Many respondents stated that they had taken precautions: 87% did so after the interviewer had detailed possible precautions one by one. When possible precautionary measures were listed by the interviewer, “precautions taken” scored 2.02 reflecting that precautionary behaviour, albeit frequent, focuses on a small number of preventive measures taken at a single time. Whatever the way of asking the question, the main precautions that the respondents stated they had taken (hand washing and avoiding crowded places) were in agreement with official recommendations . In several surveys, hand washing was the main precaution that the respondents said they had taken [4, 27]. This precaution was taken to the same extent on Reunion Island (59%) and in France (59.7%) but to a lesser degree in the Netherlands (36%). Avoiding gatherings (crowded places) was mentioned more often by the respondents on Reunion Island (34%) than in France (14.6%)  or the Netherlands (8%) .
The specific behaviour of young adults is an important result of our study. Young adults (18–29 years) appeared as being the most active in terms of prevention, in contrast to other studies which reported elderly people to be more active in applying preventive measures [5, 28, 29]. Young adults were 8 times more likely to undertake preventive measures than other age groups. At the same time, the proportion of the population reporting influenza like illness in our sample was higher among young people and decreased with age. This feature was in agreement with results of the virological survey conducted on the same population in the context of the CoPanFlu prospective study that revealed that young people had been far more frequently infected by influenza A (H1N1) than elderly respondents . One interpretation might be that young adults could have reacted well once they realized that they were more exposed to influenza A (H1N1) than other age groups. The research into behavioural attitudes over an extended period during influenza epidemics shows that people are highly adaptive and that their attitudes evolve over time [4, 5].
Moreover, the most educated respondents had likely adjusted their behaviour to the changing general belief about the real severity of the epidemic. Those people most informed about the projected lower severity of the epidemic, felt it less necessary to take effective precautions. Conversely, a low standard of education was found to be associated with precautionary behaviour in our study as in a similar study conducted in the Netherlands during a human avian influenza epidemic .
The vast majority of respondents said that they were well informed (78%) and showed that they were knowledgeable. Knowledge of modes of transmission scored a satisfactory 4.47 (on a scale of 0 to 6) especially since for each wrong answer one point was deducted from the total score. This was true for 40% of the respondents who mentioned the mosquito as a potential vector of transmission. The recent chikungunya epidemic in Reunion Island may explain this link between mosquito and influenza A (H1N1). Finally, a question remained as to the transmission of the infection from pigs to people. In fact the zoonotic transmission was exactly the opposite in Reunion Island and was that of a reverse zoonosis (ie: transmission of the virus from humans to animal) as large and prolonged contamination of swine herds in Reunion Island occurred as a consequence of the H1N1 pandemic .
In our study, multivariate analysis showed that vaccination against seasonal influenza appeared related to a more general preventive attitude, as has also been observed in other studies . This suggests that seasonal vaccination applied before the start of the epidemic, as an initial step towards prevention, had promoted preventive behaviour during the epidemic. Similarly a previous study has shown that undertaking influenza vaccination in the past, greatly facilitates vaccination in subsequent years . Vaccination against seasonal influenza was not part of the recommendations against influenza A (H1N1) and no specific vaccine against influenza A (H1N1) was available during the whole H1N1 outbreak in Reunion Island. Nevertheless, 37% of respondents, regardless of their age, believed in the efficacy of the vaccine against seasonal influenza and 14% stated that they had taken this precaution to protect them from influenza A (H1N1). This attitude fitted the logic of previous messages: a vaccine that includes strains that circulated in previous years is also supposed to provide a protection (even partial) against the newly introduced strain. This was what transpired during the epidemic . Vaccination against seasonal influenza was more widely adopted in main France than in the overseas department of Reunion Island (25.6% versus 14%). Of note, however, is the fact that in France, where the specific vaccine against influenza A (H1N1) was available, the use of the new vaccine was limited with a coverage of only 27.4% .
The number of preventive measures regarded as effective scored 3.04 (on a scale of 0 to 6), confirming good knowledge of preventive measures and confidence in the precautions suggested in the local prevention campaigns. The score for the effectiveness of preventive measures remained a predictor of precautionary behaviour in the model. The optimism through belief in a large number of precautions regarded as effective was also shown in another study conducted in the United Kingdom .
Perceived severity and vulnerability with regard to influenza A (H1N1) were regarded as moderate with average scores of 6.05 and 5.26 respectively (on a scale of 0 to 10). As our survey was conducted after passage of the epidemic wave, the perception of risk and severity of influenza A (H1N1) might have been minimized. A previous survey conducted in three phases over the course of an epidemic showed that perceived severity decreased at the end of the epidemic . However, these perceptions remain higher on Reunion Island than in France where perceived severity and vulnerability were on average 3.64 and 2.95 respectively . Finally, the severity of influenza A (H1N1), although rated as moderate, may have been slightly overestimated by our sample owing to the fact that it contained a large proportion of women whose perceived severity and perceived vulnerability were observed to be higher. This finding is shared by other studies [5, 33]. Influenza A (H1N1) was regarded as more serious than seasonal influenza by most of the respondents (58%) but did not cause major worries compared to other concerns. Health concerns focused on chronic diseases; cancer was the main concern of the respondents, as had been observed in a previous study in France . Moreover, influenza A was presented in the media as a serious epidemic, meaning that a parallel could have been drawn with the unquestionably more severe vector born Chikungunya epidemic. This potential danger was confirmed by concern over a future chikungunya epidemic that scored 7.1 (on a scale of 0 to 10). The perceived risk of influenza A (H1N1) on Reunion Island must be interpreted in the context of the previous chikungunya epidemic of 2006. Although this epidemic occurred three years before, it rapidly spread on a large scale, was severely symptomatic in a large fraction of the population (muscular and articular pain) and brought about chronic symptoms in a significant proportion of infected persons (266,000 cases, 246 persons hospitalized in intensive care, some 40 maternity-neonatal infections and a total of 243 deaths) . Comparatively, the influenza A (H1N1) epidemic was regarded as far less severe both in terms of its scale and its health consequences.
