Background
Despite the proven success of childhood vaccination programs in reducing or eliminating numerous vaccine-preventable diseases (VPDs), vaccine hesitancy continues to be an issue throughout the United States. Vaccine hesitancy is a delay in acceptance or refusal of vaccines despite their availability and is well-understood to be driven by a complex set of factors that include social norms, previous experiences, and personal beliefs among others [
1]. While vaccine hesitancy is not a new phenomenon [
2] in the US, recent increases in the number of parents that refuse vaccination for their children [
3], as well as recent outbreaks of VPDs such as pertussis and measles [
4,
5], have resulted a renewed interest in understanding vaccine-related behavior. Moreover, this phenomenon is not constrained to the US; numerous countries around the world are also contending with issues stemming from vaccine hesitancy, including reduced levels of coverage and corresponding VPD outbreaks [
6,
7].
In the US, each state and the District of Columbia
1 develops and implements their own childhood vaccination laws, regulations, and procedures, which are generally enforced at school entry [
8]. For children with a contraindication for vaccination, all states have a provision for a medical exemption [
9]. Further, nearly all states allow parents to obtain a nonmedical exemption (NME) for their child based on personal, philosophical, or religious beliefs. Currently, only California, Mississippi, and West Virginia do not provide this option. In states having an NME provision, the reasons for which they can be granted vary (e.g., personal and/or religious beliefs), as does the relative ease in which they can be procured. The restrictiveness of state-level NME requirements has been shown to affect the number of parents exempting their children from vaccination, as states with more restrictive policies tend to have fewer NMEs [
10,
11]. Vaccine hesitancy in the US, as expressed through the use of NMEs, remains a highly contested topic across the country. In recent years, more than half of all states have considered and/or passed legislation that modified their existing NME requirements, including efforts to make NMEs both easier and more difficult to acquire [
3,
9,
12].
Much of the research regarding vaccine refusal and NMEs has focused on understanding
who are using NMEs, finding that parents choosing to acquire NMEs for their children tend to be well educated, high-income, and white, and that schools having high NME rates tend to be located in neighborhoods having similar characteristics [
13]. Another active area of research has been to examine
where parents choosing to use NMEs reside and whether they cluster geographically [e.g.,
14‐
16]. Understanding where and why geographic clusters of low vaccination coverage have formed is important, as these regions can be at risk of losing herd immunity and outbreaks of VPDs [
17‐
19]. Herd immunity is the indirect protection provided to those without immunity when overall vaccination coverage in the population is high, resulting in a decreased risk of disease transmission within the population [
20]. Although children with NMEs cannot be assumed to be fully unvaccinated (given limitations of some surveillance systems) [
21], numerous studies have shown a link between NME use and VPD outbreak risk [e.g.,
14,
22‐
24].
While the use of NMEs is usually associated with the refusal of one or more vaccines, vaccine hesitancy includes a range of potential parental behaviors regarding vaccines, from having or expressing concerns about vaccination (without action), to delaying the recommended schedule, to outright refusal [
25]. It is generally accepted that those that refuse vaccines make up a very small proportion of all vaccine-hesitant parents [
26]. Yet, the ability of those opposed to vaccination to disseminate their message and engage with other vaccine hesitant parents has been bolstered by the increasing availability and use of the Internet [
2]. One particular concern is that parents will advance along the spectrum of hesitancy (e.g., from resistant to refusal) as the debate surrounding vaccines, vaccine safety, and parental rights continues to play out online [
27].
Understanding vaccine hesitancy and VPD outbreaks in the US and other countries is and has been an active area of scientific inquiry; however, there has been less emphasis placed on how vaccine-related behavior changes over time and across space. There has been limited research on how vaccine refusal manifests throughout a region and whether there are observable spatial patterns that can shed light on the processes that drive hesitant or resistant parents to become vaccine refusers. The goal of this research is to initiate such an analysis by examining vaccine refusal over time via an analysis of the changing spatial patterns of NME use in a large study region. Evaluating NMEs for children entering kindergarten from year to year presents an interesting opportunity to understand this phenomenon, as there should be a different group of parents sending children to school each year.
