Background

Although the twenty-first century saw a rapid decline in global mortality attributable to infectious diseases, they continue to account for high morbidity and mortality in low-income countries [1]. Epidemics of infectious diseases may arise and propagate faster than before [2] due to increased social mixing and exposure to wild animal reservoirs and challenges with timely detection and containment [3, 4].

A 2016 analysis suggested that 22 of the 25 most epidemic-vulnerable countries are in Africa, particularly concentrated across the Sahel region, and that vulnerability correlates with recent or ongoing conflict [5]. Low-income countries are generally the least well-prepared [6], particularly in regions at elevated risk of emerging zoonotic infections [7].

An individual’s subjective judgement of a health threat—or risk perception—is central to key health behaviour change theories, including the Health Belief Model [8], the Protection Motivation Theory [9], the Extended Parallel Process Model [10], and the Risk Perception Attitude framework [11]. These theories generally assume that risk perceptions are an essential precursor of protective health behaviours. While this assumption has not been consistently borne out in individual studies, meta-analyses suggest a modest to moderate influence of risk perception on health behaviours [12,13,14,15].

Three theoretical approaches seek to explain risk perception. The psychometric paradigm in psychology theorizes that cognition and emotion play a role in the formation of risk perceptions, by influencing information processing and judgment for decision-making [16]. The cultural theory of risk in sociology and anthropology posits that risk is non-objective and perceptions are determined by an individual’s sociocultural reality [17]. Among multidisciplinary models, the Social Amplification of Risk Framework (SARF) ties technical assessments of risk with psychological, sociological, and cultural perspectives, modulated by social and individual factors [18].

Risk communication is a fundamental intervention in epidemic responses, and is defined by the World Health Organization as the “the real-time exchange of information, advice and opinions between experts, community leaders, or officials and the people who are at risk”, with the implicit assumption that this process will instigate appropriate individual perceptions and inform behaviour [19]. However, risk perceptions are subject to other influences [20], including individual numeracy [21], prior experiences and imminence of the threat [22].

Highly epidemic-vulnerable countries are likely to also experience insecurity, poverty and underperforming health services [5, 6]. Here, populations and responders are often confronted with concurrent and competing risks to life, against limited resources [23]. This context is likely to influence how the general population health professionals tasked with their care and protection, perceive and make decisions about health risks. Studies in low-income multi-hazard contexts indicate that environmental risk perceptions and prioritisation are influenced by hazard characteristics (e.g. chronicity) [24], individual factors (e.g. socioeconomic status) [24], and collective coping capacity [25, 26], and that risk perceptions vary within groups and over time [26].

There is insufficient evidence on the effectiveness of epidemic risk communication interventions in low and middle-income settings [27]. A thorough understanding is required of how risk perceptions of epidemics are constructed by individuals from the general population and health professionals, the factors influencing these risk perceptions and how they interact in a context of high vulnerability to epidemics. Such insight is essential for informing effective and contextualised epidemic risk communication interventions. This review aimed to examine the existing evidence on risk perceptions of epidemic-prone diseases among the public and health professionals in highly epidemic-vulnerable countries. We also examined how risk perception has been conceptualised and measured by researchers in these settings.

Methodology

The review is designed and reported as per the PRISMA Statement [28]. The inclusion and exclusion criteria are described in Table 1.

Table 1 Inclusion and exclusion criteria
Full size table

Search strategy, study screening and selection

We searched seven bibliographic databases to cover the multiple disciplines of risk perception research: EMBASE, Global Health, MEDLINE, PsycINFO, Africa-Wide Information, CINAHL Plus and Web of Science. The search terms covered three concepts: risk perception, epidemic-prone diseases, and eligible countries. Since the concept of ‘risk’ does not translate directly into many languages spoken in the targeted countries, we included search terms for ‘risk perception’ that have been used to study similar concepts, or that hold neutral (e.g. likelihood) or positive connotations (e.g. safety). We also perused related systematic reviews to identify additional synonyms for these concepts [29,30,31]. Additional file 1 shows the detailed list of search terms and search strategies used. The search was not limited by language although data extraction was limited to English results. The search was restricted to citations published on or after January 2011, and was conducted on 28 December 2021.

