Disentangling Racial, Ethnic, and Socioeconomic Disparities in Treatment for Colorectal Cancer

Colorectal cancer is the third most commonly diagnosed cancer in the United States (1). Only 39% of patients with colorectal cancer are diagnosed at a localized stage, for which the 5-year survival rate is about 90% (2). Incidence and mortality from colorectal cancer have declined in recent decades thanks to increased screening and scientific advances in treatment. However, improvements have not been equal for all groups (3, 4). Minority and socioeconomically disadvantaged patients with cancer are at an increased risk for later stage at diagnosis due to screening nonattendance, longer diagnostic intervals, treatment delay, and nonadherence to proper treatment, all of which can lead to poorer cancer outcomes and reduced survival (5–8). Compared with non-Hispanic white patients, age- and stage-adjusted, cancer-specific mortality among patients with colorectal cancer has been shown to be 24% higher among black men and 19% higher among black women; Hispanic men and women were not shown to be at significantly higher risk (9). Colorectal cancer–specific and overall mortality is also disproportionately higher among people with a lower socioeconomic status (10).

The National Comprehensive Cancer Network (NCCN) Clinical Practice Guidelines are the most widely recognized and used guidelines in oncology clinical policy around the world. These evidence-based, consensus-driven guidelines were developed to ensure clinicians can provide their patients with preventative, diagnostic, and supportive services that lead to the best outcomes (11). Prompt treatment is believed to improve survival in patients with colorectal cancer, and treatment should be initiated as soon as possible after diagnosis (12). Previous research has considered failure to initiate treatment within 60 or 90 days to be treatment delay (13).

Mechanisms that contribute to disparities in receipt of timely and appropriate care for cancer arise from factors at the patient-, provider- and health system-level. Minority and immigrant patients face cultural and language barriers, lower health literacy, as well provider implicit bias (14–18). Socioeconomic disparities result from both neighborhood-level contextual factors such as access to quality health care services and lack of health-promoting resources, and individual-level factors such as lack of employer-provided health insurance or having a usual health care provider, as well as availability of time to attend appointments.

However, poorer quality care has also been observed by race and ethnicity among individuals with adequate health insurance (14–18). Socioeconomic barriers could interact with (moderate) or explain (mediate) some, or all, of the racial/ethnic disparities in receipt of high quality care. Although several studies have described racial and ethnic disparities in treatment for colorectal cancer, no previous study has disentangled the relative contribution of SES in these disparities. The objective of this study was to investigate the interrelationship between patient race/ethnicity and neighborhood SES on outcomes reflecting two dimensions of care quality for colorectal cancer: receipt of appropriate treatment and timely initiation of treatment among those who are properly treated.

The California Cancer Registry (CCR) is California's statewide population-based cancer surveillance program administered by the California Department of Public Health's Chronic Disease Surveillance and Research Branch (19). By law, all occurrences of cancer among Californians are required to be reported to the CCR, ensuring the population is representative of all of California (19). CCR variables include patient demographics, geocoded residence at the time of diagnosis, and tumor characteristics including stage, and receipt of surgery, chemotherapy, and radiation. Residential urbanicity based on the 2010 U.S. census was assigned to each patient at the census block group level.

This study was reviewed and approved by Institutional Review Boards (IRB) at San Diego State University, the University of California San Diego, and the California Department of Public Health Committee for the Protection of Human Subjects (the statewide IRB that grants permissions to use data from the CCR).

Patients diagnosed with first primary stage I, II, or III cancers of the colon or rectum (excluding cancers of the rectosigmoid junction), as defined by the American Joint Committee of Cancer 7th edition, diagnosed between 2009 and 2013, and alive at the time of diagnosis were included in the initial study population (N = 44,697). Patients diagnosed with stage II colon cancer (n = 12,036) were excluded because treatment guidelines for stage II colon cancer differ by high versus low risk, a variable not available in the tumor registry. All Asian American, Native Hawaiian, and Pacific Islander groups as well as American Indians (n = 4,457) were excluded due to insufficient samples sizes to allow for stratification by nSES. Patients belonging to the following groups were also excluded because of insufficient sample sizes: 2+ races (n = 734); transsexual or transgender (n = 11). Our final study population included 26,870 patients.

The primary outcome was timely, guideline concordant treatment (GCT) according to the 2016 NCCN guidelines (20). Deviation from GCT was defined three ways: (i) Undertreatment: receiving less than the minimum site- and stage-specific recommended treatment by NCCN; (ii) >60-day treatment delay: first definitive therapy (surgery, chemotherapy, or radiation) initiated more than 60 days after the date of diagnosis; and (iii) >90-day treatment delay: first definitive therapy initiated more than 90 days after the date of diagnosis (Fig. 1).

The independent variables of interest were patient race/ethnicity and nSES. Race in the CCR is abstracted from the medical records. Ethnicity in the CCR is derived using the North American Association of Central Cancer Registries (NAACCR) Hispanic-Latino Identification Algorithm which uses a combination of last name, maiden name, Spanish/Hispanic Origin, Birthplace Country, primary race, sex, and Indian Health Service (HIS) linkage (21). Race and ethnicity were combined as non-Hispanic white (NHW), non-Hispanic black (NHB), and Hispanic (including those who identify as white or black race). Although individual-level SES is unavailable in the CCR, nSES, a contextual-level measure of deprivation, is classified by the CCR using a multi-component composite score based on demographic, social, economic, and housing data collected through the American Community Survey (22) at the block-group level.

Covariates for confounding control included cancer site (colon or rectum), stage at diagnosis (local, regional, or remote), year of diagnosis, biological sex, age at diagnosis, insurance type at diagnosis (not insured, private insurance, Medicaid, Medicare, military, or other/not otherwise specified), marital status at diagnosis (single/never married, married or domestic partner, separated or divorced, or widowed), whether or not the reporting facility with the earliest date of admission had an ACOS-approved cancer program, and residential urbanicity (metro, urban, rural). All covariates were selected a priori based on variables shown to be associated with receipt of GCT in published literature (23–26).

To classify treatment delays, full dates for diagnosis, surgery, radiation, and chemotherapy were required. Of our 26,870 patients, 2,976 (11.08%) had an incomplete date (i.e., month and year only) of diagnosis, surgery, radiation, and/or chemotherapy. The middle of the month was imputed for these incomplete dates. Of our 26,870 patients, 1,794 (6.68%) had either a missing date (completely missing or only year) or missing covariate information (Supplementary Table S1). To resolve this, we used multiple imputation (PROC MI) using SAS version 9.4 (SAS Institute, Cary, NC) with 7 imputed datasets, a valid statistical procedure for recovering missing data to create complete datasets that can then be analyzed through standard procedures (27).

All analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC).

Descriptive statistics were used to describe our population overall, and the prevalence of undertreatment and treatment delay by patient race/ethnicity and nSES. An intercept-only (empty) outcome model was used to evaluate block group as a potential cluster variable.

To explore the influence of our two primary exposure variables: race/ethnicity and nSES on each of our three outcomes: undertreatment, >60-day treatment delay, and >90-day treatment delay, we ran a series of adjusted, moderated, and mediated models. These models are distinct in how they handle our primary exposure variables, but similar in choice of model fit and covariates for control.

In our adjusted models, model 1 was minimally adjusted for age, sex, cancer site, stage, and year of diagnosis. Model 2 was further adjusted for marital status, cancer program, residence, payer, and mutually adjusted for race/ethnicity (in nSES models), and nSES (in race/ethnicity models). In our moderation models, we used logistic regression models to estimate the joint effect of race/ethnicity and nSES (treated as a composite variable), using NHWs living in the highest nSES as the reference group, with control for all of the model 2 adjustment variables. This joint effect model is similar to adjusted model 2, but the use of a composite exposure variable allowed for an interaction between race/ethnicity and nSES.

To investigate the mediating effect of nSES on racial/ethnic disparities, we used a mediation analysis technique developed by VanderWeele and colleagues (28, 29) to estimate the effect of non-white race/ethnicity on undertreatment, >60-day treatment delay, and >90-day treatment delay, as mediated by lower (Q1–Q2) versus higher (Q3–Q5) nSES. The mediation analysis allows us to quantify the direct and “mediated” effects. The mediation models estimate the natural direct effect (NDE), natural indirect effect (NIE), and the total effect (TE). The TE is the effect of non-white race/ethnicity on undertreatment and treatment delay. Some or all the TE could be explained by our mediator, nSES, and the TE can be split into two components: the NDE and the NIE. The NDE expresses the effect of non-white race/ethnicity on undertreatment and treatment delay, after removing (adjusting for) the effect of nSES. The NIE expresses how much of the effect of non-white race/ethnicity on undertreatment and treatment delay is mediated through nSES. The percentage mediated is the total effect of non-white race/ethnicity on undertreatment and treatment delay that is explained by nSES. The parameter estimates and confidence intervals were obtained using PROC CAUSALMED and PROC MIANALYZE to combine imputed datasets. We used a logistic model, adjusted for site, stage, year of diagnosis, sex, age, marital status, cancer program, residence, and payer, with standard errors calculated via the bootstrap method, resulting in effect interpretation on the odds ratio scale (28, 29).

Statistical significance was assessed at P < 0.05.

The study population (N = 26,870) was 70.3% NHW, 7.7% NHB, and 22.0% Hispanic. Rectal cancer accounted for 31.7% of tumors, with lower prevalence among NHBs (25.1%) and higher prevalence among Hispanics (35.7%). Hispanic patients were younger, on average than NHWs and NHBs. Hispanic and NHB patients had lower neighborhood socioeconomic status, with 53.3% of NHBs and 57.0% of Hispanics classified as lowest or lower–middle nSES, compared with only 26.7% of NHWs. Hispanic patients were more likely to be uninsured (3.4%) or be covered by Medicaid (12.3%) compared with NHWs (1.1% uninsured; 3.7% Medicaid) and NHBs (1.8% uninsured; 9.1% Medicaid). NHB patients were less likely to be treated at ACOS-approved cancer centers (39.3%) compared to Hispanics (49.0%) and NHWs (52.6%). NHB patients were less often married (39.9%) compared with NHWs (53.6%) and Hispanics (55.1%). Most patients lived in metropolitan areas, with only 4.0% of NHWs and <1% of NHBs or Hispanics residing in urban or rural geographies (Table 1).

Overall, 25.9% of patients were classified as undertreated, 26.02% NHWs, 26.94% of NHBs, and 25.27% of Hispanics. More than 60-day treatment delay was experienced by 12.9% of all patients, 11.7% of NHWs, 17.0% of NHBs, and 15.5% of Hispanics. More than 90-day treatment delay was experienced by 7.1% of all patients, 6.5% of NHWs, 9.2% of NHBs, and 8.1% of Hispanics. Undertreatment was experienced most often by patients in the lowest quintile nSES, and least by patients in the highest quintile (30.8%–22.5%, respectively). More than 60-day treatment delay ranged from 17.2% in the lowest quintile nSES to 9.3% in the highest and >90-day treatment delay ranged from 10.4% in the lowest quintile nSES to 4.6% in the highest (Table 2).

In empty models, census block group accounted for <5% of the outcome variance, and therefore was not included as a cluster variable. In minimally adjusted models NHB [OR = 1.39; 95% confidence interval (CI) = 1.23–1.57) and Hispanic (OR = 1.17; 95% CI = 1.08–1.27) patients had increased odds of undertreatment compared with NHWs. In model 2, after adjustment for nSES and other suspected intermediate variables, the odds ratios were attenuated for NHBs (OR = 1.16; 95% CI = 1.02–1.31) and nullified for Hispanics (OR = 1.01; 95% CI = 0.93–1.10). Treatment delay outcomes were more strongly associated with race/ethnicity; with both NHBs (>60-day OR = 1.78; 95% CI = 1.57–2.02; >90-day OR = 1.73; 95% CI = 1.46–2.04) and Hispanics (>60-day OR = 1.50; 95% CI = 1.38–1.64; >90-day OR = 1.47; 95% CI = 1.31–1.64) at increased odds in minimally adjusted models. These positive associations were attenuated, but persisted, even after full adjustment in model 2. In minimally adjusted models, compared with highest quintile nSES, lowest (OR = 1.84; 95% CI = 1.77–1.91), lower–middle (OR = 1.44; 95% CI = 1.39–1.50), middle (OR = 1.34; 95% CI = 1.29–1.39), and upper–middle (OR = 1.15; 95% CI = 1.10–1.19) quintile nSES patients had increased odds of undertreatment, linearly increasing with decreasing nSES. These positive associations were attenuated, but persisted, after full adjustment in model 2. The treatment delay outcome was also more strongly associated with nSES with evidence of a linear trend: with lowest quintile nSES (>60-day OR = 2.10; 95% CI = 1.86–2.37; >90-day OR = 2.59; 95% CI = 2.21–3.04), lower–middle (>60-day OR = 1.71; 95% CI = 1.52–1.93; >90-day OR = 1.89; 95% CI = 1.60–2.22), middle (>60-day OR = 1.42; 95% CI = 1.26–1.61; >90-day OR = 1.64; 95% CI = 1.40–1.93), and upper–middle (>60-day OR = 1.34; 95% CI = 1.19–1.51; >90-day OR = 1.38; 95% CI = 1.17–1.62) quintile nSES at increased odds in minimally adjusted models. Like the race effect models, these positive associations were attenuated, but persisted, after full adjustment in model 2 (Table 3).

In our joint effects analysis (Table 4), compared with NHWs in the highest quintile nSES, NHWs, NHBs, and Hispanics in the lowest (NHW OR = 1.55; 95% CI = 1.35–1.78; NHB OR = 1.78; 95% CI = 1.43–2.20; Hispanic OR = 1.53; 95% CI = 1.33–1.77), lower–middle (NHW OR = 1.29; 95% CI = 1.14–1.45; NHB OR = 1.33; 95% CI = 1.04–1.69; Hispanic OR = 1.32; 95% CI = 1.13–1.54), and middle (NHW OR = 1.23; 95% CI = 1.10–1.37; NHB OR = 1.61; 95% CI = 1.23–2.10; Hispanic OR = 1.26; 95% CI = 1.06–1.50) quintile nSES, as well as NHBs in the upper–middle (OR = 1.44; 95% CI = 1.07–1.95) quintile nSES, had increased odds of undertreatment. These racial/ethnic disparities in undertreatment are attenuated for NHWs and Hispanics in the upper–middle quintile nSES and nullified for NHBs and Hispanics in the highest quintile nSES. Treatment delay effects were stronger than those for undertreatment, especially for >90-day delay. Compared with NHWs in the highest quintile nSES, NHWs, NHBs, and Hispanics in the lowest, lower–middle, middle, and upper–middle quintile nSES had marked increased odds of 90- and 60-day treatment delays. However, these racial/ethnic disparities in treatment delay were nullified for NHBs in the highest quintile nSES. We observed a linear trend of increasing risk of undertreatment and delay with increasing nSES in NHWs and Hispanics, but not in NHBs.

For the mediation analysis, we estimated the total effect of non-white race on undertreatment (OR = 1.15; 95% CI = 1.07–1.24), >60-day delay (OR = 1.49; 95% CI = 1.37–1.61), and >90-day delay (OR = 1.43; 95% CI = 1.28–1.60). Being in the lower or lowest quintiles of nSES explained 45.71% of the effect of non-white race on undertreatment, 22.69% of the effect on 60-day delay, and 31.97% of the effect on 90-day delay (Table 5).

The objective of this study was to investigate the interrelationship between patient race/ethnicity and socioeconomic status on receipt of high-quality care for colorectal cancer in California. Our study rests on the premise that GCT has been shown to improve survival. Therefore, understanding mechanism that led to improved GCT may lead to improved survival among patients with colorectal cancer. We were interested in examining disparities in two distinct measures of care quality, receipt of all recommended treatment and timeliness of treatment initiation, and quantifying the contribution of socioeconomic deprivation on the racial and ethnic differences for each measure. To our knowledge, this is the first study to extensively quantify the contribution of SES in racial/ethnic disparities in receipt of timely appropriate treatment for CRC.

In our minimally adjusted models, Hispanics and NHBs were at increased risk for undertreatment and treatment delay. After further adjustment with nSES, these associations were attenuated. nSES was an important risk factor for both undertreatment and treatment delay with lowest SES patients being twice as likely to experience >90-day treatment delay compared to the highest SES groups. Our joint effects analysis revealed that lower nSES patients of any race were at substantially higher risk for undertreatment and treatment delay, and that racial/ethnic disparities are reduced or eliminated among non-white patients living in the highest SES neighborhoods. Our mediation analysis revealed that lower nSES explained 46% of undertreatment among non-white patients, 23% of 60-day and 32% of 90-day treatment delay. This is a key finding. Racial and ethnic disparities persisted after accounting for neighborhood socioeconomic status, and between the two, race/ethnicity explained a larger portion of the total effects.

In line with our findings, several studies have shown that although most colorectal cancer patients do receive proper treatment, blacks had lower rates of consultation, and among those who were consulted, received surgery, chemotherapy, and radiation less often than whites (23–25, 30). Furthermore, considering only NCCN centers, 90% of stage II–III rectal cancer patients received proper treatment but only 81% received that treatment within 9 months (31).

Poor quality of care contributes to reduced survival for patients with cancer. Recently, Oh and colleagues examined 15-year trends in survival disparities by race/ethnicity and nSES using the same dataset as our study (3). They observed that although racial and ethnic disparities in survival are improving over time, there were no changes in disparities by nSES and increasing disparities by health insurance status. Also, a 2018 study using CCR data revealed through a mediation analysis that 16% to 30% of the effect of race/ethnicity on survival was explained through stage at diagnosis, leaving a larger portion attributable to race/ethnicity (32).

Barriers in receipt of health care and preventative services include financial barriers (inadequate health insurance, low personal income, and high poverty rates), physical barriers (lack of transportation and geographic access to health care facilities), and personal barriers (cultural and linguistic factors, discrimination, and provider bias; ref. 14). The top four barriers identified within the Latino community were low knowledge and awareness of colorectal cancer, language barriers, lack of insurance, and undocumented legal status; additional barriers included seeking healthcare only when sick, fatalism, denial that colorectal cancer can occur, other needs more pressing than preventative care, and use of home remedies rather than biomedical care (33).

Our findings support the idea that disparities in colorectal cancer patients' access to early detection services, receipt of timely and high-quality care, and survival can be explained, at least in part, by socioeconomic inequalities present within and between racial and ethnic groups. A 2006 article using CCR data found that after adjustment, later stage diagnosis was more likely among Hispanics relative to NHWs and among persons in the two lowest SES quintiles relative to the highest quintile (34). Two recent studies revealed, using SEER and the National Cancer Database, that patients with lower SES were less likely to receive any recommended treatment and that there was a higher rate of nonadherence in Medicaid, Medicare, and uninsured patients (26, 35). Furthermore, within each ethnic group, patients with high SES experienced longer survival times compared with middle and low SES patients (36). Previous research has also noted that patients with lower SES and subsequently a lack of health insurance or nonprivate insurance, are more likely to not receive proper care.

Possible reasons why minorities experience undertreatment and treatment delays include it was recommended less often, patient refusal, and higher prevalence of comorbidities (23, 30, 20). Discrimination was observed in a 2012 study examining colorectal cancer screening practices and found that general practitioners were less likely to recommend screening to immigrants and most reported that immigrants were less likely to participate (37). Furthermore, physicians whose patients are primarily minorities have been found to have less training and expertise in performing screening procedures, less access to clinical resources, and receive less insurance reimbursements (38).

Others before us have observed that racial/ethnic disparities are only partially explained by patient SES (39–40). Especially among Hispanics, colorectal cancer disparities have been attributed to language barriers. A 2014 study found differences in colorectal cancer screening attendance between English- versus Spanish-speaking Hispanics (40). A study that took place in North Carolina concluded that Hispanics appear to be aware and interested in screening but culturally and linguistically appropriate programs to address barriers such as lack of access to resources, including lack of tailored colorectal cancer information, cost uncertainty, and stigma are needed (41). Our study was unable to address language barriers, but we believe it would be important to include in future research of colorectal cancer treatment disparities.

nSES is generally considered to be a contextual variable, capturing a patient's environment and access to healthcare, among other factors (22). A block group is the smallest geographic unit of aggregation published by the census and contains between 250 and 550 housing units. Thus, to the extent that neighbors are similar in terms of their education and income levels, neighborhood-level SES may be representative of individual-level SES (42). However, both individual-level and neighborhood-level SES have been shown to be important contributors of health, with both distinct and shared mechanistic pathways (43, 44). We also accounted for Medicaid insurance, which may be a better proxy for individual deprivation but could be situational in nature rather than an accurate reflection of socioeconomic position. We observed that 6% of patients overall (9% for Blacks and 12% for Hispanics) were covered by Medicaid, which includes some, but not all the patients classified as living in the lowest quintile SES neighborhoods. Thus, for the interpretation of our results, fully adjusted models for neighborhood SES reflect a contextual interpretation of SES, while holding individual-level deprivation constant.

