RxFX - Mobile
Figure 1. Posterior distribution of Daily Sleep Quality
Figure 2. Posterior distribution of Fitbit Active Score
Figure 3. Posterior distribution of Daily Calories Burn
Figure 4. Posterior distribution of Intercept
Figure 5. Variable Importance Random Forest (MSE= 6.97%)
Making data inspired decisions regarding health care treatments and affects on individuals would help the many people suffering from complex conditions. The severity of many of these conditions such as arthritis, depression, hypertension, ADHD, anxiety and diabetes can be influenced by hundreds of factors in daily life.
Machine learning and other data science activities can be used to analyze hidden factors that can be improving or worsening a patients health care. But, in order to understand the factors and their relationships, we first need to obtain as much data as possible. Publicly available API's enable us to obtain more and more relevant transactions and interactions, making this endeavor easier.
With adequate data and machine learning techniques, studies are enabled to help individuals discover which hidden factors are worsening or improving their conditions.
What it does
This web application does the following:
- Facilitates the entry and collection of any kind of health factors for you as an individual.
- Perform a wide variety of advanced analytics based on unique individual data and/or the aggregated community data.
- Provide a visualization of the results of the analytic studies.
First time data contributing users are encouraged to set up automated data ingestion feeds from their personal digital lives. Think social media, fitness data, pharmacy data, health service information, etc. In addition, users can set up reminders and notifications to manually supplement their data with transactions not yet digitally recorded.
Aggregated user data is used to determine the factors that most influence any given aspect of health using data science techniques such as machine learning. Studies of the effects of interactions can be generated at rates previously unknown to medical practitioners.
Using the Analytics Dashboard, data streams can be combined and visually analyzed by physicians and evaluate one's progress over time and explore the data to find insights. Users are notified of ways they can improve their life via personalized real-time mobile notifications. Users who wish to optimize quantifiable aspects of their life are able to search and examine a list of products that are most effective at helping the average user achieve a particular health and wellness goal.
The general public can review the studies.
Data Storage - User data is securely transmitted and stored on HIPAA compliant servers
How we built it
Using cloud9, and material design principles, coding HTML, CSS3, Angular, R, .pharmacokinetic modeling techniques
Challenges we ran into
Creating pleasing design elements that simply the flow of what is a fairly complex tool.
Performing multiple studies using machine learning techniques (Bayesian and Random Forest, etc) repeatedly with different hyper parameters until useful insights could be gleaned.
Learning to develop modular code so that design element changes could be more effectively introduced.
Accomplishments that we're proud of
Participating in fun and worthwhile project!
Refactoring individual developer code to make the various web components have consistent design.
What we learned
From the background Bayesian analysis, we learned that the Daily Sleep Quality has a significant impact on people's mood the next day, controlled by daily calories burned and daily activity. For every one point increase in sleep quality score, the mood score will increase 0.621. (HDI: 0.515,0.739) The posterior distribution of coefficient of sleep quality score is shown above in Figure 1.
Another useful method we implemented is Random Forest, which was used to indicate that daily sleep quality is the most important factor that can influence user's mood. The variable importance plot was shown in Figure 5.
What's next for RxFX
Continue exploration of wearable technology and couple it with the emerging fields of non-traditional interfaces and smart home technology to help the elderly and others. Imagine the smart home technology in an assisted living facility that detects you have woken up and are in the kitchen. The smart home engages an Alexa skill that has been set up. "Good morning Don, what are you having for breakfast today"? The non-traditional interface perhaps makes it more amenable to record activities and transactions as you become loath to type on a computer at five in the morning.