Article Text

Download PDFPDF

Race, neighbourhood characteristics and disparities in chemotherapy for colorectal cancer
  1. Y Hao1,2,
  2. H Landrine3,
  3. A Jemal1,
  4. K C Ward4,
  5. A R Bayakly5,
  6. J L Young Jr4,
  7. W D Flanders4,
  8. E M Ward1
  1. 1Surveillance and Health Policy Research, American Cancer Society, Atlanta, Georgia, USA
  2. 2Centers for Public Health Research and Evaluation, Atlanta, Georgia, USA
  3. 3Behavioral Research Center, American Cancer Society, Atlanta, Georgia, USA
  4. 4Department of Epidemiology, Rollins School of Public Health, Emory University, Atlanta, Georgia, USA
  5. 5Georgia Comprehensive Cancer Registry, Georgia Division of Public Health, Atlanta, Georgia, USA
  1. Correspondence to Yongping Hao, Centers for Public Health Research and Evaluation, Battelle, 2987 Clairmont Road NE, Atlanta, Georgia 30329, USA; haoy{at}battelle.org

Footnotes

  • Competing interests None.

  • Ethics approval The study had the approval of the IRB of Georgia Department of Human Resources, Office of Commissioner (project number: 070101).

  • Provenance and peer review Not commissioned; externally peer reviewed.

View Full Text

Statistics from Altmetric.com

Race and neighbourhood characteristics have been known to be associated with disparity in cancer staging with higher proportion of unstaged1–3 or advanced-stage cases4 5 among minority races and/or in socioeconomically disadvantaged neighbourhoods. Our focus in this study, however, is to investigate whether similar disparity in treatment exists among those cases which have received full staging—specifically, cases assigned as stage III colon and stage II/III rectal cancers, for which adjuvant chemotherapy has been recommended.

Use of adjuvant chemotherapy for stages III colon and II/III rectal cancers has increased since the 1990 NIH consensus statement recommending postoperative 5-fluorouracil/levamisole for these patients.6–8 Considerable disparities in receipt of this treatment persist nonetheless, with older patients less likely than younger ones,7–11 and Blacks less likely than Whites8 10 12 to receive recommended chemotherapy. Whether chemotherapy treatment also varies with neighbourhood socioeconomic status (SES), residential segregation or rurality remains largely unknown. The few studies that have examined neighbourhood SES have yielded inconsistent results; some found no relationship between neighbourhood SES and treatment of colon8 12 or rectal8 12 13 cancers, whereas others found that colon cancer patients residing in low-SES neighbourhoods were significantly less likely to receive chemotherapy than those in higher-SES communities.10 13 The three studies that examined neighbourhood rurality found similarly inconsistent results; two (in California and Louisiana) found no differences between patients residing in rural versus urban communities,11 12 whereas the third (in Washington) found significant rural–urban difference for colon cancer patients.13 In this study, multilevel analyses were used to explore the association of neighbourhood characteristics and racial disparity in receipt of chemotherapy for stages III colon and II/III rectal cancers diagnosed 2000–2004 in Georgia. Analyses of racial disparities nested within different neighbourhood settings can highlight the geographic areas of disparities where interventions are mostly needed.

Materials and methods

Data

Data on all invasive stages III colon and II/III rectal cancers diagnosed 2000–2004 were obtained from the Georgia Comprehensive Cancer Registry (GCCR). Georgia is one of the states funded by the National Program of Cancer Registries of the Centers for Disease Control and Prevention and has been collecting all cancer cases diagnosed among Georgia residents since 1 January 1995. We did not include data from earlier years, because 2000 was the first data year that GCCR received gold certification from the North American Association of Central Cancer Registries. Cancer cases were geocoded to the census tract level according to patient address at the time of diagnosis. Neighbourhood (census tract) poverty and segregation (% Black residents) were obtained from 2000 census data, and neighbourhood rurality data obtained from 2003 Rural–Urban Commuting Area (RUCA) codes14 and linked to the GCCR data using census tract identifiers. Cases (18%) that lacked an exact street address (ie, rural routes, PO Boxes) were goecoded to the centroid of the ZIP code, which ranged from 40.2% in rural, 22.8% in suburban and 9.6% in urban areas. However, the per cent of cases missing exact addresses are similar for Black and White patients (17.2% and 18.5%).

