Doctors should be empowered to treat patients to the best of their abilities. But EHRs (Electronic Health Records) are keeping them from doing their job.

For every hour that doctors spends with a patient, they have to spend two additional hours with the EHR. This is disruptive to physician workload, and adds unnecessary burden. It reduces the amount of attention that physicians can give to patients, reducing quality of care. The excessive workload contributes to physicians becoming disillusioned and burnt out. In fact, nearly 2/3 of US physicians feel burnt out, depressed, or both - one of the highest rates among all professions - and 56% of them blame documentation (Medscape, 2017). Doctors are humans too, prone to exhaustion. This contributes to medical errors being the third highest cause of death in the US, second only to heart disease and cancer (BMJ, 2016).

Doctors need a tool that will drastically reduce screen time, and maximize patient face time.

What it does

Meno BOT sits between the physician and EMR systems, such as Epic. Think of it as an intelligent, always-vigilant medical scribe. From the conversation that the doctor has with the patient, it will automatically bring up most relevant information for treatment. It will extract the most relevant points from the doctor's speech, and upon approval it can automatically populate the EMR system that the doctor uses.

This drastically cuts down on the time that doctors have to spend documenting, allowing doctors to focus their energy treating patients, not dealing with paperwork.

How We built it

We used Web Speech API to convert voice into text. We trained our (Machine Learning training speech processing tool) on sentences and commands that a physician would use during patient engagement, so that it will extract most relevant information from the sentence ("intent"). Code was written in Facebook's React framework. MobX was used for state management.

Challenges I ran into

1) training We learned firsthand how painstakingly long it takes to train to handle multiple intents and complex sentences. Because medicine is an area that requires high accuracy, we had to lower our expectations for the complexity of commands and information extraction that our bot could process.

2) Integration with existing EHR platforms Our vision for the project was that the bot would extract the most relevant information and with physician approval, automatically input them into the preexisting EMR system that the doctor uses. We found that due to limited information about these preexisting systems, we were not able to implement this fully. However, each of these EMR systems have predesignated fields, which would make this process relatively quick to implement.

Accomplishments that I'm proud of

When we first spoke with Dean Davidson of School of Nursing at Johns Hopkins, she hinted at EHR and data standardization as one of the largest issues in healthcare today. Admittedly, our initial reaction was "that sounds like too large of a problem for us to solve in 36 hours!" But listening to and speaking with some inspirational mentors, we were reminded that it's not acceptable to treat crucial issues as 'unsolvable,' especially at the institution where since its inception, visionaries have been realizing their visions of what medicine should be.

So, we're proud of at least presenting a proof-of-concept solution to an incredibly complex yet critical issue, instead of backing away and finding 'easier' problems to solve.

What We learned

We learned about the various challenges that doctors face associated with EHRs and the magnitude of the problem in the healthcare system. We also learned how to use and the challenges associated with training the language processor as well.

What's next for Meno BOT

We would augment our solution to the challenges we faced mentioned above.

Share this project: