County-level estimates of human papillomavirus vaccine coverage among young adult women in Texas, 2008

Abstract

Purpose

The purpose of this study is to describe the county-level geographic distribution of human papillomavirus (HPV) vaccine coverage among young women aged 18–26 in Texas using multilevel, small area estimation.

Methods

Multilevel (individual, county, public health region) random-intercept logit models were fit to HPV vaccination data (receipt of ≥ 1 dose Gardasil®) from the 2008 Behavioral Risk Factor Surveillance System and a number of secondary sources. Using the parameters from the final model, we simulated HPV vaccine coverage in each county.

Results

Indirect county-level estimates ranged from 1.9–23.8%, with a weighted state average of 11.4%. The counties with the highest and lowest coverage estimates were Orange County, TX and Webb County, TX respectively. Significant correlations were observed between HPV vaccination and age, Hispanic ethnicity, and the percentage of uninsured at the county and public health region levels.

Conclusions

Small area analyses have been used in a variety of settings to assess a variety of health outcomes, and as shown in this study, can be used to highlight geographic disparities and opportunities for intervention in HPV vaccine coverage.

INTRODUCTION

The U.S. Food and Drug Administration (FDA) approved the use of Gardasil® (HPV4), a quadrivalent vaccine against four prevalent strains of human papillomavirus (HPV), for females 9–26 years old in June 2006 ¹. The Advisory Committee on Immunization Practices soon thereafter recommended routine vaccination of females aged 11–12 and catch-up vaccination of females aged 13–26². Since 2008, over 30 peer-reviewed articles have examined HPV vaccine uptake in the U.S. and other countries. Most have focused on adolescent females. Surveillance of young adult women has not kept pace, leaving gaps in our knowledge of uptake and its predictors in this population. Although national uptake is known (by 2007, 10% of women aged 18–26 had received ≥1 dose HPV4; ^{3, 4}), no studies to our knowledge have explored geographic variation in uptake among young adult women. This information is critical for 1) identifying areas with low vaccine uptake, 2) developing geographic-specific interventions to increase vaccine uptake, and 3) strategic resource allocation, program planning, and policy-making.

The purpose of this study was to describe the geographic distribution of HPV vaccine coverage in Texas. We applied a multilevel, small area model of HPV vaccine initiation using the 2008 Texas Behavioral Surveillance System (TX BRFSS^{5, 6}) to provide county-level estimates of HPV vaccine coverage among females aged 18–26 years. Given the results of previous studies, we expected to find a negative association between age and HPV vaccination and a greater likelihood of vaccination among whites than among racial/ethnic minorities^4,7–9.

METHODS

Small area estimation (SAE) operates on the assumption that areas with the same characteristics will have similar outcomes. Thus, statistically we could borrow information from the population-based model to derive local estimates. Place-based random-effects are increasingly included in these population-based models to allow for location-specific patterns. In this study, we utilized multilevel (individual, county, and health service region (HSR)) SAE to derive county-level estimates of HPV vaccine coverage among Texas females aged 18–26. In Texas, there are 254 counties nested within 11 HSRs, which represent state-designated boundaries for the administration of public health services and resources. Hierarchical level data and a random intercept term were included to improve model fit and ensure the correct selection of group-level covariates and parameter estimates.

We used multiple data sources to facilitate our methodological approach. The individual-level dataset used to obtain the outcome (i.e., ever received ≥ 1 dose HPV4) and level 1 covariates (i.e., age, race/ ethnicity, and educational attainment) was the 2008 TX BRFSS. The group-level covariates (i.e., county and HSR, where available) to be tested were obtained from the U.S. Census Bureau (percentage of uninsured adults aged 18–64 and percentage of total population in poverty), the Texas Cancer Registry (cervical cancer incidence rate), the Association of Religion Data Archives (evangelical religious adherence rate), and U.S. Office of Management & Budget (rurality index). The auxiliary dataset, which enumerates the population in each county by age, sex, and race/ethnicity, was obtained from the National Vital Statistic System¹⁰. This dataset serves as the foundation for predicting vaccine coverage for both sampled and unsampled counties.

We constructed a series of increasingly complex, multilevel random-intercept logit models of HPV vaccination (ever received ≥1 dose HPV4, yes or no) in MLwiN Version 2.20¹¹. Adding covariates sequentially by level allows one to examine the contribution of covariate sets and how their addition changes the parameter estimates of the previous model. A total of 277 sampled women aged 18–26 from 61 Texas counties were included. Weights were incorporated at each level and scaled as recommended by Carle (Method A;¹²) and Goldstein¹³. A variance component term was included at the county and HSR level to ensure the correct selection of covariates and the accuracy of their parameter values.

