Skip to content

Simulating and analyzing Covid-19 transmission and hospital trends per region in Illinois.

License

Notifications You must be signed in to change notification settings

numalariamodeling/covid-chicago

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Modelling the COVID-19 pandemic in Illinois

This repository includes simulation model and analysis scripts for modelling the COVID-19 pandemic in Illinois per covidregion.

1. Model overview
2. Software used
3. Postprocess and analyse simulation outputs
4. Data sources
5. Model updates
6. Resources

1. Model overview

1.1 Compartmental model structure (emodl file)

A basic SEIR model was extended to include symptom status (asymptomatic, presymptomatic, mild and severe symptoms), hospitalization, critical illness, and deaths, while allowing to track detected and undetected cases separately. In the model, the susceptible population is exposed (infected) at a constant rate described by the transmission probability and contact rate with the infectious population. After latent period of few days, the exposed population becomes infectious and moves either to the asymptomatic or pre-symptomatic compartments. At the end of the incubation period pre-symptomatic population develops either mild or severe symptoms. Mild symptomatic cases recover at a similar rate as asymptomatic cases, while all severe symptomatic cases that ‘should’ need professional care move to the hospitalization compartment. Hospitalized cases either recover or develop critical illness and then recover, or die. Once recovered, we assume that the population stays immune throughout the simulation period. The asymptomatic presymptomatic, mild symptomatic, severe symptomatic infections, undetected hospitalized are the infectious compartments with reduced infectiousness for the detected sub-compartments due to self-isolation. We assume that hospitalized cases that are detected are not infectious. model Simulations run per Emergency Medical Service Area (EMS) and are aggregated for restore regions, and for Illinois. As of the 22nd of July, the 'covid regions' are used. For simplicity, the term 'EMS' is kept in the modelling files.

Show SEIR structure with vaccinations

When specying 'vaccine' as one of the intervention scenarios, the whole compartmental structure is mirrored for the vaccinated population. The vaccinated population is assumed to be less infectious and fewer infections are symptomatic (reducted fraction mild and severe symptoms). Whereas for the fraction of the vaccinated population that develops severe symptoms, the transition through the hospital stages is the same as for not-vaccinated population. Per default the fraction_severe is reduced by 100-95% (see parameter table).

model (Note, the post-ICU compartment not shown in flowshart but included in the emodl file)

1.2. Model parameters

Most of the parameters are derived from literature, local hospital data as well as doublechecked with other models used in Illinois (i.e. UChicago). The starting date, intervention effect size, and the transmission parameter "Ki"are fitted to death data.

Show parameter tables

1.2.1 'reaction paramaters'

All the parameters are sampled from a uniform distribution as specified in the experiment config (yaml) file

parameter name
Ki Transmission rate (contact rate * infection probability)
Ks Progression to presymtomatic ( fraction_symptomatic / time_to_infectious))
Kl Progression to asymptomatic ((1 - fraction_symptomatic ) / time_to_infectious))
dAs detection rate of asymptomatic infections
dSym detection rate of mild symptomatic infections
dSys Detection rate of severe symptomatic infections
Ksym Progression to mild symptoms
Ksys Progression to severe status ( fraction_severe * (1 / time_to_symptoms))
Kh Hospitalization rate
Kh_D Hospitalization rate minus delay in detection
Kr_a Recovery rate of asymptomatic infections
Kr_m Recovery rate mild symptomatic infections
Kr_m_D Recovery rate mild symptomatic infections minus delay in detection
Kr_h Recovery rate of hospitalized cases
Kr_c Recovery rate of critical cases (progression to H before discharge)
Kr_hc Recovery rate of critical cases in med/surg after ICU stay
Kc Progression to critical
Km Deaths
Kv_l Vaccinations (S -> S_V) for vaccine intervention model only

