Decoding Doctors’ Notes with Natural Language Processing

May 23, 2019

Over the last decade, the use of Electronic Health Records (EHRs) collect patient and population data in a centralized digital format. Their proliferation has made healthcare analytics easier and more effective, with large amounts of patient information stored as discrete data. However, a lot of information documented by a physician is not easily accessible for analysis, even if its stored in the EHR. Clinical notes, radiology reports, and discharge summaries, for instance, all contain valuable information for quality care, but are hard to analyze due to their lack of standardization. Natural language processing (NLP) is an important technique that can unlock unstructured, free-form data.

One recent success with using NLP in health informatics has been at Mercy, a St. Louis-based health system. Mercy set out to track heart failure device performance. Across their system, Mercy had over 100,00 patients with such devices. However, as much as 40 percent of the useful data were stored as unstructured free form data. Many key cardiological measures and symptoms are only reported in notes, making it difficult to get a clear picture of a patient’s health from the discrete data on their chart alone. However, providers do not document their notes in the same way, using different phrases for the same symptoms, creating challenges in turning notes into actionable data through traditional means. NLP helped Mercy identify synonyms for symptoms and detect patterns in naming that ensured that they could gather reliable data over years of physician notes and leverage 34 million notes going back seven years. Mercy’s use of NLP gave them actionable data for their cardiology study, but it is only one example of what NLP can do for EHRs.

From the relatively simple, such as identifying information to prevent HIPAA violations, to the more complex, such as extracting detailed drug information from clinical notes or helping identify self-harm risks among veterans with PTSD, NLP has helped hospitals improve patient outcomes and save money. However, NLP is not without its challenges. Training NLP takes large quantities of data, and accuracy largely depends on the quality of the input and corrections made over time. ASR Analytics is leveraging advanced analytic tools, like NLP, to help hospitals and other sectors of the healthcare industry get the most out of their unstructured data. Training an NLP can be time consuming and labor intensive, but, as Mercy and many other healthcare providers and researchers have found, the hard work can improve care and results.

To learn about how ASR Analytics can help your business solve problems through data analysis, contact us here



Tags: