Hidden in plain sight: Clinicians may not spot EHR cancer risk info

A new study in the Journal of the American Medical Informatics Association found that patients may have risk factors for heritable breast and ovarian cancers present throughout their electronic health records – but because the information is scattered, treating clinicians may not recognize it as actionable.

“The full story of a patient’s risk for heritable cancer within their record often does not exist in a single location. It is fragmented across entries created by many authors, over many years, in many locations and formats, and commonly from many different institutions in which women have received care over their lifetimes,” wrote the research team from the University of Washington School of Medicine.

In their study of EHRs from 299 women, the team found that 8% met national guidelines for a referral for a genetic risk evaluation, but half of those patients had not been referred.

“Had the scattered risk factors for each patient been presented together to a treating provider with knowledge of NCCN guidelines, more women might have been referred to a medical geneticist or genetic counselor, and might have engaged appropriately in a discussion of the risks and benefits of genetic testing,” wrote the researchers.


As the researchers note, individuals at increased risk for heritable breast and ovarian cancer benefit from referrals to cancer genetics professionals. But because risk factors can be scattered over EHRs, providers may not recognize them for what they are when viewed as a whole.

The team reviewed the complete EHRs of 299 women who had been seen more than five times or hospitalized more than twice in the University of Washington Medicine health system between April 2018 and April 2019. 

Six medical students trained in EHR use reviewed the records of the sample, which included UW Medicine health system records, records and documents from other institutions, and handwritten questionnaires completed by patients. The records included structured and unstructured data.

The most common risk factors were a family or personal history of ovarian or breast cancer; researchers often found family history information in note narrative text rather than in the EHR family history tool.

“In this random sample of women selected from among those cared for in our health-care system, we were able to identify many women whose EHRs contain risk-factor information meeting national guidelines for further genetic risk evaluation, yet half of these women had no record of a referral in their EHR,” wrote the study authors. 

They noted that identifying the patients did not require any additional outreach efforts and that all the relevant information was in their chart already. 

“Finding this information took trained reviewers far more time than most busy clinicians can reasonably devote to a chart review,” wrote the researchers. They suggested that method including image and natural language processing would be helpful in finding actionable information dispersed throughout EHRs.


Moves to more robustly integrate artificial intelligence and NLP into EHR analytics have been underway for a few years. 

Last year at the HIMSS Machine Learning and AI for Healthcare event in Boston, Varun Gupta, IT director, advanced analytics and data management at Mount Sinai Health System, showcased how clinicians and case managers can use NLP algorithms to access social data captured throughout EHRs.

“It’s all there in unstructured format. And there’s a huge opportunity to use that data to find out insights. That’s where natural language processing comes in,” said Gupta.


“To our knowledge, this is the first report showing the amount of cancer risk-factor information in the EHR and seeking to determine its potential impact on patient referrals,” wrote the researchers. 

“As others have noted, primary care providers are overwhelmed with other clinical priorities that prevent the systematic documentation and use of family health history tools. If providers gather risk-factor information in the course of care – and often they do not – it may be entered in a fashion most expeditious for the pace of clinical practice; specialized structured data capture tools may not be used because of the time required to use them,” they added.


Kat Jercich is senior editor of Healthcare IT News.
Twitter: @kjercich
Email: [email protected]
Healthcare IT News is a HIMSS Media publication.

Source: Read Full Article