- Top of page
- Materials and methods
The relationship between socioeconomic status (SES) and health/disease has been the focus of numerous investigations over the past 20–25 years. The basis for these investigations is often through the theoretical framework of social determinants of health [1, 2]. Marmot, Wilkinson, and House reported the statistically significant effect of income and education on survival in the general population [1-5]. Marmot's explanation for this relationship in affluent countries, such as the USA, is that the relationship between income and survival is the result of relative differences in social participation and opportunity to control life circumstances . Much of the literature focuses on income; however, Marmot argues that education may be a better indicator of factors linked to social position that are important to health and survival given that the effect of income on mortality is markedly reduced when education is included in predictive models .
In spite of the numerous published papers that have specifically examined the relationship between SES and cancer survival, questions persist because of inconsistent conclusions [7-32]. Possible explanations for this inconsistency include differences in the research question being asked (e.g., impact of SES on survival in the general population or impact in a clinical trial population), the patient population (e.g., homogeneous or heterogeneous histology or stage, as well as the national, racial, and ethnic composition), sample size or power considerations, the data source (e.g., census, regulatory, Surveillance, Epidemiology and End Results Program, clinical trial treatment trial, or patient reports), and the measure of SES (e.g., income, education, or occupation).
Such methodological differences are evident in research published in the past decade that has focused on the relationship between SES and survival among patients with breast cancer. Based upon education level reported to the French census, Menville reported that higher mortality observed among highly educated breast cancer patients had attenuated since the 1970s . In contrast, Bouchardy , Dalton , and Thomson  reported that patients with lower SES had poorer prognosis. Bouchardy used patient-reported occupation to define SES among her population-based Swiss cohort that primarily had early stage breast cancer. Dalton linked Denmark's registry of patients who received adjuvant protocol treatment with an administrative database that contained each patient's education level. Thomson derived SES from Scotland's census data. Unlike multivariate analyses reported by Bouchardy and Dalton, the effect of SES disappeared in Thomson's study after adjustment for prognostic factors. Albano used data from the National Center for Health Statistics to show that the relative risk of death from breast cancer is highest among patients with 12 or fewer years of education . Gordon used census-derived measures of education and income to show that low SES was associated with poorer survival among White, node-negative breast cancer clinical trial participants but not among African Americans [31, 32].
Nearly 20 years ago, the Cancer and Leukemia Group B (CALGB), a national cooperative group funded by the National Cancer Institute explored the relationship between SES and survival of cancer patients enrolled on eight CALGB studies . After adjustment for known prognostic factors including cancer type, performance status, age, and protocol-specific factors, analyses showed that clinical trial participants with low income or only a grade school education had poorer survival than patients with higher SES. This paper is part of a larger project to examine a large database of background forms that CALGB had routinely collected between 1990 and 1998, and to ‘validate’ findings of Cella  concerning the relationship of education and survival within larger, homogeneous cancer patient populations that participated in CALGB trials .
This paper focuses specifically upon the relationship between education and survival among patients with breast cancer who participated in one of 10 CALGB-coordinated clinical trials initiated between 1987 and 1998 (Table 1). Analyses assessed the effect of education on survival separately within early stage and metastatic patient subgroups. The strength of this research is that it uses a patient-reported measure of SES (education), explores the relationship between education and survival within one cancer type, and has power(i.e., sufficient patient numbers) to detect clinically important effect sizes within the unique context of clinical trials where initial therapeutic choices are not influenced by SES.
- Top of page
- Materials and methods
The analyses described in this paper are based upon the experiences of all patients accrued to one of 10 breast cancer studies that provided education data (Table 1). These 6166 patients constituted 74% of the enrolled patients. A comparison of the characteristics of patients included and not included in these analyses showed no significant difference relative to age (data not shown). Among patients who had the opportunity to provide education data, the racial composition of the group of patients who did and did not provide education data did not significantly differ.
The study cohort included 5146 early stage patients and 1020 metastatic patients. The majority were White (84%), and 52% were 50 years of age or older (Table 2). A greater proportion of metastatic patients were African American (17% vs 10%), 50 years of age or older (72% vs 48%), postmenopausal (82% vs 48%), or had less than a high school education (19% vs 11%) than that observed among early stage patients (p < 0.001 for all comparisons). African American patients were less likely to be a high school graduate than non-African Americans (72% vs 90%; p < 0.0001), as were older patients (78% vs 90%;p < 0.0001) and separated/widowed patients (72% vs 89%;p < 0.0001).
