Getty Images/iStockphoto

Expanded Patient Medical History Can Improve Clinical Notes

Researchers added additional patient history information, such as family history timelines, to determine a patient’s disease risk and predict outcomes.

Adding age and patient history can improve clinical notes to support both clinical and transitional research studies, according to a study published in BMC Medical Informatics and Decision Making.

Adding patient clinical history, such as age and time information can help determine a patient’s disease risk, predict patient health outcomes, and understand disease progression. For example, a patient has increased cancer risk when she has one or more family members who have cancer.

Clinical histories are becoming more prevalent in EHRs. However, specific age and time clinical events, such as risk factors, surgical interventions, and past diagnoses for patients and family members are not often documented or available for research.

Researchers dissected 138 de-identifed discharge data summaries of past medical history, past surgical history, family history, and social history. According to a separate study, most patients are comfortable with sharing their EHR data and biospecimens for research, but may have sharing preferences.

Study authors said these data summaries have a higher chance of containing age and specific clinical events than acute event history. With this data, researchers developed two natural language processing (NLP) models to integrate this data to address named entities, attributes and values, and relationships between the named entities.

Researchers captured age, procedure, and social determinants of health categories with 8 percent, 10 percent, and 23 percent of all annotated named entities to those categories, respectively. Researchers also observed an agreement of 85 percent for both age and time.

“Age and temporally-specified mentions within past medical, family, surgical, and social histories were common in our lung cancer data set; annotation is ongoing to support this translational research study,” explained the researchers.

In the study, researchers expanded upon the current models to support age, time, and family history information.

“Annotator training and feedback yielded some notable observations,” wrote the study authors. “Firstly, AGE and TIMEX3 classes were placed in consistent locations through the text and followed easily identifiable patterns likely explaining fewer inter-annotator inconsistencies. Other classes required extensive training to achieve acceptable IAA for attribute annotation.”

To come to these observations, researchers added biological information that assessed disease risk that was linked to genetic heritability. The research team also added temporal information that is related to age and placed on a patient timeline.

Looking forward, researchers said further work can be done by integrating similar models into existing models to extract medical named entities.

“The long-term goal is to develop a hybrid rule-based and deep learning NLP system to automatically extract age and temporal information, for the construction of longitudinal clinical profiles for any patient, including our lung cancer cohort,” concluded the study authors.

“We then build a prototypical NLP tool to assess the amount of work necessary to extract such new information, and to serve as a foundation for a future automation efforts.”

Most patients are comfortable with sharing their EHR data and biospecimens for research purposes but may have sharing preferences based on researchers’ affiliations and specific data items, according to a study published in JAMA Network Open.

Of the 1246 study participants, only 3.7 percent declined to share their health information with their own healthcare providing institution. A total of 352, or 28.3 percent, declined to share with nonprofit institutions, and 590, or 47.4 percent, declined to share with for-profit institutions.

A total of 291 participants, or 23.4 percent, indicated that they were willing to share any data with any researcher, while 46 participants (3.7 percent) were not willing to share any data with any institution.

A total of 909 patients (72.9 percent) said they would be willing to share their data selectively, or had a general preference for sharing within a certain institution. 

Dig Deeper on Health IT optimization

Cloud Computing
Mobile Computing