We conduct an analysis in the state of California, where the NME rate for incoming kindergarteners increased from less than 0.5% in 1996–97 to more than 3% in 2013–14
2 [
28]. As a result of the increasing use of NMEs and the corresponding decrease in vaccination coverage within the state’s school system, California passed and implemented two laws aimed at curbing the use of NMEs. AB2109 was implemented prior to the 2014 school year and made NMEs more difficult to acquire by requiring all parents to receive counseling from a health care provider prior to obtaining a valid NME. SB277 was implemented prior to the 2016 school year and removed the NME provision entirely. We constrained our analysis to only consider the time period from 2000, the earliest year with publically available data, to 2013, the final school year prior to the implementation of AB2109. By restricting the analysis to this particular time period, we are able to examine the changes in vaccine-related behavior that were unaffected by the large-scale policy changes limiting California parents’ ability to acquire NMEs for their children.
We already know that state-level NME use increased in California from 2000 to 2013. Others have examined spatial clustering of NMEs in California over recent time periods in relation to pertussis outbreaks [
14], school characteristics and medical exemptions [
15], and policy changes [
29]. This research is solely focused on the local spatial patterns of NME use and how they evolved over time. For example, we are interested in establishing whether the statewide increase in NME uses was largely driven by smaller increases distributed across the entire state or spatially isolated pockets of substantive change. We address the following research questions:
1.
Did NME use become more spatially clustered over time?
2.
Did the location of local spatial clusters of NME use change over time?
3.
What were the general temporal patterns of NME use?
By examining the spatiotemporal nature of the changes in NME use, we aim to uncover whether vaccine refusal acted as a spatially diffusive or contagious process over this time period, as suggested by others [
15,
29]. The first two research questions focus on the changing spatial patterns of vaccine-related behavior in an effort to unravel the role that “space” or “location” may play over time. In particular, we are interested in whether NME use in California demonstrated some form of spatial structure, which would shed light on the role of local behavior, e.g., spatial diffusion, wherein NME use within one region may stimulate hesitant parents in nearby regions to become vaccine refusers. The third research question places an emphasis on the temporal trends of NME use; this question explores the statewide increase in NME use as a function of various local-level changes over the study period.
Discussion
By evaluating the spatiotemporal patterns of NME use over a 14-year time period for a state-level study area, this analysis helps to shed light on how vaccine refusal evolves over time across a large spatial extent. While NME use in California rose from 0.73% to 3.09% over this time period, our grouping analysis showed that 50% of the regions had consistently low NME rates (at or near 0%) over time. These findings show that the vaccine-related behavior of parents in many California regions was not substantially affected by the changes in NME use occurring in other regions over this time period. Another 35% of all regions had moderate increases in NME use, mirroring the statewide behavior. The remaining 15% of all regions in the state had large increases in NME use over this time period. Hence, the largest increases in vaccine hesitancy appear to have occurred in a relatively small proportion of regions throughout the state. In these regions, parents appeared to be influenced to use NMEs due to the relatively large proportion of other parents that already chose to use NMEs in prior years [
17].
Part of understanding the potential risks associated with vaccine refusal is determining where parents choosing NMEs reside, given current understanding how geographic clustering of communities having high NME rates and low vaccination coverage affects herd immunity and VPD outbreak risk. The results of the global clustering analysis clearly demonstrated that NME use in California became more spatially clustered over this time period. The local clustering results provided information regarding how the spatial clusters evolved over time, notably showing the expansion of the High NME use clusters and the contraction of the 0% NME use clusters. Hence, not only did NME use increase statewide over this time period, it became more geographically clustered overall, which produced expanding local clusters of high use. In these local geographic clusters, even if some of the students with NMEs had actually been vaccinated, there was an increased probability that unvaccinated (or undervaccinated) individuals could come into contact with one another, increasing the risk of disease transmission [
38] and the potential for outbreaks [
39]. This is especially salient when examining school age children, as schools offer an environment where large numbers of children interact on a regular basis. Another notable finding pertaining to disease transmission risk is the contraction of the 0% NME use clusters over time. Particularly, these locations could be viewed as relatively safe haven from potential outbreaks, as there were no students exempted from vaccination requirements at the specific location, nor in its immediate neighborhood. In the year 2000, for example, 27.84% of all schools (
n = 2017) were identified as being in a 0% NME use cluster, but that number fell to only 5.62% of schools (
n = 392) by 2013 as NME use expanded geographically.
