Disparities in cervical cancer elimination time frames in the United States: a comparative modeling study

Abstract

Population-level estimates in time frames for reaching cervical cancer elimination (ie, <4 cases per 100 000 women) in the United States may mask potential disparities in achieving elimination among subpopulations. We used 3 independent Cancer Intervention and Surveillance Modeling Network models to estimate differences in the time to cervical cancer elimination across 7 strata of correlated screening and human papillomavirus vaccination uptake, based on national survey data. Compared with the average population, elimination was achieved at least 22 years earlier for the high-uptake strata and at least 27 years later for the most extreme low-uptake strata. Accounting for correlated uptake impacted the population average time frame by no more than 1 year. Consequently, national average elimination time frames mask substantial disparities in reaching elimination among subpopulations. Addressing inequalities in cervical cancer control could shorten elimination time frames and would ensure more equitable elimination across populations. Furthermore, country-level elimination monitoring could be supplemented by monitoring progress in subpopulations.

The combined effects of vaccination against human papillomavirus (HPV) infection and cervical cancer screening have set the United States on a path to achieve the World Health Organization’s (WHO’s) goal of eliminating cervical cancer as a public health problem (ie, incidence of <4 per 100 000 women)¹ in the next 15-25 years.² Population-level estimates of cervical cancer elimination timing can facilitate planning and mobilize stakeholder and community engagement but do not specifically address potential disparities in reaching elimination among subpopulations. In the United States, neither screening nor vaccination uptake are homogeneously distributed across the population,³ raising the question as to whether achieving a population average elimination metric may mask disparities in reaching elimination across subpopulations. Furthermore, emerging evidence from screening behaviors among vaccinated cohorts suggests that HPV-vaccinated women in the United States are more likely to receive timely screening than unvaccinated women and vice versa,^4,5 potentially exacerbating disparities in reaching elimination. Prior US analyses (eg, Spencer et al.⁶) often focus on sociodemographic groups with high burden of cervical cancer such as minoritized race, rurality, or area poverty. Although these analyses can help narrow focus on subpopulations with varying levels of preventive behavior, vaccination and screening adherence are generally considered separate inputs. Consequently, such analyses are unable to capture potential correlation of HPV vaccination coverage and screening adherence within measures. Through a comparative model-based analysis within the Cancer Intervention and Surveillance Monitoring Network, we provide the first insights into the impact of correlations in HPV vaccination and cervical screening behaviors on the time frames for reaching cervical cancer elimination in the United States. These findings are important, as they (1) show to what extent a national elimination target may mask elimination time frame disparities across subpopulations, (2) shed light on how the average national vaccination time frame may be affected by correlated screening and vaccination uptake, and (3) provide a first step toward more targeted public health interventions that may expedite elimination and reduce disparities in cervical cancer control in the United States. Importantly, our analyses are primarily designed to conceptualize how correlated heterogeneities affect elimination time frames and are not meant to provide exact elimination time frame predictions or suggestions for policy.

We used 3 independently developed hybrid individual-based Cancer Intervention and Surveillance Monitoring Network models of HPV transmission and cervical carcinogenesis (Harvard, Policy1-Cervix, and Sexually Transmitted Diseases Simulator - Microsimulation Screening Analysis [STDSIM-MISCAN]). The models differ with respect to the type and number of health states, HPV genotype categorizations, histological cancer types, model cycle length, and data sources used to parameterize the model prior to fitting to the US setting and have been described in previous comparative modeling analyses for the United States.^2,7 Common US model inputs include hysterectomy rates and all-cause mortality.⁷ The models were calibrated to HPV and cervical disease outcomes in the United States, achieving a good fit to common empirical targets.^2,7

We stratified the US female population across 7 strata of cervical cancer screening behaviors, ranging from screening annually to never screening based on data from Kaiser Permanente Northern California.⁸ For each screening stratum, we assigned either a national-average HPV vaccination coverage (baseline scenario) or a stratum-specific HPV vaccination coverage (correlated scenario) (Figure 1). Average vaccination coverage was based on historic vaccination uptake by age 26 years as applied in a prior analysis.² For the correlated scenario, we estimated the stratum-specific vaccination coverage by drawing values from an inverse cumulative density function of the beta distribution with a mean equal to the average vaccination coverage and a calibrated variance so that the differences in vaccination uptake reflect the observations from a US national health interview survey study showing higher vaccination coverage among those who are up-to-date with screening and vice versa⁴ (see Supplementary Methods; Tables S1 and S2; Figure S1 for more details). In the models, women aged 21-65 years were screened using cytology, and screen-positive women were managed according to guidelines.⁹ To minimize unnecessary complexity, we did not simulate switches to HPV primary screening or co-testing. We assumed compliance to recommended colposcopy with biopsy (54%) or treatment for precancer grade 2 (47%) or grade 3 (63%) independent from the screening strata.¹⁰

Figure 1.

