SAVR- Stroke and Cardiovascular Risk Assessment using Wearable Technology
Presented By Roshan Pillai, BASIS Scottsdale
Cardiovascular risk is the risk of a person developing cardiovascular diseases, often sudden. Heart disease is the leading cause of death for both men and women. Most of the current warning tools utilize the threshold level wherein, a warning is given when a patient’s symptom crosses a threshold. Unfortunately, this threshold method does not take into account other signs and symptoms demographic, or social factors, nor does it attain a 360-degree view of a patient’s
risk. Cardiovascular issues can appear rapidly and without warning. Heart attacks (Myocardial Infarction) and Strokes (Cerebrovascular Accidents) can cause irreversible damage and focus on prevention or early diagnosis is still needed.
Also, the current risk calculators and available tools do not provide results based on input from wearables/ live data and also do not incorporate the latest technology that’s available to the users to include information from their medical records. I created an app that utilizes an interoperable based approach where a patient/user can take their everyday data from a wearable, integrate the information available on their health app/health records and with user input, including their reported medical symptoms to screen and find out their relative risk for stroke/cardiovascular disease.
My design incorporated integration with Wearable Technology, the ability to integrate live data, and allow user input, as well as the ability to access health records (via PHI: Protected Health Information) in a HIPAA compliant manner to protect patient privacy. I formulated a hierarchical project architecture to create this application. My first layer was the data collection layer. Once I had obtained the data, I converted the data to .csv files for easier analysis. The next layer was the analytical model layer where I utilized machine learning, created a statistical model, and then combined the two models to create my algorithm. Once my algorithm generated both a viable and accurate structure for prediction, the next step was implementation. Lastly, I constructed an Intelligence Layer to access Wearable Technology as well as create SAVR (Stroke and Vascular Risk), my mobile application. For my machine learning algorithm, my unique approach was to mesh both statistical analyses through the use of Ensembling Models. The machine learning method used was XGBoost, an open-source, R and Python-based software library which is a versatile tool made to work through most regression, classification, and ranking problems as well as user-built objective functions. I chose XGBoost because it’s a library designed and optimized for boosted tree algorithms which are specifically tuned towards training an algorithm for software functionality. Also, XGBoost is known for its model performance/ accuracy and execution speed; both critical for this project. I then did the statistical model to add another layer to my algorithm accuracy. I did a linear regression and created a statistical model algorithm using Python. The algorithm plots all of its analyzed data then finds the most accurate linear line to its predictions. Then, I performed a linear regression to find out how my application should weigh different factors. After I generated both a machine learning model and a statistical model, I used Ensembling Models to combine the two. The combined algorithm revealed how different risk factors must be designed to create an accurate prediction for a patient.
Lastly, to build the SAVR Application, I used Ionic, a hybrid app development platform. Ionic is an open-source Java-Script-based Language. With Ionic, I combined the Analytical Model Layer and the Intelligence Layer to create the Presentation Layer, where the user could input their data and get a predicted Risk Score. This was SAVR, the wearable app that I hope that one day will be a tool that helps save lives.