Applying mathematical models to support decision-making in the COVID-19 pandemic

Mathematical epidemiological models are useful for informing such decisions through exploring and understanding the consequences of a variety of plausible scenarios regarding assumptions about virus and disease characteristics, health system capacity, and the timing and strength of PHSM.9–11 We provide a background to how epidemiological models have been applied to analysing the COVID-19 pandemic in online supplemental text A1 in the appendices. In summary, mathematical models allow prediction and comparison of key epidemiological outcomes such as the daily incidence of infections (including severe infections that require hospitalisation and intensive care) and deaths. There is considerable uncertainty associated with these predictions, particularly as COVID-19 is a novel viral disease in humans, for which the evidence on transmission and clinical outcomes continues to evolve, and as such many of the parameters and processes that govern its spread and clinical outcomes are unknown or poorly quantified.10–12 Moreover, models make simplifying assumptions in their representation of reality, and they evolve and adapt over time with the emergence of new data and evidence.12 Nonetheless, the qualitative behaviour of models can effectively address certain questions of current importance to policy-makers. For example, models can compare different scenarios that represent alternative PHSM strategies and give insight into their relative benefits,10 13 such as flattening the epidemic curve and reducing the expected burden on the healthcare system.14 Models can warn about future waves of infection, subject to their assumptions about infection, reinfection and immunity.15 Importantly, modellers need to work closely with decision-makers and the media to ensure that model predictions are interpreted and used appropriately, that the associated uncertainty is clearly communicated12 and that the context and motivation of the developers and users are transparent.10 This collaboration is also important for collection of model inputs as the reliability of any prediction is conditional to the quality of the epidemiological data feeding the model. Moreover, while a model is a useful tool for supporting decision-making, its predictions should be interpreted across a realm of other tools and data sources.

Participatory approach to modelling the COVID-19 pandemic in the EMR

Public health officials in EMR countries noted how mathematical epidemiological models were informing public health decisions in the countries that were hit earlier in the pandemic.16 17 As a result, some of them have expressed an interest in receiving technical support from WHO for mathematical modelling to interpret the potential epidemiological spread of the outbreak, given different PHSM scenarios. Accordingly, the WHO Eastern Mediterranean Regional office (EMRO) established a modelling support team in mid-March as part of the Information Management pillar in the COVID-19 Incident Management Support Team.18 This is the first time the WHO Regional Office used modelling as a tool in an ongoing outbreak response. Here, we present an innovative approach that was developed by the EMRO modelling support team to conduct participatory modelling analyses together with public health professionals from the EMR to support decision-making.

The primary objective of the modelling support team was to address the imminent decision-making needs faced by policy-makers. Another objective was to provide training on mathematical epidemiological modelling, including the underlying concepts and methods, to promote awareness of how models work and how they can be used effectively and to work towards building technical capacity for mathematical modelling in the EMR.

A participatory approach was adopted to conduct country-specific modelling analyses that were relevant to the PHSM being considered for the country. As part of this approach, policy-makers from the Ministries of Health in the countries and WHO country offices (WCOs) were actively involved throughout all stages of the modelling process.13 19 20 The WHO EMRO modelling support team also worked with the WHO Headquarters, researchers from academia and other United Nations (UN) agencies to investigate different modelling approaches, help interpret model results and guide the analysis of policy-relevant questions (see online supplemental figure A3 in the appendices). Participatory modelling can facilitate the translation of model findings into decision-making in a number of ways. First, it can guide the modelling analysis to identify and address questions that are relevant and important to public health in the country. It can also encourage shared learning among researchers, modellers and policy-makers, and foster common understanding and expectations.19 Importantly, the involvement of within-country analysts and policy-makers builds trust, ownership and support,19 20 meaning that the findings of the analyses will be more likely to be considered in decision-making.

Policy questions posed by policy-makers in EMR

After consulting with policy-makers from countries, by the end of March 2020, the EMRO modelling support team identified the most urgent questions to address through modelling exercises: (1) How many infections and deaths are expected at different times? (2) What are the expected healthcare system requirements? (3) What is the effect of PHSM on the spread of the epidemic? To be suitable for supporting rapid decision-making in countries in the EMR, many of which are low/middle-income countries, the modelling tools used needed to be readily adaptable to country-specific settings and PHSM policies, with data inputs swiftly attainable and requiring low computational resources. It was also necessary to use an interpretable modelling approach that could be readily communicated in non-technical terms to participants from a variety of disciplines.