Schematic depicting 7 strata of screening behaviors ranging from screening annually to never screening for (A) baseline assumptions without correlated screening and human papillomavirus vaccination uptake and (B) correlated assumptions for screening and human papillomavirus vaccination uptake. Correlated vaccination uptake was based on an inverse beta distribution model (see Supplementary Methods) resulting in the same population-level average (ie, 75.1%). Population weights reflect the percentage of the population used in the population-weighted (national) elimination time frame. See Supplementary Material for additional details. Pop = population.

For each model, we estimated the time to cervical cancer elimination using age-standardized (2000 US female population aged 0-99 years¹¹) cancer incidence under the baseline scenario reflecting a weighted average of the 7 screening strata assuming national average HPV vaccination coverage (Figure 1, A). We quantified the difference in the projected elimination year compared with the national baseline (ie, earlier, delayed, or never achieved) for each stratum individually and for the weighted average correlated national scenario assuming strata-specific vaccination coverage (Figure 1, B). As the level of herd immunity for unvaccinated women within each stratum is unknown, in our base case we applied the average age-specific herd immunity effects projected for unvaccinated women estimated from our prior US elimination analysis.² In a sensitivity analysis, we considered an extreme scenario of no herd immunity benefits to unvaccinated women. We assumed vaccination of females to occur on average at age 17 years, initially with the quadrivalent vaccine (for birth cohorts born 1980-1998) and switching to the nonavalent vaccine (for birth cohorts born in 1999 and later). We assumed both vaccines provided 100% protection against acquisition of vaccine-targeted HPV-16 and -18, and that the nonavalent vaccine additionally provided 95% protection against HPV-31, -33, -45, -52, and -58.

The 3 models found that correlated disparities in vaccination and screening uptake did not substantially affect the predicted national average elimination time frame, with the Policy1-Cervix predicting a delay of 1 year, the STDSIM-MISCAN model predicting no change, and the Harvard model predicting 1 year earlier (weighted average in Figure 2). However, the models projected substantial differences in elimination time frames across the 7 strata of correlated screening and vaccination uptake. The elimination threshold was delayed 27-35 years in 2 of the models and never achieved in 1 of the models, for the most extreme low-uptake stratum, whereas the time frame ranged by more than 20 years earlier up to 8 years later for the other over- and underscreening strata, respectively (Figure 2). Projected disparities were only modestly reduced when assuming average instead of correlated vaccination uptake across the strata (Figure S2).

Figure 2.

Differences in cervical cancer elimination (incidence <4 per 100 000 women) time frame in the United States for strata with correlated vaccination and screening uptake compared with the national (weighted average of the strata) without correlated vaccination and screening uptake. The graph shows absolute differences in elimination year for each population stratum compared with the projected elimination year for the national weighted average of each stratum assuming equal vaccination coverage across strata (observed national average of 75.1%, ie, without correlated behavior). STDSIM-MISCAN = Sexually Transmitted Diseases Simulator - Microsimulation Screening Analysis. ^aAlready eliminated is defined by having achieved an incidence of less than 4 per 100 000 without vaccination (ie, screening only).

These findings are subject to several simplifications. First, we assumed no assortative mixing in sexual behavior within the strata by applying the same herd immunity effects for each stratum. In sensitivity analysis (Figure S3), we show that under the extreme assumption of no herd immunity, population disparities in reaching elimination are more profound, and 1 model (Harvard) predicting elimination would not be reached for the weighted average population. Although the assumption of no herd immunity is unrealistic, as herd effects would still be expected within subpopulations with low levels of vaccine uptake,¹² these findings demonstrate that our main analyses are sensitive to herd immunity assumptions and that our predictions are conservative with regards to the elimination time frame disparities to the extent that assortative mixing exists within the different strata. Furthermore, the population average herd immunity is based on a relatively high vaccination coverage level (75.1%), which could at least partially explain why disparities predicted by the 3 models were only modestly reduced when there was no correlation of behaviors (Figure S2).

