The influence of COVID-19 risk perception and vaccination status on the number of social contacts across Europe: insights from the CoMix study

The COVID-19 pandemic has led to unprecedented wide-ranging effects across the globe on social life and mental health, and has adversely affected the global economy [1‐4]. This has resulted in individuals adopting differential coping behaviours to deal with the psychological distress caused by the pandemic [5‐7]. Since its emergence, it has led to massive loss of lives worldwide with an estimated 6.9 million confirmed COVID-19 deaths as of May, 29th 2023 [8], and an approximated 14.9 million excess deaths associated with the pandemic in 2020 and 2021 [9]. Furthermore, there is mounting evidence that some of the people who suffered from the COVID-19 disease experience prolonged adverse health effects with continued multisystemic symptoms weeks and months post infection with a substantial impact on health and wellbeing [10, 11]. From March 11, 2020, the time COVID-19 was declared a global pandemic by the World Health Organization [12], different countries have experienced different waves of the pandemic. This in turn has led to implementation of a range of interventions to alleviate pressure on the healthcare systems as well as to control the pandemic [13, 14]. Before the introduction of the COVID-19 vaccines, governments relied on adopting various Non-Pharmaceutical Interventions (NPIs) to curtail the transmission dynamics of SARS-CoV-2 [15]. Since the introduction of the COVID-19 vaccines, different governments have adopted different vaccine roll out strategies. As vaccine uptake is increasing globally [16], governments are monitoring and adjusting NPIs depending on the epidemic situation in their respective countries [14]. Due to the continuing uncertainty on the future trajectory of the pandemic, as has been demonstrated by the newly emerging COVID-19 variants of concern such Omicron and its sub-variants [17, 18], global efforts on both vaccination and proportionate implementation of relevant NPIs are essential to reduce cases without too much negative social and economic impacts [19, 20].

Since NPIs mainly encompass social distancing measures, adherence to these measures might be influenced by attitudinal and demographic determinants [21]. During the past pandemics, the control of fast spreading infectious diseases has relied, partially on populations’ risk perceptions, both at individual and societal level [22]. Thus, investigating the key attitudinal determinants influencing behavioral responses is pivotal to continue guiding the implementation of appropriate strategies. Risk perception is a key component of the Health Belief Model (HBM) [23] and the theory of Protection Motivation and Self-efficacy (PMS) [24]. The HBM framework emphasizes that individual’s likelihood to adopt health preventive behaviours are mainly based on their risk perceptions [23]. Whilst the PMS theory postulates that the implementation of the recommended health protective behaviours is based on individual’s risk perceptions and self-efficacy to adopt them [24]. Several empirical studies have explored the relationship between risk perceptions and the adoption of health protective measures during the current COVID-19 pandemic [25‐32] and previous pandemics [33, 34] and found that risk perceptions play a significant role in the adoption of health protective measures. The most utilized health protective behaviours in these studies include physical distancing, frequency of hand washing, wearing of face masks, and avoiding public places. Risk perception has yielded significant relationships with the number of social contacts from two recent studies, one in Belgium [31] and the other in UK [35] during the COVID-19 pandemic.

In addition to the NPIs which are mainly focused on reducing close person-to-person contacts, following the initiation and continued uptake of COVID-19 vaccination, it is imperative to explore whether vaccination alters social contact behaviour given the inherent uncertainties in vaccine waning as well as protection against emerging variants [36]. However, the literature on the relationship between COVID-19 vaccination status and perceptions and social contact behaviour during a pandemic is very limited. Thus, given the crucial role of contact behaviour in the dynamics of SARS-CoV-2 virus [37, 38], it is important to analyse how specific COVID-19 perceptions and vaccination status relate to the number of social contacts in a wider geographical context for the ongoing management of the pandemic. Utilizing data collected under different phases of the pandemic and also under different intervention measures from multiple countries is crucial to correctly disentangle possible transient and/or country specific effects.

Here, we present analyses of the influence of COVID-19 risk perceptions and vaccination status on the number of social contacts of individuals using longitudinal social contact data collected during the COVID-19 pandemic as part of the CoMix study [39]. This is pivotal to continue enhancing the understanding between risk perceptions and social contacts as the world continue pushing towards a post-acute phase of the pandemic.

