Time series anomaly detection in helpline call trends for early detection of COVID-19 spread across Sweden, 2020

Abstract

Abstract Timely detection and surveillance of disease community spread is a potent tool for implementing effective public health interventions. This study investigates the National Telehealth Service (1177 helpline) across 18 regions in Sweden in 2020 to identify early signals of community transmission of COVID-19 at the beginning of the pandemic. Focusing on calls related to key COVID-19 symptoms (cough, fever, and breathing difficulties in adults), we analyze their frequency and distribution across referral categories, comparing them to 2019 data. We employ an explainable time series anomaly detection algorithm using daily call data to identify the first collective anomalies across regions. The results show that anomalies in call data were correlated with, but preceded, the first confirmed case infected in Sweden by a median of 7 days (IQR: 2.5–10.5) and the first hospitalized case infected in Sweden by a median of 13 days (IQR: 7.25–16). They also preceded the estimated onset of community spread, indicated by the absolute confirmed cases (median: 24.5, IQR: 18.25-32.5), and severe outcomes defined by hospitalizations (median: 33, IQR: 27.25-44). These findings showcase how helpline call monitoring, using time series anomaly detection, can aid early outbreak detection.