From the March 2017 issue of HealthCare Business News magazine

When CHF is not the primary diagnosis, the only mention of the condition might be within an unstructured section of a patient’s notes, which could be easily overlooked when a hospital is creating a patient’s discharge summary. The patient’s follow-up treatment plan may fail to include appropriate therapies for managing the condition, such as lifestyle modifications or adjusted medication regimens. NLP can extract discrete values for left ventricle ejection fraction from progress notes and echocardiogram reports to identify any missing CHF diagnoses and ensure symptoms are managed.

Important insights about a patient can be found in discharge summaries that indicate if there is a close family member available to provide support, if the patient uses a walking cane or wheelchair or is living alone. Such factors can be extracted using NLP to transform them into discrete data. Providers can then combine CHF-related data with details on a patient’s lifestyle and environment, and leverage predictive analytics to produce a real-time estimate of the patient’s 30-day hospital readmission risk. Follow-up care plans can be further customized to minimize readmission risk.


In order to assign risk under evolving value-based reimbursement models, payers must analyze patient populations and stratify patients based on known health and environmental conditions. Often, critical patient information, including CHF status and risk factors, is hidden in unstructured formats held in data “lakes.” NLP technology gives payers the ability to extract key data and integrate it with conventional data warehousing and analytic solutions for further analysis.

One large payer, for example, has improved its risk assessment process with the implementation of NLP. Unstructured data from exported EHR data in continuity of care document (CCD) format, including nurse notes from member calls and emails and PDF documents, are all stored in a Hadoop data lake. The payer uses NLP to identify CHF risk factors such as family history and smoking status from the unstructured data. The NLP results are then integrated with the structured data in a Netezza data warehouse. This allows both structured and unstructured data to be used in risk stratification models. As a result, they are able to glean vital insights to more precisely assess population risk.

About the author: Simon Beaulah is Linguamatics’ senior director of healthcare and is responsible for the company’s health care products and solutions, including applications for clinical risk models, population health and medical research.

Back to HCB News

March 2017 Magazine

Read more stories from the March 2017 issue

More Voices