The GCCR is a statewide, population-based cancer registry that compiles data on demographics, cancer diagnosis, stage and first course of treatment using medical records from both hospital and non-hospital sources. Primary invasive colorectal cancer (CRC) cases were identified according to International Classification of Disease for Oncology (ICD-O-3) topography codes C18.0–18.9, C19.9 and C20.9.15 Cancer cases were coded into stages I–IV according to the AJCC Cancer Staging Manual, 6th edn.16 When available, pathological staging was selected over clinical staging. Cases that lacked stage information were coded as unstaged (7%), ranging from 5% in 2004 to 8% in 2000. There is no difference in unstaged cases between Blacks and Whites (7.8% and 6.8%) and men and women (6.6% and 7.7%); however, unstaged cases are slightly higher in rural than urban patients (8.9% vs 6.3%) and varied by age (from 5% in ages <65 to 17% in ages ≥80).

Among the staged, there were 5435 stages III colon and II/III rectal cancer cases diagnosed in Georgia between 2000 and 2004 (27% of total CRC cases); 98% of these had documented surgery (5297 cases) and were eligible for this study. The GCCR does not routinely collect co-morbidity data or contact physicians for information on adjuvant treatment given at physicians' offices (hospital cancer registries perform this task). The chemotherapy data collected by the GCCR does provide codes, however, to indicate various reasons for treatment nonreceipt.17 Less than 2% of cases were documented as not receiving chemotherapy due to co-morbidity (n=28), death (n=7) or patient/guardian refusal (n=40). A total of 296 cases were excluded because receipt of chemotherapy was coded as unknown (n=116) or recommended but not documented as received (n=180). This yielded 4926 cases. An additional 25 patients residing in census tracts without a RUCA code were excluded. We also excluded all race/ethnicity except non-Hispanic White and Black (n=153) to restrict analyses to non-Hispanic White (White) and non-Hispanic Black (Black) patients (n=4748).

Study variables

Clinical outcome variable

Receipt of chemotherapy (yes/no) was the outcome variable; “yes” included patients who received a single agent, multiple agents or agent not otherwise specified and “no” included patients whose medical records indicated that chemotherapy was not administered as part of the planned first course of therapy, along with those for whom the receipt of chemotherapy was not documented. Our interest was in factors associated with receiving any chemotherapy (ie, guideline therapy (5-fluorouracil/levamisole) and 5-fluorouracil/leucovorin, as well as with other cytotoxic agents (eg, oxaliplatin, capecitabine and irinotecan)) known to be effective in reducing tumour relapse, improving survival and reducing all-cause mortality.7 9 18–20 Approximately 55% of the patients in this study were documented as receiving chemotherapy during the 5-year period 2000–2004. Although there are concerns about patient and hospital-related chemotherapy underreporting in cancer registries,12 21 22 prior studies found moderate to high agreement between registry data and medical chart reviews (κ=0.72)23 and Medicare claims (κ≥0.73)24 on chemotherapy for all cancers.

Individual-level predictors

The four individual-level predictors (obtained from the GCCR) were patients' sex, race, age at diagnosis and anatomic site. Age at diagnosis was grouped into six categories: 15–54, 55–64, 65–69, 70–74, 75–79 and 80 years. The median age was 65 years. According to the ICD-O-3 anatomic codes, colon was coded C18 and rectum included C19.9 and C20.9.

Neighbourhood-level predictors

The three neighbourhood-level predictors were poverty, segregation and rurality. Neighbourhood poverty was measured as the percentage of census tract (CT) residents below the federal poverty line. Prior studies reveal that this measure is superior to other area-based measures of SES (eg, median home value) in sensitivity to and predictive validity regarding SES-related health disparities, cancer disparities in particular.25 26 The categories of percent residents below the poverty line (BPL) used in prior studies were used here: <5% BPL, <5–9.9% BPL, 10–19.9% BPL and ≥20% BPL,25 26 where CTs with ≥20% BPL are defined as federal poverty areas.27 Segregation was measured as CT% Black residents because this may capture CT segregation better than indices (eg, dissimilarity) meant to be used at the metropolitan statistical area level.28 CT % Black was grouped into <10%=Low, 10–29.9%=Moderate and ≥30%=High, based on the statewide distribution of Blacks. Neighbourhood rurality was assessed using the 10 standard RUCA codes14 collapsed into three groups: Urban=RUCA Code 1, Suburban=RUCA Codes 2–6 and Rural=RUCA Codes 7–10.29

Analytic strategy

The 4748 Black and White patients (Level 1) were nested in1403 CTs (Level 2). Multilevel logistic regression explored the role of neighbourhood and individual variables in receipt of any chemotherapy. SAS PROC GLIMMIX was used to model receipt of chemotherapy (yes/no) using the binomial logit link. Model parameters were estimated using maximum likelihood procedure with Newton–Raphson Ridge Optimisation algorithm.30 Tests of significance for comparison were two-sided at a level of p<0.05.