We employed a multilevel generalized linear mixed effects model (GLMM) to explore the associations between the risk perceptions and vaccination status with the number of social contacts [44]. The participants observations from each survey round (level-1) were nested within participants (level-2), and participants were nested within countries (level-3). Random effects were used to correct for correlations in this nested nature of the study. We also performed exploratory modeling using GLMM where the country was considered as a fixed effect and the participants as random effects to explore cross-level effects of the country and individual perception variables [45]. In both modeling approaches, the number of social contacts is modelled using (1) a negative binomial distribution, accounting for possible overdispersion in the counts, and (2) zero-inflation component to deal with excess zeroes in the number of social contacts. The intra-class correlation, which is a quantity measure of the between and within group variation [46], was used to gain insights on heterogeneity in number of social contacts between and within the countries. Throughout our analyses, the number of contacts were truncated at 100. The models included the individual risk perception variables and vaccination status (vaccinated versus not vaccinated) and adjusted for the participant’s household size, gender, age, day of the week (week day versus weekend), self reported high risk status, history of COVID-19 infection, employment status, and stringency index as potential confounders. The stringency index was considered in four levels: low (0-40), moderate (41-55), high (56-70) and very high (71-100). The vaccination status was primarily defined based on the first dose of any of the COVID-19 vaccines since some surveys were conducted in the initial phases of the vaccination where few individuals were fully vaccinated. History of infection was defined based on whether participants had previously tested for the virus. The models were fitted using maximum likelihood estimation. We used R version 4.1.1 and the glmmTMB package (version 1.0.2.1) [47] for all statistical analyses.

Model building was performed for each individual perception variable. This was informed by preliminary exploratory analyses of a model including all the perception variables and models for each individual perception variable. Hence in total, we had 3 models for the multilevel generalized linear mixed effects corresponding to each individual perception variable (i.e. perceived severity, perceived susceptibility, perceived risk to vulnerable) and 3 models for the generalized linear mixed effects including the country as a fixed effect. The significance of the model variables was assessed through Type III Wald tests and a significance level of 5% was considered.

The main objective of this study was to explore the influence of COVID-19 vaccination status and the COVID-19 related risk perceptions on the number of social contacts. We analyzed longitudinal data collected in 16 European countries using a multilevel generalized linear mixed effects model to account for within and between participants variations while controlling for the hierarchical structure of the data. Furthermore, we also performed cross-level analysis to explore the relationships between both the perceptions and vaccination status with number of social contacts in the different countries.

The results indicated that perceived severity and vaccination status played a crucial role in modulating the number of social contacts. More specifically, we found that individuals who had high levels of perceived severity reported fewer social contacts as compared to those who had low and neutral levels of perceived severity. The observed associations are consistent with the results found in analysis of CoMix data limited for Belgium [31], and UK [35]. The Belgian CoMix study encompassed two longitudinal surveys, one between April 2020 and August 2020, and the other between November 2020 and April 2021. The results indicated that in the first survey, participants who had low and neutral levels of perceived severity reported 70% and 56% more contacts as compared to those who had high perception of severity. Whilst in the second survey, the participants with low and neutral perceptions on perceived severity reported 62% and 76% more contacts than those who had high perceived severity. The UK CoMix study [35] on the other hand utilized data collected between March 2020 and March 2021. Applying clustered bootstrapping to obtain the mean number of social contacts, the results indicated that participants who had low levels of perceived severity reported more contacts than those who had high levels of perceived severity. The similarity in the observed relationships in these studies further highlights the crucial role of perceptions in modulating social contacts.

The predicted number of contacts in individuals who were not vaccinated against the SARS-CoV-2 were lower than individuals who had received a vaccine. These differences were consistent in all the countries included in this study and is also consistent with a recent study that found that vaccinated individuals generally had more contacts than the unvaccinated [52]. This implies that vaccination against COVID-19 played a crucial role in shaping the social contact behaviour during the COVID-19 crisis. This could be due, among other factors, to the minor restrictions that vaccinated individuals were subjected to in several European countries, where the vaccination certificate was one of the necessary condition to access public or indoor areas. In such a circumstance, the increase in contacts could be related also to the higher potential of social interactions of vaccinated individuals. However, as our analysis showed that individuals change their risk perception after vaccination, this could indicate the possible influence of Peltzman effect, whereby individuals alter their risk perceptions after access to preventive measures, leading to greater engagement in riskier behaviour (e.g., neglecting preventive measures in terms mask-wearing and social distancing) [53‐55]. Thus, the increase in contacts is likely due to a combination of greater potential for social interactions and spontaneous behavioral change. These apparent changes in behaviour following COVID-19 vaccination could have important implications in the context of the COVID-19 pandemic which has been characterized by the emergence of variants of concern such as Omicron and its sub-variants [17, 18]. These mutations initiated uncertainties in the effectiveness of vaccines in conferring protection to the vaccinated [36]. Thus, an increase in social contact behaviour following vaccination could be disastrous for viral transmission dynamics in the event of emergence of an immune escaping variant.

