As I wrote before, many patients tell their health care providers quite a bit about what is going on their lives, both good (graduations, new jobs, etc.) and bad (broken vehicles, problems with the landlord, family level stressors, etc.) While sometimes just chit-chat, more frequently these life details are an implied prelude to explaining a health issue (anxiety) or reason for being late or no-showing appointments. Quite often the patient will tie these stories into their wellbeing. For example, when reviewing home blood glucose monitoring over the course of the month with a patient, I’ve had more than one explain their high readings in the latter half of the month as being due to running out of cash/EBT card value for healthy groceries. These patients start out each month carefully following a lower carb, higher fiber diet. However, they are on limited incomes using, for example, SSI, SNAP benefits or just the first paycheck of the month to cover their monthly food expenses. Once their grocery budget for the month runs out, they resort to inexpensive processed foods and value menu items. Across a wide variety of health challenges, patients know what the barriers are to trying to remain healthy and they lay it out pretty clearly for us. For example, “I’d like to get treated Hep C, but I can’t get to the GI doctor’s office on the bus line.”
As clinicians we try to make adjustments to the patient’s plan of care—prescribing “sliding scale” medicine to cover the higher blood sugars the second half of the month, recommending “buy this/not that” off the value menu, writing a new referral to the oversubscribed GI clinic at the hospital which is on the bus line—but we rarely capture these issues (food & transportation insecurity) as structured data in our notes and electronic health records. As a result, the magnitude of these problems and the ability to correlate them to health outcomes is limited. Also, in today’s world where there are increasing investments to address SDOH/HRSN, particularly in expanded health teams, we lose the consistent opportunity to ensure our patient’s get connected with these team members who may have much better solutions (e.g., farmer’s market vouchers and food pantry referrals, medical ride sharing access, etc.)
We could add to the clinician burden requiring them to look up and assign codes to each of the social issues mentioned by their patients. Alternatively, we can use technology, in this case natural language processing (NLP), to recognize these issues, and present them for both coding and action on the patient’s behalf. Developed out of a need to be able to demonstrate how families involved in the child welfare system were doing—both good and bad—Augintel’s NLP that is trained to understand descriptions of SDOH in text, can do just that.
Using the standardized domains developed under the auspices of the Gravity Project, Augintel’s NLP model tagged language related to these codes in the notes of 615 of 1,000 individuals served by one clinical organization. When applied to the nearly 120K clinical notes for these 1,000 persons, 12% discussed inadequate finances, 6% food insecurity and 4% inadequate housing, for example. Looking just at those 615 persons, the prevalence of these issues was 86% for inadequate finances, 69% for food insecurity and 65% for inadequate housing.
The model can also be used to recognize issues that have yet to be codified by the Gravity Project, for example neighborhood safety. (How many times have I been told by patients it is not safe to walk in their neighborhood due to unleashed dogs, no sidewalks, gangs or unstable homeless people?) Whether the intervention is a connection to a Silver Sneakers program which sponsors walks at the local mall, the enforcement of leash laws or community policing, these too are critical health related needs that must be captured and addressed.
We are very excited about the potential of Augintel’s model to assist clinical organizations, care coordinators, researchers and policy makers in using the information told to us by patients about their SDOH/HRSN. Not only can it identify the issues, but it can also track interventions and progress on remediating these health-inhibiting challenges. Look for further discussion of these in my next posts on the benefits of employing NLP for health related social care.
.jpg)
.png)
