metadata-NotreDame-FRED.txt

team_name: NotreDame-FRED
model_name: NotreDame-FRED
model_abbr: NotreDame-FRED
model_contributors: 'guido espana <guido.espana@nd.edu>,  rachel oidtman,  sean cavany,  alan
  costello,  anneliese wieler,  anita lerch,  carly barbera,  marya poterek,  quan
  tran,  sean moore,  alex perkins <taperkins@nd.edu>  '
website_url: https://github.com/confunguido/covid19_ND_forecasting
license: cc-by-4.0
team_model_designation: secondary
ensemble_of_hub_models: false
methods: FRED is an agent-based model developed for influenza with parameters modified
  to represent the natural history of COVID-19
team_funding: GE and TAP were supported by a RAPID grant from the National Science
  Foundation (DEB 2027718).
data_inputs: NYTimes Daily reported deaths count
methods_long: "We used an existing agent-based model, FRED (Framework for Reconstructing\
  \ Epidemic Dynamics) which was developed by the university of Pittsburgh in response\
  \ to the 2009 influenza pandemic (Grefenstette et al. 2013). We modified the model\
  \ parameters to represent the natural history of COVID-19 and calibrated a set of\
  \ parameters to reproduce current trends of deaths due to COVID-19 in the US. FRED\
  \ simulates the spread of the virus in a population by recreating interactions among\
  \ people on a daily basis. To accurately represent the population of each of the\
  \ states simulated, we used a synthetic population of the US that represents demographic\
  \ and geographic characteristics of the population in 2010 (Wheaton 2012). Each\
  \ human is modeled as an agent that visits a set of places defined by their activity\
  \ space. This activity space contains places such as households, schools, workplaces,\
  \ or neighborhood locations. Transmission of SARS-CoV-2 can occur when an infected\
  \ person visits the same location as a susceptible person on the same day. Numbers\
  \ of contacts per person are specific to each place type. For instance, school contacts\
  \ do not depend on the size of the school but the age of the agent. Infected agents\
  \ have a probability to stay at home if they develop symptoms. Those who do not\
  \ stay at home continue their daily activities. Public health interventions are\
  \ included in the model to represent the changes in agents\u2019 behavior in response\
  \ to an epidemic.  In this study, we limited the interventions to school closure\
  \ and shelter in place. Schools are closed on a specific date to represent state-level\
  \ guidelines (IHME COVID-19 health service utilization forecasting Team and Murray\
  \ 2020). In the case that schools are closed, students limit their activity space\
  \ to household or neighborhood locations. Shelter-in-place interventions reduce\
  \ each agent\u2019s activity space to their household at a compliance level from\
  \ 0-100%, which was estimated as part of the model calibration. Agents who do not\
  \ comply with the shelter-in-place orders continue with their daily routines. We\
  \ used state-level orders to determine the date at which people are advised to shelter\
  \ in place (IHME COVID-19 health service utilization forecasting Team and Murray\
  \ 2020).  See the methods file in our repo for more detailed methods and model limitations.\n\
  Updates 2020-05-11 - We updated our likelihood used in our calibration, switching\
  \ from a Poisson to a negative binomial distribution in order to allow for greater\
  \ over-dispersion in the distribution of deaths. - We changed the way that we calculate\
  \ the proportion of the population that is sheltering in place. We no longer use\
  \ the dates at which government advisories were issued. Instead, we now exclusively\
  \ use the google mobility data (https://www.google.com/covid19/mobility/) on journeys\
  \ made to residential locations. We calibrate the maximum compliance to shelter\
  \ in place, and use a generalized additive model fitted to the google data to forecast\
  \ movement in the coming 5 weeks. For more details on this approach, see the methods\
  \ file in our repo."