Comment

AI detects patients at highest risk of COVID-19 mortality

Peer-reviewed research published in Nature Scientific Reports

First published: 20 Aug 2021 | 3 min read
Author: Angus Reed

During the pandemic, digital technologies came to the forefront with the aim to alleviate the overwhelming strain on hospital resources. At Huma, we recognised an opportunity to use artificial intelligence (AI) to support and influence crucial clinical decision-making during the COVID-19 pandemic. Our research was published in Nature Scientific Reports.

We set to work developing a dynamic model that identifies patients who are at highest risk of mortality upon contracting COVID-19. We built the technology using over 10 years’ worth of UK Biobank data collected from 11,245 people who tested positive for COVID-19, employing a leave-one-out approach for validation to maximise the dataset. We found that 91% of the time, the model predicted the patients who went on to have the most serious complications.

Using AI, our risk model assesses patients’ medical records, identifying those who have pre-existing conditions that may be linked to fatal COVID-19; for example, previous acute kidney failure or respiratory failure. At the same time, the model captures live patient information during the early stages of the disease, such as dynamic changes in vital signs and symptoms. This double-pronged assessment means clinicians can more easily detect which COVID-19 patients are likely to need urgent attention based on both historical and real-time physiological data and thereby focus their care accordingly.

This project has shown us the value of AI in capturing real-time data and providing accurate risk models that will pave the way to more preventative medicine. We believe risk scores like this will play a vital role in the future of digital health if applied across a wide range of conditions, including cardiovascular disease, diabetes, and rare illnesses. We are entering into an era in which real world, living data is becoming more capturable, and this COVID-19 risk model is an example of how we can intelligently and responsibly use patient information to provide better care and help people live longer, fuller lives during the pandemic and beyond.

To learn more about our team’s work, watch the video https://vimeo.com/589753943 and read the peer-reviewed paper published in Nature Scientific Reports here: https://go.nature.com/3z1O4Dj

Read Next