Fax: (713) 500 9264
Meta-analysis of racial disparities in survival in association with socioeconomic status among men and women with colon cancer
Article first published online: 23 APR 2007
Copyright © 2007 American Cancer Society
Volume 109, Issue 11, pages 2161–2170, 1 June 2007
How to Cite
Du, X. L., Meyer, T. E. and Franzini, L. (2007), Meta-analysis of racial disparities in survival in association with socioeconomic status among men and women with colon cancer. Cancer, 109: 2161–2170. doi: 10.1002/cncr.22664
- Issue published online: 18 MAY 2007
- Article first published online: 23 APR 2007
- Manuscript Accepted: 1 FEB 2007
- Manuscript Revised: 19 DEC 2006
- Manuscript Received: 16 NOV 2006
- colon cancer;
- socioeconomic status;
Few studies have addressed racial disparities in survival for colon cancer by adequately incorporating both treatment and socioeconomic factors, and the findings from those studies have been inconsistent. The objectives of the current study were to systematically review the existing literature and provide a more stable estimate of the measures of association between socioeconomic status and racial disparities in survival for colon cancer by undertaking a meta-analysis.
For this meta-analysis, the authors searched the MEDLINE database to identify articles published in English from 1966 to August 2006 that met the following inclusion criteria: original research articles that addressed the association between race/ethnicity and survival in patients with colon or colorectal cancer after adjusting for socioeconomic status. In total, 66 full articles were reviewed, and 56 of those articles were excluded, which left 10 studies for the final analysis.
The pooled hazard ratio (HR) for African Americans compared with Caucasians was 1.14 (95% confidence interval [95% CI], 1.00–1.29) for all-cause mortality and 1.13 (95% CI, 1.01–1.28) for colon cancer-specific mortality. The test for homogeneity of the HR was statistically significant across the studies for all-cause mortality (Q = 31.69; P < .001) but was not significant across the studies for colon cancer-specific mortality (Q = 7.45; P = .114).
Racial disparities in survival for colon cancer between African Americans and Caucasians were only marginally significant after adjusting for socioeconomic factors and treatment. Attempts to modify treatment and socioeconomic factors with the objective of reducing racial disparities in health outcomes may have important clinical and public health implications. Cancer 2007. © 2007 American Cancer Society.
Colorectal cancer is the third most common cancer diagnosed (excluding skin cancer) and the second most common cause of cancer death in the United States.1–3 It has been estimated that 148,610 men and women will be diagnosed with and 55,170 will die of cancer of the colon and rectum in 2006.1, 2 Although the mortality from colon cancer has decreased over time,1–3 incidence and mortality still vary greatly by race/ethnicity.1–8 For example, the age-adjusted death rate of colorectal cancer from 2000 to 2003 was 27.3 per 100,000 African Americans and 19.3 per 100,000 Caucasians in the National Cancer Institute's Surveillance, Epidemiology, and End Results Program (SEER) areas,3 whereas the relative 5-year survival rate was 54.7% for African Americans and 65.1% for Caucasians.2, 3 Numerous studies have indicated that the increased mortality in African Americans with colon cancer may be attributed in part to more aggressive tumors and to more advanced stage at diagnosis,9–12 differences in treatment,13–19 tolerance of treatment,8 provider characteristics,20, 21 and screening and postdiagnosis surveillance.2, 8, 22–24 For example, several population-based studies demonstrated that African Americans with stage III colon cancer were less likely than Caucasians to undergo definitive surgery and receive adjuvant chemotherapy as part of standard therapy.15–18
Socioeconomic status (SES) is an important known determinant of outcomes in cancer patients.5, 7 SES impacts cancer outcomes through numerous ways, including health care resources and psychosocial factors. However, relatively few studies have examined the impact of SES on treatment and survival among patients with colon cancer.25–39 Only a handful of studies have addressed racial disparities in survival for colon cancer by adequately incorporating both treatment and SES31, 33, 35, 37 in addition to factors like cancer stage, tumor grade, and comorbid conditions. Among this limited number of studies,28–39 the findings have been inconsistent. One of the main reasons for these inconsistent findings may be that previous studies had relatively small numbers of study patients. Therefore, we undertook a meta-analysis to synthesize the findings from previous studies. This approach can be helpful in increasing the sample size and, thus, enhancing the study power by combining numerous studies. In this meta-analysis, our objectives were to systematically review the existing literature, have a more objective appraisal of the evidence, examine the heterogeneity between the studies, and provide more stable estimates than those provided by individual studies of the association between colon cancer survival and race after accounting for SES.
