Prioritizing US Cervical Cancer Prevention With Results From a Geospatial Model

Purpose To determine if differences in screening and vaccination patterns across the population may accentuate ethnic and geographic variation in future burden of disease. Methods Using Cancer in North America data provided by the North American Association of Central Cancer Registries, county cervical cancer incidence trends from 1995 to 2009 were modeled for the entire United States using ecologic covariates. Rates for health service areas were also modeled by ethnicity. State-level incidence was mapped together with Papanicolaou (Pap) screening, past 3 years (women ≥ 18 years old), and three-dose human papillomavirus (HPV) vaccine coverage (girls 13 to 17 years old) to identify potential priority areas for preventive services. Results US cervical cancer incidence decreased more during the periods 1995 to 1999 and 2000 to 2004 than during the period 2005 to 2009. During these 15 years, the most affected areas became increasingly confined to Appalachia, the lower Mississippi Valley, the Deep South, Texas, and Florida. Hispanic and black women experienced a higher incidence of cervical cancer than both white and Asian and Pacific Islander women during each period. Women in 10 of 17 states/districts with a high incidence (≥ 8.14/100,000) reported low Pap testing (< 78.5%), HPV vaccine coverage (< 33.9%), or both prevention technologies. Conclusion The decline in cervical cancer incidence has slowed in recent years. Access to HPV vaccination, targeted screening, and treatment in affected populations is needed to reduce cervical cancer disparities in the future.


INTRODUCTION
Cervical cancer incidence and mortality in the United States have declined steadily in the decades since the adoption of cytology-based Papanicolaou (Pap) testing. 1 Recently, cervical cancer death rates seem to have plateaued, and a slight decrease has been reported in the use of Pap testing. 2 Cervical cancer disparities (the uneven distribution of burden to areas of lower socioeconomic status and minority ethnic groups) are increasing. [3][4][5][6] Cervical cancer incidence is elevated among older and black women 7,8 and in impoverished areas such as Appalachia, the lower Mississippi Valley, and the United States-Mexico border area. 9 Reduced access to screening services is a driving factor in these areas. Nationally, 3-year Pap screening completion is . 80% for women 21 to 65 years of age, with lower levels of screening among immigrants, minorities, the poor, and older women. 10 Approximately 47% of girls born in the United States in 2000 had received at least one dose of human papillomavirus (HPV) vaccine by age 13 years. 11 Safety concerns and views on sexual activity influence parental decisions regarding the vaccination of adolescent girls. 11 Although the safety and efficacy of the HPV vaccination is well established, in 2013, 38% of girls 13 to 17 years of age in the United States had completed the three-dose HPV vaccine series, 11 far less than optimal. 12 Because the reduction in cervical cancer incidence as a result of HPV vaccination will take decades to be realized, 12 surveillance is needed to identify communities with a higher burden of cervical cancer and to direct cervical cancer screening and vaccination services where they are needed most. 1 To define high-priority areas for preventive services, we estimated small-area cervical cancer incidence, including by ethnicity, and mapped state-level incidence with percentage screening and HPV vaccination coverage.

Incidence Data
The North American Association of Central Cancer Registries (NAACCR) provided a Cancer in North America data set with county-level incidence and population denominators for this analysis. A total of 36 of 50 state registries, all of which met NAACCR silver or gold standards for tumor registration data quality, gave active consent to include their incident cases in the analytic data set. For the years 1995 to 2009, complete or partial countylevel cervical cancer incidence data were reported by NAACCR registries in areas covering 74% of the

Spatiotemporal Model
We developed a model similar in composition to one used commonly to predict current-year cancer incidence in the United States 13 to estimate county-level cervical cancer incidence in 5-year intervals from 1995 to 2009. 14 The multivariable logistic regression model included a set of covariates selected through a forward and backward process. A generalized linear model was used (PROC GLIMMIX, SAS 9.3, SAS Institute, Cary, NC) with three random terms to account for spatial autocorrelation (longitude and latitude of the county), temporal autocorrelation (year of diagnosis), and residual autocorrelation of covariates. Missing cervical cancer incidence at the county level was modeled on the basis of the reported incidence in counties with comparable attributes. The incidence for counties with observed data was also updated according to model predictions to slightly adjust reported rates. The time periods of interest for modeled all-ethnicity county-level incidence were 1995 to 1999, 2000 to 2004, and 2005 to 2009. To present incidence among ethnic groups, data were aggregated at the health service area (HSA) level, 15 reducing instability from small counts at the county level. HSAs are either a single county or a cluster of contiguous counties that are relatively self-contained with respect to hospital care. Data were suppressed when there were 16 or fewer modeled incident cases. The accuracy of predicted rates for areas with missing data depends on how well covariates in the model predict the actual incidence.