1.2.2 Transmission and disease parameters

Parameter Description Value Unit Source
Initial infectious population Number of infectious population that initiates the local transmission 10 N Assumed
Ki (transmission rate) Rate at which susceptible become infectious (contact rate * probability of infection). rate Fited to data from pre March 21 (EMResource and Line List data)
Ki multiplier (transmission rate multiplier) This parameter adjusts the initial transmission rate over time to reflect changes in mitigation policies, lockdowns and mask wearing as well as other factors that affect transmission. Fited to data every week using monthly time events for change in transmission.
Date of imported infection Date when local transmission started Feb 13 - Feb 27 date Fit to data from pre March 21 (EMResource and line list)
time_to_infectious Time from being exposed to become infectious (2.4 , 3.5) days Li et al 2020
time_to_symptomatic Time from becoming infectious to onset of symptoms (3.0, 4.5) days Li et al 2020 and Jing et al 2020
time_to_hospitalization Time from possible detection to hospitalization (3, 6) days Huang et al 2020
Time to critical Time between hospitalization and critical illness (4, 6) days NMH, Huang et al 2020
Time to death Time between critical illness and deaths (4, 6) days Yang et al 2020
Recovery time asymptomatic Time until an asymptomatic infection is cleared 9 days Cevik et al 2020
Recovery time mild symptomatic Time until mildly symptomatic case recovers 9 days Cevik et al 2020
Recovery time hospitalized Time until hospitalized cases (severe symptomatic) recover (4, 6) days NMH, Lewnard et al 2020 and Wang et al 2020
Recovery time critical (time varying) Time until critical cases (severe symptomatic) recover and return to H before discharge (8,10) Bi et al 2020
Recovery time postcrit Time until critical cases that stayed in med/surg after ICU stay are discharged (1,4) Assumed, informed by NMH data
Fraction symptomatic Fraction of infections that develop either mild or severe symptoms (0.5, 0.7) Oran and Topol et al 2020
Reduced fraction symptomatic Lower fraction of infections that develop either mild or severe symptoms (reduced for vaccinated population) (0.8, 1) Assumed
Fraction severe symptomatic Fraction of symptomatic that develop severe symptoms (0.02, 0.1) Salje et al 2020
Reduced fraction severe symptomatic Lower fraction of symptomatic that develop severe symptoms (reduced for vaccinated population) (0, 0.05) Assumed
Fraction critical (time varying) Fraction of severe symptomatic infections that require intensive care (0.2, 0.35) Lewnard et al 2020
CFR (time varying) Case fatality rate (0.01, 0.04) Wang et al 2020
Reduced infectiousness of detected cases Fraction of detected cases that isolate and are removed from the infectious population (0, 0.3) Assumed
Reduced infectiousness of asymptomatic cases Fraction of asymptomatic cases that are less infectious (currently disabled/not used) (1, 1) /
Reduced infectiousness of vaccinated cases Fraction of vaccinated cases that are less infectious (only used in vaccine intervention model) (0.1, 0.2) Assumed
Detection probability of asymptomatic case Used for contact tracing simulations, per default asymtomatic cases are not detected (0, 0) Assumed initial value, increase informed from Illinois specific data
Detection probability of mild symptomatic case (time varying) Initial value of the detection rate, which is increasing over time (0.05, 0.2) Assumed initial value, increase informed from Illinois specific data
Detection probability of severe symptomatic case (time varying) Initial value of the detection rate, which is increasing over time (0.2, 0.5) Calculated from IL data
Impact of transmission mitigation policies, lockdown and mask-wearing reflected in transmission rate parameter Fiteed to data and updated every week using monthly 'transmission changepoints'

Note: List also on Box!

1.2.3 time-varying parameters (intervention scenarios)

The time-event option in cms allows to change a paramter at a given time point (days) (which will stay constant afterwards if not resetted using a stop time-event). Time-event are used to define reduction in the transmission rate, reflecting a decrease in contact rates due to social distancing interventions (i.e. stay-at-home order). The time event can also be used to reflect increasing testing rates by increasing the detection of cases (i.e. dSym and dSys for increased testing at health facilities, or dAs and dSym for contact tracing)

Current scenarios include any combination of those listed below:

  • Baseline (continued mitigation)
  • Reopen (discontinued mitigation)
  • Rollback (itensified mitigation)
  • Triggered rollback (itensified mitigation based on threshold i.e. in critical cases)
  • Bvariant (increase in transmission rate and disease severity)
  • Vaccinations (splitting compartments into not-vaccinated vs vaccinated with substantially reduced infectiousness and severity for vaccinated population)

For details, click here here

1.3. Model outcomes

Currently the model includes 28 compartments of which 43 outcome variables are generated. The outcome variables or channels, as referred to in the py plotters, are different aggregates of the main compartments by type (i.e. all detected, all severe symptomatic) and includes cumulative counts for the calculation of incidences during postprocessing. A ranking 'observeLevel' was introduced to select subsets of the outcomes. The primary outcomes are those required for the weekly deliverables, the secondary are the related outcomes that are not required for the standard outputs and tertiary are those that can easily be calculated outside the model, such as prevalence, or outcomes rarely used such as infectiousness by symptomatic type and detection level. The outcomes indicated with a * were removed from the direct model outcomes, and are calculated in the postprocessing, daily counts are can be calculated from all 'cumulative' outcomes.

Show table

alphabet. No name description observeLevel compartments included
1 asymp_cumul Number of all asymptomatic infections that happened (cumulative) primary asymptomatic, RA, RAs_det1
2 asymp_det_cumul Number of all detected asymptomatic infections that happened (cumulative) secondary As_det1, RAs_det1
3 asymptomatic Number of asymptomatic infections secondary As
4 asymptomatic_det Number of detected asymptomatic infections secondary As_det1
5 crit_cumul Number of all critical cases that happened (cumulative) primary deaths, critical, RC2,RC2_det3
6 crit_det Number of critical cases that are detected primary C2_det3, C3_det3
7 crit_det_cumul Number of all detected critical cases that happened (cumulative) primary C2_det3, C3_det3, D3_det3, RC2_det3
8 critical Number of severe symptomatic infections that are hospitalized that are critical primary C2, C3, C2_det3, C3_det3
9 death Number of COVID-19 deaths in the population primary D3, D3_det3
11 death_det_cumul Number of all detected COVID-19 deaths in the population that happened primary D3_det3
12 detected Number of detected COVID-19 infections regardless of symptomaticity primary As_det1, Sym_det2, Sys_det3, H1_det3, H2_det3, H3_det3, C2_det3, C3_det3
13 detected_cumul Number of all detected COVID-19 infections regardless of symptomaticity (cumulative) primary As_det1, Sym_det2, Sys_det3, H1_det3, H2_det3, C2_det3, C3_det3, RAs_det1, RSym_det2, RH1_det3, RC2_det3, D3_det3
14 exposed Number of exposed (infected not yet infectious) in the population secondary E
15 hosp_cumul Number of all hospitalizations due to COVID-19 that happened (cumulative) primary hospitalized, critical, deaths, RH1, RC2, RH1_det3, RC2_det3
16 hosp_det Number of detected COVID-19 hospitalizations primary H1_det3, H2_det3, H3_det3
17 hosp_det_cumul Number of all detected COVID-19 hospitalizations that happened (cumulative) primary H1_det3, H2_det3, H3_det3, C2_det3, C3_det3, D3_det3, RH1_det3, RC2_det3
18 hospitalized Number of severe symptomatic infections that are hospitalized primary H1, H2, H3, H1_det3, H2_det3, H3_det3
19 infected Number of all infected in the population primary all exposed, pre-symptomatic, symptomatic, hospitalized, and critical (N- (susceptible + deaths + recoveries))
20 infected_cumul Number of all that were infected (cumulative) secondary infected, recovered, deaths
21 infected_det Number of all infected that are detected secondary infectious_det, H1_det3, H2_det3, H3_det3, C2_det3, C3_det3
23 infectious_det Number of all infectious that are detected tertiary As_det1, P_det , Sym_det2, Sys_det3
24 infectious_det_AsP Number of all non-symptomatic that are infectious and detected tertiary As_det1, P_det
25 infectious_det_symp Number of all symptomatic that are infectious and detected tertiary Sym_det2, Sys_det3
26 infectious_undet Number of infectious that are not detected tertiary As, P, Sym, Sys, H1, H2, H3, C2, C3
27 presymptomatic Number of presymptomatic infections primary P, Pdet
28 presymptomatic_det Number of all detected presymptomatic infections secondary Pdet
29 prevalence* Number of infected (cumul) over total population tertiary infected / N
30 prevalence_det* Number of detected infected (cumul) over total population tertiary infected_det / N
31 recovered Number of recovered COVID-19 cases in the population primary RAs,RSym, RH1, RC2, RAs_det1, RSym_det2, RH1_det3, RC2_det3
32 recovered_det Number of detected recovered COVID-19 cases in the population primary RAs_det1, RSym_det2, RH1_det3, RC2_det3
33 seroprevalence* Number of recovered (cumul) over total population tertiary (infected + recovered) / N
34 seroprevalence_det* Number of detected recovered (cumul) over total population tertiary (infected_det + recovered_det) / N
35 susceptible Number of susceptibles in the population primary S
36 symp_mild_cumul Number of all mild symptomatic infections that happened (cumulative) primary symptomatic_mild, RSym, RSym_det2
37 symp_mild_det_cumul Number of all detected mild symptomatic infections that happened (cumulative) primary symptomatic_mild_det, RSym_det2
38 symp_severe_cumul Number of all severe symptomatic infections that happened (cumulative) primary symptomatic_severe, hospitalized, critical, deaths, RH1, RC2, RH1_det3, RC2_det3
39 symp_severe_det_cumul Number of all detected severe symptomatic infections that happened (cumulative) primary symptomatic_severe_det, hosp_det, crit_det, deaths_det
40 symptomatic_mild Number of mild symptomatic infections primary Sym, Sym_det2; Sym, Sym_preD, Sym_det2 ; Sym, Sym_preD, Sym_det2a, Sym_det2b
41 symptomatic_mild_det Number of detected mild infections in the population secondary symptomatic_mild_det
42 symptomatic_severe Number of severe symptomatic infections secondary Sys, Sys_det3; Sys, Sys_preD, Sys_det3 ; Sys, Sys_preD, Sys_det3a, Sys_det3b
43 symptomatic_severe_det Number of detected severe symptomatic infections secondary symptomatic_severe_det

Notes:

  • when using the vaccination scenario, these outcome channels include both vaccinated and not vaccinated compartments, whereas additional (same) outcomes are generated for the vaccinated population, denoted with suffix _V.
  • prevalence, seroprevalence and infection fatality ratio calculated using calculate_prevalence, example use in plot_prevalence.py
  • daily new counts are calculated from the cumulative outcomes using calculate_incidence, this function is integrated into the load_sim_data function when reading in trajectoriesDat.csv
  • look up how these are defined in the emodl here

1.4. Age and spatial model structures

1.4.1. Age-structured model

The "age_model" duplicates each compartment of the simple or the extended model for n age groups. To allow the age groups to get in contact with each other at different rates, the Ki (contact rate * probability of transmission) needs to be specified for a all age-combinations.

Contact matrix

The contacts between age groups were previously extracted for running an EMOD model from Prem et al 2017. Script that extracts the contact matrix values.

1.4.2. Spatial model

The "spatial_model" uses a special syntax as described here. To generate or modify the emodl files use the locale specific emmodl generator

1.4.3. Spatial age-structured model

A test verion is available under emodl file. To generate or modify the emodl files use the locale-age specific emmodl generator

2. Software used

The Compartmental Modeling Software (CMS) is used to simulate the COVID-19 transmission and disease progression. The CMS language defines 5 main type: species, observations, reactions, parameters and functions, in addition time-events can be added as well as state-events. Multiple compartments, called ‘species’ can be defined. The movement of populations between compartments is called reaction. The model runs with different solvers, including spatial solvers. The model is written in 'emodl' files and model configurations are written in 'cfg' files. The output is written into trajectories.csv files.

The latest model description in emodl format is written in the extendedmodel.emodl file (note original emodl with history in extendedmodel_cobey.emodl.

2.1 Run simulations

The runScenarios.py is used to run multiple simulations given a configuration file with the parameters. The script builds off a default configuration file extendedcobey.yaml and substitutes parameters with the values/functions in the user-provided configuration file using the @param@ placeholder. Multiple trajectories.csv that are produced per single simulation are combined into a trajectoriesDat.csv, used for postprocessing and plotting.

Show runScenarios examples

Base model
  • python runScenarios.py --model "base" -r EMS_1 --scenario "baseline" -n "userinitials"
  • python runScenarios.py --model "base" -r EMS_2 --scenario "baseline" -n "userinitials"
  • python runScenarios.py --model "base" -r EMS_3 --scenario "baseline" -n "userinitials"
  • python runScenarios.py --model "base" -r EMS_4 --scenario "baseline" -n "userinitials"
  • python runScenarios.py --model "base" -r EMS_5 --scenario "baseline" -n "userinitials"
  • python runScenarios.py --model "base" -r EMS_6 --scenario "baseline" -n "userinitials"
  • python runScenarios.py --model "base" -r EMS_7 --scenario "baseline" -n "userinitials"
  • python runScenarios.py --model "base" -r EMS_8 --scenario "baseline" -n "userinitials"
  • python runScenarios.py --model "base" -r EMS_9 --scenario "baseline" -n "userinitials"
  • python runScenarios.py --model "base" -r EMS_10 --scenario "baseline" -n "userinitials"
  • python runScenarios.py --model "base" -r EMS_11 --scenario "baseline" -n "userinitials" Note: the base model structure is not being updated and to run single regions it is recommended to use the locale model as described below.
Locale model
  • python runScenarios.py --model "locale" -r IL --scenario "baseline" -n "userinitials"
  • python runScenarios.py --model "locale" -r IL -sr EMS_1 --scenario "baseline" -n "userinitials"
  • python runScenarios.py --model "locale" -r IL -sr EMS_1 EMS_5 EMS_11 --scenario "baseline" -n "userinitials" Note: -sr can take any number or combination of the 11 subregions in the format of EMS_1 to EMS_11
Age model
  • python runScenarios.py --model "age" -r EMS_1 --scenario "baseline" -n "userinitials"
  • python runScenarios.py --model "age" -r EMS_2 --scenario "baseline" -n "userinitials"
  • python runScenarios.py --model "age" -r EMS_3 --scenario "baseline" -n "userinitials"
  • python runScenarios.py --model "age" -r EMS_4 --scenario "baseline" -n "userinitials"
  • python runScenarios.py --model "age" -r EMS_5 --scenario "baseline" -n "userinitials"
  • python runScenarios.py --model "age" -r EMS_6 --scenario "baseline" -n "userinitials"
  • python runScenarios.py --model "age" -r EMS_7 --scenario "baseline" -n "userinitials"
  • python runScenarios.py --model "age" -r EMS_8 --scenario "baseline" -n "userinitials"
  • python runScenarios.py --model "age" -r EMS_9 --scenario "baseline" -n "userinitials"
  • python runScenarios.py --model "age" -r EMS_10 --scenario "baseline" -n "userinitials"
  • python runScenarios.py --model "age" -r EMS_11 --scenario "baseline" -n "userinitials" Note: This model is not maintained and will be integrated into the locale model.
Age-locale model (testing)
  • python runScenarios.py --model "agelocale" -r IL --scenario "baseline" -n "userinitials"
NU model
  • python runScenarios.py --model "nu" -r NU -c "nu_undergrad.yaml" -e "nu_undergrad.emodl" -n "userinitials"

