Differences in social activity increase efficiency of contact tracing

Abstract

Abstract: Digital contact tracing has been suggested as an effective strategy for controlling an epidemic without severely limiting personal mobility. Here, we use smartphone proximity data to explore how social structure affects contact tracing of COVID-19. We model the spread of COVID-19 and find that the effectiveness of contact tracing depends strongly on social network structure and heterogeneous social activity. Contact tracing is shown to be remarkably effective in a workplace environment and the effectiveness depends strongly on the minimum duration of contact required to initiate quarantine. In a realistic social network, we find that forward contact tracing with immediate isolation can reduce an epidemic by more than 70%. In perspective, our findings highlight the necessity of incorporating social heterogeneity into models of mitigation strategies.

Supplementary information: The online version supplementary material available at 10.1140/epjb/s10051-021-00222-8.

Fig. 1

Simulating the spread of COVID-19 on the contact network. Here, a zoom view on the geographical positions of a few individuals (based on GPS coordinates) during a typical work day and for a representative run of the epidemic model. Regions of contact (defined by signal strength exceeding the $-85$ dBm cutoff) are shown as diffuse clouds of pink. Snapshots shown are at day 2, 23 and 44 of the outbreak

Fig. 2

a A small subset of a contact network for 1 week. Link thickness indicates the cumulative contact time, with links with less than 2 h cumulative activity being omitted. Black lines represent the links recurring from the previous week, whereas the red lines are new links. b Top: histogram of contact events over a single day (semi-logarithmic plot). The coefficient of variation is $c_V=1.03$ and the mean is $\mu =131$. Bottom: histogram of contact events over a 7 week period, divided by the number of days to obtain an average daily rate (semi-logarithmic plot). Here, $c_V=0.95$ and $\mu =86$. Both plots show a marked heterogeneity, demonstrating that contact heterogeneity is approximately a quenched disorder on the timescale of a few weeks. c Our agent-based model of COVID-19 spreading on a contact network. Individuals in the susceptible state may be exposed by those in the presymptomatic as well as infected states. The exposed-presymptomatic triplet of states together comprise the gamma-distributed incubation period

Fig. 3

The effects of social heterogeneity on an unmitigated epidemic. The red curves show the incidence, measured as the sum of exposed and infectious individuals (whether symptomatic or not). The blue curves indicate the attack rate, i.e. the cumulative fraction of the population who have been exposed to the disease. In both cases, the curves correspond to the true, edge swapped and randomized networks, in order of increasing brightness. Each trajectory represents an average of 50 simulations

Fig. 4

The effects of social heterogeneity on contact tracing at different thresholds. Comparison of exposed + presymptomatic + infected (red) and recovered (blue) individuals in the three networks types. The testing rate is set at 0.5 times the rate for leaving the symptomatic infectious stage, giving a 25% probability of being tested while infected

Fig. 5

Contact tracing effectiveness. Disease parameters are identical to those of Fig. 3. a Rate of testing vs final size of epidemic and average number of days spent in quarantine per person. The contact threshold is set at 15 min. The rate of testing is measured in units of the rate for leaving the infected state, meaning that a rate of testing of 1 corresponds to a 50% chance of being tested during the infectious period. b Contact threshold vs final size of epidemic and average number of days spent in quarantine per person. The rate of testing is set at 0.5 times the rate for leaving the symptomatic infectious stage, giving a 25% probability of being tested while infected. For each value of the parameter, 50 simulations were run