Application of the CoMo model in the EMR

The team found that the COVID-19 International Modelling (CoMo) model developed by the CoMo Consortium21 offered a user-friendly interface to modelling impacts of PHSM on the time trajectories of infections (including severe infections requiring hospitalisation and intensive care) and deaths. The CoMo model was able to address the questions posed by policy-makers in the region, had the flexibility to be adapted to the regional context and had a computationally efficient online application that allowed the model to run quickly on large populations.

The CoMo model21 is an age-dependent deterministic Susceptible–Exposed–Infectious–Recovered model that predicts daily incidence of COVID-19 infections in the population under different assumptions about PHSM. Details of the CoMo methodology are provided in online supplemental text A1 in the appendices. In summary, the model considers five levels of infection severity: asymptomatic infections, symptomatic infections, and infections requiring hospitalisation, intensive care treatment and ventilated intensive care treatment. Infection severity and associated mortality are age-dependent, in that the proportion of infected individuals requiring hospitalisation, and the proportion that die, varies with age. The CoMo model incorporates a hospital submodel that indicates when hospital treatment requirements, including treatment in hospital beds, intensive care units and ventilators, exceed the capacity of the country’s healthcare system.

The CoMo model incorporates an explicit representation of the PHSM that have been commonly used to mitigate the spread of COVID-19. The model considers PHSM that target infected individuals, including self-isolation of symptomatic individuals and self-quarantine of members of their household, screening of the contacts of individuals with a positive diagnostic test result and mass testing. The model also considers PHSM that aim to limit transmission throughout the wider population, including school closure, workplace closure, physical distancing measures, border closure, shielding elderly individuals, handwashing and mask-wearing. For each intervention, the user provides inputs about the coverage, defined as the proportion of the total population to which the intervention applies. The user also inputs the intervention timing and duration, and the adherence of the population to the intervention. These coverage and adherence parameters act on age-specific social contact rates in home, work, school and other community environments, which are estimated in the model using the country-specific contact matrices developed by Prem et al.22 A useful feature of the CoMo model is that it allows the coverage of each intervention to vary over time, meaning that the intervention can have a stronger effect over some time periods and a weaker effect over other time periods. We have used this feature in applying the model to EMR countries to represent relaxations of PHSM (such as during Eid and Ramadan periods) and strengthening of measures in response to resurgences in infections.

Participatory modelling process

The EMRO modelling support team, together with the CoMo Consortium, has been adapting the CoMo model to investigate COVID-19 spread dynamics in countries in the EMR, as well as interpreting the results and conveying the models’ caveats to the countries. Throughout this process, the team has assisted countries in identifying data sources, gathering country-specific data, documenting model outputs and preparing policy briefs. The team supports capacity building by delivering training and demonstrations of the models and the online tool. Establishing participatory collaborations with EMR countries to adapt and run the CoMo model involved the following stepwise approach:

1. Introduce modelling approach: The EMRO modelling support team introduces the modelling approach to the interested countries using an audiovisual presentation that provides a non-technical overview of the CoMo model, including the main inputs, outputs and limitations. This is supported by social media videos in three regional languages.23–25 The introductory sessions and the videos convey basic information about epidemiological models and the uncertainty about model parameters and structure.

2. Assemble collaborating team: A team of core collaborators that will participate in the modelling analysis is assembled. This includes public health professionals from the country Ministry of Health and the WCO as well as the EMRO modelling support team.

3. Collect model input data: Once the collaborating team embarks on a collaboration, the data collection template and a reference manual describing the model parameters and their definitions are shared. The country Ministry of Health and the WCO are supported during the collection of country-specific model parameters by convening data collection webinars and sharing data sources.

4. Review input data: The EMRO modelling support team corroborates the data inputs by cross-checking with published literature and the country Ministry of Health and the WCO. The team also cross-checks the assumptions being made about the strength and timing of the PHSM implemented in the country with Google mobility reports26 and Oxford government response tracker.27 Any discrepancies are discussed with collaborators from the country to develop a common understanding and agree on any revisions.

5. Conduct initial modelling analysis: Modellers from the EMRO modelling support team and the country conduct an initial modelling analysis to produce preliminary results. The analysis employs a scenario-based approach that develops customised scenarios which compare different PHSM implementation strategies based on the questions posed by the countries. The team drafts a technical report containing details about model inputs, methods, results and limitations that is then shared with country Ministry of Health and the WCO.

6. Review initial results: The country Ministry of Health and the WCO review the technical report and discuss with the EMRO modelling support team any modifications or updates required.

7. Update modelling analysis: The EMRO modelling support team updates the modelling analysis and produces an updated technical report which is shared with the country Ministry of Health and the WCO.