MATERIALS AND METHODS
We searched the National Library of Medicine MEDLINE literature database to identify articles that were published in English language from 1966 to the fifth week of August 2006 that met the inclusion criteria for this meta-analysis: original research articles that addressed the association between race/ethnicity and survival in patients with colon cancer or colorectal cancer after adjusting the analysis for SES. We performed 3 separate searches combining terms for colon cancer (colon or colonic neoplasms, colorectal cancer, or colorectal neoplasms) and survival (survival analysis, survival, survival rate, proportional hazards models, Cox regression, or mortality) with 1) SES or social class, 2) ethnicity (ethnic groups, minority groups, European Continental Ancestry Groups, or ethnicity), or 3) race (Continental Population Groups, African Continental Ancestry Group, blacks, or African Americans). We reviewed the abstracts of all 296 articles that resulted from these search procedures. From those 296 articles, we excluded 24 review articles and 206 abstracts that did not report a time-to-event survival analysis of the association between survival and either race/ethnicity or SES. For those articles that did not have abstracts in MEDLINE, we included them in the full review. In total, 66 full articles were reviewed, 56 of which were excluded because they did not meet our inclusion criteria. We also reviewed all reference lists in these full articles to search for any potential studies that may not have been identified in our MEDLINE search. This resulted in 2 additional articles that met our inclusion criteria.32, 39 Those 2 articles were not found in the MEDLINE search because they were not indexed with Medical Subject Headings related to race or SES. In addition, several studies discussed racial disparities in survival but did not provide any risk measures using time-to-event analysis for these 2 population comparisons40–42 and, thus, were not included in the analysis.
For the 12 studies that met the inclusion and exclusion criteria,28–39 information was abstracted and recorded on a structured information collection form by 1 investigator (T.E.M.) and was verified by another investigator (X.L.D.). Discrepancies were resolved through discussion. Abstracted information included the study setting, sample size, outcome (all-cause or colon cancer-specific mortality), race/ethnicity (definition and source), SES (definition and source), other variables in the multivariable models, and the crude and adjusted measures of association with 95% confidence intervals (95% CIs).
For several articles that did not specifically report the mortality hazard ratio (HR) in the text or tables, HRs and 95% CIs were estimated based on the information presented in the articles. For example, Govindarajan et al.37 reported the HR and 95% CI in a figure but not in a table or text. We used a ruler to estimate the HR and 95% CI, assuming that the figure was drawn to scale. Dayal et al.29 reported a P value <.001 rather than a 95% CI for the HR, so we calculated a conservative estimate for the 95% CI assuming a P value of .001. Although Akerley and colleagues did not report an HR adjusted for SES, they stated in the text that adding SES to the model did not change the HR.30 We recorded their unadjusted HR and 95% CI. For the study by Potosky et al.,35 who reported an HR in African Americans as the reference group, the HR and 95% CI were recalculated using Caucasians as the reference group to be consistent with all other studies. Rogers et al.38 did not adjust for SES but restricted their study population to those who had been enrolled in both Medicare and Medicaid for at least 1 year. Because enrollment in Medicaid requires a low income level, all participants in their study were considered to have low SES.
Because our main objective was to compare the survival of African Americans and Caucasians with colon cancer, 2 studies were excluded because they did not have comparisons of survival between African American and Caucasians.28, 39 This left 10 studies in the final analysis.29–38 However, study populations for 3 of those 10 studies had some overlaps. In their study, Merrill et al32 sampled patients with colorectal cancer cases aged ≥65 years from the 2 SEER areas in San Francisco-Oakland and Seattle-Puget Sound from 1985 to 1992; whereas, in their study, Potosky et al35 covered patients with colorectal cancer aged ≥20 years from the 9 SEER registries, including San-Francisco-Oakland and Seattle-Puget Sound, from 1987 to 1991. The study by Mayberry et al31 covered patients ages 20 years to 79 years from the SEER areas in Atlanta, New Orleans, and San Francisco-Oakland from 1985 to 1986. We performed a sensitivity analysis by excluding the study by Merrill et al,32 because their study population overlapped with 2 other studies, and we observed that the results remained almost unchanged. In addition, because the study populations were overlapping only for those aged ≥65 years and in only 2 geographic areas, all 3 studies were left in the final analysis.
