Using the Health Belief Model to understand intention to vaccinate for Lyme disease in the United States

A growing number of Lyme disease (LD) cases in the U.S. are reported in states neighbouring those with high‐incidence (>10 cases per 100,000 population) rates. Considering the evolving epidemiology, high‐incidence counties in many of these “neighbouring states,” and the forthcoming vaccines, understanding the drivers of vaccination intention is critical, particularly how drivers of intention in neighbouring states vary relative to regions currently classified as high incidence.


| INTRODUC TI ON
Lyme disease (LD) is the most common vector-borne disease in the United States.Recent estimates suggest that nearly 500,000 people in the U.S. are diagnosed and treated for LD yearly (CDC, 2022a).The mainstay of LD prevention is a range of personal protection methods recommended by public health officials, including the use of repellents, checking for ticks after being outdoors, and prompt removal of any attached ticks.However, these methods have been found to have limited effectiveness because the individual effort required to apply them consistently is challenging (Schwartz et al., 2022).Given LD prevalence, disease burden, and lack of consistently efficacious protection methods, a safe and effective vaccine will be an important means to mitigate the disease's impact.As new vaccines against LD are in development (CDC, 2023), understanding the intention to vaccinate against LD in the coming years will be especially critical to ensure that this forthcoming prevention tool is used equitably by all at-risk populations.
Currently, 15 states and the District of Columbia in the Northeast, mid-Atlantic, and upper Midwest regions of the U.S. are considered high incidence (≥10 cases per 100,000 population) for LD (CDC, 2022b).However, the geographic distribution of LD in the U.S. has expanded markedly over the past two decades.In 2001, there were 10 high-incidence states, which has increased to the current 15 states and the District of Columbia by 2021 (CDC, 2022a;John Hopkins, 2023).Similarly, from 2005 to 2019, there was a 71% increase in the number of counties classified as high incidence for LD (CDC, 2022a;John Hopkins, 2023).There has also been an increase in the number of counties with established populations of the Ixodes scapularis vector (CDC, 2018;Eisen et al., 2016).Although most cases of LD occur in high-incidence states, case counts in states bordering those with high incidence (hereafter referred to as "neighbouring states") have also increased (Schwartz et al., 2017), and it is expected that the geographic distribution of LD will continue to grow, ultimately resulting in additional high-incidence states.Neighbouring states generally have fewer initiatives to educate the population on LD risk, severity, and prevention since they are not broadly considered high-incidence regions by public health officials.However, most of these states have counties and regions within them where the incidence rate is ≥10 cases per 100,000 population (CDC, 2022a).
Given this and the potential for the availability of a safe and efficacious vaccine, we recognize that there is a need to understand and identify in advance, key drivers of LD vaccination intention between people residing in high-incidence states and neighbouring states to ensure an efficient and effective public health response.
Studies have yet to describe the intention to vaccinate against LD for individuals residing in neighbouring states although some prior studies have described intention to vaccinate in high-incidence regions.For example, one study found that 60% of adults and 71% of caregivers of children residing in high-incidence states were likely or very likely to get vaccinated against LD (Stark et al., 2024).Hook et al. (2022), found that 64% of individuals residing in Connecticut, Maryland, Minnesota, and New York would be willing to receive a LD vaccine.Likewise, Niesobecki et al. (2019) found that 84% of individuals residing in selected counties in Connecticut and Maryland would be willing to receive a LD vaccine if it were available.Key drivers of LD vaccination for people residing in neighbouring states have not been studied using a similar systematic framework.The Health Belief Model (HBM) is a conceptual framework widely used in public health research to determine key drivers associated with individual health-related behaviours, including but not limited to intention to vaccinate (Rosenstock, 1974).The HBM uses the constructs of perceived susceptibility (an individual's assessment of risk), perceived severity (anticipated consequences of an outcome), perceived benefits (effectiveness of actions to prevent disease), and perceived barriers (obstacles or costs associated with action), alongside cues to action (external prompts) to understand health behaviours.Researchers have previously used the HBM to understand the key drivers of intention to vaccinate against a wide range of diseases including seasonal Influenza (Chen et al., 2011;Fall et al., 2018), HPV (Donadiki et al., 2014;Gerend & Shepherd, 2012), and LD (Stark et al., 2024).Stark et al. (2024), the first study to use the HBM to investigate LD vaccination intention in adults and caregivers, found that both adult and caregiver intentions to vaccinate were positively influenced by cues to action, perceived susceptibility of LD, and perceived benefits to vaccination.While the results from this study health belief model, Lyme disease, risk, vaccination