A series of models were examined and four of those presented: Models 1, 2 and 3 used neighbourhood-, individual- and both neighbourhood- and individual-level predictors (respectively), whereas Model 4 tested treatment differences between Blacks versus Whites within each category of neighbourhood rurality after adjusting for other variables. Men, age group 15–54, White, <5% BPL, Low % Black and rural (to be compared with both suburban and urban neighbourhoods) were used as the reference categories in these models. In addition, models included both colon and rectum cancer cases with rectum used as the referent category. The OR and median odds ratio (MOR) were used to measure association and variation in the receipt of chemotherapy, respectively. The MOR is a recommended measure of area-level variation due to its direct comparability with the OR. MOR quantifies the variation in receipt of chemotherapy between CTs by comparing two patients with the same covariates from two randomly chosen CTs.31 A MOR of one indicates no difference between CTs in the probability of receiving chemotherapy. The MOR was derived from the variance as described by Merlo et al.32

Preliminary analyses included tumour grade as a predictor because prior data (up to 1995) from both the National Cancer Data Base and the Surveillance, Epidemiology and End Results program indicate that patients with more poorly differentiated tumours are more likely to receive chemotherapy than those with well-differentiated tumours.8 However, consistent with more recent (1995–2001) findings from Southern Netherlands, no relationship between tumour grade and receipt of chemotherapy was found here (analyses not shown)33; hence, tumour grade was excluded from the multilevel analyses.

We also examined potential bias in neighbourhood level variance estimates due to small numbers of cases per neighbourhood (n=5 on average, range=1–16). A small simulation study was conducted using the actual data to define the distribution of covariates and the estimated parameters to define the “true” relationships for the purpose of simulation. In these simulations, the neighbourhood level variance tended to be underestimated (about 10%), while the standard error of the estimate was approximately unbiased (analyses not shown). Thus, the actual neighbourhood variance might be larger than the estimated value. Further, estimates for the fixed effects and standard errors should be approximately unbiased according to previous simulation studies designed to assess the data sparseness problem.34

Results

Table 1 depicts population distribution by CT level characteristics among Georgia Black and White residents. In 2000, Whites and Blacks account for 65% and 29% of Georgia's total population (8 186 453), respectively. Compared to Whites, Blacks disproportionally reside in the poorest (39.2% vs 12.5%), most segregated (75.4% vs 18.9%) and urban (71.7% vs 60.1%) neighbourhoods.

Table 1

Population distribution by neighborhood characteristics among Whites and Blacks, Georgia, USA, 2000

Table 2 provides basic data on receipt of chemotherapy among the study population (n=4748). As shown, about 55% of patients (n=2607) were documented to have received chemotherapy during the 5-year period 2000–2004. Without adjusting for covariates, chemotherapy receipt was significantly lower for women than for men (51.5% vs 58.0%, p<0.0001) and for colon versus rectum (50.4% vs 62.7%, p<0.0001); chemotherapy receipt also varied by patient age (p<0.0001), but not patient race, and by CT poverty (p=0.0326) and CT rurality (p=0.0012), but not by CT segregation.

Table 2

Descriptive data on adjuvant chemotherapy (AC) for stages III colon and II/III rectal cancers in Georgia, USA, 2000–2004

Table 3 presents multilevel models predicting receipt of chemotherapy. Model 1 (CT predictors only) reveals that CT poverty and rurality are significant predictors of chemotherapy after mutually adjusted and additional adjusting for segregation, which is consistent with results shown in univariate analysis for the three CT predictors (table 2). Model 2 (individual-level predictors only) reveals that all individual-level variables (age, race and anatomic site) except for sex were statistically significant. Model 3 (neighbourhood and individual-level predictors) reveals that after adjusting for the individual-level variables, CT rurality continued to be a significant predictor whereas CT poverty did not. There is little or no change in OR for sex, specific age categories, anatomic subsite or race after adjustment. Specifically, the odds of receiving chemotherapy for older patients (ages 55–64, 65–69, 70–74, 75–79 and 80) were 14% to 92% lower than those for patients aged 15–54; the odds for patients with colon cancer were 32% lower than for patients with rectal cancer (95% CI 0.60 to 0.78).

