Evaluating vaccine allocation strategies using simulation-assisted causal modelling

Installation

The infection dynamics simulation code is included as a submodule. Clone this repository and initialize submodules with

git clone git@github.com:akekic/covid-vaccine-evaluation.git
cd covid-vaccine-evaluation
git submodule init
git submodule update 

Install the submodule with

pip install -e ./covid-vaccine-dynamics

Install the main package with

pip install -e .

Convert all notebook files into .ipynb

jupytext notebooks/*.py --to .ipynb
jupytext data_preprocessing/*.py --to .ipynb

The requirements.txt file contains all the required packages without pinned versions. One tested set of versions is (including requirements of the submodule):

arviz==0.12.1
cachetools==4.2.4
CairoSVG==2.5.2
ConfigArgParse==1.5.3
covid19-inference==0.3.6
filelock==3.8.0
jupytext==1.14.1
matplotlib==3.5.3
notebook==6.4.12
numpy==1.21.6 
numpyro==0.10.1
pandas==1.4.4
pymc3==3.11.2
pytest==7.1.3
requests==2.28.1
statsmodels==0.13.2
tabulate==0.8.10
tqdm==4.64.1
xarray==0.19.0

Download Infection Dynamics Traces

The traces for the infection dynamics simulation need to be put in covid-vaccine-dynamics/data/traces. The full trace files are large and can take some time to download. In order to quickly test the code base, you can download a smaller version of the trace files:

Option 1 (quick): download small trace file for main contact matrix

Download the small trace file (41 MB) for the main contact matrix (C_mat=80) and place it in the corresponding data folder with

cd covid-vaccine-dynamics/data/traces
wget https://gin.g-node.org/jdehning/covid-vaccine-dynamics-data/raw/a88a6e8d25fbe895aa08ca55c3641b6f90fc656e/traces_small/run-begin=2020-12-20-end=2021-12-19-C_mat=80-V1_eff=70-V2_eff=90-V3_eff=95-influx=0.5-draws=500.zip
cd ../../..

Note: the small trace file should only be used for testing purposes. The estimated uncertainties will be unreliable. The full trace files are required to reproduce the results shown in the paper.

Option 2 (slow): download full trace files for all contact matrices

Download the full trace files (5.1 GB in total) for all contact matrices (C_mat=70, C_mat=80 and C_mat=90) and place them in the corresponding data folder with

cd covid-vaccine-dynamics/data/traces
wget https://gin.g-node.org/jdehning/covid-vaccine-dynamics-data/raw/a88a6e8d25fbe895aa08ca55c3641b6f90fc656e/traces/run-begin=2020-12-20-end=2021-12-19-C_mat=70-V1_eff=70-V2_eff=90-V3_eff=95-influx=0.5-draws=500.zip
wget https://gin.g-node.org/jdehning/covid-vaccine-dynamics-data/raw/a88a6e8d25fbe895aa08ca55c3641b6f90fc656e/traces/run-begin=2020-12-20-end=2021-12-19-C_mat=80-V1_eff=70-V2_eff=90-V3_eff=95-influx=0.5-draws=500.zip
wget https://gin.g-node.org/jdehning/covid-vaccine-dynamics-data/raw/a88a6e8d25fbe895aa08ca55c3641b6f90fc656e/traces/run-begin=2020-12-20-end=2021-12-19-C_mat=90-V1_eff=70-V2_eff=90-V3_eff=95-influx=0.5-draws=500.zip
cd ../../..

The entire installation process can take up to 2 hours, mostly depending on how fast the data can be downloaded.

Data preparation

The raw input data is stored under data/input-data. The data is prepared with the script data_preprocessing/israel_data_processing.py. The script is run with

cd data_preprocessing
python israel_data_processing.py

This creates two pickled dataframes in data/preprocessed_data:

• israel_df.pkl,
• israel_constants.pkl.

