Influence of Community-Level Factors on Health Care Provider Recommendations for HPV Vaccine

Human papillomavirus (HPV) vaccine receipt varies across the United States. In 2022, approximately 76.0% of adolescents received at least 1 dose of the HPV vaccine nationally,¹ and state-level coverage of at least 1 dose ranged from 61.0% to 94.6%.² Socioeconomic factors, such as area-level poverty, racial and ethnic composition of neighborhoods, and urbanicity, are associated with differences in the rates of HPV vaccine receipt.^3–9 Health care access alone does not fully explain these patterns in vaccine receipt.¹ A better understanding of the potential drivers of this geographic heterogeneity is needed to identify neighborhoods and communities where interventions can have the greatest impact.

Health care providers (HCPs) play a critical role in vaccine uptake among adolescents. Receipt of an HCP recommendation is one of the strongest predictors of HPV vaccine receipt among adolescents.¹⁰ HCP recommendation practices may be influenced by the characteristics of the neighborhood in which the provider practices.¹¹ A systematic review of disparities in HPV vaccine recommendations found lower rates of HCP recommendations in the South and in more rural areas.¹² There are multiple possible mechanisms for this influence. The HCP’s own knowledge, attitudes, and beliefs may be influenced by the area in which they live and practice. One ecological study found lower receipt of an HCP recommendation for the HPV vaccine in states with higher levels of religious ideology.¹³ However, the influence of community-level factors on HCP knowledge, attitudes, and beliefs about vaccines has not been fully explored.

Another potential mechanism is that the HCP may consider neighborhood characteristics and make assumptions about their individual patients. Spatial stigma is a concept that hypothesizes that individuals who live in disadvantaged areas are further stigmatized because of their association with that area.¹⁴ For example, an HCP working in an area with a high level of poverty or other markers of low socioeconomic status (SES) may perceive a patient as being at higher risk for HPV and therefore recommend the vaccine more strongly. HCPs may also recommend vaccines differently based on the perceived characteristics or attitudes of the patient in front of them.^3,12,15–20 A study examining HPV vaccine receipt among adolescent boys found that Hispanic and non-Hispanic Black boys had higher odds of HPV vaccination in high-poverty areas than their counterparts in low-poverty areas.⁴ Interactions between patient characteristics and community-level factors have not been examined for HCP recommendations.

In this study, we characterize how HCP recommendations for the HPV vaccine vary by community-level characteristics. Additionally, we hypothesize that community-level factors (ie, area-level socioeconomic characteristics) interact with patient sociodemographic characteristics to influence receipt of an HCP recommendation. A better understanding of the drivers of geospatial variation in HCP recommendation practices for the HPV vaccine can provide insight into potential inequities in vaccination uptake. These results can be used to inform where and to what populations evidence-based provider education and communication training interventions should be targeted.

We conducted a multilevel analysis of nationally representative data from the Centers for Disease Control and Prevention (CDC) National Immunization Survey-Teen (NIS-Teen), the US Census Bureau, and the Immunization Action Coalition (IAC) from 2008 to 2019 to examine how community-level factors at the zip code level are associated with provider recommendations of the HPV vaccine. Data collection for NIS-Teen was authorized by the Public Health Service Act and sponsored and conducted by the CDC’s National Center for Immunization and Respiratory Disease. Data collected by the National Center for Health Statistics (NCHS) may be used only for statistical reporting and analysis. To merge NIS-Teen and Census Bureau data we obtained restricted use zip code data from NIS-Teen. To access restricted use variables, we submitted an application to the NCHS Research Data Center (RDC) review committee. Once approved, the restricted use data were merged with publicly available data by RDC staff. We then accessed the final merged data through the Yale University RDC. Analysis of de-identified survey data was deemed exempt by the Yale University Institutional Review Board.