After fitting the empty model, we added all level 1 covariates to the model. Retaining them regardless of statistical significance, we then evaluated level 2 covariates one by one. County-level covariates with p-values ≤0.10 were retained for further examination. Next, we evaluated level 3 (HSR level) covariates one by one, retaining only those with p-values ≤0.10. First order marginal quasi-likelihood was used to estimate all models in MLwiN. The preliminary model included all level 1 covariates and statistically significant (p-value ≤0.05) group-level covariates. Because the individual-level educational attainment was not available in the auxiliary dataset, it was removed in the final model and later adjusted for at the county level, as suggested by Congdon¹⁴. The regression coefficients for educational attainment from the preliminary model were used to calculate the weighted total of the regression coefficients in county j (i.e., variable L). This weighted total is used to adjust the county-level estimates by their respective educational characteristics in the simulation process described below.

Using the estimated regression coefficients and standard errors from the final model, we simulated 10,000 datasets for each regression coefficient from the normal distribution and applied them to the logit model to estimate the vaccination rate for each Race- and Age-specific group in county j (RAj), adjusted for county j’s educational characteristics (multiplied by L). Subsequently, we calculated the mean and standard deviation of the 10,000 RAj adjusted probabilities. To obtain summary estimates for each county, we took the population weighted mean of the 10,000 RAj adjusted probabilities (technical documentation available upon request). Thus, we weighted the probability for each demographic subgroup by their respective population distribution in county j. HSR random effects were not estimated due to insignificant variability in the final model and were ignored in the simulations. County random effects were only estimated for the 61 sampled counties in the 2008 TX BRFSS (non-sampled counties were assigned the simulated mean value). Simulations were conducted using SAS Version 9.2 ¹⁵.

RESULTS

In the TX BRFSS study population of 277 young women aged 18–26 years, the mean age was 22.7 years. Other demographic characteristics were as follows: 42.5% were non-Hispanic white, 8.3% were non-Hispanic black, 38.3% were Hispanic, 19.8% had less than a high school education, and 38.9% reported a lack of insurance. Compared to the state level estimates obtained from U.S. Census Bureau, BRFSS demographic estimates were similar.

In 2008, 12.0% (95% CI: 6.2, 17.7) of Texas women aged 18–26 reported having ever received at least one dose of HPV4 (direct estimation from the 2008 TX BRFSS). At the county-level, indirect estimates ranged from 1.9–23.8% (Figure 1; county-specific estimates available upon request). Aggregated to the state level, we obtained a population weighted average of 11.4% (95% CI: 10.3, 12.5), a value which, as expected, lies within the confidence interval of the direct estimate of HPV vaccine coverage in Texas. The county with the lowest vaccine coverage (Webb) was about 10 percent lower than the direct state estimate, while the county with the highest vaccine coverage (Orange) was about 12 percent higher than the direct state estimate.

Figure 1.

County-level estimates of HPV vaccine coverage among young women, aged 18–26: Texas: 2008

Although the purpose of model-based SAE is to provide small area estimates, model results are also important because they contribute to the formulation of the indirect estimates. Table 1 provides the results of the series of increasingly-complex multilevel models previously described. Compared to non-Hispanic whites, Hispanics were significantly more likely to be vaccinated against HPV (OR = 1.83, 95% CI: 1.35, 2.46). For age, we found an inverse association (OR = 0.80, 95% CI: 0.66, 0.95), indicating that the odds of receiving ≥1 dose HPV4 decreased with increasing age. This association was consistent across all models. Among tested group-level covariates, we found significant inverse associations between county- and HSR-level lack of insurance and HPV vaccination. The statistical significance and relative effect size of these covariates (i.e., age, Hispanic ethnicity, and county- and HSR-level lack of insurance) did not change with the removal of education from the final model.

Table 1.

Random-intercept logit model of HPV vaccination (≥ 1 dose HPV4) among young women, aged 18–26: Texas, 2008

