The application of our detailed search strategy across three electronic article databases returned 42 papers focused on the incorporation of endogenous human self-protective behaviour within infectious disease models that met our stringent inclusion/ exclusion criteria. All of these papers were published between 2002 and 2015, with a spike in 2011/ 12, with a clear Western (more explicitly, North American) bias in the country of origin; only eight of the papers included in this review originated from outside of North America, with only two of these eight papers originating from outside Western Europe (both China). The majority of the included papers did not focus on a specific disease, although the most commonly modelled was influenza. Consistent with the disease focus, the most commonly modelled protective behaviour was also influenza-related (vaccination). Our included papers therefore seem to reflect a relatively recent increase in infectious disease models incorporating endogenous human protective behaviour that is likely related to the A/H1N1 pandemic.
The broad range of models employed in the included papers precludes any firm conclusions regarding best practice for modelling both human behaviour and infectious disease spread. Nevertheless, it was clear from our data extraction that a substantial proportion of papers employed a dual-model method; using compartmental models (e.g., Susceptible-Infectious-Recovered, Susceptible-Infectious-Susceptible models depending on the disease) to represent disease spread, and economic-style models/games (e.g., cost-benefit calculations) to represent behavioural decision making. Moreover, reflecting the importance of social considerations in the modelling of infectious disease spread, a number of models employed social modelling components (e.g., contact networks; behavioural imitation) to represent the spread of disease or human behaviour.
A range of different cognitive and social constructs (with an emphasis on the cognitive) contributed to the modelling of human behaviour across papers. The cognitive constructs typically focused on the relative costs and benefits of remaining susceptible or engaging in protective behaviour. These included: perceived or actual risk (i.e., of infection or vaccine complications), the social costs of protective behaviour/ benefits of remaining susceptible, and the health costs of remaining susceptible. On the other hand, the social constructs typically focused on either behavioural comparison/ imitation of others within the model, the social consequences of engaging in protective behaviour, or the normative acceptability of engaging in a given behaviour. Very few of the included papers made explicit reference to psychological health behaviour theories when discussing human behaviour, relying instead upon literature from behavioural economics and infectious disease modelling.
Finally, just under half of the papers included in our review made reference to behavioural data in their modelling. Among these papers, there were instances of more thorough incorporation of behavioural data, such as population surveys of vaccination acceptability/uptake and travel behaviour, within the included papers.
Through synthesising the outcomes of our review with the psychological behaviour change and health protection literatures, we develop three central recommendations for how modellers can ensure that human behaviour is incorporated in their infectious disease models in a realistic and representative fashion: The role of psychological theory; the importance of the social world, and the use of behavioural data.
The role of psychological theory in identifying predictors of human behaviour
Broadly speaking, the emphasis on cognitive components within the included papers corresponds well with the psychological literature on health behaviour change. For example, the Integrative Model of Behavioural Prediction includes behavioural belief, perceived risk, normative belief, and efficacy belief components [
70]. Similarly, the emphasis on cost-benefit calculations for engaging in protective behaviour is a feature of multiple psychological models of health behaviour change (for example, the HBM [
71]; Protection Motivation Theory (PMT) [
72,
73], and; the Extended Parallel Processing Model (EPPM) [
74]). The social constructs identified in our review (particularly behavioural imitation and social norms) are also represented in several psychological theories of behaviour change (e.g., Social Cognitive Theory [
75]; Social Learning Theory [
76]; Theory of Planned Behaviour [
10], and; the Integrative Model [
70]). However, as previously noted, very few of the included papers made explicit reference to these theories. As a point of reference, recent work has identified 83 theories of behaviour change from across the social sciences [
12]. Furthermore, although there is overlap in the constructs used [
12], different models have been designed to reflect contextually-specific predictors of behaviour. For example, the HBM (cited most commonly by papers included in this review [
44,
48,
54,
55]) was designed to help understand the predictors of preventative behaviour in responses to a health threat [
12], thus making it thoroughly appropriate for application within the current context. However, both PMT and EPPM were also designed to help understand predictors of behaviour in this context, but with a particular focus on emotional responses (i.e., fear [
12,
72‐
74]). There is, therefore, a wealth of theoretical literature concerning predictors of behaviour and behaviour change within the social sciences that could be drawn upon to inform the modelling of self-protective health behaviour.