NIS-Teen is a nationally representative survey of the parents/guardians of adolescents aged 13–17 years conducted annually by the CDC to describe trends in vaccine uptake among the adolescent population in the United States. The survey sampling procedures used by NIS-Teen are described in detail elsewhere.^21–27 Briefly, NIS-Teen conducts 2 phases of data collection for a large probability sample of teens. First, a random-digit dialing telephone survey of households with an age-eligible adolescent is conducted. A priori target sample sizes are determined for 62 estimation areas including all 50 states, the US Virgin Islands, Puerto Rico, and 10 additional metropolitan areas. Household interviews are conducted with an eligible parent or guardian. In the second phase, the adolescents’ HCPs are contacted to obtain provider-verified immunization records. Approximately 60% of eligible teens have adequate provider data, and this analysis was limited to that sample. This analysis uses data from 2014 until the most recently available pre–COVID-19 pandemic year (ie, 2019). In addition to the primary outcome of interest (receipt of a recommendation), we also included several patient-level factors as potential confounders. The variables included from NIS-Teen were selected based on factors that had been previously identified in the literature as associated with provider recommendations or adolescent vaccine receipt: adolescent race and ethnicity, maternal education, household income, age of the adolescent, insurance coverage, and sex of the adolescent.^{1,12,19,28,29} 

Information from the 2010 US Census³⁰ was used for community-level measures of socioeconomic factors of interest identified through a review of the literature: median income, education, proportion of the population identifying as white, proportion unemployed, proportion publicly insured, proportion below the poverty level, and population density.^3,4,12,31,32 Census data are publicly available at the zip code tabulation area (ZCTA) level. ZCTAs are a Census Bureau measure used to map onto zip codes, which are a proprietary geographic product of the US Postal Service.³³ Census variables were categorized into quartiles for most analyses. For interaction analyses, the relevant census variables were converted into bivariate variables using a priori cutoffs. The percentage of the population under the poverty level was categorized into high poverty (≥20% of the population below the poverty level) and low poverty (<20% of the population below the poverty level) based on previous definitions.³⁰ The percent of the population that was white was categorized into “majority white” and “majority minority” using 50% white as the cut-off.

State-level school entry vaccine requirements are known to influence provider recommendation practices, including spill-over effects.^11,34–36 Therefore, we also included state vaccine requirements for meningococcal, tetanus, diphtheria, and acellular pertussis and HPV vaccines. Information on state vaccine policy was obtained from the IAC (now Immunize.org), which maintains a database of state-level middle school entry vaccine requirement policies by year.³⁷ 

US Census data and NIS-Teen data were linked using ZCTA for the Census data and participant zip codes for NIS-Teen data. A masked version of the ZCTA was included in the merged dataset provided by the NCHS for the purposes of conducting multilevel analyses without the actual zip codes of NIS-Teen participants to protect confidentiality. Immunization policy data were linked with NIS-Teen and Census data by state Federal Information Processing Standard code, a numerical code assigned to each state.

The data were combined across all years and limited to those respondents with adequate provider data for a total of 121 614 (approximately 20 000 observations per year). The weights for provider-verified data were recalculated per the instructions in the NIS-Teen user guide for merging data across years.^21,27 

The primary outcome of interest was receipt of a recommendation from an HCP for the HPV vaccine. Parents/guardians could respond “Yes,” “No,” “I Don’t Know,” or refuse to answer. For the purposes of this analysis, any respondents who did not respond were categorized as missing. Participants who responded “I Don’t Know” were combined with “No.” Bivariate analyses were conducted using Wald chi-square tests to evaluate the association between all parent/guardian characteristics and community-level factors of interest with an alpha level of 0.05 and were conducted using the appropriate procedures for complex survey data in SAS 9.3 (Cary, NC) (ie, PROCSURVEYFREQ).³⁸ A collinearity diagnostic was run, and no collinearity of concern was identified between the selected variables.

To examine the association between community-level factors and receipt of a recommendation, we used a model that included the main effects for all community-level factors of interest and adjusted for patient characteristics and policy-level factors. To examine the interaction between community-level factors and patient characteristics, we separately added 4 interaction terms to the model described above. The interaction terms evaluated were determined a priori based on the literature about parent/guardian characteristics and community-level factors associated with adolescent vaccine uptake and included interaction between community-level racial composition and adolescent race and ethnicity, between community-level poverty and adolescent race and ethnicity, between community-level poverty and adolescent poverty status, and between community-level poverty and maternal education.^3,12,32 A Wald chi-square test was run to evaluate if any of the interaction terms were statistically significant with an alpha level of 0.05.

