Studying the impacts of COVID-19 mitigation policies on health outcomes: An agent-based modelling approach

How can policymakers better prepare for future pandemics? One way of doing so may be revisiting past pandemics: which policies did (or did not) work well? Looking to the COVID-19 pandemic may be particularly instructive, given the pandemic’s unprecedented reach and the varied policy responses that it prompted. 

More specifically, studying the impact of COVID-19 pandemic policy responses on non-COVID-19 health outcomes may be important. This may shed light on the appropriate balance to strike between total lockdown and letting disease spread. Evidence suggests that mitigation policies (e.g. lockdowns) were successful at reducing lives lost to COVID-19, and at preventing health systems from being overwhelmed.1 But mitigation policies also often took their own tolls on populations’ health.2 Restrictions on personal freedoms and isolation from loved ones may have increased stress and contributed to more, and more extreme, unhealthy behaviors.3 This can be seen in lockdown-related increases in: feelings of social isolation, mental health issues, delayed or forgone medical care, interpersonal violence, and weight gain, among others.4–7 Despite this initial evidence, it is not yet fully understood whether the benefits of reductions in COVID-19-related mortality outweighed the cost of these secondary health issues.

Simulation modelling may be useful here. Simulations would enable the comparison of actual scenarios during the COVID-19 pandemic with hypothetical ones. More specifically, this project proposes using agent-based modelling (ABM), a stochastic simulation technique, to examine pandemic mitigation policies’ relationships to non-COVID-19 health outcomes. To date, several studies have examined these relationships with simulation models.8,9 However, research examining pandemic mitigation policies’ impact on non-COVID-19 health outcomes using ABM is scant. 

This represents a knowledge gap, as ABMs enable the examination of actual and hypothetical lockdown scenarios. For instance, the Netherlands saw three lockdowns, with restrictions on closing times, most recently until January 2022. With simulations, we will be able to examine the extent to which other pandemic mitigation policies – for instance, no lockdowns, only one lockdown, or the same number of lockdowns for shorter durations – may have impacted health outcomes. We will also be able to examine whether these impacts persisted over time, even after lockdowns ended.

Also,  ABMs can take into account interactions with other individuals (‘agents’), and therefore better reflect real-world dynamics. For instance, social isolation and mental health were perhaps buffered by the presence of family members in the household, (although they may have increased the likelihood of COVID-19 transmission). This interactive aspect of ABMs may help to explain their popularity in simulating COVID-19 transmission.10–12 ABMs are also well-suited to the study of more socially-determined health outcomes, due to this potential to more realistically capture social dynamics.13 However, ABMs have not yet been used to study non-COVID-19 health outcomes that may have been impacted by pandemic mitigation policies. Using ABMs to study these relationships is thus a key contribution to understanding pandemic dynamics.  

Further, by using ABMs, causal loops can be able to be taken into account. This is particularly important in modelling socially-determined health outcomes during the pandemic. For instance, when looking at mental health, it is possible that experiencing and adhering to COVID-19 measures decreased mental health due to a disruption of everyday life, increases in social isolation, and heightened concerns over the pandemic itself.14 In turn, there is preliminary evidence that people with poorer mental health were more adherent to COVID-19 mitigation measures, thereby creating a feedback loop.15 This aspect of ABMs has already been identified in the literature: Murch et al., in a study simulating the pandemic’s impact on health outcomes with a discrete time simulation engine, noted that their modelling strategy failed to incorporate feedback loops, and that ABMs would address this shortcoming.14