Low to medium-low risk perception for dengue, chikungunya and Zika outbreaks by infectious diseases physicians in France, Western Europe

For this study, the French Infectious Diseases Society (SPILF) kindly helped us by sending to the 685 hospital physicians registered on their “Infectio-flash” Discussion List a questionnaire we developed on their perception of the current and future (10-year) risk of introduction, sporadic cases and epidemics for DEN, CHIK and ZIKA, three important vector-borne diseases that are transmitted to humans by two species of mosquitoes (Ae. aegypti and Ae. albopictus) [28]. We only considered mainland France, and excluded all French overseas territories in order to focus on the risk of disease emergence in mainland France, where Ae. albopictus has settled during the last couple of decades, and Ae. aegypti is absent. The presence of Ae. albopictus in France is monitored at the departmental level [8]. We selected these three diseases due to: i) the strong human transportation connections between mainland France and its tropical overseas territories: French Guiana and French West Indies (DEN, CHIK and ZIKA), Reunion Island (DEN, CHIK), and French Polynesia (DEN, CHIK and ZIKA); ii) the huge numbers of international tourists visiting France each year (up to 83 million visitors in 2016); and iii) the social, economic and political impacts of the CHIK epidemics on Reunion Island in 2005–2006, and in Emilia-Romagna, an Italian region close to southeastern France. All these conditions render certain regions of mainland France potentially vulnerable to these new emerging diseases.

The questionnaire (see Additional file 1: Appendix I) was uploaded on a dedicated Google Forms website and the link was sent to all ID physicians registered with SPILF. Members were informed of this scientific investigation and its issues, were totally free to reply to it, or not, and when replying to the online questionnaire consented to the terms and conditions of this study. The questionnaire was completely anonymous, and practitioners were referenced with a personalized digital code. We did not collect sensitive data, in accordance with current ethics rules (see at: https://ethiquedroit.hypotheses.org/1717#more-1717; see also article 89 from the European rules, April 27th 2016 and article 40.II January 6th 1978).

The sample population (see Additional file 1: Appendix II for further details) was 47,5 years old on average (47 years old for the overall SPILF population, as of 2018], with a median age for MD thesis dissertation being 31 years old (30 years old), a sex ratio of 61/39 (male/female) (48/52), and training in infectioliogy and duration of internship of 1 year; 4 to 5 years of medical specialization plus 1 year of post-internship (for people trained between 1984 and 2017 which included all respondents to the questionnaire (from 2018 to the present, training is only 5 years and infectiologists receive an educational degree called a DES diploma).

Pre-versions of the questionnaire were sent to different public health authorities and medical staff members (regional public health agency - Occitanie, Santé publique France, welfare system - Paris Hospitals, regional hospitals...) in order to improve questions’ accuracy and intelligibility. Even if our questionnaire was not pre-tested on a subset of participants, its validity and reliability were determined according to feedback exchanges on improvements to the questionnaire with these different public health and medical personnel.

Questions and sub-questions were grouped into eight different categories: i) estimation of the total number (current and future) of imported DEN, CHIK and ZIKA cases, in the department and nationwide; ii) perception of sporadic autochthonous case development of DEN, CHIK and ZIKA (current and future) in the department and nationwide; iii) global perception of autochthonous epidemic events of DEN, CHIK and ZIKA (current and future) in the department and nationwide; iv) estimation of the level of concern about the risk of sporadic DEN, CHIK and ZIKA cases (current and future) in the department and nationwide; v) estimation of the level of concern about the global (all three diseases together) risk of epidemic events (current and future) in the department and nationwide; vi) perception of the severity of the clinical consequences (symptoms, complications, mortality…) of DEN, CHIK and ZIKA epidemics; vii) perception of the socio-economic impact of DEN, CHIK and ZIKA; and, viii) qualitative estimation of the communication by public health authorities on DEN, CHIK and ZIKA.