Model Inputs
Demographic inputs of cervical incident cases were non-Hispanic ethnicity (hereafter referred to as white, black, American Indian and Alaska Native, and Asian and Pacific Islander), Hispanic ethnicity (all ethnicities), and age. County of residence was geocoded on the basis of latitude and longitude. County-level population estimates were obtained from the US Census Bureau Summary File for each year from 1995 through 2009. Incidence data, stratified by age (, 44, 45 to 64, and > 65 years) and year of diagnosis were retrieved using SEER*Stat 8.1.2 (Information Management Services, Calverton, MD).
County-level covariates included in the model were county-level rural-urban density data, 16 an Area Health Resources covariate enumerating the number of hospital-based physicians at the county level, 15 and data on the percentages of the county population who were black, Asian and Pacific Islander, and American Indian and Alaskan Native. Cervical cancer mortality data reported by the National Center for Health Statistics were also included for each county in the United States. Socioeconomic covariates incorporated into the model were the percentage of the county population with income below the poverty level and the percentage of the population > 25 years of age with > 4 years of college education. The model also contained a variable indicating whether the county was in a National Program of Cancer Registries-funded area. The model provided estimates of county-level cervical cancer incidence for the entire United States, including areas and years with missing data as well as those with reported data. To illustrate geographic distributions in rates, modeled county estimates were mapped using 2000 Census county designations (ArcGIS 10.1; ESRI, Redlands, CA). In the all ethnicities combined model uncertainty related to small numbers was addressed with spatial smoothing. The population-weighted, nonparametric algorithm used universal Kriging after detrending. 17 Data were approximately normally distributed and no transformations were applied, although first-order surface trend was removed.
Smoothing was not applied to incidence maps of specific ethnic groups because of the potential for instability.

State-Level Incidence Trends
State-level modeled incidence trends on the basis of estimated age-standardized incidence were analyzed using joinpoint regression analysis (Joinpoint 3.5.0; Information Management Services). The technique fit a series of joined straight lines on a logarithmic scale for annual age-standardized rates. 18 Using data from the model for each year, state trends were estimated for 5-year fixed intervals, or annual average percent change (AAPC), using weighted annual percent changes from joinpoint models.

Identification of Priority Areas
Three state-level variables were visualized: modeled cervical cancer incidence during the period 2005 to 2009, state-level proportions of women 21 to 65 years of age who had had a Pap test during the past 3 years 19 estimated from pooled 2008 to 2010 responses to the National Health Interview Survey and Behavioral Risk Factor Surveillance System, and HPV vaccination coverage by state (obtained from the National Immunization Survey of Teens performed during 2013) 20 defined as the percentage of girls 13 to 17 years old who had received three doses or more of either the bivalent or the quadrivalent vaccine.
State-level cervical cancer incidence was mapped in tertiles (low, midlevel, and high incidence). Low incidence was 5.17 to 6.75, midlevel incidence ranged from 6.76 to 8.13, and high incidence was 8.14 to 9.76 cases per 100,000 women. Low statelevel Pap screening coverage was defined as , 78.5%, midlevel as 78.5% to 81.4%, and high as 81.5% to 88.5%. Low state-level HPV vaccination uptake was defined as , 33.9%, midlevel as 33.9% to 40.1%, and high as 40.2% to 56.5%. States in the bottom third distribution for threedose HPV vaccine receipt (, 33.9%), Pap screening (, 78.5%), or both vaccine and Pap screening were depicted with horizontal, vertical, and crosshatched lines, respectively. The status of state Medicaid expansion as of January 2016 was also assessed. 21

RESULTS
Modeled cervical cancer incidence and observed death rates per 100,000 women from 1995 to 2009 are presented in Table 1. Incidence and death rates were highest among non-Hispanic black followed by Hispanic women in each time

Ethnic/racial groups with highest cervical cancer incidence rates
Racial groups with lowest cervical cancer incidence rates Kentucky, and New Jersey). Many of the 16 states with a midlevel cervical cancer incidence (shown in orange) were adjacent to high-incidence states. One midlevel-incidence state (Missouri) had low uptake of both Pap screening and HPV vaccination. Among the midlevel-incidence states, five (New Mexico, Wyoming, Nebraska, Oklahoma, and Ohio) had a low percentage of Pap screening, and five (Alaska, Kansas, Maryland, North Carolina, and Georgia) had low vaccine uptake. States with the lowest cervical cancer incidence (shown in gold) were located in New England, the northern tier of states, the four corners region, and Virginia. In three low-incidence states (Montana, Idaho, and Utah), a low percentage of women received Pap screening, and a low percentage of girls received three doses of HPV vaccine. There was also a low percentage of Pap screening in Arizona, North Dakota, and South Dakota, and the percentage of girls in Virginia who received HPV vaccination was less than 33.9%.
As of January 2016, six of 17 jurisdictions with the highest cervical cancer incidence had not expanded Medicaid coverage (Alabama, Florida, Mississippi, South Carolina, Tennessee, and Texas). Seven of 16 states with a midlevel cervical cancer incidence had not expanded Medicaid (Georgia, Kansas, Missouri, North Carolina, Nebraska, Oklahoma, and Wyoming). Among the 18 states with the lowest cervical cancer incidence, three had not expanded Medicaid (Idaho, Montana, and Utah).