Table 3

Multilevel models predicting adjuvant chemotherapy for stages III colon and II/III rectal cancers in Georgia, USA, 2000–2004

It should be noted that the Black–White difference, which is not significant in univariate analysis (table 2), becomes significant after adjusting for sex, age and anatomic site (table 2, Model 2). Stratified analyses indicate that age is clearly a confounder for race (results not shown). Briefly, higher proportion of Blacks than Whites was diagnosed for stages III colon and II/III rectum cancers in ages 15–54 (31% vs 22%) and 55–64 (25% vs 23%), but lower proportion of them received chemotherapy: 64% versus 73% in ages 15–54 and 61% versus 69% in ages 55–64, respectively. Similarly, although higher proportion of Black patients resided in urban areas, lower proportion of them in younger ages received recommended treatment: 62% versus 74% in ages 15–54 and 58% versus 70% in ages 55–64, respectively. This also demonstrates why the Black–White difference among urban patients, which is not significant in univariate analysis (table 2), becomes significant after adjusting for age (table 3, Model 4); whereas the racial disparity among rural patients, which is significant in univariate analysis (table 2), turns out to be not significant after adjusting for age (table 3, Model 4).

Model 4 examines the association of patient race within rural versus suburban versus urban settings. We did not further evaluate the association of patient race within categories of poverty and segregation because neither of them significantly contributed to disparities in receipt of chemotherapy after adjusting for individual level variables and neighbourhood rurality (table 3, Model 3). Model 4 reveals significant treatment disparities for patients residing in rural neighbourhoods compared to their urban and suburban counterparts irrespective of patient race; likewise, the model reveals that Black–White disparities are confined to and strongest among urban patients. Specifically, odds of receiving chemotherapy for urban and suburban patients were 38% (95% CI 1.09 to 1.74) and 53% (95% CI 1.20 to 1.94) higher (respectively) than for rural patients. Among those residing in urban neighbourhoods, the odds of receiving chemotherapy for Black patients were 24% (95% CI 0.62 to 0.94) lower than for their White counterparts. Receipt of chemotherapy did not significantly differ between Blacks and Whites residing in suburban or rural areas. This pattern also emerged in models restricted to urban only, suburban only and rural only patients (results not shown). The inclusion of spatial autocorrelation between census tracts in models did not alter the results either (results not shown). Also shown (table 3, bottom) is that receipt of chemotherapy continued to vary across neighbourhoods even after controlling for the individual- and neighbourhood-level factors (0.151, SE 0.053); this neighbourhood variation in receipt of chemotherapy was not explained by neighbourhood poverty, segregation or rurality. The MOR of 1.44 in Model 4 indicated that a patient's odds of receiving chemotherapy would be about 44% higher (in median) if a patient with the same covariates moved from a randomly chosen neighbourhood to another randomly chosen neighbourhood with a higher probability of receiving chemotherapy.

Discussion

This study has several important results. First, patients with stages III colon and II/III rectal cancers (CRC patients) who reside in rural areas were significantly less likely than their urban and suburban counterparts to receive chemotherapy; this result is inconsistent with prior studies that found no rural–urban differences in receipt of chemotherapy.11 12 This difference in results may reflect differences in measurement; studies that found no differences used a two-group classification (rural vs urban) whereas we used the three-group classification of rural versus suburban versus urban, and the latter may be more sensitive to geographic variation in access to healthcare. 29 35–38 In addition, associations between rural residence and chemotherapy utilisation may differ in different geographical regions and healthcare systems. Our finding is consistent with a larger literature on the barriers to treatment among rural cancer patients. Such studies have found that rural cancer patients experience more difficulty than their urban cohorts in accessing new, more effective therapies and specialised services due to poorly designed and/or shrinking health infrastructures, socioeconomic hardship and geographic barriers (eg, the burden of long-travel distances).38 39 Thus, our findings may highlight resource needs of CRC patients in rural areas.

Second, we found that Black CRC patients were significantly less likely than their White counterparts to receive chemotherapy but that this treatment disparity was confined to urban Blacks and Whites alone. This finding might be viewed as somewhat inconsistent with prior studies that have found general Black–White disparities.8 10 12 However, those studies did not examine the nested association of racial disparities within rural/urban areas; hence, their findings for Blacks (as a whole) might have been a function of disparities for urban Blacks alone. The latter possibility is supported by the fact that the majority of the US and of the Georgia Black populations (72%) reside in urban areas. Foremost, among the factors that might contribute to the racial treatment disparity within urban areas is lower access to or utilisation of quality health care. A growing body of literature indicates that living in poor neighbourhoods has deleterious effects on heath and access to health care40–42; such effects might be greater for Blacks than for Whites in part because Blacks (39%) disproportionately reside in high poverty (especially urban) neighbourhoods relative to Whites (13%).43 Similarly, prior studies indicate that many healthcare settings in urban Black neighbourhoods lack advanced diagnostic and imaging resources and medical specialists.44–47 Consequently, urban Blacks often receive lower-quality screening, diagnostic and treatment services than Whites44–50 and often from physicians who are less competent than those in urban healthcare settings for mostly White patients.35 46 In any event, our findings suggest that interventions to reduce CRC treatment disparities by race might benefit from targeting urban areas.

