Keeping track of electronic medical records, especially across various clinical systems and over time, is difficult, not user-friendly, and sometimes impossible. Compacting longitudinal EMRs into a format that can be readily accessible, interpretable, and saved into a simple system can revolutionize the field of medical record keeping.
hCode, a novel frontier for medical record collection and personalized medicine
Based on the combination of parametric survival model and machine learning approach, hCode is an individualized "QR code" for health which uses dimension reductionality to pack medical information into a single picture. This pictorial code is generated by an interactive web app which collects various health variables (eg. blood pressure) to transform data into pixels which store medical information. These hCodes are easily stored on mobile devices by patients and can be presented to clinics or in emergency settings to understand one's medical history and even for personal reference of actionable health items. Additionally, hCodes integrates information from past codes with new health data to create updated hCodes for storing medical information longitudinally.
How I built it
We modeled multi-dimensional medical records information using a combination of survival analysis, machine learning, functional principal component analysis, and JIVE (joint and individual variance explanation) for encoding longitudinal healthcare variables. The model was trained on more than 10,000 Americans from NHANES, a publicly available, nationally representative data set. Based on the model (published earlier this year), we were able to calculate the phenotypic age, and further show the difference between phenotypic age and chronological age, which indicates whether one’s physiological conditions is older or younger than their chronological age. To empower the compression of health data into hCode, we also developed a base matrix which can be easily accessible by the individual and health care providers.
Challenges I ran into
Reading through multiple papers to derive medical parameters and generating machine learning models was quite a learning adventure. It also was difficult to break down a dense mathematics paper and translate the findings into an interactive visualization to present complex data in a user-friendly, intuitive way.
Accomplishments that I'm proud of
It was super exciting to develop a tool that integrates newly published research on a hot topic. Also, we are very proud to overcome hurdles in rendering and deploying an interactive web app and output.
What I learned
So much! It took us several hours just to hack the data from the latest published literature and build the base matrix for hCode. It was the first time for all of us to create a web app. We also learned a bunch of complex math to make the backend generate the hCode. Reading through some hot new research on the field was very exciting too.
What's next for hCode
Real-time collection of data by linkage to health apps on mobile devices and Fitbits. Increased security measures to protect health information of users. Additional functions to integrate more health parameters and newly updated clinical recommendations based on medical information collected. Also a mobile app!