The examples above show an abbreviated version, accepting most defaults. The table below shows all available options and their defaults.

runScenarios options

The minimum command is python runScenarios.py. If no arguments are specified, per default a 'baseline' simulation runs for all 11 regions in Illinois and the file name will include a random number for unique identification. The table below shows an overview of all arguments currently enabled.

Show runScenarios options

no argument long argument short model specific required help choices default
1 --masterconfig -mc FALSE FALSE Master yaml file that includes all model parameters. "extendedcobey_200428.yaml"
2 --running_location -rl FALSE FALSE Location where the simulation is being run. If None (not provided) the script tries to determine running location based system variables "Local", "NUCLUSTER" "None"
3 --region -r FALSE FALSE Region for which to run simulation. E.g. 'IL' 'IL','EMS_1', 'EMS_2', 'EMS_3', 'EMS_4', 'EMS_5', 'EMS_6', 'EMS_7', 'EMS_8', 'EMS_9', 'EMS_10','EMS_11','NU' "IL"
4 --subregion -sr TRUE FALSE Subregion for which to run simulation. any combination of EMS_1 to EMS_11 i.e. 'EMS_11' , 'EMS_1' 'EMS_2' , 'EMS_1' 'EMS_5' 'EMS_11' ['EMS_1', 'EMS_2', 'EMS_3', 'EMS_4', 'EMS_5', 'EMS_6', 'EMS_7', 'EMS_8', 'EMS_9', 'EMS_10','EMS_11']
5 --experiment_config -c FALSE FALSE Config file (in YAML) containing the parameters to override the default config. This file should have the same structure as the default config. example: ./experiment_configs/sample_experiment.yaml If not provided, the default experiment_config is selected based on model specification "None"
6 --emodl_template -e FALSE FALSE Template emodl file to use. If not provided, the emodl_template is generated based on model AND scenario specification. If no scenario specification is given it uses the baseline scenario! "None"
7 --model -m TRUE FALSE Model type (see choices) "base", "locale","age","agelocale","nu" "locale"
8 --scenario -s DEPENDS FALSE Intervention scenario to use. Might differ for locale and other models. 'Any combination of "baseline", "rollback","triggeredrollback", "reopen","bvariant", "vaccine"' (details here "baseline"
9 --paramdistribution -dis TRUE FALSE Use parameter ranges or means (could be extended to specify shape of distribution) (used only for locale/spatial model) "uniform_range", "uniform_mean" "uniform_range"
10 --cfg_template -cfg FALSE FALSE Template cfg file to use. For more details visit https://docs.idmod.org/projects/cms/en/latest/solvers.html "model_B.cfg", "model_Tau.cfg", "model_RLeapingFast.cfg", "model_RLeaping.cfg","model_FD.cfg","model_DFSP.cfg","model_SSA.cfg" "model_B.cfg"
11 --name_suffix -n FALSE FALSE Adding custom suffix to the experiment name. If not specified, a random number will be used f"_test_rn{str(today.microsecond)[-2:]}"
12 --post_process -p DEPENDS FALSE Whether or not to run post-processing. Note default on NUCLUSTER vs Local varies "dataComparison", "processForCivis" "None"
13 --sample_csv -csv FALSE FALSE Name of sampled_parameters.csv, any input csv will be renamed per default to 'sampled_parameters.csv' "None"
14 --observeLevel -obs FALSE FALSE Specifies which outcome channels to simulate and return in trajectoriesDat.csv "primary", "secondary", "tertiary" "secondary"
15 --expandModel -expand FALSE FALSE Specific for test delay, defines where to allow test delay (As,Sym, Sys) "uniformtestDelay", "testDelay_SymSys", "testDelay_AsSymSys" "testDelay_AsSymSys"
16 --trigger_channel -trigger TRUE FALSE Specific channel name of trigger to use (used only for locale/spatial model) "None", "critical", "crit_det", "hospitalized", "hosp_det" "None"
17 --fit_params -fit TRUE FALSE Name of parameters to fit (testing stage, currently supports only single ki multipliers), ki_multiplier_4 to ki_multiplier_13 (currently supports only 1 at a time) "None"

The configuration file is in YAML format and is divided into 5 blocks: experiment_setup_parameters, fixed_parameters_region_specific, fixed_parameters_global, sampled_parameters, fitted_parameters. The sampled parameters need the sampling function as well as the arguments to pass into that function (function_kwargs). Currently, only a few sampling/calculation functions are supported. More can be added by allowing for more libraries in generateParameterSamples of runScenarios.py.

Note that the user-supplied configuration file is used to provide additional or updated parameters from the base configuration file.

2.3 Inputs:

  • Master configuration: YAML file that defines the parameter input values for the model (if not specified uses the default extendedcobey_200428.yaml)
  • Running location: Where the simulation is being run (either Local or NUCLUSTER)
  • Region: The region of interest. (e.g. EMS_1, or IL for all EMS 1-11 inclued in the same model)
  • Configuration file: The configuration file with the parameters to use for the simulation. If a parameter is not provided, the value in the default configuration will be used. (e.g. sample_experiment.yaml)
  • Emodl template (optional): The template emodl file to substitute in parameter values. The default is extendedmodel.emodl. emodl files are in the ./emodl directory.
  • cfg template (optional): The default cfg file uses the Tau leaping solver (recommended B solver).
  • Suffix for experiment name added as name_suffix (optional): The template emodl file to substitute in parameter values. The default is test_randomnumber (e.g. 20200417_EMS_10_test_rn29)

Region specific sample parameters (i.e. using estimates parameters per regions)

Age extension and age specific parameters

Sampled parameters

As described in 2.1. and 2.2 parameters are sampled from the base configuration files when running python runScenarios.py. The sample_parameters.py script allows to: (1) generate csv file from configuration files without running simulations (2) load and modify an existing sampled_parameters.csv (change or add single or multiple parameter) (default location experiment_configs\input_csv) (3) replace single values for one or more parameter use python dictionary --param_dic {\"capacity_multiplier\":\"0.5\"} (4) combine with multiple values for one or more parameters define additional csv file --csv_name_combo startdate_Ki_sets.csv (5) combine with multiple values for one or more parameters defined according to YAML --yaml_name_combo example.yaml

Show examples

Running examples:

  • nsamples: optional, if specified if overwrites the nsamples in the base configuration, if loading an existing csv the first n samples will be selected (i.e. when selecting samples from an excisting csv file, could be modified to be random if needed)
  • emodl_template: the emodl template is required to test whether the parameter csv table includes all required parameters defined in the desired emodl file to run
  • example 1: python sample_parameters.py -rl Local -r IL --model locale --experiment_config spatial_EMS_experiment.yaml --emodl_template extendedmodel_EMS.emodl -save sampled_parameters2.csv
  • example 2: python sample_parameters.py -rl Local --model locale -save sampled_parameters_1000.csv --nsamples 1000
  • example 3: python sample_parameters.py -rl Local --model locale -load sampled_parameters_1000.csv -save sampled_parameters_1000_v2.csv --param_dic {\"capacity_multiplier\":\"0.5\"}
  • example 4: python sample_parameters.py --model locale --csv_name_combo sampled_parameters_sm7.csv -save sampled_parameters_sm7_combo.csv -(sampled_parameters_sm7.csv not under version control, but would for example include 10 values for social multiplier 7 for all 11 regions, the base sample parameters are repeated for each of the 10 rows of the additional csv)
  • example 5: python sample_parameters.py -e "example.emodl" -load "example_csv_base.csv" -yaml "samp_params_combos_example.yaml"
    • See the example yaml file (default directory is ./experiment_config/) for how to specify the additional parameters.

