Discussion
This review identifies and summarizes the quantitative evidence about the possible relationship between HCWs' knowledge, beliefs and attitudes about vaccines and their intentions to vaccinate. The results of the included studies clearly show that these relationships do exist, although unfortunately, the data does not allow us to make conclusions about a causal link, mainly because all but one of the studies are cross-sectional. Only one retrospective case–control study could show a causal link between beliefs, attitudes and intentions to vaccinate, but even this study should be considered as cross-sectional because it evaluated HCWs' beliefs and attitudes at a given point in time, and these variables may change over time.
Given the range of the inclusion criteria, the included studies differed widely in their evaluations of the variables, methodologies, and statistical analysis techniques. Therefore, the results cannot be integrated to quantify the magnitude of the associations and must be evaluated individually. Even so, it appears that all the studies show associations in the direction postulated by the SIEVE experts, although some associations were statistically significant, and others were not.
Knowledge measurement was only based on questions on vaccination contraindications, which may imply that researchers were especially worried about the importance of this factor in regard to the intention to vaccinate. On the other hand, the variety of topics that were explored such as beliefs and attitudes is wide, as is shown in Table 6, though few were repeated in various studies. This may denote interest of the researchers on showing the importance of this type of factors on the intention to vaccinate. In regard to beliefs, the topics that made reference to efficacy and security of the vaccines, severity of the vaccine preventable disease and use of alternative medicine theories, appeared in various studies and seem to be perceived by the researchers as important factors. The associations found with the intention to vaccinate, show this importance. As for attitudes, only "Low confidence in Public Healthcare information or national recommendations" factor was explored in more than two studies (in 3 out of 12 that measured attitudes). This may reflect that there doesn't exist a clear investigation line intended to prove the importance of a specific factor. It's also interesting to highlight that the knowledge, beliefs and attitude themes explored in the HCW in the included studies, do not differ from the barriers towards vaccination explored in the general population.
The results of this study must be understood in the context of the limitations of the methodology used. Including observational studies in systematic reviews presents specific challenges, as observational studies have inherent biases (mostly selection and information bias) and vary in their study designs. Taking these concerns into account, we have attempted to be as rigorous as possible in the methodology of this review.
We tried to reduce identification bias by performing the literature search in four databases in addition to a manual search, selecting studies in seven languages and making personal contact with authors when necessary. We did not search for unpublished articles or for "grey literature", and there may be unidentified articles in databases we did not search. Consequently, there may be a risk of publication bias. One limitation was that we chose to include only the studies that started collecting data after the publication of the article by Wakefield et al. This article should not imply that changes occurred in the relationship between knowledge and the intention to vaccinate, and therefore, there may have been studies before this date whose results were equally valid for understanding the current situation. On the other hand, the included studies were those published up to June 2009, so at the time of its publishing our study won't include the latest evidence. In spite of this, having not found other systematic reviews on this topic, the results may be of interest to investigators, policy makers and healthcare professionals. In addition, we only included data from developed countries. In other words, there may have been relevant studies from developing countries that were not included and also that the results cannot be extrapolated to these countries. HCWs' perceptions of vaccines and of vaccine-preventable diseases may be different in developing countries because they face different disease burdens, and this is the reason why these studies were not included.
One difficulty we encountered was classifying and distinguishing the information related to knowledge, beliefs and attitudes, despite the definitions we established, as shown by the low Kappa index for inter-observer agreement. It is also likely that this difficulty in delimiting the variables caused the low Kappa index for the abstract selection, and we believe that an intentional training of the researchers in identifying and classifying the variables during the pilot study would have improved the concordance.
The tools used to determine the risk of bias and the quality of the studies included were based on the Newcastle-Ottawa scales, which are widely used for observational studies. The deficiencies in the methodology and reporting of many of the studies are reflected in the low scores on the quality scales, questioning the reliability of the studies reviewed. With regard to the sample selection, almost all of the studies used an acceptable sampling method, but most failed to report the comparability of the respondents and the non-respondents, which may imply there was a self-selection bias. Many included studies used subjective measures (e.g., self-reporting, unverified intentions or behaviors), which can lead to information bias. Care must be taken in interpreting such information, as there is a tendency for respondents to provide what they believe to be socially acceptable answers. Some studies did not control for confounders, such as demographic factors, meaning that the variability in the intentions to vaccinate may be incorrectly associated with knowledge, beliefs or attitudes. The studies also failed to use validated instruments to measure attitudes, knowledge and beliefs. In addition, there are factors related to immunization that fall in these three areas but were not measured in the included studies, so these studies may offer only a partial viewpoint of the relationship between HCWs' attitudes, knowledge and beliefs and intentions to vaccinate.
This review fills a gap in the literature, and thus, despite the limitations of our methodology, we believe that the benefits of illuminating this relevant topic overcome the limitations. Qualitative studies, which were not considered in this review, could be the objective of a future review. A gold standard research study is a broad-based population study that controls for confounding factors and biases and uses validated tools to measure knowledge, beliefs, attitudes, and intentions to vaccinate or vaccine coverage. The knowledge, beliefs and attitudes that may determine the intention to vaccinate are those stated at the moment when the intent to vaccinate is measured. The fact that these variables change over time may explain the absence of longitudinal studies that are capable of demonstrating causality. Despite this lack of evidence, various studies have evaluated the impact of interventions on the intent to vaccinate or the practice of vaccination, and others have even assessed their cost-effectiveness.
Some years ago, it was stated that the success of vaccination programs depends on strong professional commitment, that it is important to have written clinical guidelines to strengthen, instruct and support professionals at the time of vaccination, and that effective use of information technology would be beneficial. On the last two points, great strides have been made; it may be time, as the SIEVE experts state, to test and implement strategies to incentivize health professionals.