Approximation of Influenza-like illness rates using sleep and cardiorespiratory data from a smart bed

ABSTRACT

Introduction: Pathophysiologic responses to viral infections affect sleep duration, quality, and concomitant cardiorespiratory function. Real-world, longitudinal monitoring of sleep metrics using a Smart Bed could prove to be invaluable for infectious disease detection. Previously we leveraged sleep metrics from a smart bed to build a COVID-19 symptom detection model. Analysis of pre-pandemic data with this model indicated that our results may generalize to detecting symptoms of other influenza-like illnesses (ILI). Here we investigated whether seasonal ILI trends reported by US Center for Disease Control and Prevention (CDC) can be approximated from aggregation of individual ILI symptom predictions. Materials and Methods: An IRB approved survey with COVID-19-specific questions was presented to opting-in Sleep Number customers from August to November 2020 in the USA. COVID-19 test results were reported by 3546/9370 respondents (249 positive;3297 negative). Sleep duration, sleep quality, duration of restful sleep, time to fall asleep, respiration rate, heart rate, and motion level were obtained using ballistocardiography signals from the smart bed. Longitudinal seep data from January 2020 to December 2020 from 122 of the positive and 1603 of the negative respondents were used to develop an individual-level COVID-19 symptom detection model. The model produces a probability of experiencing COVID-19 symptoms for each sleep session. Pre-pandemic sleep data from January 2017 to December 2019 from 4187 responders (1820 sleep sessions per night on average) were used to assess the ability of the developed model to generalize to ILI symptom detection. Weekly rates of high-scoring sleep sessions between January 2017 and June 2018 were fitted to the weekly ILI rates as reported by CDC using a negative binomial model. Subsequently, Pearson correlation coefficients were calculated for the predicted and reported rates between July 2018 and December 2019. Results: Correlation between the predicted and CDC reference was 0.91 (+0.04 compared to the baseline model). Correlation restricted to the influenza season (week 40 of 2018 to week 20 of 2019) was 0.87 (+0.13 compared to the baseline model). Conclusions: The sleep metrics measured with a smart bed platform are a unique source of longitudinal data, collected in a real-world, unobtrusive manner. This system may serve as a valuable asset in predicting and tracking the development of symptoms associated with a wide variety of respiratory illnesses, including influenza and COVID-19. Acknowledgements This study was funded by Sleep Number Corporation.