Oct 17, 2020
mPath: Fixing our Broken Healthcare System with Artificial Intelligence
Made at the TKS Hackathon 2020. Thank you to Aahaan, Richa, Manroop, Anush, and Kabeer for making this wonderful team and product possible.
Something that every human shares is our sense of altruism. We care for our young, our old, our disabled, and our sick. We go to medical school to learn how to save lives, become doctors, nurses, and caretakers. We build hospitals for everyone to get the help they need and design world-class machines to run tests and provide life support.
But once a patient gets inside a hospital, how long does it really take to access medical services?
Take the case of Marianne Porter, a New Brunswick woman who recently passed away from kidney failure. She waited 11 hours, yes, 11 hours, for a doctor to see her in a hospital emergency room. Although she was facing a time-sensitive condition, there were simply too many people waiting in the ER before her. The hospital needed to sort everyone into the appropriate wards before administering in-depth care. Sadly, Marianne succumbed to her illness before she could see a doctor. One delayed blood test was the difference between her life and death.
Unfortunately, Porter is one of many patients whose experiences show the horrific shortcomings of hospital systems. In Ontario, Canada, the average emergency room wait time is 16.3 hours, 2 hours up from 2017. During the flu season, this average rises to 18.2 hours and 34% of patients are admitted within the province’s target time of 8 hours.
Do you sleep in your waiting room chair while you wait?
Do you eat breakfast, lunch, and dinner from the overpriced vending machine?
How do you manage the pain?
How do you support your family members?
The leading cause of these long wait times is that many people seek treatment at hospitals without considering alternative treatment locations, such as family doctors or urgent walk-in clinics. For good reason, people default to going to hospitals just in case a family doctor or clinic doesn’t have the equipment to help them.
Even though it is smart to seek help as soon as possible, 1 in 5 people who go to the ER could be getting sufficient care from either urgent care walk-in clinics or family doctors.
What if there was a way to reduce the number of people in the ER and therefore reduce the wait time by 20%?
Introducing mPath
Motivated by the potential to make an impact on billions of people around the world, we’ve designed mPath, an application that:
- identifies how critical a patient’s symptoms and condition are,
- then guides patients to get the care they need in the fastest time possible.
This app is accessible:
- On mobile devices
- As a web app
- Through QR codes located in hospital emergency rooms
Across all these platforms, mPath’s proprietary machine learning model helps direct you to the nearest care facility such as a family doctor, urgent care walk in clinic or a hospital, with the shortest wait time.
How does mPath work?
This system takes a 2 pronged approach, in hospitals, as well as at home. The first portion focuses on individuals and families at home. This system is based around the Ontario healthcare system, but can and eventually will expand worldwide. The Ontario prototype allows patients to track data with their health cards as well as provides a feedback loop for the algorithm.
Patient care is our ultimate priority, and to ensure complete security, our application will be encrypted in accordance with HL7 protocol.
Users can enter their symptoms into the powerful AI-based algorithm, which would immediately give a recommendation based on the model trained off the Clear Triage, medically-acclaimed, database. The easy to use interface and multi-language option allows accessibility to users of all ages and backgrounds. Users also have the option to link any biotech devices, such as an Apple Watch, to the application to provide additional context into your symptoms. In fact, the app’s UI/UX helps patients easily input their symptoms in a variety of languages. Of course, this method requires extra time to input information, so you set it up when you aren’t in need of urgent help or in a life-threatening scenario.
As for the algorithm itself, a semi-supervised learning model was used with structured and unstructured data. NLP is used to filter stop words, ID/categories are then converted into hot vectors, and Word2Vec is integrated for word embedding and string clustering. Finally, when all converted, they are run a MLP, which outputs an intelligent
The algorithm uses the certified Schmitt-Thompson Protocols, which is the industry standard for medical triage structure and delivery. The protocols consist of the symptoms and key identifiers according to level of triage. The data was web-scraped from the Clear Triage Database to create the training and validation data. Using NLP, long short-term memory paired with adversarial models, and a multilayered perceptron the algorithm reaches a conclusion. NLP is used to filter out artifact parts of speech, then the trained adversarial-modified LSTM model finda correlations in our string values, like symptoms and medical history. Outputs of the NLP and LSTM word embedding and integer values like age and weight are fed into our final neural net, which outputs our predictions.
Upon getting a recommendation from the algorithm, the system uses GPS navigation to identify the closest location to ensure convenience for the user. Since artificial intelligence will never truly replace human instinct, the application will also provide a percentage alongside each prediction of where to go. For instance, upon request, a user will see that mPath is 98% sure suggesting going to a walk-in clinic, and 2% sure of going to a family doctor.
We emphasize that the app is not a substitute for emergency services such as 911.
The second portion of the system will be deployed in hospitals. This system runs a similar algorithm, that helps identify where they should be cared for, and redirects them if necessary.
This web app will be accessed through a QR code at the hospital front desk. Patients would have to scan the code, and fill out the form; similar to the home system. The system conducts a triage analysis, and if it determines that the patient could get care from a family physician, or an urgent care clinic, they would be given the option to either go to the clinic and receive treatment significantly earlier, or wait at the hospital.
Executable
mPath is a project that can be very easily incorporated into the healthcare industry. For years, hospitals have used a triage system to determine if a patient needs immediate medical attention, or if it is safe for them to wait. By incorporating artificial intelligence, this process streamlines the emergency room waiting process and allows every patient to get care within a reasonable timeframe. This also improves the quality of care medical officials can provide, with the stress of time partially reduced; we anticipate that “hallway healthcare” will also be reduced.
Conclusion
The mPath team has spoken with doctors around the world and called urgent care clinics and seen positive approval of mPath. The intersection of healthcare and artificial intelligence creates one of the most impactful fields you can go into and we look forward to growing our team.
