By the time the COVID-19 pandemic forced shelter in place warnings around California, the virus was already well past containment. First-line medical responders were already running short of space and supplies, and yet the pandemic was still in it's early stages. The situation got bad fast, in large part due to a lack of actionable information. Who is sick, how sick, where should resources go, when should a care giver step back and isolate?
Using wearable sensors to generate continuous temperature recordings and nightly heart rate, heart rate variability, and respiration rate recordings, collaborators at UCSF gathered physiological data and combined these with self-reported daily symptom questionnaire responses. Roughly 50,000 people joined the effort, donating months of continuous data and over a million symptom reports. The goal is now to build models of illness progression, and more importantly, onset and severity prediction.
UCSD's contribution to TemPredict and beyond:
As the lead at UCSD, I am collaborating with Dr. Ilkay Altintas at SDSC to make our data available in a convenient, well-structured model to enable easy collaborations. Beyond this crisis, we want our learnings to serve as a template for how large biological time series data can be managed to optimize discovery. More efforts like this one are coming, and knowing how to handle all that data is its own challenge, aside from analyzing it for biomedical applications.
We welcome those interested in aiding either the data management or biomedical analysis efforts.
Since then many additional papers have grown from these data. The large and diverse data set continues to support algorithms for specific outcomes. It also allows us to develop generalizable tools to support de-biasing in algorithms so that health algorithms can be more trustworthy over diverse populations. This research has also let us develop tools and methods that support more efficient algorithm development methods.