The growing body of SP literature on vaccination highlights the increased interest in the use of choice-based experiments, to elicit preferences for a variety of vaccines and to understand factors influencing vaccine decision-making of different groups of individuals. A total of 42 studies were identified in this review, capturing preferences of three different target groups, covering fourteen vaccines or vaccine programs and with the majority published after 2010. Given the limited amount of studies assessing preferences of health advisors, this review focused on examining and comparing preferences for vaccine attributes of vaccinees and representatives (including health advisors). The former generally focused on preferences of adults and adolescents, while the latter mainly captured parental preferences for childhood vaccines.
Among the 42 included studies, sixteen studies were of high-quality and could be included in the comparison of vaccine preferences. Irrespective of target group captured, outcome-related attributes, such as vaccine effectiveness, vaccine risk and protection duration, were most frequently reported, followed by attributes covering the monetary cost of vaccines. Outcome- and cost-related attributes were also most commonly statistically significant across all studies, indicating that the same factors are generally preferred across different groups of individuals. Correspondence was also observed for least preferred attributes, since attributes related to a vaccines’ access were least valued by both target groups. However, it should be noted that elements of accessibility might already be included in other attributes (e.g. cost, time). In the study of Verelst et al. [79] vaccine accessibility was for instance uni-dimensional and incorporated not only the availability of vaccines, but also its monetary costs. Therefore, it should be interpreted cautiously.
The overall finding is in line with a recent review of Lack et al. [21], which focused on HPV-vaccination and found that vaccinees, parents and providers have the strongest preferences for attributes related to vaccine outcomes. Comparable patterns were also identified among earlier reviews of CAs [20] and DCEs [19] capturing vaccine preferences. Michaels-Igbokwe et al. [19] indicated for instance that attributes related to degree/duration of protection and risk were most often statistically significant across DCEs studying preferences for childhood and adolescent vaccines. In addition, attributes included in DCEs generally addressed features of vaccines, while neglecting service (i.e. process) or contextual aspects (i.e. other). The latter was observed to a lesser extend in this review, as nearly half of the high-quality studies incorporated attributes describing vaccine coverage rates, waiting times, access to vaccines, locations, information provision or social support. However, time and service delivery were not statistically significant in studies among respectively vaccinees and representatives. A more plausible explanation for the underrepresentation would hence be that aspects of a vaccine process are simply less important for vaccinees and representatives in making vaccine decisions. This hypothesis is supported by findings of Guo et al. [45], which outlined that service convenience and quality were less ‘dramatic’ than vaccine features.
Despite the overall correspondence in preferences of target groups, current findings showed that outcome-related attributes were more often statistically significant in studies targeting vaccinees (esp. vaccine risk), while cost-related attributes were more often statistically significant in studies of representatives. This indicates that the level of evidence for outcomes and costs slightly differed between both target groups. However, outcome and cost parameters were statistically significant in both target groups, indicating no differences in preferences of vaccinees and representatives. Instead differences for cost might be (partly) explained by the definition of this domain. Particularly among studies targeting representatives, cost-related attributes were operationalized differently. They were for instance defined as ‘type and value of parental reward (received when full schedule of vaccinations is completed)’, ‘out-of-pocket cost’ and ‘payment for one doctor visit’. This might have affected the way in which respondents interpreted attributes and eventually the way in which they valued vaccine scenarios. Studies targeting vaccinees mainly used consistent wording (i.e. ‘out-of-pocket cost’).
Due to the use of a relatively strict quality-threshold, recommended by the developers of the checklist, 26 lower-quality studies were not considered in the comparison analysis as they were assumed to be susceptible to bias. However, the robustness analysis confirmed findings of the main analysis and only showed a slight increase in preference for attributes covering disease risk. This was observed for both target groups and suggests that epidemiological and affective factors, such as the susceptibility to and severity of diseases, also affect vaccine decisions. Qualitative research on vaccine behaviour and the Health Belief Model confirm this finding and indicate that the valuation of and satisfaction from vaccines is linked to perceptions of vaccines and associated diseases [84–86]. Nonetheless, this discrepancy could also be caused by the conceptual overlap that was identified in the quality assessment. Four lower-quality studies included more than one risk-related attribute, while none high-quality study did so. As a consequence, disease risk was reported more often in lower-quality studies than in high-quality studies. According to Mandeville et al. [31], the occurrence of overlap could distort parameter estimates, as attributes (and effects) are not distinct and do not vary independently of each other. Respondents might for instance experience difficulties in distinguishing between attributes and in interpreting them separately.