We calculated pooled HRs and 95% CIs that compared the survival experience in African Americans with the survival experience in Caucasians using the meta (package: sbe_16_2.pkg) command in Intercooled Stata (version 8.0; Stata Corporation, College Station, Tex)43, 44 and specified a random-effects model.45 The Q-test statistic was used to test for heterogeneity among the studies,46 and the Egger and Begg regression asymmetry tests were used to test for the presence of potential publication bias using metabias (sbe19.pkg).47, 48 We used a P value <.05 to indicate both heterogeneity and publication bias. The pooled HRs were calculated separately for all-cause mortality and for colon cancer-specific mortality. HRs that were included in the pooled estimate were adjusted for various confounders, such as age, sex, SES, and other variables, as presented in Table 1. When multiple HRs were presented in the original articles, we included the HR and 95% CI that were adjusted for the most extensive confounding variables available.
|Reference||Study type||Year of DX||Last date of follow-up||Setting/location||No. of cases (deaths)||Age group, y||Tumor stage||Primary SES measure||SES data source||Other variables in the analysis|
|Rogers et al., 200438||RC||1984–1994||12/2001||Hospitalized and enrolled in both Medicare and Tennessee Medicaid||272 (233)||697 (579)||969 (812)||>65||I–IV||Medicaid beneficiaries||Tennessee Medicaid||Age, sex, nursing home status, residence in an urban or rural setting, stage, anemia, comorbidities|
|Govindarajan et al., 200337||RC||1984–1997||NA||State-funded university hospital in Arkansas||190 (NA)||427 (NA)||617 (NA)||>18||0–IV||Education, income and poverty at zip code level||1990 Census of Population and Housing for Arkansas||Cancer site, stage, treatment, age, sex|
|Lee-Feldstein et al., 200236||RC*||1987–1993||12/1996||Sacramento region of the California Cancer Registry||121 (NA)||1078 (NA)||1329 (722)||>65||I–IV||Education at census tract||1990 Census||Health insurance, stage, sex, age, tumor location, hospital type|
|Potosky et al., 200235||RC||1987–1991, 1995||12/1998||SEER Atlanta, Connecticut, Detroit, Iowa, New Mexico, San Francisco-Oakland, Seattle-Puget Sound, Utah (except 1995) and Los Angeles (only 1995) areas||NA||NA||2145 (1127)||>20||II, III||Income at census tract level and race-specific||1990 Census||Cancer site, tumor extent, lymph node status, grade, treatment, age, sex, comorbidities, y of DX, SEER region|
|Marcella and Miller, 200134||RC||1987–1993||12/1994||Pennsylvania State Tumor Registry Expanded Cancer Incidence File||3784 (1347 CRC deaths)||58,020 (17,014 CRC deaths)||61,804 (10,213 AC deaths and 18,361 CRC deaths)||NA||In-situ, local, regional, distant, unknown||Race- and sex-specific education and unemployment||1990 Census||Cancer site, stage, grade, age, sex|
|Roetzheim et al., 200033||RC||1994||12/1997||Florida Cancer Data System||566 (267)||7966 (3277)||9551 (3883)||NA||In situ, local, regional, distant, unstaged||Education and income at census tract||1990 Census||Cancer site, stage, treatment, age, sex, insurance, urban/ rural place of residence, comorbidities, smoking|
|Merrill et al., 199932||RC||1985–1992||12/1994||SEER San Francisco-Oakland and Seattle-Puget Sound areas||NA||NA||15,352 (NA)||>65||In-situ, I–IV, unknown||Education at census tract||1990 Census||Cancer site, stage, age, sex, comorbidities, SEER region, insurance status|
|Mayberry et al., 199631||PC||1985–1986||12/1990||Cancer registries in Atlanta, New Orleans, or San Francisco- Oakland||454 (221)||521 (201)||975 (422)||20–79||I–IV||Poverty, education, and occupation at individual level||Personal interview||Stage, tumor characteristics, treatment, age, sex, location, symptoms, comorbidities, health behavior|
