Using NLP to find critical healthcare outcomes hidden in plain sight in the EHR
"Insanity is doing the same thing over and over again and expecting different results."
I’m sure most of us have heard this infamous quote, wrongfully attributed to Albert Einstein. Wrongful attribution or not, it’s a valid point. Why do we do this in healthcare?
The way we approach healthcare is the insanity problem
Let’s start analyzing this by looking at healthcare outcomes and how they are broken down.
- Socioeconomic factors (40%): education, job status, family/social support, income and community safety
- Physical environment (10%): living conditions, working conditions,
- Health behaviors (30%): tobacco use, diet and exercise, alcohol use, sexual activity
- Health care (20%): access to care / quality of care
Our actual healthcare spending in the US is within that last 20%. In 2019 it was 3.8 trillion This accounted for 17.7% of the gross domestic product. With all that money being utilized I certainly expect better outcomes. But this is not our story…in fact the U.S. has continually had the lowest life expectancy and highest suicide rates among high income countries for many years. We know structured data has many great attributes i.e. helping with billing. The problem is, part of the promise of the electronic health records, the “holy grail of medical technologies”, was that these were supposed to provide better medical outcomes. However, many studies seem to show the conversion from paper to electronic means creates more issues rather than alleviates problems. Take for example this one issue - safety alerts within the EHR for medication problems. Safety alerts for actual issues were only represented about 54% of the time in 2009. Nine years later representation had only improved to 66%. This is unacceptable.
Sanity check- it’s time for a new approach- addressing factors within the 80%
Let’s address the 80 percent. The medical field has simply been ignoring these other important outcomes. We are in a time where doctors barely know who their patients are and where one doctor rarely has a clear picture of what the other physicians on a patient’s “team” are doing. Social determinants of health (SDOH) are included in this 80 percent. As defined by the Centers for Disease Control, they are the conditions in the places where people live, learn, work, and play that affect a wide range of health and quality-of-life risks and outcomes.
Ok I’m back to reality…but where do I begin?!
One way would be to identify key areas of SDoH in your population if these are not already known. This information will be mainly found in text-based documentation within clinical notes. Look for patterns within clinical notes by utilizing Natural Language Processing (NLP). If something is indeed a concern, good providers will usually document it. You will find many key areas such as transportation, food insecurities, housing and employment issues etc. Identification is key: you can’t address an issue if you don’t know it is there. For example, if you find out that 20% of your population has food insecurities then you can address the issue with the right resources, like implementing a food pharmacy.
It’s not just the sane thing, it’s the right thing to do- It’s economically important
A National Institute of Health assessment around Food Insecurity showed that if the average consumption of fruits and vegetables would increase by about 0.4 servings per day, it would prevent about 1.93 million cardiovascular disease events and save $39.7 billion in health care costs.
So beyond the fact that this is the sane thing, it is the RIGHT thing to do and why we ALL need to care.