Voluntary risk mitigation behaviour can reduce impact of SARS-CoV-2: a real-time modelling study of the January 2022 Omicron wave in England

Throughout the COVID-19 pandemic, human behaviour has modified and controlled transmission. In the initial epidemic growth phase in early 2020, government-defined social distancing, including school closures and stay-at-home orders, was widely used worldwide to prevent the exponential growth in numbers of cases. However, as the pandemic progressed and personal biosecurity measures have become available, the role of individual choice in using face coverings, rapid home tests and vaccination has become increasingly important.

A challenge for predicting future SARS-CoV-2 burden is accurately capturing the population response to measures and advice. The first impositions of stay-at-home orders were widely adhered to [1, 2]. Behaviours changed rapidly upon the imposition of restrictions but were slower to revert to ‘normal’ once restrictions were lifted, with a high proportion of individuals still practicing biosecurity measures such as wearing face coverings and limiting travel [1]. Such behaviours are influenced by multiple factors including trust in the government, individual perception of risk, media reporting and personal interactions, and the importance of different factors varies over time [3].

Behavioural dynamics, which affect both transmission potential and reporting of cases [4], are increasingly included in infectious disease models, although with varying degrees of detail. In particular, the availability of Google Mobility data [5] during the COVID-19 pandemic has been used as a proxy for changing mobility and contact behaviour in transmission models [6‐8]; however, assumptions about future behaviour are still required. Surveys of intended behaviours have the potential to aid future predictions but may overestimate realised behaviour [9].

Here, we report on the results of a rapid, prospective survey of intended behaviours over the 2021 winter festive period. The purpose of the study was to estimate the impact of voluntary behaviours on SARS-CoV-2 transmission and numbers of cases and deaths due to the Omicron SARS-CoV-2 variant, first identified in November 2021. In the UK at the time, there were minimal legal restrictions on behaviours, although due to the Omicron variant, the UK government announced a ‘plan B’ on 8 December 2021, in which face masks were compulsory in indoor public spaces and home working was encouraged.

At the time this work was undertaken in December 2021, there was still considerable uncertainty about the transmission potential and severity of the Omicron variant compared to the previously dominant Delta variant. The rapid growth, with cases due to Omicron doubling every 2 to 3 days, suggested a substantial advantage over other variants [10, 11]. This transmission advantage could be due to an increase in transmission potential and/or immune escape mutations which render immunity from vaccination and previous infections less protective against infection than before [12, 13]. Emerging evidence suggests that infection with Omicron is less likely to lead to severe disease and death [14, 15], although with considerable uncertainty.

Data here suggest that a high proportion of people intended to practice risk mitigation behaviours to prevent the spread of COVID-19 during the Christmas 2021 period (from 20 December 2021 to 2 January 2022), with approximately 80% reporting using home testing kits. We estimate that these realistic risk-mitigating behaviours, including mask-wearing and regular home testing, could lead to a 40% reduction in the Omicron effective reproduction number and a 28% reduction in cumulative numbers of deaths. Nonetheless, these intended risk mitigation behaviours are not sufficient to reduce the effective reproduction number below one, suggesting that epidemic control of Omicron cannot rely solely on self-imposed restrictions, and a further reduction in social contacts of 25–50% is required for control. Survey results also suggest that announcements of ‘Plan B’ restrictions had limited behavioural impact in our sample and would therefore have limited epidemiological impact. Under our baseline scenario, we estimated approximately 50,000 further deaths due to Omicron alone, which is a 30% increase in the total number of deaths in England between March 2020 and December 2021.

This study was conducted in late 2021, one of a number of models presented at SPI-M-O that explored scenarios for the January 2022 SARS-CoV-2 ‘Omicron’ wave [33, 34]. Scenario modelling, as opposed to prediction or forecasting, is used to understand epidemic drivers such as vaccine coverage, the impact of variants or population behaviour change [35]. With our individual-based approach, we were able to quantify the impact of ‘plan B’ and voluntary behaviour change. Our approach is less computationally intensive than dynamic models, and by using theoretical results, we provide a fast, transparent and intuitive understanding of how behaviour translates to hospital admissions and deaths. A drawback of our approach is that it is not dynamic, so we cannot capture the ongoing effects of vaccine waning, vaccine rollout or changing behaviours. We also cannot estimate trajectories of cases and deaths. The main source of uncertainty in our modelled estimates is uncertainty in the severity of disease, both intrinsic severity (the probability of severe disease in a naïve, unvaccinated individual) and the probability of severe disease in individuals with prior immunity that suffer breakthrough infections.

Our results use surveys of intended behaviours to modulate the transmission potential of SARS-CoV-2. Other approaches for incorporating behaviour into models include using mobility data, such as Google Mobility [5], social network and mobile phone data [36, 37], which have been used during the COVID-19 pandemic. Such routinely collected, geolocated data can fill data gaps from under or absent sampling, for example before other systems are established. Each source of behavioural data offers a different perspective, with varying biases and representativeness [38]. A comparative study showed that the COVID-19 reproduction number was overestimated by mobility data but underestimated by survey data; in general, survey data provided more reliable estimates of transmission changes than mobility data [39].

Furthermore, self-reported future intentions may not reflect actual behaviour. A survey conducted during the pandemic found that while 70% of individuals intended to fully self-isolate with COVID-19 symptoms, only 50% actually did [9]. Underreporting of other risk mitigation measures is less clear: 70% of our cohorts intended to use face coverings, which is less than measured use [40]. Eighty percent of our cohorts intended to use home testing kits before seeing family and friends, which is greater than the 40% [41] and 50% [42] of individuals who reported using home testing kits between April 2021 and June 2021. However, home testing kit use increased by 63% in December 2021, compared to the rest of the year [43], suggesting that intended use was broadly reflective of actual use. Risk mitigation behaviour also varies with socio-demographic factors such as gender and age [41]. While established longitudinal cohorts are designed to be broadly representative, they nevertheless suffer from respondent bias with over-representation of females and higher socioeconomic status [18]. Our surveys did not include many children or young adults, and ALSPAC participants are mostly resident around Bristol. There are likely to be additional unmeasured biases in the surveys and understanding how risk mitigation behaviour varies by age, ethnicity and socioeconomic status would be valuable for improved characterisation of the epidemiology in different communities.

This study was used in real-time to quantify the necessity of additional restrictions over-and-above voluntary measures. This work demonstrates the power of utilising existing longitudinal cohorts for rapid elicitation of behavioural information in combination with population-levels transmission models. Fostering such links improves the accuracy and utility of disease models used for public health action and decision making especially for rapidly spreading infections.