We then estimate the severity factorisation

python severity_factorisation.py
cd ..

This creates the severity factorisation for different assumed waning timescales:

• regular waning: data/factorisation-with-fit
• fast waning: data/factorisation-with-fit-fast-waning
• no waning: data/factorisation-with-fit-no-waning.

Hello world

To run a scenario run the main script with

python main.py --config=configs/<VACCINE_ALLOCATION_STRATEGY>.yml

where <VACCINE_ALLOCATION_STRATEGY> is one of the following:

• observed,
• uniform,
• young-first,
• elderly-first.

This saves the results in run/YYYY-MM-DD_HH-MM-SS_<VACCINE_ALLOCATION_STRATEGY>/. The run output file format is shown here.

Reproduce results

Counterfactual vaccine allocation strategies

To reproduce the results for the different vaccine allocation strategies run

python experiments/policy_exp.py --config configs/policy-exp.yml

which saves the results in run/YYYY-MM-DD_HH-MM-SS.FFFFFF_policy_exp/. The run can be analysed with the notebooknotebooks/va_strategy_experiment_analysis.ipynb. Runtime: 5.5 hours (using 64 cores), 44 hours (on regular PC, estimated assuming 8 cores).

Impact of vaccine uptake

Run

python experiments/vaccine_acceptance_exp.py --config configs/vaccine-acceptance-exp.yml

which saves the results in run/YYYY-MM-DD_HH-MM-SS.FFFFFF_acc_exp/. The run can be analysed with the notebooknotebooks/uptake_experiment_analysis.ipynb. Runtime: 13 hours (using 64 cores), 104 hours (on regular PC, estimated assuming 8 cores).

Simulating other disease types

Run

python experiments/risk_profile_exp.py --config configs/risk-profile-exp.yml

which saves the results in run/YYYY-MM-DD_HH-MM-SS.FFFFFF_risk_profile_exp/. The run can be analysed with the notebooknotebooks/risk_profile_experiment_analysis.ipynb. Runtime: 17 hours (using 64 cores), 136 hours (on regular PC, estimated assuming 8 cores).

Impact of immunity waning

Run

python experiments/waning_exp.py --config configs/waning-exp.yml

which saves the results in run/YYYY-MM-DD_HH-MM-SS.FFFFFF_waning_exp/. The run can be analysed with the notebooknotebooks/waning_experiment_analysis.ipynb. Runtime: 11 hours (using 64 cores), 88 hours (on regular PC, estimated assuming 8 cores).

Run output files

run_output_directory
├── result.npy
├── result_samples.npy
├── factorisation_data
│   ├── observed_infection_data.csv
│   ├── observed_severity_data.csv
│   ├── population_data.csv
│   ├── risk_factors.csv
│   ├── time_dependence.csv
│   ├── vaccination_data.csv
│   ├── vaccine_acceptance_data.csv
│   ├── vaccine_efficacy_waning_data.csv
│   └── waning_curve.csv
├── factorisation_data
│   ├── age_group_names.npy
│   ├── age_groups.npy
│   ├── D_a.npy
│   ├── observed_vaccinations.npy
│   ├── P_a.npy
│   ├── P_t.npy
│   ├── vaccination_statuses.npy
│   ├── week_dates.npy
│   ├── week_dates_scenario.npy
│   ├── weeks.npy
│   ├── weeks_extended.npy
│   ├── weeks_scenario.npy
│   └── weeks_scenario_extended.npy
├── severity_factorisation
│   ├── f_0.npy
│   ├── f_1.npy
│   ├── g.npy
│   ├── infection_dynamics.csv
│   ├── infection_dynamics_samples.npy
│   ├── median_weekly_base_R_t.npy
│   ├── median_weekly_eff_R_t.npy
│   └── vaccine_efficacy_params.npy
└── vaccination_policy
    ├── U_2.npy
    ├── U_2_full.npy
    ├── u_3.npy
    └── u_3_full.npy