Impacts
• Lyme disease incidence in states bordering those with high incidence has been increasing.To ensure an efficient and effective public health programme if any of these states were to emerge as high incidence in the future, understanding the vaccination intention of residents of these neighbouring states will be critical.
• In this study, we applied the Health Belief Model to Lyme disease vaccination intention for adults and caregivers of children residing in states bordering those with high incidence.
• We found that cues to action had the largest influence on vaccine intention for adults and caregivers, highlighting the critical role of healthcare providers and public health authorities.
were informative, the results were only generalizable to persons residing in high LD incidence states.Given the expansion of LD incidence, the objectives of this study were to (1) describe the drivers of LD vaccine intention among U.S. adults and caregivers of children residing in neighbouring states using the HBM approach, (2) compare the drivers of LD vaccine intention between adults and caregivers residing in neighbouring states, and (3) compare the drivers of intention to vaccinate for adults and caregivers residing in neighbouring states with those of U.S. adults and caregivers of children residing in high-incidence states.

| Study design
Building on Stark et al. (2024), we used the same study methodology by deploying two online surveys in U.S. neighbouring states. 1 The first survey focused on recruiting adults, who were subsequentially instructed to answer survey items on their own behalf (hereby referred to as the "Adult" sample).The second survey focused on recruiting adults who reported being the primary caregiver for a child under the age of 18 (hereby referred to as the "Caregiver" sample).
Respondents in the caregiver sample were prompted to answer the vaccine intention and HBM items on behalf of their child.The surveys were distributed separately to the two samples to prevent selection effects in respondent recruitment.
The surveys were conducted using a representative survey panel maintained by Qualtrics (Qualtrics, Provo, Utah; Boas et al., 2020), which recruited and compensated respondents and supported the programming and distribution of the survey instrument.
The sampling frame was defined as U.S. states neighbouring high-incidence LD states, and the population size of each state within the geographic area (i.e., the sampling frame) was retrieved from the U.S. Census Bureau (2023).Sample sizes of 800 for both adults and caregivers were estimated to achieve a margin of error of plus or minus 3%, using a confidence level of 95%.The surveys were piloted in early October 2023, and the final data were collected from October 2 to October 24, 2023.Study materials were reviewed and approved before the surveys were fielded by Advarra, an independent Institutional Review Board service.

| Participant eligibility
Participant eligibility was consistent with Stark et al. (2024).All respondents were required to live in a neighbouring state and be 18 years of age or older.Quotas for age, sex, and race/ethnicity were used to ensure a representative sample.To both approximate the population's urban distribution and ensure sufficient representation from respondents in rural and suburban areas where LD incidence is higher, a quota required that only 25% of the sample reside in a large city (Miller et al., 2012).
Specific wording for the demographic characteristic questions is provided in Table S1.

| Measures
All survey items were replicated using the questions asked by Stark et al. (2024), who in turn had adapted these items from earlier research.Vaccine intention and most HBM constructs (i.e., perceived susceptibility and severity to LD, perceived benefits to vaccine uptake, and cues to action) were measured with two separate survey items.The construct of perceived barriers was divided into two sub-constructs: costs and safety concerns of receiving a LD vaccine.
Vaccine intention and the HBM constructs were measured with fivepoint Likert scales (i.e., 1 = very unlikely…5 = very likely for vaccine intention, or 1 = strongly disagree…5 = strongly agree for HBM constructs).The specific wording of these items and the frequency of responses can be found in Table S2.
Data collected from high-incidence states by Stark et al. (2024) were also used to compare with the data collected by this study.This allowed us to compare vaccine intention, HBM constructs, and the influence of HBM constructs on vaccine intention between neighbouring and high-incidence states.

