Passive detection of COVID-19 with wearable sensors and explainable machine learning algorithms

Eric J Topol,Giorgio Quer,Matteo Gadaleta,Vik Kheterpal,Katie Baca-Motes,Steven R Steinhubl,Jennifer M Radin,Edward Ramos

doi:10.1038/s41746-021-00533-1

Eric J Topol, Giorgio Quer + Show 6 more

Open Access

https://doi.org/10.1038/s41746-021-00533-1

Copy DOI

Abstract

Individual smartwatch or fitness band sensor data in the setting of COVID-19 has shown promise to identify symptomatic and pre-symptomatic infection or the need for hospitalization, correlations between peripheral temperature and self-reported fever, and an association between changes in heart-rate-variability and infection. In our study, a total of 38,911 individuals (61% female, 15% over 65) have been enrolled between March 25, 2020 and April 3, 2021, with 1118 reported testing positive and 7032 negative for COVID-19 by nasopharyngeal PCR swab test. We propose an explainable gradient boosting prediction model based on decision trees for the detection of COVID-19 infection that can adapt to the absence of self-reported symptoms and to the available sensor data, and that can explain the importance of each feature and the post-test-behavior for the individuals. We tested it in a cohort of symptomatic individuals who exhibited an AUC of 0.83 [0.81–0.85], or AUC = 0.78 [0.75–0.80] when considering only data before the test date, outperforming state-of-the-art algorithm in these conditions. The analysis of all individuals (including asymptomatic and pre-symptomatic) when self-reported symptoms were excluded provided an AUC of 0.78 [0.76–0.79], or AUC of 0.70 [0.69–0.72] when considering only data before the test date. Extending the use of predictive algorithms for detection of COVID-19 infection based only on passively monitored data from any device, we showed that it is possible to scale up this platform and apply the algorithm in other settings where self-reported symptoms can not be collected.

Highlights

Frequent monitoring to quickly identify, trace, and isolate cases of SARS-CoV-2 is needed to help control the spread of the infection as well as improve individual patient care through the earlier initiation of effective therapies[1]
Passive monitoring is possible with commercial sensor devices measuring biometrics such as resting heart rate[5], sleep[6] or activity, which have been shown to be effective in the detection of COVID-19 versus non-COVID-19 when incorporated in combination with self-reported symptoms[7]
We developed a deterministic algorithm to discriminate between symptomatic individuals testing positive or negative for COVID-19, analyzing changes in daily values of resting heart rate, length of sleep and amount of activity, together with self-reported symptoms[7]

Summary

INTRODUCTION

Frequent monitoring to quickly identify, trace, and isolate cases of SARS-CoV-2 is needed to help control the spread of the infection as well as improve individual patient care through the earlier initiation of effective therapies[1]. Individual sensor data in the setting of COVID-19 has shown promise in identifying pre-symptomatic infection[8], the need for hospitalization[9], correlations between peripheral temperature and self-reported fever[10], differences in the changes in wearable data between individuals with COVID-19 versus influenza-like-illnesses[11], and an association between changes in heart-rate-variability and infection[12]. These studies focused on a specific device brand, or on a predefined set of signals. The algorithm uses selfreported symptoms when they are available, or otherwise makes its inference based on sensor data only, adapting to different engagement levels of the individuals in the study

RESULTS

METHODS

CODE AVAILABILITY