When running simulations with an pre-existing csv file, specify

  • --sample_csv (name of csv file in experiment_configs\input_csv ). Note: the specified csv will per default be renamed to "sampled_parameters.csv", hence the "sampled_parameters.csv" in input_csv is overwritten each time. A copy of the sample parameters can be retrieved from the simulation folder.

2.6. Setup

Running simulation requires the CMS software and python. The model runs on Windows and Linux and on the computing cluster at Northwestern University ('Quest'). The detailes are described below.

Show setup description

Software requirements and packages

The requirements.txt includes name and version of required python modules.

Local environment setup

Use a .env file in the same directory as your runScenarios script to define paths to directories and files on your own computer. Copy the sample.env file to .env and edit so that paths point as needed for your system.

Running on OS X or Linux

The CMS software is provided as a compiled Windows executable, but can be run on Unix-like systems via wine. If you do not have wine installed on your system, you can use the provided Dockerfile, which has wine baked in. To build the Docker image, run docker build -t cms. Set DOCKER_IMAGE=cms in your environment or your .env file to use it.

Running on Quest (NUCLUSTER)

Information related to running on quest can be found in the readme in nucluster

3 Postprocessing and visualization of simulation outputs

Via the --post_process argument in the runScenarios command additional scripts will run directly after simulations finished. Batch files are generated for data comparison, process for civis steps and basic plotter (age, locale emodl regardless of the --post_process argument. Batch files are only generated for the most important postprocessing files and additional batch files or postprocesses may be linked to runScenarios of needed. To see all the available postprocessing scripts, go to the plotters folder (see readme in folder for details). When adding the flag --post_process "processForCivis" in the runScenarios.py submission command, the batch files related to the weekly deliverables are automatically executed.

Postprocessing

Locally, there are two main batch files:

  • run_postprocess.bat is a wrapper batch file that runs postprocessing files from the list below (general postprocessing not for weekly deliverables)
  • run_postprocess_for_civis.bat is a wrapper batch file that runs postprocessing from the list below (postprocessing for weekly deliverables)

For postprocessing on Quest, please read the readme in nucluster

The postprocessing includes the following steps below:

Show postprocessing scripts

  • 0_runCombineAndTrimTrajectories.bat calls combine_and_trim.py combines and trims the simulation output csv files (trajectories.csv files)
  • 0_locale_age_postprocessing.bat calls locale_age_postprocessing.py to plot trajectories for pre-specified outcome channels per age group.
  • 1_runTraceSelection.bat calls trace_selection.py calculating the negative log-likelihood per simulated trajectory, used for thinning predictions and parameter estimation (- 1_runSimulateTraces.bat calls simulate_traces.py extracts fitting parameters,identified as parameters that vary (needs sample parameters to be fixed), and produces besttrace.csv and best_ntraces.csv, optionally starts a follow up simulations.)
  • 2_runDataComparison.bat calls data_comparison_spatial.py comparing model predictions to data per region over time
  • 3_runProcessTrajectories.bat calls process_for_civis_EMSgrp.py generates the result csv dataframe (i.e. nu_20201005.csv) and generates descriptive trajectories per channel and region
  • 4_runRtEstimation.bat calls estimate_Rt_forCivisOutputs.py that runs the Rt estimation, the Rt columns are added to the result csv dataframe (i.e. nu_20201005.csv), produces descriptive plots
  • 5_runOverflowProbabilities.bat calls overflow_probabilities.py calculates the probability of hospital overflow and produces the (i.e. nu_hospitaloverflow_20201005.csv), also adds total number of beds additional script
  • 6_runPrevalenceIFR.bat calls plot_prevalence.py
  • 7_runICUnonICU.bat calls plot_by_param_ICU_nonICU.py
  • 8_runHospICUDeathsForecast.bat calls hosp_icu_deaths_forecast_plotter.py
  • 9_runCopyDeliverables.bat calls NUcivis_filecopy.py that generates the NU_civis_outputs subfolder and copies all relevant files and adds the changelog.txt. Note: the changelog.txt will need manual editing to reflect new changes per week.
  • 10_runIterationComparison.bat calls iteration_comparison.py that generates the iteration comparison plot (last 3 weeks)
  • 11_runCleanUpAndZip.bat calls cleanup_and_zip_simFiles.py to clean up simulations (deletes per default single trajectories !! and optinally zips and/or deletes simulation folder

ATTENTION-1: if 1_runTraceSelection.bat was run and the output csv files are located in the experiment folder, all subsequent scripts and plotting scripts will per default filter the simulated trajectories, if not explicily set to False in the load_sim_data function call. ATTENTION-2: 1_runSimulateTraces.bat only needed for parameter estimation!

