Forward
The benefits are there for the taking but there is a need to convert unstructured data into a more organized format to make it more usable. This conversion poses a challenge to organizations due to the volume and variety of unstructured data that is produced on a daily basis. Barriers to leverage EHR data exist because of unstructured data volumes that are still unparsed. Due to the present form of our data, it cannot be used to make constructive healthcare decisions. The need to leverage unstructured data is vital in the shift from a fee-for-service to a fee-for-value care model.
It is here that natural language processing (NLP) can help organizations create a difference.
While the use of NLP is recent in the medical domain, it shows promise in elevating unstructured clinical records. One study applied an NLP-based system to convert clinical texts into the Unified Medical Language System (UMLS) code and obtained commendable results in recall and precision. NLP has also found use in automated cohort selection and veterinary medicine. By analyzing free-text notes and accumulating insights from the process, NLP-based solutions can effectively utilize all the available data to make the most informed decisions.
Choosing the Right NLP Partner
It is a challenge for healthcare firms to construct clinical NLP capabilities from scratch. Such a task is expensive, time-consuming, complicated, and expertise-intensive, especially in machine learning skills. NLP also requires high-quality datasets to work with, which can be achieved through proper data annotation. The input for NLP workflows is always a text corpus, the nature of which defines the kind of annotation and labeling that needs to be done.
For instance, in most clinical scenarios, the goal for organizations is to understand the domain, intent, and sentiment of the input at hand. Holistic data annotation and labeling can help organizations make the most of the unstructured data at their disposal.
ezDI’s NLP platform (ezNLP) is designed to help organizations gain crucial insights from unstructured data. It employs a set of techniques that allows computers to extract individual clinical entities from relevant unstructured text. This information is then organized into a structured format such as data tables, columns, or variables that could be used for analysis. With the use of our clinical NLP-as-a-service, companies and healthcare organizations can identify medical information, such as medical conditions, diagnosis, medications, and dosage, more efficiently.
Organizations can also avail human-powered data annotation with hundreds of resources that can be scaled to thousands on a highly customizable platform that is PHI-centric. As a result, they can cater to unique healthcare use cases such as semantic search and in revenue cycle management as well as clinical trials, among others.
ezNLP uses advanced algorithms based on machine learning (ML) and deep learning (DL) to detect patterns and correlate them with data such as radiological images to recognize potential diseases and prepare reports. Other value additions such as medical and conversational transcriptions from U.S.-based transcriptionists help organizations maintain an accurate record. Our service can also link the extracted information to medical ontologies such as ICD-10-CM/PCS, RxNorm, SNOMED, CPT, and more.
Our automated and patented clinical NLP platform is backed by years of effort and is based on hundreds of millions of diverse patient records covering all specialties to offer accurate predictions and recommendations at scale. In addition, it is accessible through a simple application programming interface (API) call, making the process of extracting valuable and actionable insights from the unstructured text completely automated. Our use of API results in lower costs, time, and effort to process unstructured text volumes. This is backed by a transparent, truly SaaS, a business model with no upfront commitments or minimum fees.
To learn more about ezDI’s AI-based mid-revenue cycle management solutions visit www.ezdi.com and to see the live product demo of our Clinical Documentation, Coding, Compliance/Auditing, Quality Measures, Encoder, and Enterprise Analytics request a live demo.