When examining characteristics of high- and lower-quality studies, it is observed that all high-quality studies were conducted in HICs and applied MXL/RPL or LCM. In addition, they mainly focused on vaccines against sexually transmitted infections, while lower-quality studies were characterized by a broader range of vaccines, countries and econometric models. High-quality studies were also more likely to pilot test surveys and to express outcomes in WTP or predicted vaccine uptake. The latter is in contrast to Clark et al. [17] who focused on general health preferences and observed a decline in the use of monetary values and probabilities. However, probabilities are particularly useful in vaccination, as herd immunity is an important (but yet hard to achieve) externality which can only be acquired when vaccination coverage passes a certain threshold [87–89]. In addition, adult and traveller vaccines might require (co-)payments, which can be adequately captured in monetary values such as WTP [32]. The trend of using more sophisticated designs and appropriate software, observed in the review of Soekhai [18], is reinforced by current findings. A last remarkable observation refers to the variability in sample sizes and threshold(s) used in studies reporting statistical significance. On average, high-quality studies used larger sample sizes compared to lower-quality studies. High-quality studies with larger sample sizes (≥ 500) were also more inclined to use multiple thresholds (i.e. alphas), whereas lower-quality studies used smaller alphas for sample sizes below 500 and p < 0.05 for sample sizes above 500. Particularly the latter is contrast with previous research, which indicates that larger sample sizes are required when lowering alpha and vice versa [90, 91].
With respect to the quality assessment, this review reported an average score of 9.3, which was almost one point higher than reported by Michaels-Igbokwe et al. in 2017 (score: 8.4) [19]. This may suggest that choice-based experiments have improved elements of design in the last couple of years. However, no improvement was observed in our quality scores per period. Furthermore, the quality assessment indicated that industry-funded studies scored remarkably lower than non-industry funded studies. This addresses the need to get insight into industry sponsorship and used methodology. As observed by previous reviews using the 13-criteria checklist [19, 31], a variety of scores was administered and none reached the maximum score of 13. All included studies failed at least one criterion. Weaknesses were particularly observed on elements of choice task design, experimental design and conduct. This underlines once again the technical requirements for all four stages and highlights the need to improve scientific rigour across choice-experiments in health.
Strengths and weaknesses
A strength of this study is the use of a formal quality assessment tool [31] to critically appraise the methodological quality and internal validity of included studies. Despite the evidence indicating that quality scores not necessarily reflect a study’s validity, the tool has demonstrated its usability in previous reviews in health [19, 31, 92]. A quality threshold of 75% was used, as it was assumed that only studies who satisfied at least 10 out of 13 criteria were able to exclude most threats to validity occurring along the four stages of individual choice experiments. This promoted the internal validity of this review, since conclusions regarding drivers of vaccine decisions were based on attributes identified in high-quality studies only [31, 36, 93]. In addition, robustness of current findings was tested and confirmed in the comparison of high- and lower-quality studies. Based on this analysis, it could be ascertained that findings were largely not affected by exclusion of lower-quality studies [37]. The comprehensiveness of the search is also a strength. As indicated in the method section, the primary search was updated and related reviews were screened. Only two additional studies were identified, confirming the accuracy of key words used and suggesting that the primary was all-encompassing [27]. However, data was extracted from published literature and relied on what was reported in articles and available supplementary material. Like in any review, reporting and publication bias could hence not be eliminated [93, 94]. A last strength refers to eligibility criteria. In contrast to previous research, no in- or exclusion criteria were formulated based on vaccine topic or site examined. Included studies hence covered a variety of vaccines/programs, populations and settings. This heterogeneity promoted the representativeness and improved generalisability (external validity) of current findings [37]. Due to the limited research on preferences in low-resource countries and of healthcare professionals, both were still underrepresented which may hamper generalisability of results to these particular populations and settings.
Beside strengths, some limitations could also be identified. This review was limited to choice-based experiments and excluded studies without a component of choice. Scaling approaches in which respondents were asked to rate or rank vaccine scenarios (e.g. best-worst scaling), were for instance not taken into account. Eleven studies were excluded based on this criterion (eight primary search, three snowballing), most of which focused on vaccines against HIV [95–102]. In addition, key steps of this review were performed by a single researcher. This may have induced reporting bias as data was not extracted and interpreted by two researchers independently [36, 93, 94]. To minimize the occurrence of inconsistencies/mistakes, all steps were closely monitored and checked by a second researcher and ambiguities were discussed and agreed upon. Due to the heterogeneity of studies included, a variety of attributes were used. To reduce the number of attributes, the commonly used classification of outcome, cost and process was used [21, 103–105]. However, multiple attributes could not be classified properly, and a fourth category, other, needed to be added. This category included many different attributes (and domains), which ranged from disease-specific to societal or contextual factors. This variety hampered appropriate naming and interpretation of the category as a whole. Besides, the decision was made to include health advisors into the representatives’ group, because both referred to individuals that make vaccine decisions for others and there were only four studies focusing on health advisors. Particularly given the latter, it was considered inappropriate to create a separate group. In addition, analyses revealed that no new domains were introduced by the studies of health advisors, which suggests that the inclusion of both had no influence on the findings about representatives. The last, yet most important limitation refers to the approach to identify factors driving vaccine decisions. Due to the range of vaccines, attributes, choice tasks, populations and outcome measures within the included studies, it was not possible to pool results into a combined effect by means of a meta-analysis [106]. Due to reproducibility problems, it was also not feasible to assess the relative importance per attribute and thereby to indicate whether an attribute was more important than another. Instead, drivers were based on frequency of reporting and statistical significance of domains. Although both measures give an indication about the importance of attributes, the adequacy is discussible. Statistical significance is not only contingent upon the set of attributes used, but also on the way in which it is defined [36, 90, 91]. An attribute could for instance be considered statistically significant in one study (at p < 0.10), while it is not in another that used a smaller alpha (e.g. p < 0.01). The frequency of reporting domains was also skewed by studies including more than one attribute of the same domain. This distortion as well as the variability in significance levels hampered interpretability and may have induced bias on outcome level. Nonetheless, this was tried to minimize by accounting for significance levels (alphas) used and the amount of studies reporting certain domains.