Our study shares, with several other univariate analysis studies, three main factors influencing the adoption of precautionary behaviour: perceived severity, perceived vulnerability and perceived self-efficacy [2, 4, 5]. These results are in accordance with the Protection Motivation Theory (PMT)  and Health Belief Model (HBM) . These factors are not always retained following a multivariate analysis since they are closely related to age, gender and standard of education, all factors that are confusing in a multivariate analysis. The socio-demographic profiles of the samples impact the model, revealing distinctive features specific to each target population target. Our sample could have played this role, highlighting the contrast in behaviour between elderly and young adult respondents. The research into behavioural attitudes over an extended period during influenza epidemics shows that such attitudes evolve over time and that people are highly adaptive [4, 5]. The authors agree that such perceptions (severity, vulnerability, perceived self-efficacy) are underpinned by anxiety which is itself dependent on the messages conveyed by health authorities and which are in turn relayed by the media [2, 4, 5, 34].
Our study has a number of limitations. Firstly, our sample was not representative of Reunion Island’s population due to the inclusion of a high proportion of elderly people and women. Hence whether conclusions could be generalized to the whole population of Reunion Island may be questionable. The over-represented elderly people and women weigh towards a selection bias. An explanation is that the interviewers preferred home-based respondents, thereby limiting refusals but making the choice of the household’s reference person more difficult. Analysis by sex and age group do not reveal significant statistical differences except a higher perception of severity by women. However, selection bias (women and elderly over-represented) has simply revealed, in the multivariate analysis, the contrasted attitude of young people. Apart from this result, elderly people did not appear to have a particular attitude. Elderly people could have overestimated severity of influenza A (H1N1) because influenza is known as an important cause of morbidity and mortality among elderly people . However, our study did not reveal a higher perception of the risks among elderly respondents. The elevated tendency towards isolation among our target population does not seem related to the composition of our sample although elderly respondents may naturally be more inclined towards isolation owing to their generally reduced mobility. However, the young and elderly people in our sample declared that they taken this precaution to an equal extent (avoidance of mixing in groups: crowds, transport).
A second limitation is that the survey started at least 2.5 months after the outbreak and recall bias is a possible limitation. It seems however that this bias was mitigated. As already mentioned, there is a good correlation between infected cases of influenza A (H1N1) among respondents and their serological status . These explanatory factors are consistent with a low recall bias.
A third limitation may arise from the introduction of desirability bias by the two-phase questioning technique used to assess the implementation of precautionary measures. However, compared to self-reported precautions, the second method has allowed collecting more complete data. Indeed, the response rate is always higher when memory is prompted by exhaustively recalling all the available precautions. On the other hand, the respondents very often mentioned first the active steps they had taken individually (hand washing, wearing of a mask, being vaccinated). The avoidance type precautions (not going to concerts, not traveling on public transport, not taking children to school) are retained more after a recall. It is possible that the reported precautionary behaviors of young adults could result from a social desirability bias as the participation of an interviewer can prompt responses reporting the implementation of prevention measures. Desirability bias is unlikely once the reality of the epidemic could be clearly perceived at the end of the epidemic passage, especially if one considers the highly alarmist announcements released by health authorities and media at the start of the epidemic.
Fourthly, the proportion of the population taking precautionary behaviour is high (87%) and interpretation of differences by predictor is challenging. The lack of power of the statistical test is balanced by a high size of the sample and a high participation rate observed (95%).
Our study has also a number of strengths. Firstly, the high participation rate is unusual. For example, the participation rate, for a telephone survey, was 46% on a representative sample of the population in France (excluding French overseas territories) . In the Netherlands during an Internet-based survey (online questionnaire), the participation rate improved during the epidemic: respectively 59% at the start of the epidemic, 63%, half way through and 79% at the end . In Reunion, the survey on perceived risks and preventive attitudes was conducted on average 2.5 months (SD 1.5) after the epidemic ended. This fact may explain the high participation rate observed. A further explanation may be that the respondents in our sample had been contacted regularly during the A (H1N1) epidemic as part of the CoPanFlu-RUN protocol (virological aspects) [16, 21, 22].
A second strength of this study is that the proportion of population (20–59 years) reporting influenza like illness (42%) in our sample was comparable to the seroconversion rates observed (39.4%) in the prospective serosurvey .
Thirdly, the percentages obtained from dichotomous questions (yes/no) from our questionnaire appear consistent with the scores obtained by combining several separate responses. This observation is reassuring with regards to the relevance of our sample analyses.