Third, consistent with prior studies, we too found that older CRC patients were less likely than younger ones to receive chemotherapy. Detailed explanations for this age-related disparity have been presented elsewhere.7–11 These include the higher frequency of co-morbidities among older patients and higher patient refusal rates (due to concerns about toxicity) and (often) greater financial and geographic barriers among older patients. Similarly, the finding that colon cancer patients were less likely than patients with rectal cancer to receive chemotherapy also is consistent with some prior research.8 Likewise and consistent with prior studies, we did not find significant sex differences in receipt of chemotherapy (after adjusting for other factors). This is not surprising given that sex has not been a consistent predictor of chemotherapy among CRC patients; some studies found that women were less likely than men to receive chemotherapy,7 10 13 whereas others found either the opposite pattern8 or no sex disparity,12 as we did. Finally, neither poverty (after adjusting for individual-level factors) nor segregation was found to be a significant predictor of chemotherapy, and this too is consistent with some previous work.11 12

This study has several notable strengths such as the large number of Black patients (25% (n=1173) of the study population) and, hence, a sufficient sample size for multivariate analyses of racial-geographic disparities in chemotherapy. In addition, we analysed 2000–2004 GCCR data, and these are more complete and reliable than data from previous years. Likewise, we used census tract as the area level in the multilevel analyses and census tract-level data known to be superior to zip code, county and other geographic levels in sensitivity to neighbourhood influences on health.12 25 Similarly, this study appears to be the first analysis of cancer treatment that used a three-group classification of rurality (and, hence, enhanced sensitivity to potentially robust geographic differences29 35–38) and the first to examine the interaction between patient race and neighbourhood-level variables.

Nonetheless, the study also has several limitations. Foremost, among these is the under-collection of chemotherapy data by population-based cancer registries.21 22 24 Our findings are based on the available data, and these tend to underestimate chemotherapy usage due to lack of treatment information from physician offices. Such underestimation might possibly account for some of our findings (eg, if documented chemotherapy is more likely to be missing for urban Blacks than urban Whites). However, a study of California CRC patients showed that disparities in treatment, along with SES, largely explain the decreased survival of Black patients,51 suggesting that the above explanation is unlikely. A related concern about this study is that the inclusion of cases (18%) that lacked exact street addresses might result in potential misclassification of CT level characteristics, especially among rural CTs. Alternative approach such as “geo-imputation” would increase match rates of CTs,52 which we were unable to peruse due to lack of access to raw data. Similarly, the lack of stage and treatment information for those unstaged cases (7%), which were slightly higher in rural (8.9%) relative to urban neighbourhoods (6.3%), might affect treatment disparities in urban–suburban–rural neighbourhoods among staged cases included in the study.53 Also, the lack of information in the GCCR on patient individual-level income, education, health insurance and co-morbidity, each of which may moderate or mediate the relationships found. For example, if the largest personal SES and insurance differences are between urban Blacks and urban Whites, such differences might have contributed to the urban-racial disparity found. Similarly, like other population-based cancer registries, the GCCR lacks information on hospital characteristics (ie, volume) and the availability of oncologists, both of which might play a role in chemotherapy use. Finally, our study included CRC patients diagnosed and treated in Georgia only, and this study population might not be representative of patient- or treatment-patterns in the US.

In summary, significant disparities in receipt of chemotherapy were found for rural CRC patients irrespective of race and for Black CRC patients in urban neighbourhoods. Our findings highlight geographic areas where targeted interventions might be needed.

What is already known about this topic

  • Since the 1990 NIH consensus, adjuvant chemotherapy use for stages III colon and II/III rectal cancers has increased.

  • There are disparities in receipt of this recommended treatment among older patients compared to younger ones and among Blacks compared to Whites.

What this study adds

  • There are disparities in receipt of chemotherapy for rural patients after adjusting for important variables at both individual- and community-level.

  • The Black–White disparities in receipt of this treatment are confined to urban patients.

  • Interventions should be targeted to rural patients in general and urban Blacks.

Acknowledgments

This study was supported by the American Cancer Society. We are deeply grateful for the thoughtful insights given by the referees which have greatly improved the paper.

References

View Abstract

Footnotes

  • Competing interests None.

  • Ethics approval The study had the approval of the IRB of Georgia Department of Human Resources, Office of Commissioner (project number: 070101).

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.