Doctors need help managing the complexities of modern medicine.
Introducing the first self driving EHR
Your computer system becomes a part of your team.
Like any good team member your system should know what you are doing and why. It should catch your mistakes, anticipate your next action and reward your hard work.
You train your system by using it, then it begins to follow you, and slowly the magic starts.
With academic advisors and research collaborators from Dartmouth, University of Michigan, University of Texas, Harvard, and under the guidance of respected industry leaders we make sure the magic is not going to run out anytime soon.
Want to learn more? Read on..
Quick history of the medical document.
A static record of what was important to the doctor. A doctor sees a patient, writes a note on paper, and puts it on top of the note stack. After several years information becomes lost.
When we started using computers in healthcare it was to type notes rather than write them by hand.
In the more recent past we began generating our documents with pre-populated segments of text.
Computers help with transmitting orders and results but the encounter remains the same, or even worse due to auto generated text .
The future of healthcare is a faster and more accurate diagnosis and resolution of problems. With the ever increasing amount and types of data being generated to manage complex patients, physicians will require systems that help them meet these challenges. It will involve a self learning healthcare system that will ultimately provide autonomous patient management, under the watchful eye of a physician.
Information is of two types: human derived and sensor derived. The progression toward the future model is a progression from human derived information to sensor derived information. In the distant future we will not have to rely on us humans to describe or explain our symptoms.
Sensor derived information will allow us to avoid ambiguous and vague symptoms, getting us to the actual cause of the problem and to a quicker resolution.
The loop is: we say things about how we feel (human data) and we associate those statements with the data we collect from labs and studies (sensor data).
The sooner we connect these two data types the sooner we can trust the sensor data. This is the process of validating sensor data.
Currently, human derived information is stored in our clinical notes. The quality of this data is very poor. At best it is a partial record of the events that occurred during an office visit, but more often it is a collection of prebuilt phrases often containing inaccurate data. When we try to use this data to learn new things about the human condition and our clinical problems, we are very limited. Using today's clinical notes to validate sensor data is not optimal. We need more accurate and objective records of what actually occurs during doctor visits if we are to move forward.
By using a combination of multimedia and text, our system captures the most objective human data and then categorizes this data into meaningful elements.
There are two types of meaningful elements: clinical and procedural.
Clinical elements lead to increased knowledge of particular clinical problems.
Procedural elements allow the system to more accurately predict our actions and support them.
So what are these elements and what makes our system so novel? The primary categories are problems, todos, and encounters.
A problem is a set of related data that has a unique name and conceptual meaning.
Todos are tasks that are expected to resolve a problem.
Encounters are a log of computer actions that the doctor did during an encounter.
Computers are great for handling changeable data. Our problems, just like our lives, are in a constant state of change. A traditional EHR is a stack of frozen notes, a reasonable way to describe a past event, but not a good model to handle a life that is still in becoming. Variables describe change, and when we organize our health problems with the right kind of variables we can better fit the richness of a human life. The problem, todo, encounter model allows us to categorize meaningful elements as both clinical and operational. When we record the audio of a doctor visit and associate the action log of user events we have optimal data for training machine learning algorithms.
As users manage problems they leave a trail of actions that we can learn from. We have been building predictive models of these actions - our system is becoming aware. The system is learning from the content of the audio recordings because the system segments the speech in meaningful elements that have both a procedural and a clinical meaning. In other words, each element is a certain type of action about a certain medical concept: a question about a lab test; a statement about a symptom; a summary of information about a particular problem. We categorize procedural actions as accessing information, adding information, etc., and we categorize medical information using standardized, structured clinical data formats.
By collecting these meaningful elements and using machine learning to constantly optimize our predictive model we can evolve our self driving EHR. This is the future of healthcare.
There are many more subtle but revolutionary aspects to our system that we can show you in person, so get in touch and become part of the next generation of doctors and prepare to rediscover joy in being a doctor.
Credits:
Created with images by NEC Corporation of America - "NEC-Medical-137" • pedrosimoes7 - "Older woman profile" • www.ilmicrofono.it - "bodycare, clinic, clipboard, doc, doctor, female," • Unsplash - "computer business typing" • Pacific Air Forces - "120615-F-FD024-006" • FirmBee - "office tax business" • johnvoo_photographer - "bacteria/viruses" • ColiN00B - "dna dns biology" • Pezibear - "child girl blond" • beba - "leaves green shadow play" • Rosa Menkman - "Filtering Failure" • sipa - "clouds nature clouds form" • OakleyOriginals - "63Dad seeing patients" • kev-shine - "Technology" • Greg Burkett - "Analysis" • Unsplash - "books pages story" • Unsplash - "barley cereal grain"