A second simplification is that we only assumed correlated disparities in vaccination and screening behaviors. However, research suggests that disparities may accumulate for disadvantaged populations across the disease and intervention continuum.³ Accounting for additional correlations within the strata in our analysis and HPV acquisition risks, disease progression rates (eg, because of cofactors like tobacco use), hysterectomy rates, background mortality, quality of care, screening follow-up, or treatment uptake, the projected disparities could be further exacerbated. Therefore, our simplified and conservative projections of US cervical cancer elimination disparities call for a more thorough investigation into the impact of clustered disparities across the disease and intervention continuum, using empirical data and modeling. It is important to note that we evaluated differences in elimination time frames in the context of cytology-based screening, although both primary HPV and co-testing (cytology and HPV testing) are recommended and commonly used in the United States.¹³ Given that HPV testing can remain similarly protective at less frequent intervals, it is possible that the gap in elimination time frames between screeners and nonscreeners would be wider with greater HPV testing usage.

The substantial disparities identified in our analysis uncover the limitations of using only a single national elimination target as a policy goal. While the WHO’s national elimination threshold is an important benchmark for countries, the threshold should be supplemented with monitoring of subpopulations to ensure progress is equitable, consistent with WHO recommendations. This can improve ongoing country-level assessments and provide guidance in selecting targets for public health interventions. In addition, models used for country-level elimination time frame projections or general decision making in cervical cancer control should not only reflect national averages but also capture local contextual realities in terms of heterogeneities and clustering of risks. Furthermore, a sole focus on elimination time frames as a policy target may undermine the harms to benefits ratios for screening.¹⁴ Clearly, those who screen annually have the shortest time frames to reaching elimination; however, screening more frequently than the recommended 3-yearly schedule for cytology is not recommended, as the harms of screening start to outweigh the benefits.¹³ This is especially the case if vaccination and screening behaviors are correlated, as those who screen more frequently than currently recommended are also those who are most likely protected by vaccination, severely reducing the efficiency of screening.¹⁵

To equitably achieve cervical cancer elimination, interventions are needed that focus on those within the population who need it the most. Increasing screening participation for demographic groups or in geographic areas with disproportionately low vaccination coverage could reduce inequities while expediting cervical cancer elimination on a national level. The strata in our analysis were designed to conceptually demonstrate the importance of correlated behaviors and cannot directly be translated into specific population subgroups to be targeted for interventions. Future analysis would seek to better characterize the sociodemographic factors (and ultimately the root causes of disadvantage) most strongly related to these correlations in vaccination and screening behavior, and those groups (or underlying structural causes) would likely be the target of interventions. However, as a first step, we demonstrate the impact of this correlation to motivate future work to consider this more directly.

In conclusion, we demonstrate that there are likely substantial disparities in reaching cervical cancer elimination in the United States. Achieving equitable cervical cancer control and elimination across the population necessitates a comprehensive assessment of the clustering of disparities across the disease and intervention continuum so they can be quantified, understood, and reduced through targeted interventions, policies, and monitoring.

Author contributions

Emily Annika Burger, PhD (Conceptualization; Formal analysis; Investigation; Methodology; Writing—original draft), Erik E.L. Jansen, PhD (Conceptualization; Formal analysis; Investigation; Methodology; Validation; Visualization; Writing—original draft), Daniel de Bondt, MS (Data curation; Formal analysis; Methodology; Writing—review & editing), James Killen, BE (Data curation; Formal analysis; Investigation; Writing—review & editing), Jennifer Spencer, PhD (Conceptualization; Investigation; Methodology; Writing—review & editing), Mary Caroline Regan, MS (Formal analysis; Project administration; Visualization; Writing—review & editing), Megan A. Smith, PhD (Data curation; Formal analysis; Writing—review & editing), Stephen Sy, MS (Data curation; Formal analysis; Software; Writing—review & editing), Karen Canfell, DPhil (Conceptualization; Supervision; Writing—review & editing), Inge M.C.M. de Kok, PhD (Conceptualization; Methodology; Resources; Supervision; Writing—review & editing), Jane J. Kim, PhD (Conceptualization; Funding acquisition; Methodology; Supervision; Writing—review & editing), and Jan A.C. Hontelez, PhD (Conceptualization; Formal analysis; Methodology; Supervision; Visualization; Writing—original draft; Writing—review & editing).

Supplementary material

Supplementary material is available at JNCI: Journal of the National Cancer Institute online.

Funding

This study was supported by funding from the National Cancer Institute (U01CA199334). Emily A Burger receives salary support from the Norwegian Cancer Society (#198073).

Conflicts of interest

K.C. is the co-PI of an investigator-initiated trial of cervical cancer screening, Compass, run by the VCS Foundation, which is a government-funded not-for-profit charity. Neither K.C. nor her institution have received funding from industry for this or any other research project. All other authors declare no conflicts.

Data availability

Model outputs by time and age underpinning the analysis, and associated post-processing R script, are available for download at https://github.com/NCI-CISNET/BurgerJansenJNCI2024.