|Akerley et al., 199330||RCT||1981–1986||05/1986||12 Participating VA hospitals||25 (NA)||110 (NA)||135 (NA)||NA||IV||Education, employment status at individual level||Clinical trial procedures/ interview||Treatment, age, weight loss, body surface area, performance status, blood chemistries at entry|
|Dayal et al., 198729||RC||1977–1982||02/1985||22 Comprehensive cancer centers in the US||1072 (NA)||2545 (NA)||3617 (NA)||All ages||Local, regional, distant||Education at zip code level||1970 Census||Age, sex|
We also performed a number of sensitivity analyses. For example, to determine the influence of a single study, we calculated the pooled estimate and 95% CI by excluding 1 study at a time using the metainf (sbe26.pkg) command in Stata.49 Because 7 of the 10 studies reported HRs for survival in patients with colorectal cancer, whereas 3 studies reported HRs for patients with colon cancer only, we calculated separate estimates for the studies that included colorectal cancer and the studies that included colon cancer alone. We also calculated separate estimates for the studies that did or did not consider treatment in addition to SES in their analysis. Furthermore, we calculated pooled estimates for the studies that included and excluded patients with advanced tumor stage. Finally, we provided the pooled estimates of HRs for studies that adjusted for group-level SES versus individual-level SES.
Table 1 presents the basic characteristics of the 10 studies that were included in this meta-analysis, covering study settings, year of diagnosis, number of patients, number of deaths, primary measure of SES, data source for SES, and important variables that were considered in the analysis. In total, 96,494 patients with colon or colorectal cancer were included in our meta-analysis, which covered broad areas in the United States: SEER areas, Arkansas, Florida, Pennsylvania, Tennessee, 22 official comprehensive cancer centers across the country, and 12 Veterans Administration hospitals. These studies represented patients who were diagnosed over a period of 29 years from 1977 to 1997 who had up to 17 years of follow-up. The primary measure of SES in most studies (7 of 10 studies) was the percentage of residents who lived under the poverty line or education at the census tract or zip code level.29, 32, 33, 35–37 Two studies included individual-level education, occupation, or income collected from personal interview.30, 31
All 10 studies provided estimates on racial disparities in survival for colon or colorectal cancer that adjusted for some measures of SES, age, and sex. However, only 5 studies also adjusted for treatment,30, 31, 33, 35, 37 5 studies adjusted for comorbidities,31–33, 35, 38 and 9 studies adjusted for tumor stage or grade.30–38 Only 3 studies31, 33, 35 adjusted for all of the key variables described above on SES, treatment, comorbidity, and tumor stage or grade. Despite these heterogeneities in their methods and analyses, 7 of those 9 studies presented an elevated HR for all-cause mortality in African Americans compared with Caucasians after adjusting for SES, but only 3 studies had increased HRs that were statistically significant (Table 2).29, 33, 37
|Reference||All-cause mortality||Colon cancer-specific mortality|
|HR (95% CI)*||Combined HR (95% CI) of other studies with this study omitted||HR (95% CI)*||Combined HR (95% CI) of other studies with this study omitted|
|1. Rogers et al., 200438||0.95 (0.81–1.11)||1.13 (0.99–1.29)||—||—|
|2. Govindarajan et al., 200337||1.5 (1.25–1.8)||1.12 (0.97–1.28)||—||—|
|3. Lee-Feldstein et al., 200236||1.06 (0.81–1.39)||1.09 (0.97–1.21)||1.18 (0.83–1.68)||1.13 (0.99–1.30)|
|4. Potosky et al., 200235||1.06 (0.79–1.43)||1.15 (1.00–1.32)||1.32 (0.55–3.15)||1.10 (0.94–1.29)|
|5. Marcella et al., 200134||—||—||1.19 (1.09–1.30)||1.11 (0.97–1.26)|
|6. Roetzheim et al., 200033||1.18 (1.01–1.38)||1.13 (0.99–1.30)||—||—|