Two papers cited within the current review provide an excellent example of how infectious disease transmission modelling can incorporate a more nuanced representation of human behavioural decision making. Specifically, these models combine statistical modelling (specifically logistic regression modelling based on a combination of previous literature and survey data) with agent-based modelling techniques, to present detailed models of infectious disease transmission that incorporate the HBM [
44,
48]. However, there is an inevitable compromise between striving for a realistic presentation of human behaviour, and the requirement and constraints of modelling [
44]. Thus, despite the appeal of a nuanced psychological modelling as presented in these examples, we accept that this is not always appropriate or desirable.
The theoretical literature relating to behaviour change may instead be more useful for identifying key predictors of human behaviour that have been overlooked within infectious disease modelling. Indeed, a recent paper posits that an awareness of the main factors underlying human behaviour within psychological models may be sufficient for modelling infectious disease transmission (although the authors do acknowledge the importance of further exploring this issue [
77]). By way of an example, consider the role of emotional responding within both PMT and EPPM [
72‐
74]. The role of emotions as a theoretical domain associated with behaviour change [
13], and the relationship between emotional responses (e.g., anxiety) and behaviour change within the context of the H1N1 pandemic [
78], mark emotional responding as a potentially important predictor of behavioural responses to an infectious disease outbreak. However, no articles included in our review made explicit reference to the modelling of any emotional responses to an infectious disease outbreak. One paper that fell just short of our inclusion criteria did include fear-based responding within a model, and found that relatively low levels of fear-related flight can influence the spread of an infectious disease [
32]. In other words, although it is not necessarily prudent to consistently model complex behaviour change theories in their entirety, an awareness of and familiarity with the extensive theoretical literature on health behaviour change could help infectious disease modellers to examine the efficacy of previously understudied predictors of human behaviour within future infectious disease models.
Our first recommendation is, therefore, that infectious disease modellers should draw upon the extensive psychological literature concerning the predictors of health behaviour change when incorporating human behaviour into their models. Although the explicit modelling of complex behaviour change models in their entirety may represent a gold standard for infectious disease modelling (see [
44,
48]), this may not always be appropriate [
77]. Instead, we recommend that modellers familiarise themselves with the behaviour change literature to both: a) identify previously understudied predictors of self-protective health behaviour, and; b) test the effect of incorporating these predictors into future infectious disease models on model validity. Indeed, the importance of cross-disciplinary work to inform future infectious disease modelling has recently been highlighted within the literature [
77]. Recent work by Susan Michie and colleagues to review the behaviour change literature represents an excellent starting point for this endeavour [
12‐
14].
The importance of social constructs for modelling infection prevention behaviour
Several of the included papers make a clear attempt to incorporate complex social constructs (e.g., contact, imitation, norms) into their modelling of human behaviour. However, as for traditional psychological theories of health behaviour change, this involvement is at a relatively surface level; more could certainly be done to improve the inclusion of social constructs in the modelling of human behaviour. There is a longstanding tradition of research within social psychology that tells us that individuals can be members of a wide range of social/cultural groups that are more or less important to them depending upon the context that they are in (i.e., if you are at work you may identify yourself most strongly according to your profession, whereas if you are watching a football match you may identify yourself most strongly according to the team that you support). These ideas are conceptualised formally as part of Social Identity Theory and Self Categorisation Theory (e.g., [
79‐
82], see also [
83]). More recent research in this tradition (such as that presented above) has focused on applying these theories within the context of health behaviour (‘The Social Cure’, see [
15]), and it is this literature that is of particular relevance for infectious disease modellers.
