Methods

Data

We analyzed three sources of data:

1. Serology

In May 2020, the Andorran government carried out a voluntary serological screening for both IgG and IgM antibodies of COVID-19, in which 91% of the population participated (Royo-Cebrecos et al., 2020). The data were used to estimate the total number of people had been infected in each parish.

2. Spatio-temporal telecom data

Telecom data was provided by Andorra Telecom (AT). Since AT is the sole telecom provider in Andorra, the data covers all mobile subscribers in the country, unlike most telecom datasets where the market is fragmented. Each observation in the AT data includes a unique ID for the subscriber, a timestamp, the coordinates of the device at the time of the update, and nationality for the subscriber’s home network.

3. COVID-19 case reports

We used the official COVID-19 case reports for the country of Andorra, from the Repository by the Center for Systems Science and Engineering (CSSE) at Johns Hopkins University (Dong & Gardner, 2020).

Analysis

We computed a range of mobility metrics across the population on a daily basis between the beginning of March 2020 and the end of October 2020. These included variations of the following metrics:

• The numbers of trips being made, within each parish and between parishes.
• The numbers of people staying home each day
• The interaction potential: an aggregation of all the time spent by any pair of individuals in close proximity to one another (~50m). 

The time series of each mobility metric was compared to a proxy for transmission rate: the log growth in cases over 2 week intervals, normalised by the susceptible fraction of the population. The comparison was done over a range of lag times from -30 days (mobility behaviour leads infection rate by 30 days) to +30 days (mobility lags infection rate by 30 days).

Results and Discussion

As shown in the figure below, the various mobility metrics calculated showed broadly similar trends, dropping off sharply during March and remaining close to their minimum levels throughout April. 

The most dramatic change in mobility behaviours came immediately after the introduction of the lockdown over the weekend of March 14th and 15th (see figure). The metrics also show a gradual increase in mobility during May, corresponding with a gradual relaxation of policies, or adherence to those policies.

Figure: Spatial variation of “Interaction Potential” over the course of two days. Left: March 12th, just before introduction of lockdown. Right: March 19th, just after introduction of lockdown. Potential interactions are binned into H3 cells of resolution 11 (Uber, 2020). For privacy reasons, only cells with at least 100 potential interactions are shown.

When found that some of the mobility metrics we computed had more substantial correlations  with transmission rate than others. As shown in the figure below, the strength and direction of the correlations depended on the lag period. Four metrics (Indoor interaction potential, outdoor interaction potential, number of trips between parishes and number of people making trips between parishes) were highly correlated with transmission, leading by 18-21 days. This suggests that when the behaviors associated with these mobility metrics change, transmission rates respond about 18-21 days later. Three metrics (portion of people staying home, number of people staying home and total trips) were positively correlated with transmission with a lead of 2-3 weeks, and also moderately negatively correlated with transmission rate, with a lag of about 2-3 weeks. This suggests that (i) when these mobility behaviours increase, there is an increase in transmission rate 2-3 weeks later and (ii) when the transmission rate increases, there is a decrease in the mobility behaviours 2-3 weeks later.

These findings provide support for policies which aim to reduce transmission by encouraging people to remain distributed in their own home communities, thereby reducing both inter-community transmission and interaction potential. Furthermore, the events which draw the highest density of crowds should be identified and discouraged.

References

Dong, E., Du, H., & Gardner, L. (2020). An interactive web-based dashboard to track COVID-19 in real time. The Lancet infectious diseases, 20(5), 533-534.

Doorley, R.M., Berke, A., Noyman, A., Alonso, L.A., Ribo, J., Arroyo, V., Pons, M. & Larson, K. (2021). Mobility and COVID-19 in Andorra: Country-scale analysis of high-resolution mobility patterns and infection spread. medRxiv.

Royo-Cebrecos, C., Vilanova, D., López, J., Arroyo, V., Francisco, G., Pons, M., Carrasco, M.G., Piqué, J.M., Sanz, S., Dobaño, C. & García-Basteiro, A. (2020). Mass SARS-CoV-2 serological screening for the Principality of Andorra. Preprint: https://www.researchsquare.com/article/rs-119323/v1.

Uber (2020). Overview of the H3 Geospatial Indexing System. https://h3geo.org/docs/core-library/overview. Accessed:    2020-12-22