Care-Tracker

Streamlining remote oncology care with custom data+RAG Model

Comment

Care-Tracker by Team Oncare

Mission

Cancer treatments can have a myriad difficult side-effects that impact a patient's quality of life. As a consequence, it can be undoubtedly challenging for patients to accurately track and report side effects to their doctors. The motivation behind Oncare's Care Tracker was to create a simplified way for patients to record their side-effects on a daily basis with a keyboard or a push-to-talk voice button. We also recognised the possibility of streamlining the response process for doctors by using AI to detect and extract important keywords from their notes. We believe improving the communication and response user experience in Care-Tracker will enable better primary care management for the AI era.

What it does

Patients can use the app to log notes about their daily health and any side-effects experienced. The custom data and NLP model analyzes these notes to detect instances of symptoms like nausea, fatigue, nerve pain, etc. A summary report is generated highlighting the patient's main side effects over a given timeframe, giving both patients and doctors a potentially better communication experience in their primary care interactions. In turn, Doctors can have access to streamlined data-driven insights into the patient's condition between visits.

How we built it

The app was built using:

  • a Vanilla JS front-end
  • Python-Flask
  • MongoDB database
  • RAG + Gemini AI model to detect instances of common chemotherapy side effects in patient documents. The extracted keywords are used to classify the patient's symptoms and generate a summary report.

Challenges we ran into

  • One challenge was customising the NLP model to accurately detect side effect keywords from natural language notes. We had to provide the model a custom synthetic dataset of sample patient notes labeled with side effect keywords.
  • There were also challenges building a seamless UX for patients to log daily notes and view reports. We wanted the app to be simple enough for patients to use easily even when feeling unwell.

Accomplishments that we're proud of

We also developed a functional, responsive web app with an intuitive interface for daily logging and report viewing. Though basic, we were able to demonstrate the potential usefulness of the concept.

What we learned

  • This project taught us a lot about the possibilities and limitations of applying NLP techniques to healthcare use cases.
  • We learned the importance of custom data quality and the difficulty of disambiguating medical terminology. Though challenging, we are motivated to continue iterating on the project.

What's next for Oncare Care-Tracker

  • Going forward, we would focus on improving the model quality with how we use data.
  • We would also expand user accessibility with addition of voice-notes.
  • We aim to build this into a solution that can eventually help real patients manage side effects and improve outcomes.

Built With

Share this project:

Updates