Implications for research and policy
Implications for research
Despite the importance of ensuring the accuracy of vaccine preferences, the quality assessment showed that the choice task, experimental design and elements of conduct received less attention compared to analysis. Studies conducted in LMICs particularly reported inappropriate experimental designs, showed conceptual overlap across attributes and failed to pilot test the survey among a small sample of the target population. In 2017, Michaels-Igbokwe et al. [19] observed similar methodological patterns and already highlighted to need to improve the design and context of DCEs. Although fourteen studies published afterwards improved pilot testing, similar scores were reported on the experimental design and sufficiency of response rate. In addition, scores on conceptual overlap even decreased. This indicates that choice experiments in vaccination have not yet structurally improved their designs and conduct since then. The time lag between the conduct of a study and publication of results could play a role, as high-quality studies were on average published three years after its conduct. Improvements in choice design and conduct are notwithstanding crucial to ensure reliable estimates of vaccine preferences. As recommended by Soekhai et al. [18] this might be facilitated by formulation of guidelines to report choice experiments.
Furthermore, this review intended to study three distinct target groups. Due to the limited amount of studies on preferences of health professionals, no conclusions could be drawn for this decision-making role. As literature [9, 107, 108] showed that decision strategies particularly differ for medical professionals and vaccinees, future research could focus on preferences of healthcare professionals to assess whether similar vaccine attributes are preferred by them as well. This could potentially be done by broadening the approaches used to measure SP (e.g. add contingent valuation). Given the scope of this review, included studies were grouped based on the decision-maker targeted and not on the type of vaccine. Although Lack et al. [21] focused on HPV vaccination, limited studies were available for other vaccines/vaccination programs. Therefore, it was not possible to use a classification based on vaccine type. A classification based on broader target groups would still be possible and future research could focus for instance on childhood vaccines, adult vaccines and/or traveller vaccines. However, it should be noted that our classification already encompassed some of these groups, as vaccinees often involved decisions for adolescent or adult vaccines and representatives commonly concerned parental decisions (for childhood vaccines). Additional research could also focus on target groups other than those distinguished in this review (e.g. policy makers, based on gender) or on vaccine preferences in low-resource settings. A combination of both would also be interesting as qualitative research [109, 110] indicated that national decision-makers in LMICs particularly preferred simplified delivery mechanisms, thermostability and an extended shelf-life. In light of the current corona pandemic and ongoing search towards an appropriate vaccine, it would also be worthwhile to assess preferences for (future) vaccines against epidemic infections, such as COVID-19 and SARS, and to determine whether vaccine preferences differ based on epidemic severity.
Implications for policy
In contrast to what is suggested in previous qualitative studies [9, 11, 12], this review demonstrates that vaccine preferences show similar patterns for vaccinees and representatives. This means that broadly the same strategies could be adopted to promote and optimize vaccination behaviour. Vaccine strategies should focus on highlighting outcome-related vaccine attributes. Proper and understandable information should thus be provided about the effectiveness of vaccines, duration of effectiveness and risks associated with vaccine administration, dosing and handling. Insight into the latter is particularly important for vaccinees. Another key driver in vaccine decisions is the cost of vaccines. Therefore, effective pricing strategies should be applied (if applicable) when introducing or continuing the use of vaccines. Particularly for vaccine decisions that involve representatives (e.g. childhood vaccines), the element of cost is important. Although this study primarily focused on attributes of vaccines and vaccine programs, the robustness analysis indicated that disease risk is important for vaccinees and representatives as well. Therefore, information strategies should not only cover vaccine-related aspects, but should also inform target groups about the severity and probability of diseases. With respect to the design of vaccine programs, service or process-related factors such as location and waiting time were less important compared to vaccine scheduling and number of doses or visits. Across included studies, individuals value a reduced number of doses/visits when deciding for themselves and others, which implies that vaccine programs should try to minimize the amount of dosages as much as possible. As outlined in previous research [111, 112], combination vaccines offer the potential to reduce the amount of injections, visits and associated costs. Although most childhood vaccines already involve combination vaccines, adolescent vaccines such as the HPV vaccine still consist of multiple injections. Despite the clinical superiority of two- and three-dose HPV vaccines, Pinto, Dillner, Beddows & Unger [113] observed that there is yet some evidence indicating that one dose may provide similar protection.