National Trends and Disparities in Cervical Cancer Screening among Commercially Insured Women, 2001–2010

Early detection of cervical cancer contributes to reduced risk of premature mortality. (1, 2) Women found to have early cervical cancer have a 5-year survival rate of more than 90%, and it is likely that almost all cervical cancer deaths could be prevented by adhering to screening and follow-up recommendations (3).

Cervical cancer incidence and mortality from 2006 to 2010 differed among white (7.2 and 2.1/100,000, respectively), Hispanic (9.6 and 2.6/100,000, respectively), and black women (9.8 and 3.9/100,000, respectively) as well as among younger and older women (4). Low socioeconomic status is associated with an increased risk of cervical cancer and mortality (5).

Substantial progress in increasing cervical cancer screening was made during the 1980s and 1990s with the advent of health plan-level performance measurement and public reporting, but survey-based studies found that rates trended down during the subsequent decade (6–12). Trends in socioeconomic and racial/ethnic screening disparities are less clear; two studies detected declining racial/ethnic disparities but these were driven by declining rates among white and black women (9, 12). Disparities by poverty level remained stable from 2000 to 2010 (12).

Consensus guidelines have evolved regarding the appropriate age of initiation and intervals for Papanicolaou (Pap) testing. Recent cervical cancer prevention guidelines from 2009/2012 now recommend screening every 2 to 3 years for women ages 21 to 29 years and every 3 to 5 years for women ages 30 to 65 years (13–15). Earlier guidelines (16–18) generally advocated more frequent testing. For example, the 2003 American College of Obstetrics and Gynecology guidelines (18) recommended that annual screening should begin approximately 3 years after initiation of sexual intercourse, but no later than age 21. It was further recommended that women 30 years and older who have had three consecutive test results negative for intraepithelial lesions and malignancy could be screened every 2 or 3 years. The Healthy People 2020 goals included 3-year cervical cancer screening rates of 93% for women ages 21 to 65 years (19). These goals were published in 2010, before the 2012 U.S. Preventive Services Task Force revision of cervical cancer screening guidelines that allow for increased screening intervals.

Over the past decade, cancer screening for women has also been a target of intensive quality improvement efforts (20). Pap testing is an established quality measure often included in pay-for-performance programs from health plans (20). These data are collected and reported for health plans nationally by the National Committee for Quality Assurance (20).

In this study, our objective was to assess trends in cervical cancer screening from 2001 to 2010 among commercially insured women by key demographic and age groups.

We studied enrollment information and administrative claims data during 2001 to 2010 from a large national health plan with members in all 50 U.S. states. We linked sociodemographic variables derived from the 2000 U.S. census using zip code of residence.

We included commercially insured women ages 21 to 64 years with 35 months or more of continuous enrollment between 2001 and 2010. We excluded women with evidence of hysterectomy or cervical cancer in prior years (see Supplementary Table S1).

Our primary outcome was the presence of at least one Pap test per 3-year interval among eligible women. Thus the denominator was “1” if a woman was enrolled for a given 3-year period, and the numerator was “1” if she had at least one Pap test any time during that interval. We used Current Procedural Terminology and International Classification of Diseases, 9^th Revision (ICD-9) codes to capture Pap testing in claims data (see Supplementary Table S1). For women with 71 or more months of continuous enrollment, we examined for a Pap test in each sequential 36-month span, requiring the final span be at least 35 months in duration. Thus, a given woman could be in the denominator in multiple years during 2001 to 2010. For spans that did not align with calendar year spans, we assigned the span to the 3 calendar years during which the greatest numbers of months occurred. We assigned July to June spans to the span including the later calendar year. We used similar methods for the annual screening outcome.

To generate proxy measures of socioeconomic status, we created previously established (21) categorical variables of census block group poverty and education levels derived from 2000 U.S. census reports (22). We used a woman's most recent address on record for generating geocoded neighborhood characteristics. We classified members as residing in white, black, or Hispanic neighborhoods based on living in neighborhoods with 75% or more persons of the given race/ethnicity; we assigned census blocks with 75% or more persons of both Hispanic ethnicity and black race to the Hispanic category. We classified members from census block groups that did not fall into one of the race/ethnicity categories as from mixed race/ethnicity neighborhoods. We created age categories of 21 to 29 years and 30 to 64 years and classified members by the region of the country in which they resided (Northeast, South, West, and Midwest) based on their most recent zip code.

We examined sociodemographic characteristics of the populations in 2001, 2005, and 2010. We then calculated cervical screening rates as the number of eligible women with a Pap test in a given 3-year period. We used the logistic version of generalized estimating equations (GEE) to model individual-level Pap testing rates and disparities to account for correlations (with robust variance) over years within each individual. We ran 6 total models: one model each for overall rates by age group (21–29 and 30–64) and two models within each age group that we used to calculate rates and disparities by neighborhood poverty level and race/ethnicity. To generate estimates for the overall age group, we included in the statistical model categorical variables for all levels of race/ethnicity, poverty, U.S. region, education, and each 3-year screening period. We used the STATA margins command on the 3-year screening period term to generate fully adjusted estimates of cervical screening rates. To estimate adjusted screening rates by race/ethnicity group as well as disparities simultaneously over time, we further included the interactions between the race/ethnicity category and the 3-year period variable in a single model after controlling for other covariates above that were not the stratifying covariate. We then used the STATA margins command on the interaction term of race/ethnicity category and 3-year screening period to estimate each 3-year screening rate for the subgroup of interest (e.g., white, black, Hispanic, etc.), an approach which accounts for coefficients of all other covariates. To estimate disparities between the race/ethnicity subgroups, we again used the margins command on the category of interest, while using the margins “over” option estimate margins at unique values of the 3-year screening period variable. We used the same approach to obtain adjusted screening rates and disparities for poverty levels. We plotted these fully adjusted 3-year Pap testing rates. To characterize and plot the magnitude and direction of trends in screening and disparities in these plots, we used joinpoint regression models and annual percentage change (APC) statistics. (23) Joinpoint software uses the grid-search method (24) to fit a regression function with unknown joinpoints. It finds the number of significant joinpoints by performing permutation tests and finds P values using Monte Carlo methods (25). We allowed a maximum of one joinpoint, included standard errors derived from our GEE models, and we fit an autocorrelated errors model based on the data. For each outcome, the model contains an estimated intercept and APC. For the disparities trend models, the intercept yields an estimate of the initial baseline disparity and the APC demonstrates whether the disparity increased, decreased, or remained stable. We conducted analyses using STATA 12.1, and the Joinpoint Regression Program 4.0.4 (NCI). The Human Subjects Committee at Harvard Pilgrim Health Care approved this study.

Table 1 shows characteristics of the populations in 2001, 2005, and 2010. Between 1.1 and 1.3 million women per year were eligible for the 3-year cervical cancer screening measure with median ages ranging from 42.3 to 44.6; 8.4% to 9.9% were ages 21 to 29 years and 90.1% to 91.6% were ages 30 to 64 years. The percentages from the highest poverty and lowest education neighborhoods were 8.0% to 8.8% and 3.9% to 4.4%, respectively; 2.7% to 2.9% were from predominantly black neighborhoods and 1.1% to 1.7% were from predominantly Hispanic neighborhoods. Most women were from the South and Midwest United States.

Tables 2 and 3 include estimated initial and final screening rates and disparities as well as APCs of fitted joinpoint models. Figures 1–3 display fitted trends in 3-year Pap testing from joinpoint models as well as disparities in those trends. Supplementary Tables S2 and S3 include adjusted rates over all years of the study.

In 2001 to 2003, women ages 21 to 29 years had an estimated 3-year Pap testing rate of 81.3% (P < 0.001) and this did not change significantly over the decade (Table 2; Fig. 1). Estimated disparities by low-high neighborhood poverty level were 3.1% and 2.0% in 2001 to 2003 and 2008 to 2010, respectively, a nonsignificant decline (Fig. 3A; Table 3). Estimated disparities between women from low and middle poverty neighborhoods were small in 2008 to 2010 (low-middle 2: 0.5% and low-middle 1: −0.1%, Table 3; Fig. 3A). Initial white-black disparities were 4.0% and declined significantly from 2005 to 2007 to 2008 to 2010 to 2.8% at an APC of −0.65% (P = 0.021). White-Hispanic disparities declined from 4.3% to 0.8% over the decade, a −0.50% APC (P = 0.024).

Among women ages 30 to 64 years, estimated 3-year Pap testing rates trended down from 76.8% to 73.1% (−0.94% APC, P < 0.001; Table 2; Fig. 1) until 2005 to 2007, followed by an APC of −0.49% (P = 0.025) until 2008 to 2010. This pattern was similar among women from most categories of poverty and race/ethnicity (Table 2 and Fig. 2C). Women from low poverty neighborhoods had an uninterrupted downward trend from 78.5% to 73.1% over the decade (APC −0.78%, P < 0.001; Table 2; Fig. 2C). Women from Hispanic neighborhoods had stable estimated rates of 76.3% to 78.0% across the decade.

Table 3 and Fig. 3C and D show reductions in neighborhood race/ethnicity and poverty level disparities, which are largely driven by decreasing screening rates among women from low poverty and white neighborhoods.

Screening rates stratified by education levels were very similar to those stratified by poverty level so we chose not to display them.

