Optimisation tools for the COVID-19 pandemic

Citing this work

You can cite this repository via Zenodo with the following BibLaTeX entry:

@misc{covid_optimisation_github,
    title={Optimisation tools for the COVID-19 pandemic},
    author={Santini, Alberto},
    date={2020-05-04},
    year={2020},
    howpublished={Github repository},
    doi={10.5281/zenodo.3785566},
    url={https://github.com/alberto-santini/covid-optimisation/}
}

You can cite the paper on the optimisation of swab testing capacity with the following BibLaTeX entry:

@article{santini2020covid,
  title={Optimising the assignment of swabs and reagents for PCR testing during a viral epidemic},
  author={Santini, Alberto},
  journal={{Omega}},
  volume={102},
  year={2020},
  doi={10.1016/j.omega.2020.102341},
  url={https://santini.in/files/papers/santini-2020.pdf}
}

About

This repository was created by Alberto Santini, professor of Operational Research at Universitat Pompeu Fabra. It contains optimisation models and software to optimise decisions related to the COVID-19 pandemics.

Main links

• Interactive dashboard with results (optimisation of swab testing capacity, Italy case study).
• Technical report on the optimisation of swab testing capacity.

Contents of the repository

• Folder model-implementation contains the implemented models. The software is in C++ and uses CMake for configuration. The solver Gurobi is necessary to run the software. All software is released under the GPLv3 license (see file LICENSE).
• Folder swab-tests-data contains data for the problem of increasing swab tests capacity. Find a description of this problem and a mathematical model to solve it in our technical report.
  • Subfolder italy-case-study contains data (and the programme to generate it) relative to swabs tested in Italy during the period 01-13 April, 2020.
  • Subfolder synthetic-data contains a generator to produce simulated data, used to test how the model behaves under different scenarios.

Data Generation

This sections explains in detail how we generate data for the case studies.

Swab tests optimisation - Italy case study

We start from the publicly available data sources:

• A list of laboratories certified to test swabs for COVID-19: italy_labs_coords.csv.
• A list of factories in Italy, which certified to produce test kits: italy_factories_coords.csv.
• Data about the number of swabs tested each day, region-by-region, released by the Italian Civil Defence department: dpc-covid19-ita-regioni.csv.

We first estimate how many swabs regions would have liked to test each day, if they had enough reagents. We calculate this number from the actual number they tested, times some region-dependent multiplier which we estimate based on news and press releases (see italy-case-study/instance-generator.py for the precise numbers). We assume that, before the planner's intervention, each region would allocate an equal number of swabs to each of its laboratory. Because we start our analysis mid-crisis (April 1st) we assume that there is no safety stock of reagents neither at laboratories, nor at factories. We also assume that laboratories structural capacity (i.e., the capacity determined by staffing and test machines, but not by reagent availability) has a +25% slack compared to the number of swabs they actually tested during the period we analysed. This hypothesis means that we think (based on news and statements released by regional health authorities) that reagent availability was the main bottleneck that prevented increasing test capacity. We estimated the number of swabs and reagents that can be transferred to each region day by day, based on geographic and economic factors peculiar to each region. For example, Sardinian labs cannot exchange swabs with labs in other regions, because it would be hard to ship them out of the island. Some regions, such as Lombardia, Piemonte, Emilia Romagna or Campania have good infrastructure and are well-connected, so they are able to move more material.

Swab test optimisation - synthetic data

A detailed description of the generation process for the synthetic instances is in the technical report (Section 6.2). Because the dataset consists of 60480 instances, amounting to several GB of data, rather than providing the instances we give a script to generate them: synthetic-data/instance-generator.py. The instances created in this way will have name s-<N>.json, where <N> is a progressive number. The generator also creates a file inst-directory.csv, which lists the instance generation parameter used to create each instance.

The picture above shows an example of synthetic instance. Dots correspond to laboratories (the larger the dot, the higher the demand in the lab's region), stars to factories and lines denote regional boundaries.