Multivariate logistic regression models were run using PROC GENMOD in SAS, which implements a generalized linear model and can account for stratum-specific weighted analysis.^4,38 Backward elimination was used to identify the most parsimonious model with the inclusion criteria P is less than or equal to .10, retaining any variables necessary to ensure the model was hierarchically well formulated. All models produced adjusted odds ratio (aOR) and 95% CIs. In addition to the multilevel multivariate weighted logistic regression models described above, models were also run accounting for weighting but not clustering (PROC SURVEYLOGISTIC) and not accounting for weighting or clustering (PROC LOGISTIC) to confirm the robustness of the final modeling approach as well as models without interaction terms to evaluate main effects.

Characteristics of the study population are described in Table 1. There were 121 614 respondents with adequate provider data in the 2014–2019 data set. Almost three-quarters of respondents (72.7%) received a recommendation for the HPV vaccine during 2014–2019. Results of the descriptive bivariate analyses are in Supplemental Table 1.

Community-level education, population density, and proportion unemployed were associated with receipt of a recommendation for the HPV vaccine adjusted for patient characteristics and policy-level factors. Participants living in zip codes with the highest proportion of residents having received at least a high school degree were more likely to have received a recommendation for the HPV vaccine (Table 2; aOR for the highest quartile, 1.37; 95% CI, 1.23–1.52). Living in a higher population density zip code was associated with an increased likelihood of having received a provider recommendation compared with those living in the lowest population density quartile (aOR for the highest quartile, 1.20; 95% CI, 1.10–1.30). Participants living in areas with higher rates of unemployment were also more likely to have received a recommendation (aOR for the highest quartile, 1.14; 95% CI, 1.04–1.25). Full model results are in Supplemental Table 2.

There was statistically significant interaction between community-level poverty status and adolescent race and ethnicity (P = .0347; Supplemental Table 3). Non-white respondents living in a high-poverty zip code were 1.16 times more likely to have received a provider recommendation for the HPV vaccine compared with non-white respondents living in a low-poverty zip code (95% CI, 1.02–1.32; Figure 1). However, there was no significant difference between non-white participants living in a low-poverty zip code and white participants living in a low-poverty zip code or between non-white participants living in a low-poverty zip code and white participants living in a high-poverty zip code.

Results for the additional modeling approaches are presented in Supplemental Tables 4–7 and were concordant with the results presented above.

In this analysis of NIS-Teen data from 2014 to 2019, receipt of a provider recommendation for HPV vaccine varied by community-level socioeconomic measures. Parents were more likely to have received a recommendation for the HPV vaccine if they lived in zip codes that were more densely populated and had a greater proportion of individuals with a high school education and higher levels of unemployment. Additionally, non-white adolescents living in high-poverty zip codes were more likely to have received a recommendation for the HPV vaccine compared with non-white adolescents in low-poverty zip codes. Understanding the reasons for these differences can provide insight into potential areas for intervention to improve provider recommendation practices.

Studies have previously found that HCPs are more likely to report recommending HPV vaccine in urban compared with nonurban areas.³⁹ Similarly, lower levels of parental report of recommendation receipt have been observed in more rural areas.¹² This may reflect a difference in risk perception. HCPs in urban areas may perceive their patients as higher risk for HPV infection. One study found that Appalachian pediatricians were less likely to perceive their patient population as at risk for HPV and reported lower rates of HPV vaccine encouragement compared with pediatricians outside of Appalachia.⁴⁰ HCPs in more urban areas may also have received more education and support for HPV recommendations.⁴ A systematic review of interventions to improve HPV vaccine uptake in rural areas only found 1 study that addressed provider recommendations.⁴¹ Adolescents in urban areas may also have more consistent access to health care, leading to more opportunities to receive a provider recommendation. However, all adolescents included in this study had to have at least some access to health care to have a verified vaccination history. Studies evaluating area-level variation have also found higher levels of vaccine receipt in urban districts.^3,5,8 Our findings demonstrate the need to provide targeted education and support to HCPs in nonurban areas to improve vaccine receipt in these communities. Rates of HPV-associated cancers are highest in rural areas and are increasing,⁴² so it is critical that HCPs are recommending the HPV vaccine to their patients in nonurban areas.