| Data analysis
Data were both screened by Qualtrics and a researcher before analysis to ensure data quality.Respondents were removed if they failed an attention check, sped through the survey, or exhibited a "straightlining" response behaviour.
Cronbach's alphas were estimated to determine the internal consistencies of all HBM construct items.Items with acceptable Cronbach alpha values were combined into a single variable by taking the mean of the items within a construct (Peterson, 1994;Tavakol & Dennick, 2011).Then, t-tests were estimated to determine significant differences between the two samples (e.g., neighbouring adults vs. high-incidence adults).Results from the t-tests provide insight into variation in mean responses for vaccine intention and the HBM constructs.
Next, structural equation modelling (SEM) was used to determine the influence of HBM constructs on LD vaccine intention, similar to previous literature (Gerend & Shepherd, 2012;Rajamoorthy et al., 2018;Stark et al., 2024).Five models were estimated to determine the influence of HBM constructs on vaccine intention: (Model 1) for adults in neighbouring states, (Model 2) for caregivers in neighbouring states, (Model 3) a combined model for adults and caregivers in neighbouring states with an interaction variable for caregivers, (Model 4) a combined model for adults in neighbouring and high-incidence states with an interaction variable for vaccine attitude, and (Model 5) a combined model for caregivers in neighbouring and high-incidence states with an interaction variable for vaccine attitude.
The combined models included an interaction term between HBM constructs and an indicator variable.The indicator variable was equal to one for caregivers in Model 3 and equal to one for those residing in a high-incidence state for Models 4 and 5. Interaction terms enabled us to determine whether there was heterogeneity in the influence of HBM predictors.Specifically, the inclusion of interaction terms in Model 3 tested whether the drivers of intention significantly differed between caregivers and adults.The inclusion of interaction terms in Models 4 and 5 tested whether the drivers of intention to vaccinate differed between adults and caregivers residing in high-incidence states and neighbouring states.A construct for vaccine attitudes was also included as a moderating variable to control for overall differences in vaccine acceptance between high-incidence states and neighbouring states in Models 4 and 5.
The vaccine attitude construct was an average of two survey items adapted from the Vaccination Attitudes Examination (VAX) Scale (Martin & Petrie, 2017), "In general, I feel safe after being vaccinated" and "In general, I feel protected after getting vaccinated" (α for adults = 0.81, α for caregivers = 0.74).Additionally, Models 3-5 were estimated without the interaction terms (referred to as the "restricted" model) and with the interaction terms (referred to as the "full" model) to determine whether the interaction terms improved model fit using a likelihood-ratio test.a All items were measured on five-point scales (e.g., 1 = very unlikely -5 = very likely or 1 = strongly disagree -5 = strongly agree).
b Items adapted from Hook et al. (2022).
c Items adapted from Gerend and Shepherd (2012).

