Covasim: An agent-based model of COVID-19 dynamics and interventions

Abstract

The COVID-19 pandemic has created an urgent need for models that can project epidemic trends, explore intervention scenarios, and estimate resource needs. Here we describe the methodology of Covasim (COVID-19 Agent-based Simulator), an open-source model developed to help address these questions. Covasim includes country-specific demographic information on age structure and population size; realistic transmission networks in different social layers, including households, schools, workplaces, long-term care facilities, and communities; age-specific disease outcomes; and intrahost viral dynamics, including viral-load-based transmissibility. Covasim also supports an extensive set of interventions, including non-pharmaceutical interventions, such as physical distancing and protective equipment; pharmaceutical interventions, including vaccination; and testing interventions, such as symptomatic and asymptomatic testing, isolation, contact tracing, and quarantine. These interventions can incorporate the effects of delays, loss-to-follow-up, micro-targeting, and other factors. Implemented in pure Python, Covasim has been designed with equal emphasis on performance, ease of use, and flexibility: realistic and highly customized scenarios can be run on a standard laptop in under a minute. In collaboration with local health agencies and policymakers, Covasim has already been applied to examine epidemic dynamics and inform policy decisions in more than a dozen countries in Africa, Asia-Pacific, Europe, and North America.

Fig 1. Covasim model structure, including infection (exposure), disease progression, and final outcomes.

Yellow shading indicates that an individual is infectious and can transmit the disease to other susceptible agents. States with a dashed border are considered symptomatic with respect to symptomatic versus asymptomatic testing.

Fig 2. Example of within-host viral load dynamics in Covasim.

Each row shows a different agent in the model. Color indicates viral load, which typically peaks the day before or the day of symptom onset, before declining slowly.

Fig 3. Illustration of contact networks with multiple layers in Covasim.

(A) In reality, individual people move between household, school, workplace, and community contact layers during the day. (B) In the model (shown here with a population of 20 people, with age structure and household sizes based on Malawi data), these dynamic contacts are approximated as static average daily contacts between layers. Individuals have different numbers of connections (lines) and connection weights (line widths; default relative weights shown) for each layer.

Fig 4

Synthetic population networks for households (top), schools (middle), and workplaces (bottom). Age-specific contact matrices are shown on the left, while actual connectivity patterns for a 127-person subsample of a population of 10,000 individuals are shown on the right. All individuals are present in the household network, including some with no household connections. A subset of these individuals, including teachers, are present in the school network (circles); another subset is present in workplace networks (squares); some individuals are in neither school nor work networks (triangles).

Fig 5

Illustration of the standard object-oriented approach for implementing agent-based models (top), where each agent is a separate object, compared with the approach used in Covasim (bottom), where agents are represented as slices through a set of state arrays. Dots (…) represent omitted entries. In practice, each agent has several dozen states, and there are typically hundreds of thousands of agents.

Fig 6

Covasim performance in terms of processor usage (top) and memory usage (bottom), for the number of agents shown, simulated for 100 days. There is roughly linear scaling over three orders of magnitude of population size.

Fig 7. Covasim webapp user interface; screenshot taken from http://app.covasim.org.

Fig 8

(A) Illustrative example of a single run of a Covasim simulation. Interventions (described in the text) are shown as dashed vertical lines. (B) Full listing of the code for this simulation, including defining the parameters of the simulation (lines 4–11); defining the interventions (lines 14–23); and creating, running, and plotting the simulation (lines 26–28).

Fig 9

(A) Illustrative example of a scenario comparison using a simple custom intervention ("protecting the elderly", i.e. removing all transmission among people over age 70 after a certain date). (B) Full listing of the code for this simulation, showing the intervention definition (lines 3–6), and a compact way of creating the simulations, running them in parallel, and plotting them (lines 8–11).

Fig 10

Example transmission trees for a hypothetical population of 300 individuals with a single seed infection on day 1, with (A) no interventions, (B) testing only, and (C) testing plus contact tracing. Time is shown on the horizontal axis, with each tree representing approximately 90 days. The vertical size of each tree is proportional to the total number of infections.

Fig 11. Example calibration of Covasim to data from Seattle/King County, Washington, USA from January 27 to November 14 2020 (dashed line), with projections until December 31, including additional restrictions imposed on November 16.

(A) Number of daily COVID-19 tests, which are used as input data. (B) Calibration to the number of daily COVID-19 diagnoses. (C) Calibration to the number of daily contacts traced (weekly averages shown; data past prediction date are not available). (D) Calibration to the number of daily COVID-19 deaths. (E) Projections of the number of new infections if restrictions had not been implemented, with the restrictions as implemented, and if restrictions were implemented together with increases in testing and contact tracing. Bands show 80% forecast intervals; data are rolling 7-day averages to account for weekend reporting delays.