The relationship between education and survival is graphically summarized in Figures 1 and 2 for early stage and metastatic disease subgroups; associated statistics are provided in Table 2. Within the early stage subgroup, education had a statistically significant effect on survival (p < 0.0001) with poorest prognosis being among patients with less than a high school education in comparison with patients who completed high school (HR = 1.47; 95% CI: 1.29, 1.68). Among patients with metastatic disease, survival of patients who did and did not complete high school was not significantly different (p = 0.095; HR = 1.15; 95%CI: 0.98, 1.353).
In addition to education, the following individual factors were significantly associated with better survival among early stage patients: non-African American race, married or single women, premenopausal status, estrogen receptor (ER) positive, progesterone receptor (PR) positive, one to three positive nodes, and tumor diameter 2 cm or less. Among patients with metastatic cancer, factors associated with better survival included non-African American race, performance status (PS) = 0, ER positive, PR positive, no visceral involvement, and no bone involvement. These results are consistent with the literature concerning known prognostic factors for breast cancer.
Multivariable Cox regression analysis showed that after adjustment for known prognostic factors, early stage patients with less than a high school education were at greater risk of dying than patients who completed high school (Table 3; p = 0.0007; HR = 1.26). African American women were at greater risk of dying than non-African American women (p = 0.007; HR = 1.23), with no evidence that the effect differed among patients with and without a high school education (p = 0.453; model not shown).
Table 3. Multivariable Cox models predictive of survival
|Parameter||Hazard ratio||95% hazard confidence||Ratio limits||p-value|
|Model for early stage patients|
|Separated versus married/single||1.244||0.971||1.594||0.0838|
|Divorced versus married/single||1.178||1.031||1.347||0.0161|
|Widowed versus married/single||1.270||1.083||1.489||0.0032|
|Number of positive nodes|
|4–9 vs 0–3||1.625||1.450||1.821||<0.0001|
|10+ vs 0–3||2.741||2.374||3.165||<0.0001|
|Unknown vs 0–3||1.394||1.091||1.780||0.0078|
|Tumor diameter >2 cm||1.392||1.250||1.552||<0.0001|
|Education not high school graduate||1.259||1.102||1.439||0.0007|
|Model for metastatic patients|
|1 vs 0/unknown||1.289||1.121||1.482||0.0004|
|2 vs 0/unknown||1.424||1.245||1.629||<0.0001|
|Positive versus negative||0.627||0.534||0.735||<0.0001|
|Unknown versus negative||0.689||0.544||0.874||0.0021|
|Prior hormonal therapy||1.194||1.037||1.374||0.0134|
|Education not high school graduate||1.188||0.981||1.438||0.0779|
|Not high school graduate among African American||0.669||0.455||0.981||0.0397|
Among metastatic breast cancer patients, multivariable analysis showed that the effect of having less than a high school education varied across racial groups (p = 0.040; Table 3). The HR associated with having less than a high school education was 0.80 (95%CI: 0.57, 1.11) among African Americans and 1.19 (95%CI: 0.98, 1.44) among non-African Americans. Of greater magnitude was the statistically significant HR for race: 1.53 (95%CI: 1.24, 1.89). A model without the interaction between race and education shows that the African American main effect was statistically significant (p = 0.001; HR = 1.35; 95%CI: 1.13, 1.62) and the education main effect was not (p = 0.442; HR = 1.07; 95%CI: 0.90, 1.26; model not shown).
- Top of page
- Materials and methods
Analyses hypothesized a priori that clinical trial participation with standardized treatment plans and rigorous patient follow-up would initially negate any potential effect of social class, as measured by education, on the hazard of dying, and that after completion of ‘active’ protocol treatment, the effect of education would emerge. This hypothesis implied that education would not have an impact upon the survival of metastatic patients, given that such patients are intensively followed until treatment failure and death, and that education would have an effect among early stage patients as they are less rigorously followed long-term after protocol treatment termination.
The hypothesis was substantiated with multivariable analyses among early stage patients where the lack of a high school education and African American race were associated with a greater hazard of dying. Hazard ratios for race and education are of similar magnitude, with no evidence of an inconsistent effect of education across racial groups.