|7. Merrill et al., 199932||0.94 (0.86–1.03)||1.18 (1.05–1.33)||0.99 (0.88–1.11)||1.20 (1.11–1.30)|
|8. Mayberry et al., 199531||1.20 (0.90–1.60)||1.15 (1.00–1.32)||1.30 (1.00–1.69)||1.13 (1.00–1.29)|
|9. Akerley et al., 199330||1.51 (0.71–3.21)||1.13 (0.98–1.31)||—||—|
|10. Dayal et al., 198729||1.28 (1.10–1.49)||1.17 (1.02–1.35)||—||—|
|Combined||1.14 (1.00–1.29)†||—||1.13 (1.01–1.28)‡||—|
Four of the 5 studies showed elevated cancer-specific mortality in African Americans compared with Caucasians, and 2 of the HRs were statistically significant. Table 2 presents the HR and 95% CI for the 10 individual studies as well as the pooled HR. The pooled HR for African Americans compared with Caucasians was 1.14 (95% CI, 1.00–1.29) for all-cause mortality and 1.13 (95% CI, 1.01–1.28) for colon cancer-specific mortality. These results also were presented in forest plots (Fig. 1). The test of homogeneity of the HR was statistically significant across the 9 studies for all-cause mortality (Q = 31.69 with 8 degrees of freedom; P < .001), supporting the need for a random-effects model for the pooled results. However, the test for homogeneity of the HR was not significant across the 5 studies for colon cancer-specific mortality (Q = 7.45; P = .114). We reported the random-effects estimates for both all-cause and cancer-specific mortality regardless of the Q test P value, because the random-effects model provides a more conservative estimate of the variance.
Table 2 also provides the HR and 95% CI combined from all remaining studies when a particular individual study was removed from the estimate. All combined HR estimates were elevated in African Americans compared with Caucasians; however, 5 of the 9 HRs for all-cause mortality and 3 of the 5 HRs for colon cancer-specific mortality were not statistically significant.
Table 3 presents the findings of sensitivity analyses. For example, in 4 studies that adjusted for SES but not for treatment, the combined HR was 1.04 (95% CI, 0.89–1.22) in African Americans compared with Caucasians; whereas, in 3 studies that adjusted for both SES and treatment factors, the HR for all-cause mortality was significantly higher in African Americans than in Caucasians (1.16; 95% CI, 1.03–1.32). The HR for colon cancer-specific mortality also was significantly higher in African Americans compared with Caucasians (1.30; 95% CI, 1.01–1.67) in 2 studies that considered both SES and treatment. In the stratified analyses by individual level SES versus group level SES or early-stage cancer versus metastatic cancer, the pooled HRs were not elevated significantly in African Americans compared with Caucasians.
|Type of analyses||All-cause mortality||Colon cancer-specific mortality|
|No. of studies||Combined HR (95% CI)||Q testdf (P)||No. of studies||Combined HR (95% CI)||Q testdf (P)|
|1. All studies||Nine29–33, 35–38||1.14 (1.00–1.29)||Q8 = 31.69 (<.001)||Five31, 32, 34–36||1.13 (1.01–1.28)||Q4 = 7.45 (.114)|
|2. Studies that adjusted for SES and treatment||Five30, 31, 33, 35, 37||1.26 (1.10–1.44)||Q4 = 5.79 (.001)||Two31, 35||1.30 (1.01–1.67)||Q1 = 0.001 (.979)|
|3. Studies that adjusted for stage and comorbidity in addition to SES and treatment||Three31, 33, 35||1.16 (1.03–1.32)||Q2 = 0.43 (.807)||Two31, 35||1.30 (1.01–1.67)||Q1 = 0.001 (.979)|
|4. Studies that adjusted for SES without adjusting for treatment||Four29, 32, 36, 38||1.04 (0.89–1.22)||Q3 = 12.58 (.006)||Three32, 34, 36||1.10 (0.95–1.28)||Q2 = 6.13 (.047)|
|5. Studies that adjusted for individual level SES||Two30, 31||1.24 (0.94–1.62)||Q1 = 0.31 (.577)||One31||1.30 (1.00–1.69)||—|
|6. Studies that adjusted for group level SES||Six29, 32, 33, 35–37||1.16 (0.99–1.36)||Q5 = 27.81 (<.001)||Four32, 34–36||1.11 (0.97–1.27)||Q3 = 6.26 (.100)|
|7. Studies that were restricted to Medicaid-eligible population||One38||0.95 (0.81–1.11)||—||—||—||—|