| HBM for neighbouring states
Estimates of Cronbach's alpha (α) ranged from 0.76 to 0.88 for the HBM items, as shown in Table 2. Combined variables were created for vaccine intention and the HBM constructs by taking the mean of items within a measurement (also shown in Table 2).Relative to the Adult sample, caregivers had higher perceived LD susceptibility and severity, higher perceived benefits, more concern about vaccine safety, and cues to action were more influential.Adults reported higher barriers to LD vaccination.
SEM models were estimated to determine the influence of HBM constructs on the intention to vaccinate against LD.The SEM estimates for the HBM constructs in the Adult model provided a good fit to vaccine intention (χ 2 = 712.13***,CFI = 1.00,RMSEA = 0.00, R 2 = 0.55).As shown in Figure 1, significant predictors for intention to vaccinate against LD for adults included perceived susceptibility (0.13), perceived benefits (0.13), perceived barriers (−0.11), and cues to action (0.59).The construct for cues to action was the largest predictor of vaccine intention for adults.
The SEM estimates for the HBM constructs also pro-  = −15,000, p-value = 1.00).Thus, overall, the HBM constructs predicted intention to vaccinate against LD similarly for adults and caregivers.
This study provides insights into how adults and caregivers residing in states that border high LD incidence states might consider a LD vaccine if it were available.Although these states do not currently have a high burden of LD, the changing disease epidemiology suggests that some of these states will become high incidence in the future (Lantos et al., 2015(Lantos et al., , 2017)).This study thus provides novel insights into the drivers of LD vaccination intention generated using the HBM that can help public health officials and policymakers plan future prevention programmes.
Overall intention to vaccinate against LD in neighbouring states was moderate, with 42% of adults and 56% of caregivers likely or very likely to get vaccinated against LD.The observed difference in vaccination intentions between adults and caregivers aligns with previous research showing that parents are often more willing to vaccinate their children than themselves (Tang et al., 2016;Zikmund-Fisher et al., 2006).The HBM constructs accounted for 55% of the variance in LD vaccination intention for the adult sample and 53% of the variance in LD vaccination intention for the caregiver sample.While there were slight differences in key drivers of intention to vaccinate, the data suggest that cues to action will be the largest driver of intention to vaccinate for adults and caregivers residing in neighbouring states.This finding aligns with both Stark et al. (2024) for LD vaccine intention and other literature that has cited a recommendation from a healthcare provider as being the most influential intervention for increasing vaccine uptake for other diseases (Newman et al., 2018;Reiter et al., 2013).Thus, the influence of cues to action was expected; however, it was surprising that cues to action had significantly less influence on the intention to vaccinate for caregivers than for adults.While this warrants additional exploration, it may result from caregivers already having a higher baseline intention to vaccinate and therefore, cues to action may not be as influential as they would be if the baseline intention was lower.Alternatively, it could also be due to caregivers in neighbouring states having a greater were not negatively influenced by safety concerns like they were in high-incidence states.While this relationship also warrants additional exploration, it could be because caregivers in neighbouring states have a lower baseline intention, which could make safety concerns less pivotal in their decision-making process.
The difference in intention to vaccinate between adults and caregivers in high LD incidence states and those residing in neighbouring states was significant for both samples, suggesting that adult and caregiver intention to vaccinate in neighbouring states is meaningfully lower than it is in high-incidence states.This is not surprising since LD is less prevalent in neighbouring states, and residents of neighbouring states report lower perceived susceptibility to contracting LD compared to residents in high-incidence states (as seen in Table 3).The differences could also be a result of differences in sample demographics.Relative to Stark et al. (2024), our sample reported less formal education and lower household income.Another plausible explanation for these differences could be differences in general vaccine attitudes.As secondary analyses in Table S3 show, vaccine attitudes differed between high-incidence and neighbouring states.Specifically, adults and caregivers in high-incidence states generally feel safer and more protected after getting vaccinated than adults and caregivers in neighbouring states.However, when vaccine attitudes were added to an SEM interaction model with an indicator variable for high-incidence states, it was not found to be a significant driver of intention to vaccinate for LD.Instead, the interaction model found that adults in neighbouring states were more influenced by cues to action than adults in high-incidence states, F I G U R E 5 Prediction of caregiver intention to vaccinate from health belief model (HBM) and vaccine attitude for high-incidence state interaction terms.The Interaction model was estimated using 1656 Caregiver observations.implying that adults in neighbouring states may not know their level of risk and base their decisions on guidance from healthcare providers or other trusted professionals.A similar relationship does not exist for caregivers, suggesting that caregivers in high-incidence and neighbouring states have similar drivers of intention to vaccinate.
Although our study contributes to LD and vaccination intention literature, we acknowledge the study's limitations.Our survey was distributed through a panel, which while demographically representative of the U.S. (Boas et al., 2020), has inherent limitations (Callegaro et al., 2014;Hays et al., 2015).This study, like Stark et al. (2024), asked about the intention to vaccinate for a generic LD vaccine without specifying the schedule, efficacy, or safety profile, all of which may influence vaccination intention and behaviour.
Unlike Stark et al. (2024), our study was completed during the offpeak season for LD, which also may have influenced the intention to vaccinate positively or negatively.We also acknowledge that while the focus of this research was to understand the application of the HBM to LD vaccination intention in neighbouring states, the HBM does not capture all factors associated with vaccination intention.
Specifically, our models for neighbouring states accounted for only 53%-55% of the variation in intention to vaccinate, compared to the 60%-64% of the variance explained by the high-incidence models in Stark et al. (2024).Therefore, future research might consider developing models that explicitly include variables such as socioeconomic status, race and ethnicity, education level, and disease knowledge to understand other drivers of intention to vaccinate, variables which some research have found to be associated with vaccine uptake (Galarce et al., 2011).