We exported all citations into EndNote (Version X9, Clarivate Analytics, Philadelphia, United States of America) for screening and selection. This phase was carried out by the first author (NA) in four stages: automatic and manual removal of duplicates, screening of titles and abstracts of search results to remove ineligible studies, reviewing the full-text articles of search results to remove ineligible studies and final paper selection. When it was unclear whether or not an item met eligibility criteria during screening, the reviewer erred on the side of caution and the item was carried into full-text reviewing. The results of the screening and selection process are presented in Fig. 1.

Fig. 1
figure 1

Results of study screening and selection process

Full size image

Data extraction, quality appraisal and analysis

We extracted the following variables from each eligible study into an Excel database: information about the epidemic-prone disease(s) under study, characteristics of study population(s), study aim and objective, concept or definition of risk perception, study design and data collection method(s), results, conclusions and quality of studies.

We assessed the quality of papers using three tools: the Appraisal tool for Cross-Sectional Studies (AXIS tool) for cross-sectional quantitative studies [32], the RATS guidelines for qualitative research [33], and the Mixed Methods Appraisal Tool (MMAT) for mixed method study designs [34]. The quality appraisal tools served to highlight the strengths and weaknesses of the studies to assist in the interpretation of the findings, but no studies were excluded following quality appraisal. We used a narrative meta-synthesis and summary of the evidence to analyse the studies, due to the heterogeneous nature of eligible study designs which did not lend itself to formal meta-analysis. We categorised eligible studies into groups as shown in Table 2.

Table 2 Categorisation of eligible studies for analysis
Full size table

Analysis was guided by themes from the SARF [18]. The main premise of the framework is that portrayal of a risk source (e.g. an epidemic-prone disease) and a risk event (e.g. an epidemic and its response) interacts with psychological, social, cultural and institutional processes in ways that may lead to attenuated or amplified risk perceptions [35, 36]. The SARF provides a common terminology for comparing studies from varying disciplines, diseases and populations [37]. We described epidemic risk perceptions levels as ‘high’, ‘moderate’ or ‘low’ according to the scales used in individual eligible studies. We described all factors determined as associated or not associated with epidemic risk perceptions by individual eligible studies, and we organised presentation of factors by components of the SARF. Separately, we described conceptualisations and measurement approaches for each dimension of risk assessment (groups B above).

Results

Description of eligible studies (n = 56)

We identified fifty-six eligible studies, described in detail in Table 3. Data collection for the studies included in the review occurred between 2008 and 2020, and thirty-seven studies collected data during an active epidemic. Fifty-five studies were cross-sectional, forty-five collected quantitative data, six collected qualitative data and five used mixed methods. The majority of studies were on either Ebola virus disease (EVD) (n = 19) or coronavirus disease (COVID-19) (n = 18). Three studies compared the risk perception of two or more pathogens in the same population [38,39,40]. Thirty-three studies measured only one of the four dimensions of risk perception; perceived likelihood of infection was the most frequently reported-on dimension. The main features of the eligible studies are summarised in Table 4.

Table 3 Description of eligible studies (n = 56)
Full size table
Table 4 Main features of eligible studies
Full size table

Below, we summarise themes related to risk perception and factors influencing risk perceptions, for the general population and health professionals separately.

Epidemic risk perceptions among the general population (n = 41)

Forty-one studies included measurement or description of risk perception of epidemic-prone diseases among non-expert populations. Regardless of countries, diseases under study or whether there was an active outbreak at the time of the study, participants tended to report a high perceived severity of epidemic-prone diseases, generally above the midpoints of severity scales used by researchers [39, 41,42,43,44,45,46,47,48,49,50,51,52]. In contrast, perceived personal likelihood of contracting an epidemic-prone infection varied across studies, from low [52,53,54,55,56] to moderate [57,58,59,60,61] and high [42, 43, 47,48,49, 51, 62,63,64]. This variation persisted across countries, diseases under study and whether there was an active outbreak at the time of the study. For example, two COVID-19 studies in Ethiopia in 2020 reported contrasting levels of perceived likelihood [54, 62]. However, perceived likelihood of personally contracting infections tended to be lower than perceived severity in studies that measured both aspects of perceived severity [45, 46, 51, 52].