All the estimations were rated on a 10-point Likert scale [29] with 0 being the lowest and 10 the highest level. Reliability of the study instrument was determined using Cronbach’s α. Because we have several questions that are heavily dependent on some core questions, we calculated Cronbach’s α by two different means. Cronbach’s α taking into account all questions yielded a value of 0.938 (number of items is 72) and Cronbach’s α using only main questions yielded a value of 0.789 (number of items is 58), suggesting that the items in our questionnaire have relatively high internal consistency.

First, univariate regression models were used to investigate the relationships between the current and future risk as perceived by hospital practitioners and the different independent variables extracted from the questionnaire (see Table 1), and this for all three diseases. Second, scatter diagrams were used to visualize the plot distribution between current (x-axis) and future (y-axis) disease risk perception for the three infectious diseases. The locally-weighted scatterplot smoothing (LOWESS) non-parametric regression method was used to characterize the main perception trend. A flat plot distribution indicates the perception of a future low/absent disease risk; conversely, a plot distribution near or above the x = y line indicates a future risk perception equivalent to or higher than the current one. Non-parametric and parametric tests were used, when adequate, to evaluate correlations between responses and explanatory variables [30].

The relatively small sample size of respondents prevented using many multivariate analyses. However, general linear (GLM) and Generalized Linear Model(s) were used to analyze the influence of the different explanatory variables, and tentatively their two-way interaction terms, on the perception of future risk by developing null and minimal models [31]. In the GLMM models, variables, such as age and date of medical degree, were used as random variables, and other variables were used as fixed factors. As we did not want to produce the best-fitted explanatory models for the future disease risk perception, the dependent and independent variables were kept untransformed in multivariate models. However, the normality of distribution and homoscedasticity were checked with the Shapiro’s test. To relate the future risk perception variables to independent factors, a Gaussian and a Poisson error model were used, and factors and their interaction terms were selected by using a backward-forward stepwise elimination procedure from the general models and according to the Akaike Information Criterion (AIC) [31]. Variables were selected using the analysis of variance (ANOVA), with tests specified as “type-III” to assess the effect of each variable after accounting for all other factors [32].

The robustness of our results relative to sampling heterogeneities was tested using a modified rarefaction analysis. Random samples were generated that contained from 40 to 100% of all questionnaire data for each of the three arboviral diseases. The random sampling was repeated 10 times, and the primary analysis was run using each of these random samples. This allowed us to test the robustness of each result and exclude findings that were significant only due to the presence of outliers.

All analyses were performed using Systat ver. 13.1 (Systat Software Inc., CA) and S-Plus 4.5 (TIBCO Software Inc., CA).

The questionnaire was accessible on line between January and May 2016, and we received 80 replies (11.7% of 685). Among the 80 respondents, 33 replied to all questions, 27 omitted between 1 and 3 questions, 8 did not reply to between 4 and 6 questions, and 12 did not reply to ≥7 questions (mean number of omitted questions ± SD = 4.687 ± 9.832). This response rate was considered to be normal for this type of questionnaire, although not totally satisfactory for the category of interviewed professionals.

Their estimation of the current and future (10-year span) imported and autochthonous cases and epidemics nationwide and within their department are described in Table 1.

By plotting the current disease risk perception (x-axis) against the future disease risk perception (y-axis) (Fig. 1), we found, for all three diseases, a flat relationship for the sporadic autochthonous cases both at the departmental and national levels (Fig. 1a). By using linear parametric or LOWESS non-parametric regressions, the perception of future disease risk by ID physicians was always flat, indicating no trend for an increasing disease risk over time (Fig. 1a). Nevertheless, we observed a broad dispersion of the responses for the perception of future risks, with a higher dispersion when the current risk perception values were lower than 4 compared with values higher or equal to 4, except for imported cases of dengue at the national level (see Fig. 2 for further details). With higher levels of current risk perception, the future risk perception responses tended to become more homogeneous towards low to medium-low scores. Conversely, the perception values of the future global disease epidemic risk (all three diseases together; y-axis), both within the department and nationwide (Fig. 1b), tended to increase with the increase of the present risk perception values, with curves that more or less followed the x = y diagonal line. This last finding indicates that ID specialists are aware of the risk of epidemic appearance, but they are at the same time unable to qualify the type of risk, e.g., which specific category of arboviral diseases will spread.