|8. Studies that sampled colon cancer cases||Three29–31||1.27 (1.11–1.45)||Q2 = 0.36 (.835)||One31||1.30 (1.00–1.69)||—|
|9. Studies that sampled colorectal cancer cases||Six32, 33, 35–38||1.10 (0.94–1.28)||Q5 = 24.27 (<.001)||Four32, 34–36||1.11 (0.97–1.27)||Q3 = 6.26 (.1.00)|
|10. Studies that included only advanced-stage cases||Two30, 35||1.11 (0.85–1.45)||Q1 = 0.72 (.397)||One35||1.32 (0.55–3.15)||—|
|11. Excluding studies for which we estimated the HR or 95% CI using information presented in the article||Five30–33, 35||1.08 (0.93–1.25)||Q4 = 8.93 (.063)||—||—||—|
|12. Also excluding a study that reported only unadjusted estimates||Four31–33, 35||1.07 (0.92–1.24)||Q3 = 7.85 (.049)||—||—||—|
|13. Excluding a study that overlaps 2 other study populations||Eight29–31, 33, 35–38||1.18 (1.05–1.33)||Q7 = 16.37 (.022)||Four31, 34–36||1.20 (1.11–1.30)||Q3 = 0.44 (.931)|
We also assessed the potential presence of publication bias using the Egger test for publication bias, which appeared to be present in studies of all-cause mortality (P = .034) but not in studies of cancer-specific mortality (P = .543). This was demonstrated in funnel plots for all-cause and cancer-specific mortality (Fig. 2).
The Institute of Medicine's report in 2002 highlighted racial disparities in health care.19 Although SES is an important predictor of health outcomes and is known to explain some of the racial differences in health outcomes, few studies on racial disparities in colorectal cancer survival have accounted for socioeconomic factors. Even though we expanded our search back to 1966, we identified only 10 studies that addressed disparities between African Americans and Caucasians with colorectal cancer while taking into consideration differences in SES. Among those studies, there were remarkable inconsistencies in terms of the study design, the variables measured, the outcomes compared, and, crucially, the results obtained. Of the 9 studies that investigated all-cause mortality, 3 studies showed statistically significantly increased HRs,29, 33, 37 4 studies showed statistically insignificant increased HRs,30, 31, 35, 36 and the remaining 2 studies indicated statistically insignificant negative associations in African Americans compared with Caucasians.32, 38 To obtain a more stable estimate of the correlation between race and colorectal cancer survival after accounting for SES, we generated the combined estimates in addition to show the results for individual studies. We observed that the pooled HR for all-cause mortality was only marginally higher in African Americans compared with Caucasians (lower confidence limit, 1.00), whereas the pooled HR for colon cancer-specific mortality was marginally higher in African Americans compared with Caucasians (lower confidence limit, 1.01). These findings, together with our numerous sensitivity analyses, demonstrated that there is no strong evidence of racial disparities in survival between African Americans and Caucasians with colon cancer after accounting for racial differences in SES.
These findings, however, may be limited by the lack of information in the original studies, including patients' personal health beliefs, lifestyle factors, colon cancer screening, and posttreatment surveillance in particular, because these factors are associated with ethnicity and survival.7, 8, 43–50 In addition, several other important factors associated with race and ethnicity, such as health care provider characteristics, were not considered in the analyses, even though studies have shown that provider and physician factors explain some racial variations in survival.21, 51 Therefore, it is likely that the marginal differences in colon cancer survival between African Americans and Caucasians after controlling for SES in the meta-analysis may be explained by some of those unmeasured factors.
