While many health equity initiatives focus on aggregate ZIP code or census data to identify social determination of health (SDoH) risk factors, applying Natural Language Processing (NLP) to clinical notes enables health care organizations to identify these factors at the patient level, and therefore prioritize interventions.
NLP is not just for risk adjustment
While a lot of attention and focus on NLP is around the important benefits it provides in risk adjustment, there are other areas where it adds immense value in health care. One area that clinical NLP is also adding significant value is in health equity programs, to identify and risk stratify patients by their individual SDoH.
While many health equity initiatives focus on aggregate ZIP code or census data to identify SDoH risk factors, applying NLP to clinical notes enables health care organizations to identify these factors at the patient level, and therefore prioritize interventions.
What is clinical NLP?
NLP enables machines to comprehend text by emulating human language comprehension. Clinical NLP does this in the context of how clinicians document their patients’ findings–and understands the context in those clinical notes–differentiating family history from the patient's history, differential diagnoses from known diagnoses and present conditions from those that have been negated by the clinician. NLP systems can tirelessly analyze vast amounts of text-based data consistently and without bias. They excel at understanding complex concepts within intricate contexts, deciphering language ambiguities, and extracting key facts, relationships, or summaries. With the ever-increasing volume of unstructured data generated daily within electronic health records (EHRs), this automated approach has become essential for efficient text-based data analysis.
Using NLP for health equity
Clinicians often capture social history as part of their routine history and examination. They will record this information in the free text section of progress notes. For example, “Mrs. Smith came with her son who she lives with” or “Mr. Johnson missed his last appointment as he had no means of getting to clinic.” NLP can be used to create structured data that tags Mrs. Smith as living with family support (i.e., not socially isolated) and Mr. Johnson as having a transportation issue. This ability to characterize patients by these factors enables HCOs to prioritize and triage patients/members by their individual risk.
Recent presentations and publications have showcased the application of NLP in this precise manner. For instance, at a conference held in March of this year, an industry example demonstrated the utilization of NLP by IQVIA in identifying SDoH risk factors. Using this technology within the firewall of the health system, they were able to screen 56 percent more patients for SDoH compared to their prior screening process not using NLP. Furthermore, they identified documented SDoH in clinical notes for over 30 percent of their population, in contrast to only 0.01 percent of patients with recorded Z codes. This presentation went one level deeper by demonstrating how this enhanced workflow with NLP was impacting individual patients with an individual case study. In this case, a 20- to 30-year-old female presented to the emergency department for a headache. In the background, NLP identified that this patient had previously spoken about being a victim of abuse. The AI output was presented to a care worker with this information highlighted. They were able to take a detailed history from the patient on that topic and as a result, refer her for PTSD counseling and legal assistance. This example really drove home the potential impact that AI can have on patient care at the individual level.
As health care data continues to increase in volume, and as interoperability mandates increase the availability of this data to broader health care stakeholders–it is vital that health care organizations embrace technologies like NLP to maximize the potential of this data. While the current focus is on individual use cases like risk adjustment–the potential to expand this technology to other areas like health equity is exciting and an area health care organizations should already be including in their strategy.
If you’d like to try it for yourself, register for our SDoH APIs or click here to contact us for more information.
About the author
Calum Yacoubian, M.D., is the director for health care NLP at IQVIA. IQVIA is a leading health care data science and analytics company with a 20-year pedigree in health care NLP. Dr Yacoubian is focused on driving efficiencies that lead to patient improvement at the intersection of health care and technology. He is a clinician with eight years’ experience in clinical NLP.