The present study shows that self-reported influenza vaccination coverage in HCW is a good proxy for recorded vaccination coverage in the two previous years. The concordance, evaluated using the Kappa index and the CI of the coverage, supports this assertion. Self-reported coverage was consistently higher than coverage obtained through OH records in this study.
The finding that the proportion of unvaccinated discordant subjects was greater than the proportion of vaccinated discordant subjects may explain the tendency of self-reporting to overestimate coverage. Comparing these two groups in the different campaigns studied, in 2007, when the results show that self-reported vaccination was not a good proxy, the differences were even greater. This might suggest that either people remember better what they do than what they do not do or that, in case of doubt, people tend to be eager-to-please and thus state they are vaccinated. However, the underlying reason explaining these results needs to be further explored. With respect to the 2008 season, several factors may have influenced the fact that the coverage was not overestimated and that there was no disparity between discordant subjects. In 2008, a new model of vaccination campaign was introduced in our hospital 
, which may have better internalized by HCW, although there is not sufficient information to state that this occurred. In 2009, there were again a greater proportion of unvaccinated discordant subjects than vaccinated discordant subjects.
With respect to discordance in the self-report, taking the results of the 2009A campaign as an example, 20% of participants who self-reported being vaccinated were not, while 1.4% who had been vaccinated self-reported they had not. If it is assumed that the lack of memory is the same in vaccinated and unvaccinated subjects, the difference between these two percentages could be ascribed to being eager-to-please subjects who think they should have been vaccinated or that it is better to state they have been vaccinated because they recognize that vaccination is recommended by the hospital and the health authorities. Better characterization of these subjects could provide clues to approach groups who could potentially be convinced to be vaccinated, but the small sample of discordant subjects did not allow any pattern to be established. In the case of the 2009A campaign, contrary to what was expected given the low coverage achieved 
, no rejection effect was observed after A(H1N1)pdm09 influenza vaccination, when we expected to see vaccinated subjects who reported being unvaccinated. We also observed the eager-to-please effect, suggesting that HCW took no pride in not being vaccinated.
Comparison with other Studies and Implications
The literature review found eight studies comparing self-report with vaccination record 
. These studies were conducted in high-risk patients in whom influenza vaccination was indicated and their results () are aligned with the findings of this study which: self-reported coverage is a good proxy but tends to overestimate the coverage calculated from the vaccination record. McNemar’s test was calculated for all these studies and showed that discordance did not appear to be due to chance and that, of the total number of discordant subjects, the proportion of unvaccinated discordant subjects was greater than that of vaccinated discordant subjects. In addition, NPV were higher than PPV. Two studies 
retrospectively studied two vaccination seasons and found that self-report was a good proxy during these two seasons. High-risk patients exhibited the same behavior as our HCW, suggesting the presence of a social norm. In-depth studies of the qualitative aspects of the reasons leading HCW to be eager-to-please in order to define patterns would be of interest.
Data of published studies comparing self-reported influenza vaccination in high-risk population.
Strengths and Limitations
One strength of this study is that the sample kept the proportions of professional category and vaccination coverage of the overall population in order to avoid participation bias due to unvaccinated subjects 
. The sample characteristics were consistent with those of the whole population of hospital workers (). In addition, the coverage of the sample calculated using the OH vaccination record matched the coverage of the total number of HCW recorded in the hospital, indicating that the sample was representative and consistent, given that the CI overlapped and collected the time trend well. Other strengths were that four vaccination campaigns were analyzed together and that the number of subjects who declined to participate was very low.
have assessed the validity of vaccination records, which are also to some degree a proxy for real vaccination. It is difficult to confirm that a HCW is not vaccinated. However, we believe that our records are reliable. Most HCW can be presumed to be healthy and have few reasons for being vaccinated outside the workplace. In our hospital, the mobile unit covers all wards and is in contact with the majority of workers and also records the very-few HCW who report being vaccinated outside the hospital. A further limitation is that the Human Resources Department (HR) does not have a completely up-to-date record of the shift and location of each HCW, which would explain the number of workers not located or not known by their fellow-workers.
In our study population, self-reported influenza vaccination coverage in HCW is a good proxy for recorded vaccination coverage in the two previous years. In the different campaigns studied in the present paper and in the previous studies, the results show that, of the total number of discordant subjects, the proportion of unvaccinated discordant subjects was greater than that of vaccinated ones, suggesting that vaccination behavior influences the self-report. This explains the tendency to overestimate coverage using self-report compared with vaccination records. The sources of vaccination coverage should be taken into account when comparisons are made.