It is well known that socioeconomic factors are associated with poorer health outcomes and increased mortality.7, 50–63 First, poor SES is a major barrier for adequate health care coverage and for receiving timely and optimal treatment, thus leading to poorer health outcomes, including survival. Second, individuals with poorer SES are more likely to work in jobs with greater occupational and environmental hazards, which are associated with increased mortality. Third, poor SES has been associated with lack of adequate housing, nutrition, education, knowledge of healthy lifestyle, and health promotion and behavior. Low SES also is associated with living in poorer neighborhoods and with higher stress. In addition, several studies have shown that individuals of lower social class are more likely to receive suboptimal treatment in health care settings19, 52 and are more likely to refuse or not to comply fully with the therapy provided.19 Because a greater proportions of African Americans are exposed to the above-described hazards that commonly are associated with poorer SES, race/ethnicity often has been used as a proxy for SES.43–46 Studies on racial disparities in health outcomes that did not control for socioeconomic factors often have led to conclusions that differences in outcomes between African Americans and other races are attributable to race/ethnicity, which is an immutable characteristic. However, socioeconomic factors that are mutable are likely to be mediators in the relation between race and outcomes.7, 57, 63–65
In the current study, we observed that racial differences in colon cancer survival can be attributed mainly to racial differences in SES and that the marginal remaining disadvantage for African Americans are likely to be explained by factors related to low SES. Differences in socioeconomic factors and treatments, which are 2 of the major barriers to achieving equal outcomes for men and women with cancer, can be modified through policies and clinical guidelines. Hence, efforts to provide optimal standards of care independent of race, to improve opportunities for education and employment, to increase health insurance coverage, and to eliminate discrimination based on race/ethnicity are likely to have important clinical and public health implications.
Despite its strength in increasing the study power to detect the differences in outcomes, the meta-analysis methodology has a number of limitations. For example, there is a possibility that some studies may not have been identified in our article search, particularly those studies that were published in languages other than English. In addition, methodologies and variable measurements from the studies differed greatly, making it difficult to control fully for these differences in the analysis. Because of these confounding and biases, there may be a danger that meta-analysis of these studies may generate precise but spurious results. Furthermore, meta-analysis must rely on the results, variables, and categories provided in each individual study. For example, different variables were used to represent SES, some of which were at the group level and some of which were at the individual level. Different socioeconomic indicators may not fit all study populations.66, 67 These different SES measures may have resulted in inconsistencies in the study findings. In addition, SES covers a wide range of factors, and not all factors were measured adequately and taken into consideration in the analysis. These inconsistencies and the incomplete capture of socioeconomic factors may have led to a significant level of residual confounding, and this may explain some marginal differences between African Americans and Caucasians. Moreover, individual level and group level socioeconomic factors are independent predictors of health outcomes. Analyses that did not consider both group level and individual level SES in the analysis likely will make the results vulnerable to substantial residual confounding.
In conclusion, the evidence on racial disparities in survival for colon cancer between African Americans and Caucasians was weak after adjusting for racial differences in socioeconomic factors and treatment. Because treatment and socioeconomic factors are modifiable, efforts to eliminate racial disparities in health care and to minimize disparities in SES have the potential to reduce racial inequalities in colon cancer survival.
- 1National Cancer Institute. Cancer of the Colon and Rectum. Available at URL: http://seer.cancer.gov/statfacts/html/colorect.html. Accessed September 7, 2006.
- 2American Cancer Society. Colorectal Cancer Facts and Figures. Special edition, 2005. Available at URL: http://www.cancer.org/downloads/STT/CAFF2005CR4PWSecured.pdf. Accessed September 7, 2006.
- 3SEER Cancer Statistics Review, 1975–2003. Bethesda, Md: National Cancer Institutel 2003. Available at URL: http://seer.cancer.gov/csr/1975_2003/. Accessed September 7, 2006., , , et al.
- 19Institute of Medicine. Unequal Treatment: Confronting Racial and Ethnic Disparities in Health Care. Washington, DC: National Academy Press; 2002.
- 43Stata Corporation. STATA Statistical Software, release 7.0. College Station, Tex: Stata Corporation; 2001.
- 44Meta-analysis in Stata. In: EggerM, Davey SmithG, AltmanD, eds. Systematic Reviews in Health Care: Meta-Analysis in Context.2nd ed. Blackwell BMJ Books; 2001: 347–369., , .
- 49sbe26: Assessing the influence of a single study in meta-analysis. Stata Tech Bull. 1999; 47: 15–17..
- 55Social Epidemiology. New York: Oxford University Press; 2000., .