DISCUSSION
This study suggests that the long-term decrease in US cervical cancer incidence is slowing down. This finding is consistent with the nonsignificant decreasing trend reported during the period 2005 to 2009 in areas in the United States with highquality incidence data. 12 High-incidence geographic areas were confined increasingly to rural areas within Appalachia, Texas, the lower Mississippi Valley, and the southeastern United States. Hispanic and black women had a higher cervical cancer incidence than did white and Asian and   Pacific Islander women. In 10 of 17 states/districts with a high cervical cancer incidence, there was low use of Pap screening or HPV vaccination. Locally tailored cervical cancer vaccination, screening, and treatment efforts that target poor women living in medically underserved geographic areas are needed to maintain progress in reducing cervical cancer disparities.
As recently as the 1970s, cervical cancer was a leading cause of cancer among US women; however, the incidence has decreased in subsequent decades. 8 Trends in this study differ from sustained decreasing cervical cancer incidence trends over the past half-century, but are consistent with recent findings describing a leveling off of cervical cancer mortality 2 and incidence 12 in the United States. A possible explanation for the slowing of the decreasing trend is that women with access to health care are benefiting from preventive services such as Pap screening and HPV testing, 2 to a greater extent than are women in medically underserved groups. These underserved women, who experience a higher burden of cervical cancer, 23 include ethnic minorities, women from low socioeconomic backgrounds, 5 and women living in impoverished geographic areas. 10 Future progress to reduce the burden of cervical cancer depends on access to vaccination, screening, and treatment of these hard-toreach groups. 24 Of note, some states with the highest incidence of cervical cancer have low percentages of Pap screening and HPV vaccine uptake. Provisions of the Affordable Care Act, which require most health insurance plans to cover cervical cancer screening and HPV vaccination with no cost sharing, could improve cervical cancer prevention among low-income women. 25 In this study spanning the years 1995 through 2009, there were progressively smaller areas with an elevated cervical cancer incidence over time. Regions with the highest burden of disease during the period 2005 to 2009 were largely contained to economically deprived counties within Appalachia, Texas, the lower Mississippi Valley, and the southeastern United States. The limited progress in reducing the incidence of this cancer in areas with slow economic development or an influx of immigrant populations is consistent with findings reported in Mexico, 26 Brazil, 27 and England. 28 Lack of awareness, lack of access to health care, and cultural beliefs are barriers to cervical cancer prevention within population subgroups. [29][30][31] For instance, in Connecticut, 32 heterogeneity in the occurrence of cervical cancer precursors is reported.
Culturally competent locally targeted outreach needs to be part of cervical cancer control programs. One study of African-American women living in the high cervical cancer mortality area of Sunflower County, Mississippi, indicated that door-to-door visits to offer home self-collection HPV test kits increased participation in cervical cancer screening almost four-fold compared with clinic-based Pap testing alone. 33 In a national study, predictors of not being screened for cervical cancer included not having made a physician office visit within the past 12 months because of cost, minority ethnicity, lack of a high school diploma among residents of metropolitan areas, and selfreported fair or poor general health among nonmetropolitan area residents. 34 The heterogeneity of underserved women suggests a need for screening and HPV vaccination outreach across broad areas. 35 Other researchers have reported ethnic and geographic disparities in cervical cancer screening 34 and incidence. 36 In the United States-Mexico border area, Hispanic women were less likely than other women to have had a recent Pap test, 37 and white women in Appalachia had higher rates of HPV infection compared with the US population. 38 Although provider recommendation improves acceptance of HPV vaccination, minority and low-income women are least likely to receive such recommendations. 39 Cervical cancer prevention can be advanced through community-based interventions, 40 particularly in communities with limited access to a formal health care system. These community-based interventions may be more effective than one-size-fits-all approaches. [41][42][43] This study identified populations that would benefit from cervical cancer outreach by ethnicity, geography, and access to screening and HPV vaccination. Ongoing spatial analysis is recommended to monitor cervical cancer trends in the HPV vaccine era. 44 Study limitations include missing data for some states, which was partially addressed with geospatial modeling. Furthermore, HPV vaccination data were available only at the state level. County-level data would improve prioritization of outreach to areas with a high cervical cancer burden. Despite limitations, the analysis identifies priority areas for interventions to improve screening and vaccination rates. Although progress has been made in reducing the incidence of cervical cancer, outreach is needed in low-socioeconomic areas of the United States. Provisions of the Affordable Care Act that eliminate cost sharing for cervical cancer screening and HPV vaccination in most health plans should reduce cost as a barrier to receiving these prevention services.