By Ranjani Ramamurthy, Principal Physician Scientist and Isaac Naveh-Benjamin, Software Engineer
Before meeting a patient in clinic, a doctor tries to learn as much about her as possible. She typically scours her patient’s prior clinical notes, medication lists, preexisting conditions, labs and imaging tests. From all this information, she assembles a profile of her patient’s medical, socioeconomic and family background.
This labor is often invisible to patients, but takes the average medical practitioner as much as two extra hours per day.
Although this is labor-intensive, this time isn’t wasted. It’s critical for a doctor to know her patients as well as she can, so that she can craft an individualized treatment plan that accounts for a patient’s relevant medical, social, behavioral, and economic factors. Research suggests that that social, behavioral and economic factors contribute almost twice as much to healthcare outcomes as medical care does. Thus, they’re even more important than “traditional” medical activities like screening for disease, disease management, and clinical care.
For example, if a doctor knows that her patient lives in a homeless shelter, she immediately knows her patient is unlikely to have the tools and resources necessary to manage their congestive heart failure. They’re also unlikely to have a safe space to store their medications, have regular access to healthy low-salt food, or a scale to weigh oneself.
Although this information is vital, it’s often scattered across multiple clinical notes in the health record. Such important information generally lives in the unstructured (free-text) part of the clinical note, along with other data related to the patient’s social and family history. This makes it difficult to search for these, often most important, clinical variables.
At EmpowerMD, we’re trying to help clinicians unlock the benefits of this previously inaccessible (or hard-to-access) content. In designing our system, we’ve encoded information using rich semantic structures. This helps ensure important information is not lost.
Clinical example
To illustrate the benefits that can be unlocked via a more semantic approach, let’s consider a concrete clinical example (1).
Anna is a 55-year-old woman who comes to the clinic complaining of back pain. Normally, information about her employment, family status, and socioeconomics would on would be stored as unstructured text. However, if we were actually able to query her lifestyle data, we could determine that she is a smoker (a risk factor for vascular disease, which in turn can lead to and is often associated with back pain).
Similarly, if we could search her family history, we could discover she had gone through a painful divorce; also medically relevant, as emotional upheaval and distress often manifest as real physical pain.
We could also examine and learn her work history, living situation, and other important variables in health and well-being.
Such information is difficult to access through the electronic health record. Yet it’s vital to establishing the proper patient context. As Dr. Jim Weinstein, a spine surgeon, says, “this data is more important to me than an X-Ray or MRI. It is a better predictor of outcomes than most imaging studies.”
The case for a semantic note
At EmpowerMD, we believe there’s great potential in schematizing medical information at the point of capture. This means, quite simply, that as our ambient intelligent system analyzes doctor-patient conversations, it automatically organizes information in meaningful ways. For example, information about Anna’s smoking habits would be filed away in our internal document under “tobacco consumption”.
Because the system is ambient (that is, listening in the background), such classification occurs without any additional effort on the doctor’s part. This enables the doctor to continue to have natural conversations with their patients, without having to fiddle around with additional medical forms.
Similarly, when the doctor edits the resulting (automatically generated) clinical note, the note still looks like the traditional medical narrative. Under the hood, EmpowerMD captures any edits made by the doctor and integrates them into our semantic classification scheme. This enables doctors to have the best of both worlds – the benefit of greater semantic organization, without losing the expressivity of free text.
Applications of semantic notes
When information is stored in a semantic note, it becomes easily searchable. Instead of poring over multiple health records, the doctor can directly issue queries against the patient’s social and family history. Questions like “Does Anna have a family history of diabetes?” or “Has Anna’s employment status changed over time?” are now easily asked and answered. Over time, real-time context aware search becomes a reality.
In addition to helping doctors establish individual patient context, semantic notes can also be used to answer questions about patient populations. Looking ahead, we can envision visualization tools, built to track these traditionally unstructured variables, related to health of the patient, and those of a population/cohort of patients. Similarly, it will be possible to identify variations in care (treatment vs. outcomes) and track important gaps in screening.
In the past, some of these goals have been partially served by applying Information Retrieval (IR) techniques, such as full-text search against the corpus of EHR notes. However, the semantic approach offers tangible benefits beyond that: a vetted medical ontology at the time of encoding and saving the note, and a system that learns from the doctor’s classificatory choices in real time (unlike IR techniques, which are generally post-hoc).
In an upcoming post, we’ll dive deeper into the semantic model used by EmpowerMD and discuss how it’s updated and maintained.
Stay tuned!
Notes
(1) We’re indebted to Dr. Jim Weinstein, Senior VP of Microsoft Healthcare, for this example.