4 Data sources

5. Model updates

Updates in model structure and fitted parameters

The model is updated every week to fit to latest hospitalisation and deaths reports.

Show history of updates

  • 20210527 added ki multiplier 17 for 20210507, updated parameter fit, updated postprocessing_combine plotter with regions' best sims, corrected bug in postprocessing_combine plotter
  • 20210513 updated parameter fit, updated postprocessing_combine plotter with regions' best sims
  • 20210429 updated parameter fit, added 2nd ICU recovery parameter (recovery_time_crit_change2 and recovery_time_crit_change_time_2)
  • 20210415 added tranmission multiplier 16 for April, updated weekly fit, reactivated bvariant (now starting at 20210410)
  • 20210402 deactivated bvariant due to bug, updated parameter fit, updated plotter/nuciviscopy to include vaccine and bvariant descriptions,
  • 20210317 added ki multiplier 15 to intervention emodl yaml, set bvariant_start date after ki multiplier 15, updated scaling factor
  • 20210316 added transmission multiplier 15 for March, included bvariant, included vaccinations, updated parameter fit
  • 20210303 updated parameter fit
  • 20210217 added transmission multiplier 14 for February, added range for initial transmission rate
  • 20210210 model update: added pre-/post ICU compartments, added reduced infectiousness for asymptomatic infections, updated time-varying IL specific parameters
  • 20210204 workflow update: attached emodl generation to runScenarios.py (optional)
  • 20210203 updated parameter fit, recovery_As set to 9
  • 20210125 added ki_multiplier_12 and ki_multiplier_13 to emodl_generator_base, emodl_generator_age, emodl_generator_age_locale & ran 3 emodls
  • 20210120 updated parameter fit, added transmission multiplier 13, fixed bug in region populations, adjusted ki_multiplier_3a and ki_multiplier_3b to use means rather than range, adjusted region-specific recovery_time_crit_change1 and recovery_time_crit_change_time_1)
  • 20210114 updated parameter fit, activated transmission multiplier 12 for Dec, updated region populations
  • 20210106 added region specific recovery time critical
  • 20201216 updated parameter fit
  • 20201204 updated parameter fit, use multiplier 11 for decrease in trend
  • 20201201 updated parameter fit
  • 20201130 added transmission multiplier 12
  • 20201124 updated parameter fit
  • 20201119 updated parameter fit, added transmission multiplier 11
  • 20201110 updated parameter fit
  • 20201104 updated parameter fit
  • 20201027 updated parameter fit, added transmission multiplier 10 (previously called social multiplier)
  • 20201020 updated parameter fit
  • 20201015 updated parameter fit, reset fitting method
  • 20201007 updated parameter fit
  • 20200929 updated parameter fit, changed fitting method
  • 20200922 updated parameter fit, changed d_Sym parameters (generic)
  • 20200915 updated parameter fit, added social multiplier 7 (time event Aug 25)
  • 20200909 updated parameter fit, updated evolution of CFR
  • 20200825 updated parameter fit
  • 20200818 updated parameter fit, updated evolution of dSys and region-specific evolution of dSym
  • 20200812 updated parameter fit
  • 20200807 updated parameter fit
  • 20200804 updated parameter fit
  • 20200729 updated parameter fit, added region-specific evolution of dSym over time
  • 20200722 updated parameter fit, use covid regions instead of EMS regions for fitting (same numbering 1-11)
  • 20200715 updated parameter fit, added fifth social distancing multiplier (time event June 21st)
  • 20200706 added time-varying fraction_critical
  • 20200624 updated parameter fit
  • 20200622 adjusted increase in detection for severe and mild symptomatic cases
  • 20200622 updated model structure, added test delay in Asymptomatics and detections in presymptomatic
  • 20200616 updated parameter fit
  • 20200610 updated parameter fit
  • 20200609 separat time delay for dSym and dSys, added d_Sym_incr 1-5 proportional to d_Sys_incr
  • 20200602 updated parameter fit
  • 20200523 added d_Sys_incr4 and d_Sys_incr5, parameter fitting, including test delay per default
  • 20200521 added s_m_4, parameter fitting
  • 20200515 parameter fitting (also 20200512, 20200501)
  • 20200428 updated model disease and transmission parameters (previously 20200421, 20200419)
  • 20200428 added d_Sys_incr1-3
  • 20200421 adding scale-invariant Ki
  • 20200407 add more detected observables
  • 20200402 cobey model alignment (including presymptomatic)
  • 20200321 initial model development including (S,E, Sym, Sys, As, H, C, D, R)

6. Resources and publications

About

Simulating and analyzing Covid-19 transmission and hospital trends per region in Illinois.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages