Healthcare Access Domains Mediate Racial Disparities in Ovarian Cancer Treatment Quality in a US Patient Cohort: A Structural Equation Modelling Analysis

Ovarian cancer accounts for more deaths than any other cancer of the female reproductive system and ranks fifth in cancer deaths among women (1). In 2021, there were an estimated 21,410 new cases of ovarian cancer and 13,770 ovarian cancer deaths (1, 2). The 5-year relative survival for ovarian cancer is 49%, a rate partly driven by lack of definitive screening strategies and advanced stage at presentation. Racial disparities in ovarian cancer survival have been well described (3–5); although ovarian cancer incidence is higher among White women, survival is lower among Black women (4, 6). In the past two decades, while ovarian cancer survival improved for White women, it worsened or remained stagnant for Black women (7). The fundamental cause of the racial disparity is undoubtedly complex and multifactorial, involving aspects of tumor biology, patient demographics, and access to quality treatment (8–10). Extensive studies have documented the important role of healthcare access (HCA) in receipt of high quality, guideline adherent treatment (11–13), including measures of area-level factors driving disparities in treatment quality (14, 15). For instance, treatment at academic medical centers, and by high volume providers and gynecologic oncologists, have been associated with improved survival outcomes (16–19), while spatial and socio-demographic measures of HCA were associated with ovarian cancer mortality (14). Yet, Black patients are less likely to receive guideline-recommended treatment, which may include cytoreductive surgery, chemotherapy, and evaluation by a gynecologic oncologist, key indicators of quality care (20). Lack of access to quality care is a driver of ovarian cancer disparities, however, complete understanding of which HCA domains are most relevant in ovarian cancer treatment disparities remains elusive.

The Penchansky and Thomas healthcare access framework described multiple distinct domains of HCA, including affordability, availability, accessibility, accommodation, and acceptability (21). Racial disparities have been well documented in the various access domains, including affordability, availability, and accessibility (10, 22). Within each dimension, there are multiple levels of influence: individual (e.g., dual Medicaid eligibility), provider (e.g., specialty and volume), facility (e.g., readmission rates and quality metrics), and geographic residence (e.g., distance to treatment facility and rural/urban status). Each level of influence may influence treatment quality and utilization either directly or indirectly through the other dimensions, and the racial distribution of each domain likely varies. However, while some studies have examined specific measures within each dimension (9, 23, 24), no study to our knowledge has simultaneously explored multiple access domains to identify the unique drivers of the observed racial disparities among patients with ovarian cancer. In addition, given the multi-domain and multilevel (influences across individual and residential area level) nature of HCA measures, it is important to employ analytical methods that can: (i) accommodate the complexity of measures across multiple geographic levels, (ii) robustly define the underlying latent constructs, and (iii) incorporate pathway analysis to disentangle the impact of race versus HCA in driving ovarian cancer treatment disparities.

To achieve this goal, we used structural equation models to investigate the pathways of association for racial disparities in healthcare access domains estimable in SEER-Medicare (i.e., affordability, availability, and accessibility) and ovarian cancer treatment quality. We hypothesize that the association of Black race with ovarian cancer treatment quality will be partly mediated by HCA dimensions.

Black, White, and Hispanic patients with primary ovarian cancer of any histologic type, (ICD-9-CM and ICD-10-CM diagnosis codes 183.0 or C569) who were 65 years or older and diagnosed between 2008 and 2015 were selected from the SEER-Medicare linked dataset for this study (Supplementary Fig. S1). Prior work in this cohort reported similar rates of guideline treatment receipt and no survival disparity (25) between Non-Hispanic White and Hispanic patients. Therefore, we categorized our analytical cohort as Black (non-Hispanic) versus non-Black (White and Hispanic) to facilitate structural equation modeling of the association between race and treatment. Patients with fallopian tube or peritoneal cancers were not included due to low case numbers. Patients were excluded if they were diagnosed at autopsy or death, if they had a history of primary tumors at two or more sites other than the ovary, and if patients did not have at least 12 months of continuous enrollment in Medicare fee-for-service parts A and B before diagnosis. Patients were also required to have continuous fee-for-service enrollment in the 12 months following their diagnosis date, or until death. These selection criteria yielded a total of 8,987 patients, with 8375 non-Black and 612 Black patients.

Patient and clinical characteristics were obtained from SEER data. These included patient age at diagnosis, stage at diagnosis (stage I-IV), and cancer histology type (epithelial Type I, epithelial Type II, or other histology (26)). Validated coding algorithms were used to assess comorbid conditions and estimate the Charlson Comorbidity Index score in the 12 months prior to ovarian cancer diagnosis using inpatient, outpatient, and carrier Medicare claims files (Supplementary Table S1; refs. 27, 28).

Patients were assigned to a primary cancer treatment physician and hospital based on the plurality of their cancer treatment, inpatient and outpatient claims. A patient's primary cancer treatment provider was identified as the provider listed on the highest number of the patient's outpatient, carrier, home health, and hospice claims listing a cancer diagnosis. Physician specialties were determined from Medicare claims files using Health Care Financing Administration (HCFA) specialty codes. Physician specialties were determined using HCFA specialty codes in the Medicare claims files (29). Ties between physicians were broken by prioritizing physician specialties of interest (gynecologic oncology, medical oncology, hematology/oncology, or surgical oncology) and claim date closest to the ovarian cancer diagnosis date. The patient's primary treating hospital in the year the patient was diagnosed was defined as the facility at which the patient had the majority of inpatient and outpatient claims in that calendar year. In the case of ties, priority was given to facilities with records in the SEER-Medicare Hospital File.

To assess each patient's HCA affordability, accessibility and availability, the SEER-Medicare dataset was linked with publicly available datasets from the US Census Bureau, the Small Area Health Insurance Estimates Program, the Dartmouth Atlas of Healthcare, and the Area Healthcare Resource Files provided by the Health Resources and Services Administration. In total, 40 variables were selected across the three HCA dimensions of affordability, availability, and accessibility; selection was driven by the availability of measures that can be linked to SEER-Medicare data, which is largely comprised of area-level measures as opposed to individual patient measures. Full detail on the assessment and linkage of these measures are available in the Supplementary Methods; included measures are detailed in Supplementary Table S2. Area-level statistics were based on the patient's residence at the time of their diagnosis; when possible, the statistics were specific to the patient's year of diagnosis. If statistics for the year of diagnosis were not available, statistics for the closest year available (within 5 years) were used. Ten variables including census tract-level median household income, percentage of Black residents, percentage of residents aged 25+ with 4 years of college, and patient dual enrollment in Medicaid and Medicare were selected to estimate overall healthcare affordability. All census tract measures were derived from the 2010 American Community Survey. Six variables including patient residence at diagnosis in a metropolitan area, straight line geographic distance in miles from a patient's residence to their main treatment hospital—a proxy for travel time, patient's main treatment hospital's location in an urban or rural area, etc. were abstracted to represent healthcare accessibility. Twenty-four variables chosen to represent healthcare availability include county-level number of gynecologic oncologists per 1K residents in their year of diagnosis, Healthcare Referral Region (HRR)-level 30-day hospital readmission rates in their year of diagnosis, and patient-level main treatment hospital designated as NCI cancer center in their year of diagnosis.

All patients who had an outpatient visit with a gynecologic oncologist for their ovarian cancer diagnosis at least once in the 2 months prior to or in the 6 months following their diagnosis were identified using physician identifiers from inpatient, outpatient, and carrier claims and their HCFA specialty codes; if the patient had a claim with a physician with a specialty code indicating ‘gynecologist/oncologist’ (code = 98), they were considered to have had at least one encounter with a gynecologic oncologist in this time period; HCFA specialty codes capture approximately 83% to 90% of oncologists (29). Patient receipt of any ovarian cancer-related surgery in 12 months post-ovarian cancer diagnosis was defined consistent with previous studies (30) using Medicare inpatient, outpatient, and carrier claims. If a patient had a claim with any of the ovarian cancer–related surgical procedure codes (listed in Supplementary Table S3) in that time period, they were considered to have received a relevant surgery.

Descriptive analyses were performed to assess group differences for study variables of interest using SAS 9.4 (Cary, NC). Mean and standard deviation were calculated for continuous variables, and frequency and percentage were calculated for categorical variables. All other analyses were performed using Mplus 8. Confirmatory factor analysis was used to identify subsets of measures that best represented the three HCA domains; factor analysis methods and results are detailed in the Supplementary Methods and in previous publications. On the basis of these previous results, five variables, including census tract level education level, median household income, and household poverty rate, were selected to estimate the latent variable of overall affordability. Three variables, including patient residence in a metropolitan area, or patient residence in a metropolitan-adjacent area, and patient's main treatment hospital's location in an urban or rural area, were selected to represent the latent variable of healthcare accessibility. Variables including the number of hematologists-oncologists, physicians, and surgeons per 10,000 to 100,000 residents in the patient's county of residence at the year of diagnosis were selected for estimation of healthcare availability. Using the structural equation modeling (SEM) approach, the composite reliability (Ω) and average variance extracted (AVE) for each of the three latent variables were evaluated; Ω of greater than 0.7 and AVE larger than 0.6 represents good internal reliability (31). Our initial model proposed one additional latent variable predicted by patient's age, comorbidity score, stage, and histology type. The reliability test results were not satisfied for this latent variable (Ω = 0.113, AVE = 0.048). Therefore, patients’ clinical characteristics were included in models as four individual variables. SEM was then conducted to test the hypothesized direct and indirect associations of patient race/ethnicity and the latent HCA variables in relation to the two study outcomes: at least one outpatient visit to see a gynecologic oncologist, and receiving ovarian cancer surgery in the 2 months prior to or 6 months post-ovarian cancer diagnosis. The weighted least squares estimation suitable for binary data was used due to the presence of categorical variables (Mplus estimator WLSMV; ref. 32). Mediation analysis for associations between patient race/ethnicity and surgery were conducted and 95% confidence intervals (CI) were calculated using bootstrapped estimates (1,000 samples) of standard errors for direct, indirect, and total effects (33). Bootstrapping with 500 and 2,000 samples was also performed to test the sensitivity of estimates. To evaluate the goodness-of-fit indices for the models, the comparative fit index (CFI; 0.95 or greater), the Tucker-Lewis Index (TLI; 0.95 or greater), the standardized root-mean-square residual (SRMR; 0.08 or less), and the root-mean-square error of approximation (RMSEA; 0.08 or less) were used (34, 35). All statistical tests were two sided with a significance level of 0.05.

The data analyzed in this study were drawn from the SEER-Medicare linked database, which is available from the NCI (http://healthcaredelivery.cancer.gov/seermedicare/). Restrictions apply to the availability of these data, which were used under license for this study.

A total of 8,987 patients with ovarian cancer were included in the analysis, and approximately 7% were Black (Table 1). Black patients were younger at diagnosis, had higher comorbidity scores, and were more likely to be diagnosed at late stages compared with non-Black patients. There were racial differences in the two main study outcomes: Black patients were less likely to visit a gynecologic oncologist in the 2 months prior to or the 6 months after diagnosis (45% vs. 54%, P value < 0.0001) and were less likely to receive any ovarian cancer-related surgery in the 12 months following diagnosis (50% vs. 62%, P value < 0.0001).

The description of study constructs and factor loadings are presented in Table 2. The Affordability (Ω: 0.876; AVE = 0.689), Availability (Ω: 0.848; AVE = 0.636), and Accessibility (Ω: 0.798; AVE = 0.634) latent variables showed high composite reliability. The SEM model demonstrated a good fit to the data based on traditional SEM model fit statistics (CFI = 0.966, TLI = 0.95, SRMR = 0.049, and RMSEA = 0.045). All factor loadings between the latent HCA variables and their component variables were statistically significant, indicating that the chosen component variables for each latent variable were appropriate.

In SEM analysis (Fig. 1), Black race was directly associated with lower affordability (β: −0.36, 95% CI, −0.42 to −0.31) and availability (β: −0.06; 95% CI, −0.09 to −0.04), but higher accessibility (β: 0.08; 95% CI, 0.06–0.10). Higher affordability and availability values were directly associated with higher likelihood of seeing a gynecologic oncologist (affordability β: 0.08; 95% CI, 0.05–0.11; availability β: 0.06; 95% CI, 0.04–0.08), and higher affordability was associated with higher likelihood of receiving any ovarian cancer surgery (β: 0.07; 95% CI, 0.04–0.10). Higher accessibility was directly associated with lower likelihood of consulting with a gynecologic oncologist (β: −0.03; 95% CI, 0.05 to −0.01), and there was no significant association with receipt of ovarian cancer surgery. Seeing a gynecologic oncologist at least once was directly associated with higher likelihood of receiving ovarian cancer surgery (β: 0.33; 95% CI, 0.32 to 0.35). No significant direct associations were observed between Black race and seeing a gynecologic oncologist, or with receipt of ovarian cancer surgery. Other patient characteristics including patient age at diagnosis, comorbidity burden, cancer stage, and tumor histology were each directly associated with likelihood of visiting a gynecologic oncologist and receiving ovarian cancer surgery.

Mediation analysis of the direct and indirect effects of race on receipt of ovarian cancer surgery are presented in Table 3. The HCA latent variables (affordability, accessibility, and availability) were assessed as mediators, as was visiting a gynecologic oncologist at least once. Although there was no direct effect observed between Black race to receipt of surgery (β: −0.04; 95% CI, −0.26 to 0.15), there was a statistically significant inverse total effect (β: −0.243; 95% CI, −0.079 to −0.011). The total effect was driven by the total indirect effect from: (i) race to surgery through the mediating latent affordability variable (β: −0.03; 95% CI, −0.04 to −0.01), (ii) race to surgery through affordability and gynecologic oncologist visit (β: −0.01; 95% CI, −0.01 to −0.01), and 3) race to surgery through availability and gynecologic oncologist visit (β: −0.001; 95% CI, −0.001 to −0.001; Supplementary Fig. S2). There was statistically significant total (β: 0.10; 95% CI, 0.06–0.12), direct (β: 0.07; 95% CI, 0.04–0.10) and indirect effect of affordability on ovarian cancer surgery through gynecologic oncology visit (β: 0.03; 95% CI, 0.02–0.035), but only an indirect effect of accessibility on ovarian cancer surgery through gynecologic oncology visit (β: −0.01; 95% CI, −0.02 to −0.00). There was a direct (β: −0.03; 95% CI, −0.05 to −0.00) and total indirect effect of availability on ovarian cancer surgery through gynecologic oncology visit (β: 0.02; 95% CI, 0.01–0.02), but no significant total effect. In summary, patients of Black race had lower affordability and availability on average, both of which were associated with lower likelihood with seeing a gynecologic oncologist, which in turn was associated with lower likelihood of receiving surgery.

In novel analyses of multi-level and multi-domain measures of HCA among patients with ovarian cancer in SEER-Medicare, we observed that racial differences exist in quality treatment metrics, specifically gynecologic oncology consultation and receipt of ovarian cancer surgery. Black patients had lower measures of HCA affordability and availability, but better measures of accessibility compared with non-Black patients. Consistent with our hypothesis, the observed racial disparities in ovarian cancer treatment were primarily driven by these HCA domain differences; there was no significant direct effect of race on either of the outcomes evaluated, however, there were statistically significant indirect pathways of association through HCA domains, impacting gynecologic oncology visit and surgery. These findings highlight that the mechanisms underpinning racial disparities in quality ovarian cancer treatment can be attributed to adverse social determinants of health measures, specifically HCA, and provides important and groundbreaking clarity regarding specific drivers of ovarian cancer disparities that should be targeted through multi-level interventions to improve treatment quality for patients with ovarian cancer.

Prior studies have documented the role of specialists in receipt of guideline-adherent treatment (36–38). For instance, consultation with a gynecologic oncologist has been associated with significantly increased likelihood of receiving timely surgery after ovarian cancer diagnosis (36, 37). Given that nonadherence to guideline recommended treatment is associated with reduced survival, lack of access to a gynecologic oncologist likely contributes to observed racial disparities in survival. Further, consistent with prior studies, Black patients had greater healthcare accessibility, likely driven by larger distribution of Blacks in metropolitan areas. However, there was an inverse association between accessibility and gynecologic oncology consultation, suggesting that for Black patients, greater accessibility does not imply utilization. This utilization gap, i.e., higher accessibility but lower utilization, has been previously described and highlights the need for studies to examine other contributors to quality care for Black patients (10, 39, 40). HCA domains such as accommodation (convenience) and acceptability (quality of patient-provider interaction) are likely significant contributors. For instance, a recent study observed that 45% of Black adults reported at least one negative experience with a healthcare professional, including not being treated with respect, or being talked down to (41). Strategies to eliminate the utilization gap will require building and maintaining trust among Black and other race/ethnic minority communities, addressing the fundamental role of structural racism by promoting cultural competency and minimizing implicit bias in the clinical setting, and expanding engagement with diverse patients and communities to begin to redress the fraught historical and contemporary relationships between hospitals and the communities they serve. The accommodation and acceptability HCA domains are beyond the scope of the current analysis and require patient reported data on healthcare interactions that are not available in secondary databases. A comprehensive evaluation of all five HCA dimensions and their pathway of associations with cancer outcomes will provide much needed empirical data regarding systemic interventions to ensure equity in quality cancer treatment and eliminate disparities in survival outcomes.

There are several strengths and limitations relevant to this study. First, the analysis relied on large, population-based data of patients with ovarian cancer and outcomes ascertained using validated coding algorithms. Second, relative to other studies, the analysis included a large number of Black patients with ovarian cancer, enhancing the rigor of the SEM analysis. Third, measures of HCA dimensions relied on secondary datasets captured at multiple levels for each study participant, providing a reliable and consistent measure of the underlying construct. There are also several limitations inherent in the study design. First, the use of secondary data limited our ability to draw inferences regarding individual or personal variables that may influence patients’ ability to seek or use healthcare. For instance, we measured accessibility using a variety of measures, including geographic distance between patient residence and primary facility. It is possible that geographic distance underestimates accessibility, because it does not incorporate travel time and travel mode, important measures of accessibility only estimable from self-reported data. Secondly, several of the HCA dimensions, such as socioeconomic status, were measured at the geographic level for each patient, and did not account for individual measures. However, extensive studies have documented that the social and healthcare environment plays a critically important role in health outcomes, above and beyond individual circumstances, and is an area that deserves significant emphasis (42). Finally, our findings may not be generalizable to younger patients with ovarian cancer who are not eligible for Medicare or patients in managed care programs, because the study population is limited to patients in fee for service plans. When requiring fee-for-service enrollment, the percentage of NH-Black patients in the cohort decreased from 9% to 8%; the difference is not substantial but highlights the fact that this cohort may not be fully generalizable to patients in managed care plans. Future studies will be needed to characterize HCA dimensions among younger patients who may face greater barriers to healthcare due to lack of insurance or less generous insurance plans, leading to greater out of pocket costs, a measure of affordability.

In conclusion, racial differences in ovarian cancer treatment quality— gynecologic oncology consultation and surgery—are partly driven by HCA affordability, availability, and accessibility. Two additional domains, accommodation and acceptability, may provide powerful insights into patient-level factors that also contribute to racial disparities in treatment quality. Nevertheless, actionable strategies to promote equitable access in the three domains examined are feasible and can have substantial impact: for instance, more generous insurance policies that limit out of pocket expenses for cancer patients (addressing the financial burden of cancer care, affordability); hospitals that provide convenient and free parking, are located along public transportation routes, or explore novel strategies such as partnerships with rideshare programs to increase accessibility; and a focus on health equity in the geographic location of new clinics and hospitals to address gaps in availability of cancer care. Strategies like these will be key to mitigating disparities in ovarian cancer treatment and ensuring equity in quality cancer treatment.

L.E. Wilson reports grants from NCI during the conduct of the study; grants from AstraZeneca outside the submitted work. R.A. Previs reports other support from Myriad Genetics, Natera; and other support from Labcorp outside the submitted work. M. Pisu reports grants from NIH during the conduct of the study. M.J. Schymura reports grants from NCI/Duke University School of Medicine during the conduct of the study. B. Huang reports grants from NCI during the conduct of the study. No disclosures were reported by the other authors.

T. Akinyemiju: Conceptualization, resources, supervision, funding acquisition, methodology, writing–original draft, writing–review and editing. Q. Chen: Formal analysis, writing–original draft, writing–review and editing. L.E. Wilson: Formal analysis, writing–original draft, writing–review and editing. R.A. Previs: Writing–review and editing. A. Joshi: Writing–review and editing. M. Liang: Writing–review and editing. M. Pisu: Writing–review and editing. K.C. Ward: Writing–review and editing. A. Berchuck: Writing–review and editing. M.J. Schymura: Writing–review and editing. B. Huang: Writing–review and editing.

The authors acknowledge the helpful assistance provided by the SEER-Medicare reviewers, Information Management System coordinator Elaine Yanisko, and all the patients whose valuable data contributed to this study.

This research was funded by the NIH/NCI (T. Akinyemiju, R37CA233777). R.A. Previs is supported by a grant from the NIH K12 HD103083. B.Huang is supported by a NCI Cancer Center Support Grant (P30 CA177558).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.