Another theme was a pattern of perceiving risk of epidemic-prone diseases to others as higher than to self, and that the risk to distant individuals or communities is higher than to closer ones. For example, in a study in Sierra Leone, participants perceived the threat of EVD as highest for the country, followed by the district, community then household [50]. Another study of perceived zoonotic infection risks among market vendors also showed a perceived lower risk of infection to self, compared to the rest of the general population [65].

Among groups with a higher risk of exposure to epidemic-prone diseases, perceived likelihood of infections appeared lower than among the general population. For example, among suspected cholera patients, only a quarter thought they were at high risk of contracting cholera again—even where researchers found high levels of water contamination in their households at the time of the study [66]. Similarly, two studies showed that bushmeat hunters and vendors had reduced perceived likelihood of EVD compared to bushmeat consumers [67], and of zoonotic infections compared to vendors selling livestock or vegetables [65].

Participants also perceived some populations groups as more susceptible to risks of epidemic-prone diseases than others. For example, both internally-displaced persons (IDP) and non-displaced host communities perceived IDPs as more vulnerable to dengue fever [52]. Similarly, adult community members perceived pregnant women and children to malaria compared to others in malaria-endemic regions [48].

Information sources and channels

Respondents who acknowledged the risk of acquiring EVD in the next 6 months during an outbreak were more likely to acquire information from their community (e.g. community leaders, friends and relatives) or new media (e.g. internet, text messages), and accessed three or more information sources. Television, radio, house visits by health workers and government campaigns, and using two or less information sources appeared to have no influence on perceived risk [68]. Two studies showed inconsistent effects of newspapers, brochures and billboards on risk perception [60, 68]. Previous community experience of disease [55] and exposure to a new and unfamiliar disease [67] were associated with increased risk perception.

Individual factors

Demographic factors showed inconsistent influences on risk perception across countries and diseases. Education level [54, 60, 62, 65, 68], disease-specific knowledge [44, 54, 68, 69], rural or urban residence [41, 42, 62], marital status [54, 62], income level [54, 62], gender [41, 54, 62, 68, 69] and age [42, 54, 62, 68, 69], variably showed positive, negative or no association with epidemic risk perception across different studies. Larger family size [62] and certain occupations [60, 62] were associated with increased perceived risk in two studies. By contrast, employment status [54] was not associated with risk perception. While no specific religion was associated with risk perception [41, 62], belief in divine or spiritual protection against harm appeared to reduce perceived EVD risk [59, 67].

Disease attributes

Disease case fatality ratios and infection risks seemed to influence risk perception, indicating the role of numeracy skills [39]. The phase of an outbreak also seemed influential: an ongoing outbreak of typhoid fever was associated with a grave concern that cases would continue to increase [39], while the likelihood of acknowledging the risk of acquiring infection decreased as an EVD outbreak progressed [68]. Some disease attributes were associated with an increased risk perception among participants, specifically diseases perceived as hard to control through community infection control measures [38], unfamiliar diseases [67], and severe diseases [39, 67]. Participants cited multiple features of evident disease severity, such as rapid spread, unpredictable nature, severe or debilitating symptoms, ineffectiveness of traditional or biomedical treatments and the profound economic consequences of a debilitating illness [39, 67].

Health protective behaviours

Three studies explored the association between risk perception and a person’s belief in their ability to protect themselves from EVD, and concluded that a higher self-efficacy is associated with lower perceived risk and vice versa [58,59,60]. Another study found that vaccination against EVD lowered perceived likelihood and alleviated worry [70]. However, the relationship between risk perception and protective behaviours against EVD was not consistent; for example, one study reported that while handwashing had a positive association with risk perception, avoiding burials was negatively associated with risk perception, and avoiding physical contact with a suspected EVD case not associated with risk perception [68].