We also found that individuals living in zip codes with higher levels of education were more likely to receive a recommendation for the HPV vaccine. This finding is consistent with existing health disparities in the United States⁴³ and mirrors findings on HPV vaccine series initiation.⁸ However, this does not entirely explain the associations found in this study, as we also found that individuals living in zip codes with higher levels of unemployment were more likely to receive a recommendation for the HPV vaccine despite not finding an association between other markers of SES and receipt of a recommendation for the HPV vaccine (eg, community-level poverty and median income). Previous studies on the association between area-level SES and HPV vaccine receipt have had mixed results. Other studies using NIS-Teen data found greater odds of initiation/completion of the HPV vaccine series in areas with higher levels of poverty.^4,32 However, studies examining patterns at the county or state levels have found lower rates of initiation/completion of the HPV vaccine series associated with lower area-level SES.^8,31 These associations are complex, and further research is fully needed to understand the relationship among area-level SES, HCP recommendations for the HPV vaccine, and vaccine uptake.

Although we did not find an independent association between income or poverty and a recommendation for the HPV vaccine, we did find an interaction between poverty and adolescent race and ethnicity with HCPs being more likely to recommend the vaccine to minority adolescents in high-poverty neighborhoods compared with low-poverty neighborhoods. Again, this could reflect a tendency of HCPs to perceive risk differently based on race and ethnicity and neighborhood poverty, leading to different recommendation practices.^44–46 

Because we limited this analysis to NIS-Teen participants who had provider-verified vaccination histories, it is possible that some of the community-level associations identified in this analysis can be explained by access to health care as briefly noted above. Disparities in access to health care exist across many axes, including urbanicity, race and ethnicity, and SES.⁴⁷ Numerous studies have noted different measures of access to care (eg, insurance coverage, use of screening services, or number of visits to an HCP) are associated with adolescent vaccine receipt.^7,13,48 Individuals living in more urban areas generally have more access to health care than their counterparts in more rural areas, although this can be mediated by race and ethnicity or SES.⁴⁷ Additionally, as noted above, all NIS-Teen respondents included in this study had some access to health care.

Although the large sample size and representativeness of NIS-Teen data allow for robust analyses, there are several limitations to this study. The outcome of parental report of a provider recommendation cannot be verified, that is, whether the parent actually received a recommendation. Nor can the quality of recommendations, an important aspect known to be associated with vaccine uptake, be characterized from these data.^44,49 Furthermore, zip codes may not perfectly align with neighborhood characteristics in the same way that census tracts do, which may limit the interpretation of these findings for all communities. Participants also may not seek health care in the same zip code where they reside, or they may have moved between receipt of a provider recommendation and participating in the NIS-Teen survey, resulting in measurement error.

The statistical approach applied in this analysis was based on those used in previous examinations of NIS-Teen data and is designed to provide a conservative evaluation of the association between community-level factors and provider recommendations. The associations of interest were determined a priori: we selected only those community-level factors and interactions that had previously been identified in the literature as being associated with either provider recommendations or HPV vaccine receipt. However, we acknowledge that alternate model selection may have been more appropriate. Backward elimination for model selection can lead to locally not globally optimal models and doesn’t account for model selection uncertainty in the final estimate. Stepwise regression models have several known limitations^50,51 including the potential for biased regression coefficients and falsely narrow CIs. Alternative approaches (eg, Lasso or Elastic-Net models) with cross-validation might provide more robust results and should be considered in future research.

A better understanding of how and why HCP recommendation practices for the HPV vaccine vary at the community level can help inform interventions to improve HCP recommendations. It can also allow for a better understanding of geographic and community-level variation in HPV vaccine receipt among adolescents. Providing vaccine communication training and interventions in nonurban areas and areas with lower levels of education could improve recommendations and, in turn, vaccine uptake in these communities, which are frequently also those at greatest risk for HPV-associated cancers.⁴² Additionally, we identified that HCPs are more likely to recommend the vaccine to minority adolescents in high-poverty areas compared with low-poverty areas. Vaccine communication strategies that emphasize recommending the vaccine in the same high-quality way to all patients could help address any biases in provider recommendation practices. Ultimately, identifying and addressing the factors driving HCP recommendations of adolescent vaccines can help improve uptake of those vaccines, particularly in those communities most at risk for disease, improving health during adolescence and across the lifespan.