For clarity, and because our study generated many different statistical analyses and results, we decided to summarize the main findings in Table 2, and to focus on some specific results. Table 2 lists the main explanatory variables retained in both null and minimal linear models, based on their coefficient values, for the three distinct scenarios of arboviral disease appearance and propagation (from imported cases to autochthonous cases, and finally epidemic state for DEN, CHIK, and ZIKA), at the department and national levels. Over all, we observed that training activities and NPS awareness by respondents were, in most situations, the most important parameters by which to explain the different gradations of disease risk perception. The only exception was the perception of autochthonous case scenarios within the respondent’s department for which the presence of an international airport, the presence of tiger mosquitoes, and the identification of existing autochthonous cases were the best explanatory variables. For the epidemic state situation, longitude was an additional explanatory variable at both department and national levels. This is explained by the respondent’s distance from oceanic or Mediterranean regions that are more favorable for the establishment of a tiger mosquito population.

This is the first study on the perception of the present and future risk of vector-borne diseases (i.e., dengue, chikungunya and Zika), in a Western European country, based on an electronic survey completed by French ID physicians in 2016. Many works have focused on disease risk perception in the general population, notably in case of new emerging infectious diseases such as DEN and CHIK [33‐36], H1N1pdm flu [37‐39] and H5N1 avian flu [40‐42]; others have focused on general practitioners in France [43‐45], both general population and practitioners [46], French pharmacists [47] or risk perception in Europe and other countries worldwide [48‐50]. The appearance of various emerging infectious diseases during the last two decades (e.g., chikungunya, SARS-CoV, MERS-CoV, Ebola virus, Zika), as well as that of antibiotic-resistant bacteria, has stimulated research on risk perception in the general public and policy-makers [51‐55]. Information delivered by the media has exacerbated the general feeling, in the public and in national and regional deciders, of the importance of being able to have rapid access to clear information about disease propagation conditions, and to deliver reassuring statements to the population [56, 57]. Conversely, studies on perception risk in the different categories of health professionals, particularly those who directly deal with such diseases and infected patients, are today still very rare, or even absent.

Unexpectedly, the many different statistical models used in this study indicated that environmental, geographical and epidemiological explanatory parameters were not as important as professional training, tropical work experience and NPS awareness in explaining present and future disease risk perception in French ID physicians.

Strangely, the presence of tiger mosquitoes in different departments in the south and southwest of France was not retained as an explanatory variable in minimal models, in most cases. Indeed, the perception of arboviral disease risk was, on average, no higher among respondents working in a hospital located in a department colonized by tiger mosquitoes than among those working in the north of France, where this vector species is absent. The variable “presence of tiger mosquitoes” was significant for the analysis of risk perception only at the departmental level and for CHIK, particularly when studying the difference in risk perception between physicians working in the Hérault department where CHIK cases occurred in 2014 [58] and those from all the other departments. The presence of human autochthonous cases of DEN and CHIK during the previous years in different French departments (see Additional file 1: Appendix III) was an explanatory variable for disease risk perception at the department level, but only for Zika (see below). Moreover, like for the presence of tiger mosquitoes, latitude was never an explanatory variable in regression models, thus indicating that disease risk perception by this category of health professionals is not sensitive to a north-south gradient. Conversely, the perception of the risk of DEN, CHIK and ZIKA epidemic spread was higher among respondents working in departments located on or close to the Mediterranean or Gulf of Biscay coasts than among those working in departments close to Germany or Switzerland. Surprisingly, the presence of an international airport in the respondent’s department was not an important explanatory variable for the disease risk perception, except for DEN and ZIKA at the departmental level. This could be explained by the fact that our questionnaire coincided with the onset of the ZIKA epidemic in Latin America [59, 60] and information on ZIKA risk was at that time delivered by the French health authorities to health professionals and travelers on the occasion of the Olympic Games in Brazil, August 2016 [61]. For the year 2016, the World Health Organization also identified major DEN outbreaks in different parts of the world (South America, Philippines, Malaysia, Salomon Islands, Burkina Faso). This could also explain the effect of the ‘presence of human autochthonous cases’ variable on risk perception for ZIKA outbreaks locally. However, this does not explain why the presence of an international airport was not retained as an explanatory parameter for the risk of DEN, CHIK or ZIKA case introduction from abroad.