The sociocultural context

Among vendors, familiarity with, knowledge of and preference of a vendor’s own products, was associated with a reduced perceived risk of zoonotic infections. In one study of perceived zoonotic infection risks among market vendors in Lao, vegetable vendors reported that their products were “organic”, “healthy” and “natural”, and livestock meat vendors mentioned that their meat was mainly sourced from slaughterhouses with robust veterinary control [65]. For some bushmeat vendors, not being involved in the hunting and killing of wild animals seemed to be perceived as reducing their risk of zoonotic infections [65]. Another study amongst bushmeat handlers in Nigeria reported a low perceived risk of EVD and questioned the plausibility that well-established traditional uses of bushmeat, such as diet, spiritual fortification and treatment of disease conditions, could be risky [67].

In a multi-country study of the sociocultural features of cholera, the authors observed that in Kenya, respondents perceived women and children as more vulnerable to cholera compared to the general population, and suggested that this may be due to greater cultural sensitivity to vulnerability amongst the study participants, and a tendency to generalize the vulnerability of already-vulnerable population groups to include susceptibility to disease [41].

Table 5 summarizes the factors reported by eligible studies and their influence of epidemic risk perceptions among the general population, by element of the SARF.

Table 5 Factors reported and their influence on epidemic risk perceptions, by element of the SARF
Full size table

Epidemic risk perceptions among health professionals (n = 19)

Studies reporting on health professionals’ epidemic risk perceptions focused on how they perceived their own risk rather than the risk to communities they served. All studies but one concerned epidemic-prone infections that can readily be acquired in a healthcare setting: COVID-19, EVD, Marburg virus and pandemic influenza A (H1N1). Eighteen studies included clinical staff, six included non-clinical health facility staff, and three studies included community-based health workers. One study solely included medical students [71].

Health professionals generally reported high perceived likelihood and susceptibility to infections [72,73,74,75,76,77]. In three studies, however, only about a third considered themselves to be at risk [78,79,80]. All three studies were conducted during an active outbreak in Nigeria: two related to EVD, and one related to H1N1. Health professionals generally reported a high perceived severity of epidemic-prone diseases [71, 77, 81, 82], including high perceived disease severity should they acquire the infection themselves [72, 75].

When comparing clinical to non-clinical staff, the results of perceived risks were inconclusive. One study reported that clinical staff had higher perceived risk than non-clinical staff [74], while another study reported no significant difference in fear ratings of doctors, nurses, paramedical staff and non-clinical workers [83]. Similarly, the review findings were inconclusive with regards to whether health workers rated the risk to themselves as higher or lower than that of other health workers [75, 80].

Information sources and channels

Two studies reported that acquiring disease-specific knowledge, for example through training, alleviated fear among health workers and reduced their perceived vulnerability to EVD infection [72, 76].

Disease attributes

Health professionals reported disease attributes that increased their fear, specifically unusual clinical presentations [76], the rapid spread and unpredictable nature of an outbreak, and diseases without a pharmacological cure, such as EVD [72].

Within the clinical environment, health professionals reported that encounters with infected patients [76], witnessing colleagues die [76], and the potential to spread infection to others in the community [72] all increased their fear.

Institutional response

Health professionals reported a number of factors associated with the health system response that influenced perceived risk. These included indicators of institutional efficacy that alleviate fear, such as clear protocols and operating procedures for patient triage and isolation, the presence of experts and role models early on in the response, availability of personal protective equipment (PPE), rapid and early diagnostic tools, non-contact thermometers and sufficient handwashing and disinfection supplies and facilities [72, 76]. The studies also reported that access to vaccination [70], and vaccine research and development for diseases such as EVD [76] reduce perceived susceptibility among health workers.

Table 5 summarizes the factors reported by eligible studies and their influence of epidemic risk perceptions among health professionals, by element of the SARF.