The local spatial autocorrelation and the grouping analysis also provided interesting results regarding the potential origins of vaccine refusal in California. Notably, the regions with high NME use in the early years of the study period appeared to be small in number and relatively isolated geographically. The observed radial patterns in both the High NME cluster map and the group membership map, suggest that these isolated regions acted as seed locations that stimulated NME use in nearby regions. Specifically, the regions near the seed locations had larger increases in NME use than those further from the seed locations. The corresponding contracting radial pattern in the 0% NME cluster map provided further evidence of the phenomenon. While others have suggested NME use was a spatially diffusive process in California [
15,
29], this analysis provided a more rigorous analysis of this hypothesis and produced compelling evidence that spatial diffusion occurred over this time period.
We did not examine whether the observed spatiotemporal patterns were purely due to spatial diffusion or the result of social mechanisms that manifested as spatial patterns. For example, vaccine hesitancy could have expanded within and among groups having similar socioeconomic and demographic characteristics via social, rather than spatial, contacts and the observed spatial patterns are simply a reflection of social sorting or the self-selection of people with similar characteristics or vaccine-related beliefs into proximal geographic regions [
17]. By concentrating on geographic proximity, our results do suggest that parents choosing to get an NME may be bolstered by other nearby parents making a similar decision; however, we cannot firmly state whether this process was driven by interactions among parents living in geographically proximal regions (regardless of their social similarity or vaccine-related beliefs), by interactions among parents with similar social characteristics or vaccine-related beliefs who happen to live in geographical proximal regions, or some mixture of both. Future research that integrates socioeconomic or demographic similarity or social connections among regions may help to disentangle the roles that local spatial processes and social processes played in the overall increase in NME use and the observed spatiotemporal patterns.
Our analysis was conducted at multiple levels of data aggregation in an effort to assess the potential effects of the Modifiable Areal Unit Problem, wherein the results of statistical tests are sensitive to the scale of the observation units when using aggregated spatial data [
40]. While we focused on the Tract-level results, the results of the School, Block Group, and School District analyses were remarkably similar. This outcome supports our overall findings, as it demonstrates that the results were consistent across the multiple scales of analysis. Yet, the consistency also limits our ability to evaluate whether there is a spatial scale (e.g., family, school, local neighborhood, school district) at which the diffusion of vaccine refusal may operate; however, it does provide an interesting opportunity for future research.
One matter that was not considered in our analysis is the potential effect of school practices on the observed spatiotemporal variations in NME rates over time. Compliance enforcement and practices have been shown to be associated with school vaccinations rates [
41]. Prior to the implementation of AB2109 in 2014, NMEs were relatively easy to claim, and some California schools (11% of the 298 schools surveyed) offered a NME as a “convenience” option for parents of children not fully up to date on the school-entry vaccinations or that could not provide their child’s required vaccination records [
42]. While the vaccine-related beliefs of the parents were not assessed in the survey, an assumption is that these NMEs were potentially driven by the 1) the burden on parents of producing the required documentation or 2) the administrative burden on schools of offering a conditional admittance, which requires the school to follow-up at a later date [
29]. Increases in this school-level practice could account for some of the observed spatiotemporal patterns of NME use, especially if it were co-located in regions where vaccine hesitancy in parents was increasing. The effect of schools on NME use presents another interesting opportunity for future research.
Due to the cross-sectional nature of the data, our analysis was not able to evaluate whether parents “progressed” along the spectrum of vaccine hesitancy during the time period. However, we were able to evaluate how vaccine refusal evolved as a geographic phenomenon over this time period. We found that early in the study period, the number of regions having a high proportion of vaccine refusers was relatively small in number and these regions were isolated geographically throughout the state. Further, the use of NMEs in these isolated regions appeared to stimulate vaccine refusal in nearby regions, as the corresponding maps of NME use demonstrated a radial spatial pattern throughout much of the state.