Most of the linear models indicated that professional training components (medical school cursus, practical experience and current medicine activities) were often significant variables for explaining disease risk perception in health professionals. Having an infectiology or additional training specialty (e.g., travel medicine, epidemiology) strongly influenced their answers in minimizing their perception of arboviral disease risk. This was particularly true for the estimation of imported case risk at the departmental level, and the perception of autochthonous case and epidemic risk at the national level. Moreover, professional experience in tropical regions was an important parameter for explaining perception of arboviral disease risk in mainland France, with generally a lower level of risk perception for respondents with tropical healthcare experience. Many French practitioners traditionally spend time in French overseas territories and developing countries during and after their medical studies. NPS awareness in mainland France was also an important parameter, notably for the imported case scenarios at the departmental and national levels. Undoubtedly, knowledge of the information delivered by the NPS on local disease surveillance and diagnostic practices made respondents more aware of the real situation and lowered the perception of arboviral disease risk compared with ID physicians not aware of the plan. Thus, NPS awareness tended to make respondents more confident about their perception of risk and homogenized the questionnaire responses towards lower risk levels.

In some linear models (Table 2), the respondent’s birth year and year of medical degree and their two-way interaction terms with infectiology specialization or with NPS awareness also were important explanatory factors for disease risk perceptions. In particular, the year of medical degree was retained in the regressive models for the perception of imported case risk and, to a lesser extent, of the epidemic risk at both departmental and national levels. The importance of the birth year and year of medical degree suggests that ID physicians of different graduating classes could have received different specialized training on arboviral diseases, with older doctors giving generally lower scores. Alternatively, the younger generations of practitioners are more sensitive to emerging threats due to the recurrence of these events in the last 2 to 3 decades and their significant media coverage. In addition, the two interaction terms birth year×infectiology specialization and birth year×NPS indicated that health professionals without infectiology specialization and who were born in or after 1972 tended to give higher scores (p < 0.0001) than the rest of the responders, and that those without NPS awareness and born in or after 1972 tended to give lower scores (p < 0.0001). Overall, this suggests that the initial university training strongly impacts current training and professional awareness on risk perception.

The dispersion of the values for future sporadic cases (Figs. 1a and 2) suggests that although respondents perceived the current risk of arboviral diseases in mainland France as very low, they imagined all plausible scenarios for future sporadic case risk (from very low to high). Moreover, for higher perception values of current sporadic case risk, the values for future risk tended to converge towards low to medium estimates for the three arboviral diseases and the two scales (see Fig. 1). Nevertheless, we are conscious that a considerable reluctance to extrapolate about the future may exist when “nothing or near nothing” can be perceived today. On the other hand, the future disease risk perception values (Fig. 1b) tended to increase with the increase of the present risk perception values. Two clear patterns of arboviral disease risk perception in mainland France appeared in this study: i) respondents tended to weigh down the future risk of DEN, CHIK and ZIKA sporadic cases in a context of major uncertainty; and ii) they estimated a high level of future epidemic risk. These differences could be explained by the fact that specialists consider themselves and the national authorities effective in controlling the appearance and spread of sporadic disease cases, whereas they see as more limited their capacities to control an epidemic. In addition, for the epidemic scenario, the three diseases were pooled together. Consequently, the responders gave a global response, but were not able to qualify the type of risk: an epidemic could happen in the future whatever its etiological origin.

Interestingly, several recent models on risk of ZIKA outbreaks in the US, based on vector ecology, have suggested disease spread outside of the southernmost counties, a prediction that is inconsistent with actual observations of the ZIKA epidemic on the continent thus far [30, 33]. The findings that we present here are consistent, and would tend to indicate that ID physician perception and its variability (age of training, tropical experience…) of emerging arboviral disease threats may be an important component to be considered in regional and global health security.