This study included limitations that should be kept in mind. We classified GCT only based on information available in the CCR. For example, patient comorbidity information is not available, and the CCR does not track information about the adequacy of tumor resection, both of which could result in outcome misclassification, on a patient-by-patient basis. The degree of this misclassification could depend on race/ethnicity, as minorities have a higher prevalence of comorbidities than NHWs and are less often treated at the highest quality care facilities. Furthermore, CCR does not provide specific information on surgical techniques, radiation dose and fractionation, chemotherapy agents, and schedules, which are all components of GCT. Meaning, our measurement of GCT is susceptible to misclassification that is expected to be independent of race/ethnicity, resulting in attenuated results. Also, we used multiple imputation to address missing data which is a possible source of uncertainty. As missing data was more common in non-Hispanic Blacks and Hispanics, compared with non-Hispanic Whites, we believed the potential source of uncertainty outweighed the consequence of biasing our study population by only using complete case analysis (45). Finally, our study does not include Asian American patients, a heterogenous population for which some, but not all, members are considered disadvantaged minorities in California. This decision was made to avoid inappropriate aggregation (46–49) of Asian Americans in the absence of large enough sample sizes to investigate care outcomes for the specific Asian ethnicities of interest. Our study was not designed to evaluate socioeconomic and racial disparities in cancer screening, cancer incidence, or cancer-related mortality but these are important areas for future research questions.

In summary, the racial and ethnic disparities that exist in receipt of GCT can be partially explained by contextual socioeconomic inequalities present within and between racial and ethnic groups. Researchers and clinicians should focus their attention to minority and lower income communities that need additional health resources. These health resources might include education targeted to communities determined to be at higher risk for undertreatment or treatment delay, health material and communication development for patients with lower health literacy that accounts for linguistic factors, and additional health care facilities to increase access to health care in areas with insufficient resources or located far from the nearest resources. In addition, a continued effort needs to be made to eliminate discrimination and provider bias.

C.A. Obrochta reports grants from University of California Tobacco Related Disease Research Program and San Diego State University's Graduate Fellowship during the conduct of the study, as well as personal fees from Progenity, Inc., and Alexion Pharmaceuticals, Inc. outside the submitted work. J.D. Murphy reports personal fees from Boston Consulting Group outside the submitted work. C.A. Thompson reports grants from National Institutes of Health and University of California San Diego during the conduct of the study. No disclosures were reported by the other authors.

C.A. Obrochta: Conceptualization, data curation, software, formal analysis, visualization, methodology, writing–original draft. J.D. Murphy: Conceptualization, resources, writing–review and editing. M.-H. Tsou: Conceptualization, resources, writing–review and editing. C.A. Thompson: Conceptualization, resources, supervision, funding acquisition, methodology, writing–review and editing.

C.A. Obrochta was funded by the University of California Tobacco Related Disease Research Program (T30DT0862) and San Diego State University's Graduate Fellowship, and C.A. Thompson was funded by the National Institutes of Health (U54CA132384, U54CA132379, and KL2TR001444). The collection of cancer incidence data used in this study was supported by the California Department of Public Health pursuant to California Health and Safety Code Section 103885; Centers for Disease Control and Prevention's (CDC) National Program of Cancer Registries, under cooperative agreement 5NU58DP006344; the National Cancer Institute's Surveillance, Epidemiology and End Results Program under contract HHSN261201800032I awarded to the University of California, San Francisco, contract HHSN261201800015I awarded to the University of Southern California, and contract HHSN261201800009I awarded to the Public Health Institute. The ideas and opinions expressed herein are those of the author(s) and do not necessarily reflect the opinions of the State of California, Department of Public Health, the National Cancer Institute, and the Centers for Disease Control and Prevention or their Contractors and Subcontractors.

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.