Conceptualisation and measurement of epidemic risk perceptions (n = 56)

Studies applied variable conceptualisations of risk perception, as reflected in data collection instruments and wording of questions. For perceived likelihood, thirty-four studies conceived of this as research participants themselves contracting infection, while other studies asked participants about the likelihood of others getting infected. Twenty-seven studies used the term “risk” while other studies asked respondents about “possibility”, “probability”, or “chance”. Only three studies provided time windows in their questions, for example, risk over the next 6 months. Perceived susceptibility was conceptualised by two studies as the likelihood of contracting infections in the absence of preventive measures, and by another two studies as the comparative susceptibility among groups. For perceived severity, nineteen studies operationalised this as ‘seriousness’ or ‘dangerousness’ of the disease. Other studies asked participants about the likelihood of certain outcomes (recovery, survival, severe illness, death) should they be infected. Finally, for affective perception, thirteen studies measured ‘fear’ or ‘worry’. Two studies asked the research participant about emotions (e.g. fear or worry) towards their family members, and one study asked participants about the threat to their community, district and country.

Likert-type or Likert scales, ranging from 3- to 10-point scales, were by far the most commonly-used tool for risk perceptions across all conceptualisations. However, the use of neutral or ‘don’t know’ categories was inconsistent. Furthermore, some scales measured degrees or levels of “risk” while others measured respondents’ levels of agreement with statements. This heterogeneity in measurement modalities, measured aspects and wording limited comparability between studies. Furthermore, for several papers we could not ascertain the measurement method used.

Details of the conceptual frameworks, definitions and measurement of risk perception used by eligible studies are provided in Table 6.

Table 6 Conceptualisations, definitions and measurements of risk perception in eligible studies (n = 56)
Full size table

Quality of evidence

Of the fifty-six eligible studies, we graded forty as good, twelve as acceptable and four as poor quality. The results of quality appraisal of eligible papers are presented in Additional file 2.

Among cross-sectional studies (n = 45), the most common weakness was not categorising, addressing or describing non-responders, or commenting on potential non-response bias. Similarly, among five qualitative studies and two mixed methods studies, none reported on the numbers or reasons of those who chose not to participate.

Among qualitative studies (n = 5), there was generally a lack of information on the studies’ ethical procedures, such as for informed consent or safeguarding confidentiality and anonymity. Among mixed methods studies (n = 6), none adequately addressed divergences and inconsistencies between qualitative and quantitative data.

Discussion

To the best of our knowledge, this is currently the only systematic review to examine the evidence of epidemic risk perceptions in populations that are highly vulnerable to epidemics. The review highlights that, despite a moderate body of evidence, major gaps remain. Studies from only eighteen of the 62 eligible countries were identified. Diseases that cause frequent epidemics in these settings [30], such as measles or cholera, received little or no attention. This finding is similar to previous research suggesting that epidemics of common diseases are less likely to be responded to in a timely manner [84], or to be evaluated [30].

This review set out to identify how a context of frequent and often concurrent epidemics influences epidemic risk perceptions. Research on non-communicable and heritable diseases suggests that perceived risk of a disease influences the perceived risk of other diseases, and that the perceived risk does not necessarily correspond to the actual risk posed by a disease [85,86,87]. However, only three studies in our review compared the perception of two or more epidemic-prone diseases in the same population, and two studies explored the influence of familiarity and novelty of a disease on risk perception. Furthermore, none specifically explored the influence of the high-vulnerability context on epidemic risk perceptions. Our review highlights the need for research that explores epidemic risk perception construction in the broader context of living in a setting with frequent and multiple epidemics.

Factors influencing epidemic risk perceptions

The review findings suggest that the general population consistently perceived their likelihood of acquiring infections as lower than they rated the severity of diseases, and they were more likely to perceive the risk of infection to others as higher than to themselves. Occupational groups with high exposure to specific diseases, such as bushmeat handlers, reported even lower perceived likelihood than the general population, and similarly perceived the risk of infection to other members of their trade as higher than to themselves. This phenomenon of lower perceived likelihood, termed ‘unrealistic optimism’ [88] and described as a cognitive bias, is often observed in the general population across cultures [89]. Optimistic bias has been found to particularly occur in a comparative assessment with risk to others [90], and during active outbreaks [91]. Our findings suggest that unrealistic optimism among some high-risk occupational groups may be explained by the long-term and well-established uses of their products and services. Epidemic responders should consider how unrealistic optimism could hinder risk communication, particularly when designing communication strategies that incorporate social comparisons of risk.