By way of example, some of the papers included in this review do incorporate social norms for behavioural uptake (based on either the behavioural uptake of an individual’s contacts or population wide behavioural incidence, e.g., [
28,
29,
54,
55]). However, we know from the literature that ensuring the relevance of social norms and recommended health behaviours to one’s salient social group is important for behavioural uptake (e.g., [
84‐
86]). For example, a study of British University students found that participants were more likely to engage in health promoting behaviour (e.g., reduced alcohol consumption) to the extent that they saw themselves as British (a comparatively healthy social grouping) rather than as a University student (a comparatively unhealthy social grouping [
85]).
Similarly, behavioural imitation typically occurs within the included papers as a function of behavioural prevalence (e.g., the most adopted behaviour across the model, e.g., [
62]) or by comparing one’s own behaviour to the behaviour of a randomly selected other (either drawn from one’s contacts [
24] or from the model as a whole [
25]). However, the decision to imitate the behaviour of another individual is likely to be contingent upon social group processes. Specifically, both the degree to which one identifies with the social group that this other individual represents within a given context, and the extent to which the other individual is valued (and so has greater influence, i.e., leaders) or devalued (and so has less influence, i.e., deviants) within this group [
87]. Based on the above, the impact of social norms and behavioural comparison/ imitation is likely to vary as a function of: a) the group that individuals see as important to them in that context; b) the group membership of the individuals within both their contact network and the population as a whole who have adopted (or recommend) a given behaviour, and; c) the extent to which these other individuals are influential within a given group.
A recent paper outlining recommendations regarding the incorporation behavioural dynamics into infectious disease models has indicated the need to better understand the mechanisms underlying the relationship between behaviour and infectious disease dynamics. Specifically, the authors ask “To what extent do people themselves, their social “networks”, media opinion leaders, or health care providers affect individual behaviour?” [
77], p.25). We suggest that insights from the literature outlined in this section could help to develop models designed to answer this question. One method of achieving this could include more detailed stratification of social grouping (with a consideration of the relative importance of different groups) within a modelled population. Surveys containing questions that ask individuals to list social groups that are important to them may be one method of obtaining a more accurate understanding of the distribution (and importance) of social groups within a given population to help parameterise these models.
As previously discussed, we are aware of the tensions between theoretical fidelity and the need for model simplicity [
77]. It is, nevertheless, important to ensure that models are sufficiently realistic with regard to social and epidemiological processes to allow for the accurate exploration of potential control policies [
4]. For example, the assumption of homogenous or random mixing may be inappropriate for diseases that are transmitted via close contact [
4]. To resolve this tension, we extend a recommendation made by Funk and colleagues when considering the extent to which behaviour should be modelled explicitly [
77]. That is, we recommend that modellers interested in exploring the interplay between behaviour and disease dynamics should develop a range of models into which social constructs of varying complexity are incorporated, with the resulting outputs compared. As with the previous recommendation, this endeavour is consistent with the importance of cross disciplinary dialogue for developing future models [
77], and literature relating to the ‘Social Cure’ [
15] would be our recommended starting point.
The use of behavioural data
Over half of all of the included papers made explicit reference to data concerning human behaviour in the development of their infectious disease models. Of particular interest are the papers that made use of in-depth data sources to inform the modelling of human behaviour, including: epidemiological data concerning vaccine uptake (e.g., [
38,
43]); travel survey data (e.g., [
28]); census data (e.g., [
53]), and; health behaviour surveys (e.g., [
28,
44,
48]). By using this detailed behavioural data, modellers can help to ensure the realism of their assumptions concerning human responses to an infectious disease outbreak. Our third recommendation is for modellers to ensure that the presentation of human behaviour within infectious disease models is based on appropriate, detailed behavioural data. Although the self-report data collected by Mao and colleagues represents a good initial step towards incorporating behavioural data in infectious disease models, this data only assesses behavioural intentions rather than actual behaviour. Ideally, behavioural data should be observed directly within a target population during an infectious disease outbreak in order to ensure that the modelled behaviour is appropriate and relevant for the target population. This echoes a recommendation made in another recent review of the infectious disease modelling literature [
21], thus underscoring the importance of using good behavioural data within this context.