| CON CLUS ION
As the geographic distribution of LD continues to expand, this research, and future research that extends upon it, will help to inform public health officials to launch prevention programmes more effectively in newly impacted areas.

ACK N OWLED G EM ENTS
The authors thank Stephanie Duench for her critical manuscript review.

CO N FLI C T O F I NTER E S T S TATEM ENT
This study was supported and jointly funded by Valneva and Pfizer as part of their co-development of a Lyme Disease vaccine.
Figure2, significant predictors for intention to vaccinate for caregivers included perceived susceptibility (0.11), perceived severity (0.07), perceived benefits (0.17), perceived barriers (−0.10), and cues to action (0.49).Cues to action had the largest influence on vaccine intention for both caregivers and adults.The significant predictors across the two samples were largely similar, except for perceived severity which increased intention for caregivers but not adults.

Figure 3
Figure 3 presents the standardized path coefficients for the interaction terms between HBM constructs and an indicator variable for caregivers.The SEM estimates for HBM constructs also provided a good model fit for the combined Adult and Caregiver data (χ 2 = 1361.45***,CFI = 1.00,RMSEA = 0.00, R 2 = 0.55).The only significant interaction term was cues to action (−0.17), which had significantly less influence on the intention to vaccinate for caregivers compared to adults.While one interaction term was a significant predictor, including the interaction term did not improve SEM model fit for the combined sample data as determined by a likelihood-ratio test (log-likelihood of the restricted model = −15,539, log-likelihood of the full model = −23,040, LR 2 7

Figure 4
Figure4presents the standardized path coefficients for the interaction terms between HBM constructs, vaccine attitude, and an indicator variable for residing in a high-incidence state for adults.The SEM estimates for this model provided a good model fit (χ 2 = 1729.83***,CFI = 1.00,RMSEA = 0.00, R 2 = 0.63).Adults residing in neighbouring states were more influenced by cues to action than adults residing in high-incidence states (p < 0.05).Figure5presents a similar model but with caregivers.The SEM estimates for this model again provided a good model fit (χ 2 = 1508.72***,CFI = 1.00,RMSEA = 0.00, R 2 = 0.60).Here, we observed no statistically

E 4
Prediction of adult intention to vaccinate from health belief model (HBM) and vaccine attitude for high-incidence state interaction terms.The interaction model was estimated using 1761 Adult observations.***, **, and * denote a p-value <0.001, 0.01, and 0.05.degree of control over their children's exposure to LD environments compared to residents from high-incidence states.In contrast to the findings of Stark et al. (2024), caregivers residing in neighbouring states Demographic characteristics of survey respondents.List of items used to assess the health belief model (HBM) constructs and associated means and Cronbach's alphas.
All data analysis was conducted in Stata (Standard Edition, Version 17; StataCorp LLC, College Station, Texas, USA).Significance was reported for predictors with p-values less than 0.05.TA B L E 1a Respondents in the caregiver survey were asked to report their own age group and that of their child.bProportionsmay not sum to 100 due to rounding.TA B L E 1 (Continued)TA B L E 2