By contrast, perceived likelihood of infection was generally high amongst health professionals, though findings were inconclusive when comparing perceived risk to self with risk to colleagues. This group mainly cited concerns about their employing institutions’ ability to create a safe and effective work environment, and the effectiveness of the broader health system response, described by the SARF as the influence of the organisational response or behaviour on risk perception modification. The influence of perceived health system disaster response capacity on risk perception has been reported among health professionals in better-resourced settings, such as Singapore, Saudi Arabia and Canada [92,93,94]. However, factors other than organisational effectiveness remain insufficiently explored. These include the socio-cultural context and different information sources and channels, particularly in conditions of scientific uncertainty about the disease in question. Risk communication interventions to modify health professionals’ epidemic risk perceptions should therefore be accompanied with measures to enhance safety in the workplace.

Our findings suggest risk perception is influenced by disease characteristics, especially disease severity, familiarity, controllability and phase of an outbreak. Analogous associations feature at the core of Slovic’s psychometric paradigm [95] and Covello’s four theoretical risk communications models [96], to describe the psychological processes of risk perception formation. However, the SARF extends this further to explain how individuals or groups select specific characteristics of the risk, interpret them and communicate them to others, and how this selection varies across different settings and risks. Our review suggests that some information sources may be more influential than others, and that this variation may be due to different sources highlighting different disease attributes in their messages. Further research is needed into why certain disease characteristics become salient in settings with frequent epidemics, and how communication channels and content may mediate the relationship between disease characteristics and risk perception formation.

Review findings suggest that evidence on the influence of demographic factors on risk perception is inconclusive. This may indicate the diversity in conceptualisation and methods of measuring risk perception used by the studies in our review. Previous research suggests that age differences in risk may vary across the domain of risk under investigation—for example, different age groups may interpret disease ‘severity’ in terms of its health, social or economic consequences and therefore give different responses [97]. Similarly, gender differences in risk perception are reported to be sensitive to methodological approaches—for example, while women consistently demonstrate higher risk perceptions for all risks, gender differences are not observed when respondents are asked to rank hazards in order of severity or seriousness [98]. The findings suggest that risk communication interventions targeting a specific demographic should account for heterogeneous risk perceptions within that group.

The review suggests that there is insufficient evidence on how epidemic risk perceptions are formed or modified in these populations. Only a third of eligible studies in our review reported on factors influencing risk perceptions. In general, there was lack of depth to the inquiry in the studies. This may be due to most studies being cross-sectional and quantitative, precluding exploration of why people perceived what they did, and how and why risk perceptions varied between diseases, populations and over time. Studies among the public primarily focused on individual constructions of risk, such as the influence of disease attributes and socio-demographic variables, but few studies explored the role of information sources and channels, cultural factors, and none studied the influence of perceptions of the epidemic response. In contrast, studies among health professionals primarily investigated the influence of institutional efficacy on risk perception. Furthermore, the studies in our reported on the independent influence of selected factors on risk perception, but none explored the interaction between these factors to shed light on the complex process of risk perception formation or adaptation. Further research is needed to explore the differences in epidemic risk perceptions between population groups, particularly the social and cultural processes that intensify or attenuate perceptions of the disease risk and its manageability.

Conceptualisation and measurement of epidemic risk perceptions

Our review finds that, while epidemic risk perceptions are measured in a moderate number of studies across disciplines, there is wide variation in the conceptualisation of risk perception by researchers. Overall, the review revealed limited engagement with the concept of risk perception and only a third used conceptual frameworks or models to situate their hypotheses and findings. The authors’ conceptualisations of risk perception were mostly deduced from the study variables, instruments or results. None of the studies acknowledged the effect of question wording on how respondents may rate or describe their perceived risk [99]. This is particularly relevant in settings where studies were not conducted in the English language, since the conceptualisation of risk varies widely across cultures and languages [100].