8. Summarise policy implications: The EMRO modelling support team prepares a non-technical audiovisual presentation together with a policy brief document that targets a wide public health audience and interprets the main findings of the analysis from the perspective of decision-making, as well as conveying the uncertainty and limitations.

9. Continue collaboration: Engagement with countries continues as required to update the modelling analysis according to the progression of the COVID-19 outbreak and the control strategies under consideration in the country.

The components of this participatory modelling approach, including participants, resource requirements and methodological approaches, are summarised in online supplemental table A1 and figure A3 in the appendices. We have evaluated our experience of implementing the above nine-step process to identify benefits and limitations, and provide recommendations for the continued use of this approach (online supplemental table A2 in the appendices).

Progress to date

As of mid-January 2021, 11 countries, namely, Afghanistan, Djibouti, Egypt, Iraq, Jordan, Lebanon, occupied Palestinian territories (oPt), Pakistan, Syria, Tunisia and Yemen, have been engaged in collaborative modelling with EMRO. Of these, seven, namely, Afghanistan, Jordan, Lebanon, Pakistan, Egypt, oPt and Tunisia, had reached stage 8 of the above process and disseminated the results of the modelling analyses and policy implications to the decision-makers. Scientific committees and decision-makers in some of those countries are consulting the model’s projections, in conjunction with the epidemiological data and other data sources, when examining current policies in place and weighing possible options for instating or loosing certain measures. For instance, Tunisia’s epidemiologists are exploring options of immediate or gradual release of physical distancing and working from home measures.28 The EMRO modelling support team is currently in the process of documenting examples where model results have already been incorporated into policy. One example is the school opening strategies, such as the cyclical school opening strategy adopted in oPt.29 Another example is the implementation of 1-day lockdown per week and a daily night-time curfew in addition to keeping schools online till the end of the semester in Jordan.30 Investigating testing strategies and possibilities of under-reporting with the modelling tool is also currently being explored in some countries.

Challenges

There are a number of barriers to effective and widespread use of models in the EMR, such as the insufficient availability of national data and limited national modelling capacity. There is considerable uncertainty associated with modelling the dynamics of novel viruses,10 11 and interpreting the results can be challenging given the general limitations of the models and the input data on which they are based (online supplemental text A1 and table A2 in the appendices). The EMRO modelling support team has tackled these issues by communicating throughout all stages of the modelling collaboration the expectations about the appropriate uses and limitations of models. The social media videos that were developed by the team are presented in Arabic, English and French, and describe the uncertainty associated with epidemiological models of COVID-19 and how model results can be interpreted to guide decision-making.23–25 These videos were disseminated on different platforms and at the time of writing had received 25 359 views across all platforms. The team also encourages national capacity building by providing training on the modelling approach and caveats, as well as on using the online software and the interpretation of results. These challenges can be further addressed through encouraging active engagement in modelling as a part of public health decision-making as well as strengthening the local systems for epidemiological surveillance and data storage.

At present, several modelling analyses of the COVID-19 pandemic have been conducted for EMR countries by different international research groups, each using different modelling approaches and making different assumptions about viral transmissibility and severity, and the implementation of PHSM.31–35 This can lead to discordant sets of results and guidance being presented to policy-makers, which can be difficult to assimilate, is confusing and can potentially lead to a distrust, and loss of interest, in models. A challenge for the modelling community lies in systematically comparing and synthesising results provided by different models to provide clear and consistent guidance to decision-makers.36 Model comparison initiatives that are currently reviewing the results of multiple COVID-19 epidemiological models are a promising step towards achieving this goal.37 38 The EMRO modelling support team is involved in these activities through an international network of collaborations with academic institutions and other UN agencies (online supplemental figure A3 of the appendices).

The socioeconomic and geopolitical situation in the EMR is another unique challenge, not only for implementing PHSM but also for modelling the COVID-19 outbreak. More than half its countries are currently suffering from armed conflict and political instability.39 In addition, it is estimated that out of the 600 million people living in the region, 13 million are internally displaced, 12 million are refugees40 and 46 million are low-income labour migrants.41 These populations often reside in overcrowded spaces with poor living conditions, such as camps or camp-like settings, where it can be difficult to apply certain public health measures, such as physical distancing and self isolation. Basic adherence to personal protective measures (such as handwashing) is problematic as well due a lack of basic sanitation services. It has been estimated that the burden of COVID-19, in terms of numbers of infections and attributable mortality, has been greatly under-reported in conflict-affected countries in the EMR.42–44 A number of approaches for modelling COVID-19 dynamics in internally displaced people (IDPs) and refugee camps are being developed,45 and the EMRO modelling support team is investigating the application of such approaches to the region.