The NLP AI model extracts SDOH data from clinical records

The NLP AI model extracts SDOH data from clinical records

A study conducted by the Regenstrief Institute and Indiana University has demonstrated the potential of using natural language processing technology to extract social risk factor information from clinical records.

The NLP system developed by the research team can “read” and identify keywords or phrases that indicate a housing or financial need (for example, lack of a permanent address) and provide highly accurate performance, the agency reported.

By using NLP to search for the social determinants of health, which often lack the standard terminology found in patient electronic health records, healthcare providers can more easily find and extract this data from clinical records.

Studies published in International Journal of Medical Informaticsanalyzed more than six million clinical records from Florida patients.

System generalization and portability were evaluated, demonstrating ease and accuracy in adapting to new environments and data requirements.

In addition, advances in NLP technology can lead to more cost-effective data extraction, enabling public health perspectives and proactive interventions in addressing housing and financial needs.


Chris Harle, Regenstrief and IU Fairbanks School faculty member and senior author of the study, explains that, with more accurate or complete social information about their patients, healthcare providers may be able to tailor a patient’s medical care to take into account social needs or refer patients to other services that are appropriate for them. help deal with it.

“This approach can be reused to extract other kinds of social risk information from clinical texts, such as the need for transportation,” he says. “In addition, the NLP approach must continue to be adapted and evaluated in various health care systems to understand best practices in outreach and implementation.”

He points out that text data in healthcare varies across organizations and geographies and over time.

“Therefore, methods for extracting information from text automatically must be evaluated in various situations and, if implemented in practice, monitored over time to ensure continued quality,” says Harle.


Previously, Regenstrief Institute researchers developed three NLP algorithms to extract housing, financial and employment data from electronic health records.

The aim is to measure social determinants well enough for researchers to develop risk models and for physicians and health systems to be able to use multiple factors.

The organization has also developed an app called Uppstrom, which helps predict which patients will need referrals to social services, such as nutritionists.

The researchers say this study shows how AI and NLP models can leverage clinical data to improve care with “significant performance accuracy.”

Harvey Castro, a physician and healthcare consultant, says he agrees that the integration of NLP to extract social risk factors has tremendous potential across the healthcare spectrum.

“This can result in faster, more personalized care and proactive interventions,” he explains.

He added that its application is very broad, from the emergency room to primary care, mental health, chronic disease management, child care, geriatric care, and public health.

“NLP allows doctors to treat patients as a whole by understanding their social risk factors,” said Castro. “This could lead to more personalized care, such as connecting homeless patients directly to the social services of the ER.”

Leveraging NLP with more complex algorithms could improve understanding of patient-specific nuances while predicting possible substance abuse problems or analyzing speech patterns could help addiction intervention, he added.

“There is significant potential and wide applicability in using NLP to identify and address social risk factors, along with achieving health equity,” said Castro.

However, for NLP applications to reach their full potential, providers must step in and serve nurses, says one expert.

Nathan Eddy is a health and technology freelancer based in Berlin.
Author email:
Twitter: @dropdeaded209

#NLP #model #extracts #SDOH #data #clinical #records

Update on Dr.AI

Update on Dr.AI

Bus drivers are twice as likely to be hospitalized with severe COVID-19

Bus drivers are twice as likely to be hospitalized with severe COVID-19