A multi-layer real-time alert model for the early detection of COVID-19

31 May 2022, 14:30 
zoom & Room 206 
A multi-layer real-time alert model for the early detection of COVID-19

Eilat Lev-Ari, M.Sc. student at the department of Industrial

31 May 2022, 14:30 PM, Room 206& via zoom


Early detection of COVID-19 is crucial for reducing transmission and facilitating early intervention. Preliminary efforts for the early detection of COVID-19 mainly relied on a combination of reported symptoms and potential exposure to infected individuals. However, multiple pathogens may cause symptoms like those of COVID-19, and presymptomatic or asymptomatic cases account for 40–45% of those infected with COVID-19. Several pioneering studies have offered proactive methods for COVID-19 detection based on smartwatches and activity trackers. However, these methods rely on dedicated devices and require that individuals agree to frequently wear these devices. Here, we propose a multi-layer real-time alert model for the early detection of COVID-19. Our model is multi-layered in two senses. First, it incorporates data from multiple sources, including electronic medical records, smartphone and smartwatch sensors, self-reported questionnaires, and publicly available information. This allows combining information from multiple data sources, which are in principle independent, to increase the performance of the model. Further, if one or more of the layers are not available for a given individual, predictions can still be made. Second, the model provides a holistic view of the individual and the disease, covering aspects such as socio-demographic information, medical history, spatiotemporal patterns, clinical symptoms, and social, mental, sleep, and activity information. To evaluate our model, we analyzed data that we collected as part of the PerMed study. A cohort of 4,665 participants was followed between November 2020 and March 2022, for which 1,604 cases of COVID-19 were reported. The model obtained a remarkably high Area Under the ROC Curve (AUC) of 90% when using all layers of information, where each of the layers contributed its part. Importantly, when only objective (passively collected) layers of information were considered, the model was still able to obtain a high AUC of 88%.


Tamar Amir,
M.Sc. student at the Department of Industrial Engineering in Tel Aviv University, specializing in Data Science. Tamar holds a B.SC. degree in Industrial Engineering from Tel Aviv University. Her research focuses on Personalized Medicine and Big Data Analytics collected from Electronical Medical Records, smartphones, smartwatches, and self-reported daily questionnaires. The research is supervised by Dr. Erez Shmueli.

Tel Aviv University makes every effort to respect copyright. If you own copyright to the content contained
here and / or the use of such content is in your opinion infringing, Contact us as soon as possible >>