Introducing dbdpED, an educational initiative for digital biomarker discovery
Aiming to make digital biomarker discovery more accessible
UTILITY OF SMARTWATCH DATA
The advent and ubiquity of smartwatches and activity trackers like Apple watches, Fitbits, and Whoop smart rings have transformed the way we measure physiological signals. These trackers have enabled us to define and characterize heart rate, sleep, and activity over time. When something begins to go wrong in the body, say a viral infection or diseases like diabetes, these physiological signals start changing, sometimes slowly over time. And with the use of data analytics and machine learning tools, we can train our algorithms to detect these small changes in the body’s physiological signature to alert us of disease onset or progression.
THE DBDP
The Digital Biomarker Discovery Pipeline (DBDP) is an open-source platform seeking to make digital biomarker discovery more accessible and collaborative with open-source code, algorithms, and resources. Digital biomarkers are digitally collected data that are transformed into indicators of health outcomes. For instance, glucose levels measured from a continuous glucose monitor can be used to predict diabetic state. These digital biomarkers are being investigated for their potential in predicting conditions like diabetes, movement-related disorders, heart failure, flu, and more recently, COVID-19. You can read more about the DBDP on our website or one of our recent publications.
INTRODUCING dbdpED
Digital biomarkers currently require extensive domain knowledge and computing skills. The purpose of the DBDP is to provide code sets, functions, and algorithms for the entire digital biomarker discovery pipeline to make discovering digital biomarkers more accessible. From the input of wearable sensor data to the development of machine learning and deep learning algorithms, we have provided an open-source software resource for the digital biomarker community.
We now aim to use the DBDP to spread general awareness about digital biomarkers using an initiative called the dbdpED (DBDP — Education). Digital biomarker discovery can be a daunting task for anyone starting on this journey. The goal of dbdpED is to provide you all the tools to get started with digital biomarker discovery. We aim to spread awareness about digital biomarkers that can be developed using smartwatch data and how these could be useful for research as well as personal health management. With dbdpED, we want to arm anyone with the resources to analyze the data from their wearables to generate insights that are easy to interpret and can potentially be actionable. We aim to make these tools accessible and interpretable as well as generate interest in this field. If you wish to analyze your smartwatch data, we believe you should be able to do so, regardless of computational background or skill level. We aim to develop flexible and easy to understand guides and tutorials that can be used by anyone to analyze smartwatch data, their own or otherwise, and look for health insights.
WHO IS IT FOR
If you are interested in wearable data analysis, dbdpED is for you. You might be a school athlete looking to better analyze and visualize your performance, a clinician looking for a way to analyze wearable data from patients to discover some trends, or just someone looking to know more about what digital biomarkers are! We provide guides and tutorials that are easy to understand and will help you get information on everything you need to start digital biomarker discovery or just wearable data analysis.
COMPONENTS OF dbdpED
dbdpED will feature the following:
- “How to use” tutorials
- Code sets for wearable data pre-processing, health metric generation, statistical and machine learning algorithm application and visualizations
- Annotated codes with overview blurbs
- Different types of available visualizations and use cases for digital health data analysis
In order to make these resources truly accessible to all, we are releasing easy to understand code sets and tutorials. To encourage exploration, the code sets will be modular and will be developed for a plug-and-play approach. The DBDP features modules for exploratory data analysis, resting heart rate, glucose variability, activity classification, nutrition, sleep, and mental health with many more coming soon. We are also sharing sample data and figures to foster a better understanding of different data available from smartwatches and how to best visualize them. We will soon be releasing video tutorials on how to use dbdpED to explore digital health data and discover digital biomarkers. Check out already available tutorials here!
In the first phase of the dbdpED launch, we will focus on developing resources for high school students. An excess of complicated jargon around digital biomarker resources means it is harder for students to grasp what the field is about. With dbdpED, we focus on making this field truly inclusive. We will be launching detailed and easy to understand lesson plans for high school students and beyond. This will involve sharing the end-to-end process of digital biomarker discovery using easy to understand tutorials via blogs and vlogs. With these in-depth tutorials, we will provide users with a fun and engaging learning experience for digital biomarker discovery. The resources will be useful for anyone who is new to the field of digital biomarker discovery and wishes to learn more. All these resources will be available on the dbdpED website along with useful links to learn more about a specific method. If you are interested in being one of our pilot K-12 schools or programs, please reach out to us!
AN INTRODUCTION TO DIGITAL BIOMARKER DISCOVERY
Information from wearables is collected over long periods of time and can provide insights into an individual’s health. All this data can be exported into common file formats like .csv and .JSON and loaded into software applications like jupyter notebooks for analysis. Applications like jupyter notebooks allow you to write modular codes to explore your data. They have a huge collection of code libraries that can be utilized to perform complex operations in a simpler and concise way. We have developed numerous such code libraries that make the analysis of digital health data much simpler.
Once the data is loaded, it needs to be pre-processed to aid analysis. This is similar to cleaning and cutting your groceries before you can cook some complicated dishes using them. The size of pieces you will cut the vegetables in and the tools you will use, depend on what the vegetables are and what you plan on cooking with them. Similarly, data pre-processing can involve multiple steps depending on the data and analysis goals. The DBDP has multiple such pre-processing tools to help you with data cleaning. dbdpED will help you easily understand the purpose and utility of these modules.
Exploratory Data Analysis is usually the next step where you can generate some insightful and easy to understand figures from your data. As the name suggests, it is aimed at exploring your data before delving into the analysis. This can help establish trends in your data and help you better understand your data. The DBDP has useful and easy to use EDA modules to generate some cool visualizations from wearables data and this will be a major focus of several dbdpED tutorials!
We can now use various statistical and machine learning tools on this data to analyze it and generate useful insights. For instance, this can involve discovering if your heart rate variability data can give you insights into your health or establishing trends between sleep and mental health.
More information on these components will follow in our subsequently released tutorials which will help you get started on your journey of digital biomarker discovery.
DEVELOPING YOUR FIRST dbdpED CODE (Coming soon)
In our next dbdpED blog, we will talk more about how you can build your first ever end-to-end dbdpED code set for wearable data analysis. Stay tuned!
