haoxiangyang89/COVID_Staged_Alert

Codes and data for "Design of COVID-19 Staged Alert Systems to Ensure Healthcare Capacity with Minimal Closures"

Overview

This software supports the results in the paper: https://www.medrxiv.org/content/10.1101/2020.11.26.20152520v1

The algorithm simulates a COVID-19 epidemic for Austin, TX, and determines triggers to enact social distancing orders to avoid exceeding hospital capacity. Chosen triggers attempt to minimize the total number of days that a city is in strict levels.

All codes are tested on MacOS but should work on Windows and Linux.

Installation and running the code

• Download and unzip the code to a local path (e.g., .../COVID_Staged_Alert)
• Add both /COVID_Staged_Alert and /COVID_Staged_Alert/InterventionsMIP to your $PYTHONPATH
• The following packages are required:

• matplotlib
• pandas
• numpy
• scipy

• The installation should take less than a minute.

Guidelines for contributing

• Create new branches to test new features
• Create a pull request to merge with master

Structure of the code

threshold_policy.py

• Main module to launch the search
• Functions to execute the search and find optimized thresholds to enact lock-downs.
• Iterators for traing and testing
• Calendar generation (ad-hoc for Austin instance)

policy_search_functions.py

• Main module to run the search
• Functions to perform chance-constrained check and select the optimal policies w/o ACS setup

SEIYHARD_sim.py:

• Simulator engine
• Parallelizarion functions
• Calander utils class (SimCalendar)

epi_parameters.py

• Class EpiSetup characterizes the simulation and recompute contact matrices as needed.

intervention.py

• Class intervention defines its properties and used in the simulator.
• Helper function to create multiple interventions

policies.py

• Class MultiTierPolicy and MultiTierPolicy_ACS defines policy given the thresholds and will have functions to obtain corresponding tier given trigger statistics
• Functions to build a list of trial policies

objective_functions.py

• Construct the objective function for a simulation path

utils.py

• Timing function
• Rounding functions

intances/init.py

• Module summarizing the input of the simulator.
• Creates an instance of EpiSetup

Structure of the test scripts

main_least_squares.py

• Least-squares fit

Related input files:

instances/austin/

• austin_real_icu_lsq.csv
• austin_real_hosp_lsq.csv
• austin_hos_ad_lsq.csv
• transmission_Final_lsq.csv
• setup_data_Final_lsq.json

instances/houston/

• houston_real_icu_lsq.csv
• houston_real_hosp_lsq.csv
• transmission_Final_lsq.csv
• setup_data_Final_lsq.json

main_downsampling.py

• Donwsampling procedure (algorithm 1)
• Call downsampling.py

Related input files:

instances/austin/

• tiers5_ds_Final.json
• austin_test_IHT_ds.json

main_multitier.py

• script to obtain optimal trigger policy
• Call policy_search_functions.py

Related input files:

instances/austin/

• tiers5_opt_Final.json
• transmission_Final.csv

main_multitier_acs.py

• script to obtain optimal trigger policy and ACS setup threshold
• Call policy_search_functions.py

instances/austin/

• tiers5_acs_Final.json
• transmission_Final_orange25_yellow75.csv

Structure of the reporting scripts

plotting.py

• Plot the figures used in the manuscript

• ICU, general wards heads-in-beds
• Total daily admission
• Stacked plots

output_processors.py

• Obtain statistics based on simulation results in .p file

report_pdf.py

• generate pdf file by filling in statistics in the .tex template and compile

Demo

Run the bash file test.sh in Demo folder to obtain an optimal trigger and generate visualization and a report

• The .p file (data file) will be generated in /output
• The plots will be generated under /plots
• The report will be generated under /reporting/reports
• The run time should be within a minutes