Limitations & further considerations
Despite the detailed, in-depth nature of our review, there are inevitably limitations and further considerations that need to be borne in mind while considering our results and recommendations. First, there were methodological limitations necessitated by time and resource constraints. While we did ensure that the search strategy was identical across all three electronic databases (using the HDAS search system), we did not optimise the thesaurus terms for each individual database. Furthermore, we did not conduct forward and backward citation searching of all included papers. Despite this, our search still revealed 1988 papers (excluding duplicates), with a total of 118 papers being subjected to the final full-text assessment (following the brief full text review occurring throughout our iterative screening process). We therefore believe that the scale and nature of our search and review was entirely appropriate given our emphasis on mapping and collating the extant literature rather than producing a full systematic assessment. In addition, we were not able to achieve full multiple-review of all of the papers retained for full-text analysis. There were, however, several iterations of our review strategy, and we did subject 50 papers that the first author was unsure over to review by multiple researchers. As mentioned previously, the resulting discussions over these 50 papers further contributed to the iterative development of our final inclusion/ exclusion criteria. Moreover, the primary reason for exclusion of papers at full-text stage is presented in Fig.
1 (the details of which individual papers were excluded for which reason are available from the first author on request).
Second, given the large number of relevant papers identified following the title/ abstract check it was necessary to concentrate our review; we chose to focus on endogenous, individual self-protective behaviour. By narrowly focusing our review, it is possible that we missed out on other interesting attempts to model human behaviour in the context of infectious disease spread. For example, our criteria precluded the inclusion of papers concerning the treatment of sexually transmitted disease and papers concerning the role of human behaviour in the spread of vector-borne diseases. Interestingly, our initial search strategy was designed to capture the full array of academic literature concerning the modelling of human behaviour in response to the spread of an infectious disease. It would therefore be possible for our dataset to be used to easily conduct reviews within these (and other) contexts in future.
Thirdly, as mentioned previously, there is a Western bias in the country of origin for the vast majority of all included papers. Given the cultural homogeneity in our sample, it is important to be aware of the impact that potential cultural differences in responses to emergency situations might have on the development of an infectious disease model. In much the same way as social group membership might impact upon behaviour during an infectious disease outbreak, other research has suggested that there may be ethnic or cultural differences in willingness to engage in health-related behaviours. For instance, research by Daphna Oyserman and colleagues found that racial-ethnic minority participants saw healthy behaviour (e.g., healthy eating) as behaviour that middle-class White individuals (and not themselves) engage in [
84]. It is, therefore important for future modelling work to carefully consider the potential influence of both social and cultural influences on human behaviour in the aftermath of an infectious disease outbreak.
Finally, we are aware of two further reviews which together examine the incorporation of human behaviour within infectious disease models over the same time period as the current review [
20,
21]. Although there is some overlap in the data extracted and conclusions drawn (particularly concerning the importance of behavioural data for parameterising models [
21]), our review approaches the issues from an alternative perspective. Our primary emphasis is not on the precise mechanisms involved in infectious disease modelling, but is instead on contextualising the models within the extant psychological literature to provide recommendation for how this literature might be incorporated into future modelling efforts. By focusing more generally on the behavioural constructs that are currently modelled across the literature and how these relate to the psychological literature, our review presents a complementary analysis of the behavioural modelling literature from an explicitly psychological perspective. Given these differences, we therefore recommend that infectious disease modellers who are interested in incorporating human behaviour into their models should draw on all available reviews when attempting to develop future models incorporating human behaviour.