There’s a lot of activity going on with large technology companies and others trying to get access to EMR data to mine it for insights. They’re using machine learning and artificial intelligence to crawl notes and diagnosis to try to find patterns that may predict disease. At the same time, equal amounts of energy are being spent figuring out how to get data from the myriad of medical and consumer devices into the EMR, considered the system of record.
There are a few flaws in this plan:
- A significant amount of data in the EMR is copied and pasted. While it may be true that physicians and especially specialists see the same problems repeatedly, it’s also true that lack of specificity and even mistakes are introduced by this practice.
- As well, the same ICD-10 codes are reused. Doctors admit to reusing codes that they know will be reimbursed. While they are not mis-diagnosing patients, this is another area where there is a lack of specificity. Search for “frequently used ICD-10 codes”, you’ll find a myriad of cheat sheets listing the most common codes for primary care and specialties.
- Historically clinical research, on which recommendations and standard ranges are created, has been lacking in ethnic and sometimes gender diversity, which means that a patient whose tests are within standard range may have a different experience because that patient is different than the archetype on which the standard is based.
- Data without context is meaningless, which is physicians initially balked about having device data in the EMR. Understanding how much a healthy person is active is interesting but you don’t need FitBit data for that, there are other indicators like BMI and resting heart rate. Understanding how much someone recovering from knee surgery is interesting, but only if you understand other things about that person’s situation and care.
There’s a pretty simple and often overlooked solution to this problem: get data and information directly from the patient. This data, of a patient’s own experience, will often answer the questions of why a patient is or isn’t getting better. It’s one thing to look at data points and see whether a patient is in or out of accepted ranges. It’s another to consider how the patient feels and what he or she is doing that may improve or exacerbate a condition. In ignoring the patient experience, decisions are being made with only some of the data. In Kleiner-Perkin’s State of the Internet Report, Mary Meeker estimates that the EMR collects a mere 26 data points per year on each patient. That’s not enough to make decisions about a single patient, let alone expect that AI will auto-magically find insights.
We’ve seen the value of patient engagement in our own research and data collected, for example in identifying side effects that are predictors of post-surgical readmission. If you’re interested, in these insights, we publish them through our newsletter. In interviewing patients and providers, we’ve heard so many examples where physicians were puzzled between the patient’s experience in-clinic or in-patient versus at home. One pulmonary specialist we met told us he had a COPD patient who was not responding to medication. The obvious solution was to change the medication. The not-so-obvious solution was to ask the patient to demonstrate how he was using his inhaler. He was spraying it in the air and walking through the mist, which was how a discharge nurse had shown him how to use the inhaler.
By providing patients with useable and personalized instructions and then tracking the patient experience in following instructions and managing their health, you can close the loop. Combining this information with device data and physician observations and diagnosis, will provide the insight that we can use to scale and personalize care.