The operational definition of epidemic risk perception varied widely across studies, ranging from unidimensional or single item measures to multidimensional composite risk perception scores. Our findings indicate that most researchers measure one dimension of risk perception, usually likelihood, whereas only a minority measure a combination of dimensions, such as likelihood, severity and vulnerability. Few researchers combined measurements of probability judgements, such as likelihood and vulnerability, with consequential judgements, such as affect/feelings or severity. In their review of hazard risk perception measurement methods, Wilson et al. reported that almost half of studies measured only one dimension of risk perception, often perceived likelihood, and argued that this unidimensional approach is not particularly valid or reliable for understanding individual risk perception formation [101].

Even where different studies used the same conceptual frameworks or risk perceptions definitions, diverse measurement methods limited comparisons. It was difficult to interpret whether there were actual differences in risk perception between diseases, countries or populations, or whether observed inconsistencies were due to methodological design. For example, many eligible studies used Likert-type scales to capture risk perception responses, but the inconsistent use of ‘don’t know’ response categories by researchers complicated the interpretation of findings. Previous research indicates that a nonnegligible proportion of study respondents report not knowing their risk of diseases in studies, particularly in populations that are socio-economically disadvantaged or with health disparities [102].

While the vast majority of studies in our review were deemed of good or acceptable quality by standardised quality appraisal tools, in general, there was lack of depth to the inquiry. This may be due to the fact that the majority of studies evaluated in this review used a cross-sectional design, with most being quantitative studies, and therefore lacking in-depth and longitudinal exploration of why people perceived what they did, and if, how and why risk perceptions varied between diseases, populations and over time. Furthermore, the high level of heterogeneity in methods, tools and measurement scales in eligible studies prevented a definitive identification of factors associated with epidemic risk perceptions. Varying conceptualisations, definitions and measurements of health risk perceptions and behaviours have previously been shown to hamper cross-study comparisons [13, 20, 103].

Review limitations

Screening and selection were conducted by a single reviewer, and may have resulted in some eligible studies being missed. To mitigate this risk, the reviewer erred on the side of caution and included items with unclear eligibility in the second stage of screening. We did not include grey literature which may have provided additional and valuable insights, particularly publications by humanitarian responders serving populations in eligible countries. Due to the heterogeneity in outcomes and study methods, only a narrative analysis and synthesis was feasible. Furthermore, it was not feasible to contextualise all of the findings from the diverse set of epidemic-prone diseases, countries and population groups included in this review; instead, we attempted to identify and describe key themes that could be useful to researchers and epidemic responders. Finally, there were limitations posed by methodological weaknesses in a minority of included studies, mainly related to non-response, ethical considerations and a lack of information on inconsistencies between qualitative and quantitative epidemic risk perception data.

Conclusions

This review suggests that evidence on epidemic risk perception in countries at the highest risk of these public health emergencies is limited. Available studies afford some insight into patterns of epidemic risk perception and factors influencing its formation, but the quality and validity of these findings are affected by a lack of in-depth inquiry and exploration. There are several areas in particular that require more attention from researchers. First, risk perceptions of diseases that cause frequent epidemics in these settings, such as measles and cholera, should be given more attention and explored in-depth to better inform responses. Second, studies comparing perceptions of different epidemic-prone diseases in the same population, or comparing perceptions across different populations or settings are essential for better contextualisation of risk perception understanding. Third, research that adopts a comprehensive, theory-driven, and preferably longitudinal, exploration of epidemic risk perception construction is needed, particularly to situate risk perceptions in the broader context of living in a setting with frequent and multiple epidemics.

The review also suggests that the science of defining and measuring epidemic risk perception is still relatively underdeveloped. First, there is a need for promotion of best practices in measuring risk perceptions, such as the systematic inclusion of ‘don’t know’ categories in risk perception measurement scales. Such standardisation will facilitate comparisons among studies and allow for systematic accumulation of evidence. Second, more research that explores or measures multiple dimensions of epidemic risk perceptions is needed, such as studies that simultaneously explore perceived probability, vulnerability and severity.