Model Parameters	Model 1 OR (95% CI)	Model 2 OR (95% CI)	Model 3 OR (95% CI)	Model 4 OR (95% CI)	Model 5 OR (95% CI)
Fixed Effects
Intercept	0.13 (0.09,0.20)	0.04 (0.01, 0.14)	2.63 (0.11, 60.76)	23.59 (1.49, 374.13)	46.85 (4.18, 525.15)
Mean Centered Age	_	0.78 (0.64, 0.95)	0.78 (0.66, 0.94)	0.78 (0.65, 0.94)	0.80 (0.66, 0.95)
Non-Hispanic white	_	Ref	Ref	Ref	Ref
Non-Hispanic black	_	1.59 (0.57, 4.48)	1.89 (0.44, 8.11)	1.86 (0.51, 6.74)	1.54 (0.40, 5.95)
Hispanic	_	1.53 (1.00, 2.35)	2.08 (0.84, 5.17)	1.95 (1.39, 2.72)	1.83 (1.35, 2.46)
Other races	_	0.58 (0.18, 1.85)	0.95 (0.13, 7.00)	0.90 (0.37, 2.14)	1.15 (0.44, 2.98)
< HS1 education	_	Ref	Ref	Ref
HS graduate	_	1.87 (0.71, 4.94)	1.81 (0.46, 7.15)	1.88 (0.68, 5.21)
Some college or more	_	2.96 (1.22, 7.16)	2.58 (0.68, 9.82)	2.65 (1.09, 6.47)
Lack of insurance (County)	_	_	0.86 (0.77, 0.95)	0.87 (0.83, 0.92)	0.87 (0.83, 0.91)
Lack of insurance (HSR)	_	_	_	0.91 (0.85, 0.97)	0.92 (0.86, 0.98)

¹HS = High School

DISCUSSION

County-level estimates of HPV vaccine coverage among women aged 18–26 varied from 1.9–23.8% (Figure 1), with a population weighted state average of 11.4%. In this study, Hispanic women were more likely to be vaccinated against HPV than their White counterparts. This relationship was not found in other studies conducted nationally or in other regions^{4, 7–9}, which suggests either fundamental differences in the characteristics of Hispanics living in Texas versus other regions or conversely, better education and outreach initiatives targeting these individuals. In a previous study done by JME, Hispanic ethnicity was also a significant correlate among adolescent Texas females¹⁶. The group-level covariates of importance differed, however. For example, county-level poverty was strongly and positively associated with vaccination in adolescent females¹⁶, but not in young adult women. We suspect this is because high poverty counties have a large proportion of low income families, and income is inversely associated with eligibility for publicly-financed health insurance and services for children and adolescents. Through these publicly-funded programs (e.g., Children’s Health Insurance Program and Vaccines for Children), children and adolescents whose families meet certain income limits can received free or low cost vaccines including HPV4. Given that women aged 18 years and older are generally not covered by these programs, it is not surprising that county-level poverty does not equate to greater likelihood of HPV vaccination in this age group. Insurance coverage, on the other hand, may be very important for young women aged 18 years and older given the high cost of the vaccine series (currently $360 for 3 doses). Our research supports this association, although we were only able to include insurance coverage at the county level. Additional factors such as awareness of and access to Merck’s Vaccine Patient Assistance Program and other financing options (e.g., under/ uninsured young women can receive HPV4 from Federally Qualified Health Centers or Family Planning Clinics in Texas) may also be important, although these variables were not available for examination in our sample. Due to the self-reported nature of the survey used in this study, it is possible that the data are subject to recall bias. We should also note that our primary outcome, initiation of the HPV vaccine, is only a preliminary measure of vaccine coverage; future studies should use more current data to examine estimates of vaccine completion at the county-level. Finally, the seemingly contradictory finding that Hispanic ethnicity is positively associated with vaccination, while many border counties with large Hispanic populations have low vaccination rates is likely due to the strong influence of insurance coverage in the model.

Small sample size and privacy concerns often plague local health studies. Through the application of model-based SAE to survey data on HPV vaccination, we have shown how local health statistics can be derived and discussed their potential uses for both research and public health surveillance and control activities. In this study, we utilized a multilevel SAE framework to highlight geographic disparities and opportunities for intervention in HPV vaccine coverage among young women aged 18–26. This outcome is particularly well-suited for small area analysis because 1) receipt of the HPV vaccine and documentation of vaccination status are not required by most states (including Texas) making it difficult to accurately estimate HPV vaccine coverage in local areas, 2) geographic variation in adolescent vaccine coverage has been documented in recent studies^17,18, and 3) interest in increasing HPV vaccine uptake through local advocacy efforts, health education programs, and policies is on the rise. Although our study showed county-level differences in HPV vaccine coverage, even the county with the highest estimate had <25% of women aged 18–26 initiating HPV4, indicating that state-level policies might be more effective at bringing up coverage across the board. We propose that future small area studies on HPV vaccination examine the sensitivity of other modeling strategies in estimating county HPV vaccine coverage, estimate county HPV vaccine coverage among males, who can now receive the vaccine¹⁹, and assess whether coverage is increasing over time.