Closing Care Gaps using Social Determinants of Health

NorthShore University Health System recently combined with Edward-Elmhurst Health to become the third largest healthcare delivery system in Illinois. The health system includes nine hospitals serving an area of over 4 million residents. The health system aims to provide a service that puts patients first, and serves the communities they live in. As a first step in their health equity plan, NorthShore - Edward- Elmhurst Health have turned to technology to gain greater visibility of the health equity disparities that exist within their population, to better manage these gaps and drive more equitable healthcare.

Specifically, NorthShore - Edward-Elmhurst Health is using Linguamatics Natural language processing (NLP) technology to extract, normalize and present relevant patient level SDoH information from clinician notes to care providers giving them a more complete view of each patient’s social risk. This enables targeted intervention and more holistic patient care. It has since expanded the original use cases and is evaluating other ways this information can positively impact patient outcomes.

In the NorthShore region, significant disparities in life expectancy were observed. Using census tract data — a 22-year difference (the largest in the country) was observed between those who had the longest and shortest life expectancies. In order to address this discrepancy, NorthShore — Edward-Elmhurst Health recognized the pressing need to understand the factors contributing to these disparate outcomes.

It is well established that health outcomes are impacted significantly by factors outside of the 4 walls of a health system, and beyond the genetic predispositions to disease that we all have. A growing body of evidence points towards various factors such as our living environment, work conditions, and level of education, playing a crucial role in shaping health outcomes. These social risk factors are better known as Social Determinants of Health (SDoH).

Given the increasing understanding of the importance of these risk factors, capturing and accounting for SDoH data is becoming more widespread – both as part of internal health equity initiatives, but also within regulatory and quality frameworks (such as the 2023 IPPS Final Rule) that are becoming mandated over the coming years.

By acknowledging and addressing these social determinants, there is a potential to not only improve health outcomes but also reduce healthcare costs for a significant portion of the population.

Despite clinicians routinely capturing patients’ social history as part of admissions and other interactions with their patients, this information is rarely captured in structured or administrative data within the health system. When NorthShore reviewed the fields where SDoH can be captured within their Electronic Medical Record (EMR), only 0.1% of patients had these fields filled in. This does not mean that clinicians are not asking their patients about their SDoH — they do, but the information is usually confined to the clinician notes – such as discharge summaries, progress notes and nursing notes - sometimes referred to as unstructured data. Another challenge is the use of disparate EMR systems across the delivery system — with NorthShore having 3 separate deployments of EPIC, their acute EMR. The final hurdle that had to be accounted for was how to make any data they could surface, actionable.

To solve the first two challenges, NorthShore identified a need to use AI in the form of Natural Language Processing (NLP). This approach processes unstructured clinical notes from disparate sources, and allows them to surface any documented SDoH.

They evaluated several options, including open source and vendor provided NLP solutions. Based on a combination of healthcare expertise, transparency, and flexibility, NorthShore opted to use IQVIA Natural Language Processing (NLP), formerly Linguamatics NLP. Using IQVIA NLP NorthShore used out-of-the-box NLP models for SDoH identification. Because of its easy configurability and “open box” philosophy, the NorthShore Data Science team were able to create their own custom modification to the NLP algorithms, and extract exactly the information they needed from the medical record. They were also able to track accuracy of different SDoH metrics to gain clinician acceptance of the data the NLP provided.

During testing, the team at NorthShore was able to identify at least one SDoH risk factor in 30% of their population (up from 0.1% identified through structured fields). Having validated the NLP solution met their needs, they then sought to make the results actionable. To do this, they relied on true interdisciplinary working between clinicians, data scientists and the healthcare systems IT departments.

First, a pipeline was created between their EMR data warehouse (EDW), and their locally installed IQVIA NLP instance. The NLP algorithms run in the background — and send structured data back to the EDW. This data is then presented in a flowsheet within their EMR to emergency department (ED) social workers at the time of a patient’s admission/presentation to the ED. The social workers then validate the NLP identified SDoH care gaps and accept or reject these findings as true and current. This feedback is reviewed by the data science team, who in turn make tweaks to the NLP algorithms, should any erroneous patterns be detected.

“Staff have called this solution a gamechanger. Where they [ED social workers] were spending 80% of their time on chart review, they now can spend that 80% making an impact on the patient.”

In creating this cross functional SDoH screening process, NorthShore were then able to roll this AI-enabled workflow into clinical practice within their ED.

One month after deployment of this new workflow, the social workers in the ED have demonstrated more targeted screening and intervention in patients with potential SDoH opportunities. They have screened 56% more patients than by their prior non-AI augmented workflow and have been able to have a positive impact on the individuals attending their facility.

One example of the positive patient level impact of this technology was a female patient, aged 20 to 30, who presented to the ED with a headache. From the NLP enabled prescreening of her medical record, it was identified that she had previously been admitted and given a history of physical abuse within her prior admission. This flag was raised to the ED social care team, who then validated the NLP results, and took a trauma-based history from the patient, enabling them to elicit more details, and get the patient the further support she needed — in this case, Post Traumatic Stress Disorder counseling and legal assistance. This ability of the software to help the social care team be more targeted is part of a reason that they have called this “game-changing” technology.

While there has been huge interest and attention in social determinants of health for the last few years, there have been few examples of healthcare organizations implementing AI technologies to transform how these data are captured. The work done at NorthShore Edward-Elmhurst Health is an exemplar of the potential of AI technology, when combined with motivated and collaborative cross functional teams, to improve Health Equity outcomes and close care gaps.

With SDoH data capture mandated in increasing quality reporting programs, and being incorporated in value based care programs, such as ACOREACH, this work has real significance for healthcare organizations in 2023 and beyond. This real world example of AI for SDoH from NorthShore — Edward- Elmhurst Health should give healthcare organizations participating in health equity initiatives confidence and motivation to embrace technologies like IQVIA NLP.