Our results also indicated relatively little variation in social contact behaviour between countries, once all other confounders are accounted for. This low variation could be a result of the stringent measures implemented to limit the number of social contacts during the different phases of COVID-19 pandemic in the different countries. Interestingly, we observed substantial heterogeneity in social contact behaviour between individuals. This suggests that individual-level factors played a more substantial role in influencing contact behaviour than country-level factors. The underlying heterogeneity in social contact behaviour is consistent with results from social contact studies conducted before and during the COVID-19 pandemic [56, 57].

In our study, the relationship between perceived susceptibility and perceived risk to vulnerable with the number of contacts did not yield similar patterns across different countries. This warrants more research in order to gain insights on their relevance in influencing social contacts in the context of COVID-19. From the analyses of the perceptions, we found that perceptions of susceptibility to infections were in general the lowest in comparison with perceptions on severity and risk to the vulnerable. This is indicative of the possible presence of optimism bias, a situation characterized by individuals tending to under-estimate the probability of acquiring infections in the context of infectious disease epidemiology [58]. The lack of significant relationships between the perceived risk to vulnerable could be a result of the subjective nature of perceived risk to others. Participants who do not have vulnerable individuals in their social networks might perceive low risk as compared to those that usually interact with vulnerable individuals. A study performed during the COVID-19 revealed that individuals perceive different risks for COVID-19 on their own health as compared to others such as family, friends and the general community [59]. Another study showed that individuals who were more concerned about spreading COVID-19 to vulnerable people made more contacts at home [60]. Thus the lack of association between the perceived risk to the vulnerable and social contact behavior could suggest a more diverse meaning for this risk perception construct by different people given different experiences and interactions with vulnerable people in the community.

Our study explicitly used the number of social contacts as the response variable and considered three separate constructs related to the risk perceptions. Other studies have shown that COVID-19 risk perceptions play a crucial role on adoption of protective health behaviours [25‐32]. Although most of these studies are conducted in individual countries, a recent multi-country cross-sectional study by Dryhurst et al. [28] conducted in ten countries across Europe, Asia and America found similar associations between risk perception and the adoption of the protective health behaviours. It is worthy mentioning that most studies exploring the influence of risk perceptions on adoption of protective behaviours use cross-sectional data yielding insights on only one time point and thus the dynamical aspects of the changing pandemic situation is not taken into account. Conversely, a recent study found that risk perceptions played a significant role on the adoption of recommended health behaviours over time in the UK [29]. Thus a novelty of our study, that employed multi-country longitudinal data, was to confirm such a relation in an evolving pandemic situation and for different countries. Furthermore, as social contacts can be used to inform models of infectious diseases, our results on risk perceptions and behavioural changes following vaccination can be incorporated in future mathematical models for a more granular dynamical exploration during a pandemic.

Our work is subjected to several limitations. The reporting of the number of social contacts was done retrospectively, hence could suffer from recall bias. However, such an effect is expected to be small since participants reported contacts in the day preceding the survey day. Due to the longitudinal nature of data collection, participants could experience response fatigue posing concerns on the quality of the data collected. Assessing the possible presence and subsequent influence of response fatigue will be studied in future. This study utilized a multi-level generalized linear mixed effects model. However, there have been concerns about the appropriateness of multi-level models when utilizing multi-country data where the number of the countries is relatively small (i.e, 25 for linear models and 30 for logit models) and the number of individuals per country is large [61]. We expect a small impact on the reliability of the estimates of our individual level effects, as the relatively small number of countries only affects the estimates of the country level predictors. The study only relied on the Oxford stringency index which gives varying weights to the diverse NPIs and collapses them into a single composite index. Future work can compare analyses using different NPIs databases, including the Response Measures Database by the European Centre for Disease Prevention and Control (ECDC) and the European Commission’s Joint Research Centre [62]. Lastly, it is crucial to mention that although the surveys in each individual country were representative in terms of age, gender, and also region of residence in the panel of participants considered, the optional participation in each subsequent round of data collection could suffer from self-selection bias. The average participation was generally high in all the countries.