We examined trends in Pap testing rates and disparities from 2001 to 2010. Among commercially insured women ages 21 to 29 years, 3-year Pap testing rates remained stable at approximately 80% over the decade; disparities were small and improved for Hispanic women to a greater degree than black women. Among women ages 30 to 64 years, 3-year Pap testing rates generally declined over the decade, a pattern that was similar among women from almost all categories of neighborhood-level poverty and race/ethnicity.

Our findings are largely consistent with several other studies examining screening from 2001 to 2005. (6–8) More recently, three studies have examined cervical cancer screening from 2000 to 2010. Two surveys that did not stratify by age group found that Pap testing declined from 2000 to 2010 (10, 12) except among Hispanic women (12), similar to our results among women over the age of 29 years. A Centers for Disease Control and Prevention report found that 3-year screening rates declined among women ages 22 to 30 years, except among Hispanic women who experienced a slight increase (9).

Our study among commercially insured women shows a somewhat more favorable picture of cervical cancer screening compared with these previous surveys of both insured and uninsured women. The groups most at risk, such as younger women from predominantly black and high poverty neighborhoods, had relatively high rates of screening at baseline and did not experience declining rates or increasing disparities. Reductions in 3-year screening rates in the 30- to 64-year age group were not major but the overall 3-year screening rates we detected of 68.8% to 73.1% in 2010 were below the Healthy People 2010 goal of 90% (26). Health plans and provider groups have increasingly focused on improving cervical cancer screening rates over the last decade, so the lack of substantial increases in rates is somewhat surprising. It is possible that this represents of a ceiling effect among the commercially insured or the general lack of effectiveness of physician financial incentives in improving quality (27). It might also reflect research and clinical sentiment earlier in the decade (that preceded guideline changes) that frequent cervical cancer screening might be associated with psychologic harms (28). In addition, the rates we detected among women ages 21 to 29 years are well below self-reported 3-year screening rates of similar age insured women in one survey (approximately 93% in 2000 and 91% in 2010; ref. 9). Our study is one of the few longitudinal analyses of screening trends that uses administrative data rather than self-report. Because the insurer whose data we studied contractually requires that all claims be submitted, we are likely to have precise estimates of screening rates, raising the possibility of recall or response bias in cervical cancer screening surveys.

Our findings have several implications. Given that we detected overall screening rates well below the Healthy People 2020 goal of 93% (19), major efforts will be needed to increase rates over the next 5 years. Furthermore, it is likely that rates among noncommercially insured populations (e.g., women insured through Medicaid or without insurance) are lower. Although we did not detect major disparities by race/ethnicity or neighborhood poverty level, improving screening rates among traditionally disadvantaged populations might be especially challenging, so that targeted efforts by health plans, practice groups, and public health agencies will be vital. Finally, our findings could serve as baseline data for future studies examining the impacts of recent guideline changes (13–15) on cervical cancer screening rates and disparities over the current decade.

Our study has five main limitations. Our results are generalizable primarily to the commercially insured population under 65 years, and do not represent very low income populations such as the uninsured or those covered by Medicaid. We were unable to assess physician-level interventions and reimbursement rates unique to the health insurer we studied, but it is unlikely that these differed substantially from secular patterns or that providers would be highly aware of the specific health insurer of their patients or details of how that health insurer reimburses differentially from other private health insurers. Our measures of socioeconomic status and race/ethnicity are based on geocoded data, but such measures have been validated (21) and disparities were consistent with previous studies using self-reported data (7, 9, 12). Because we could only use a woman's most recent address to create neighborhood-level sociodemographic characteristics, our measures of neighborhood race/ethnicity and poverty would misclassify women who move between neighborhoods with differing socioeconomic status and race/ethnicity characteristics over the time span of the study. Finally, we are unable to assess longer-term trends in screening after key guideline changes that occurred in late 2009 and 2012 using our current dataset.

Our findings imply a continued need to promote cervical cancer screening in general, and especially among higher-risk age groups and demographic groups. Future studies should examine the impact of the major recent changes in screening guidelines to determine whether revised screening interval recommendations are being appropriately met.

No potential conflicts of interest were disclosed.

Conception and design: J.F. Wharam, B.E. Landon

Development of methodology: F. Zhang, B.E. Landon, D. Ross-Degnan

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.F. Wharam

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.F. Wharam, F. Zhang, X. Xu, B.E. Landon, D. Ross-Degnan

Writing, review, and/or revision of the manuscript: J.F. Wharam, F. Zhang, B.E. Landon, D. Ross-Degnan

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.F. Wharam

Study supervision: D. Ross-Degnan

This work was supported by a grant from the American Cancer Society (Principal Investigator: J.F. Wharam; 118261-RSGI-10-075-01-CPHPS).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.