Within the metastatic patient population, the relationship between education and survival is complicated by a statistically significant interaction that suggests that the effect of education on survival varies across racial groups. Within each racial group, the effect of education is not statistically significant. Among African Americans, the lack of high school education is associated with better survival (HR = 0.8). However, among non-African Americans, the lack of a high school education is associated with poorer survival (HR = 1.18), an effect that is opposite to that seen among African Americans. It is not clear whether the statistically significant interaction is a false positive result and an artifact of the non-significant effects of education that are in opposite directions, or whether this result is an indication that African American women with less education are tied into services that might support them more than less educated non-African American women. Regardless, survival of African Americans is poorer than that of non-African Americans (HR = 1.53).
The fact that education had a statistically significant effect on survival among early stage patients that did not vary across racial groups is consistent with reports by Meara that there is a gap in life expectancy among women with low and high education regardless of race . In contrast, among patients with metastatic disease, education did not have an effect on survival within racial subgroups.
This study is unique in that it focuses on the effect of education within a setting where initial therapeutic decisions are not influenced by SES. SES, however, may have resulted in observed differences in baseline characteristics of the early stage and metastatic subgroups due to influences of SES on stage at diagnosis [53-57] and access to clinical trials.
Commonly used ‘area-wide measures’ of socioeconomic data based upon census or administrative databases are unreliable as they classify all patients within a heterogeneous community as having the same SES, as measured by education or income . Dale advocates use of SES measures obtained from each individual and argues that income and education data should be obtained with the recognition that other factors such as occupation might be needed to capture the full effect of SES . Furthermore, Dale recommends that investigations focusing on SES and cancer survival should have adequate sample sizes to make scientifically and statistically sound inferences and that the investigations focus on specific cancer sites. With these criteria as benchmarks, the study described in this paper is reasonably well designed to investigate the relationship between SES and cancer survival in that SES, as measured by education, is available on the individual patient level, the sample size is large enough to assure statistically sound inferences, and a relatively homogeneous population, that is, one cancer site, has been studied. The inclusion of patient-reported income would have strengthened the study; however, such data were purposely not collected as previous pilot work had indicated that a large percentage of patients would not provide such data .
Given that race has been the focus of much of the published clinical literature concerning the effect of SES on survival, race was included in analyses as a potential confounder of the effect of education. The increased risk of death among African American women as shown in analyses is consistent with previous reports [58-60].
Both Albain  and Polite  have wrestled with the source of the race effect, whether it is biologically based or a result of SES. The provocative paper by Albain reports no racial effect on survival among patients with acute myelogenous leukemia, limited small cell lung cancer, advanced stage non-small cell lung cancer, multiple myeloma, adjuvant colon cancer, and advanced stage non-Hodgkin's lymphoma and a statistically significant negative effect of being African American on survival among sex-specific cancers (i.e., early stage breast cancer, advanced stage ovarian cancer, and advanced stage prostate cancer). The CALGB has reported similar observations for lung cancer [34, 61, 62]. The results presented in this paper complement the report by Albain as it has shown that the mortality rate associated with metastatic breast cancer is greatest among African American women. Albain concludes that tumor biology and inherited host factors contribute to differential survival outcomes by race in sex-specific malignancies.
Both race and education have been found to be independent predictors of survival among early stage breast cancer patients treated on CALGB clinical trials. Among patients with metastatic disease, race also has a significant effect on survival; however, education appears to have an effect on survival that is inconsistent across racial groups. We conjecture that education is a surrogate for social status, whereas race is a surrogate for both biological/genetic and social factors. Additional research is needed to substantiate such a statement and to gain a better understanding of the relationship between race, education, stage, clinical trial participation, and survival. An integral part of this additional research needs to be an examination of sociocultural and behavioral factors that contribute to long-term breast cancer survivors with low SES having poorer prognosis.
Regardless of the underlying mechanism for the associations between education and survival, it is clear that post-trial survivorship plans need to focus on women with low social status, as measured by education. Issues that need consideration include long-term compliance with hormone administration, management of comorbidities, cancer prevention, and detection.
- Top of page
- Materials and methods
The research was supported, in part, by grants from the National Cancer Institute (CA31946) to the Cancer and Leukemia Group B (Richard L. Schilsky, MD, Chairman) and to the CALGB Statistical Center (Stephen George, PhD, CA33601). The authors were also supported by grants (Alice B. Kornblith, PhD, CA32291; Jimmie C. Holland, MD, CA77651; Electra D. Paskett, PhD, CA77658). The content of this manuscript is solely the responsibility of the authors and does not necessarily represent the official views of the National Cancer Institute.