Survival of Blacks and Whites After a Cancer Diagnosis FREE

In the United States, 5-year survival after a cancer diagnosis is poorer for blacks than for whites—from 1992 to 1997, the figures were 52% and 63%, respectively.¹ Large, well-designed epidemiologic studies have explored possible reasons for this disparity. Differences in factors such as stage at diagnosis, socioeconomic status, and health insurance coverage, though present, do not appear to be entirely responsible.^2- 4 For example, the National Cancer Institute sponsored a Black-White Cancer Survival Study, which compared survival for blacks and whites with cancer of the colon, breast, uterus, and bladder. The poorer survival rates that were observed for blacks with all 4 cancer types persisted after adjustment for both clinical and socioeconomic characteristics.^5- 8 Because these factors do not appear to explain the racial difference in survival, the theory that cancers afflicting blacks may be fundamentally more aggressive due to biological or genetic differences has gained prominence.^9- 19

Few studies have been designed to evaluate 2 other factors, unrelated to cancer biology, which may reduce the longevity of blacks with cancer: lesser quality treatment and greater mortality due to other illnesses. Because there are numerous studies documenting differences in the quality of cancer care received by blacks and whites,^20- 25 the thought that treatment differences may ultimately underlie the discrepancies in survival has gained many proponents.^26- 28 Moreover, several studies have documented similar survival for black and white patients with cancer who have received similar treatments.^29- 33 Rates of death because of cardiovascular disease, diabetes, and other major illnesses are also greater for blacks than whites, leading some investigators to suggest that differences in population mortality should be considered.^29,34- 37

We conducted this study to determine whether there was evidence in the literature of racial disparities in survival between blacks and whites who had received the same treatments for similar stages of cancer. We postulated that if blacks had poorer survival than whites in this context, the survival differences would be explained by the excess rates of deaths due to other causes in blacks. If not, we reasoned that the magnitude of the unexplained survival gap would constitute an estimate of that which may be attributable to differences in cancer biology.

We began with a systematic review of all studies in which survival was reported for blacks and whites who had received comparable cancer treatments for similar stages of disease, with the purpose of removing treatment received and stage at diagnosis as explanatory factors. For each pair of cohorts (ie, blacks and whites within a particular study), we converted the reported survival statistics into hazards (death rates) and then calculated the ratio of the hazards for blacks relative to whites. Using the statistical tools of meta-analysis, we then combined these hazard ratios (HRs) across all studies and within predetermined categories of studies.³⁸

Next, we used life-table methods to recalculate the hazard of death for each cohort after parsing out the hazard of death due to causes other than the cancer under study. This approach allowed us to estimate the cancer-specific hazards for each cohort, while removing the bias that results from the fact that blacks experience mortality from heart disease, diabetes, and other conditions at higher rates than whites. We then calculated the cancer-specific HRs for blacks relative to whites and generated parallel-summary statistics. Some studies reported cancer-specific death rates based on ascertained cause of death; these estimates were not incorporated in our analyses.

We found English-language studies that reported on overall survival for black and white patients treated similarly for cancer, not including studies of malignant melanoma, pediatric tumors, malignancies in persons with human immunodeficiency virus infection, or premalignant conditions such as polyps of the colon. We searched MEDLINE (1966 through the third week of January 2002) using the exploded Medical Subject Headings of the National Library of Medicine terms: neoplasms, cross-cultural comparisons, blacks, racial stocks, negroid race, minority groups, medically underserved area, disease-free survival, survival, survival analysis, survival rate, outcome assessment (health care), outcome and process assessment (health care), follow-up studies, treatment outcome, prognosis and the key words neoplasms, cancer, african americans, underserved populations, survival, mortality, treatment outcome. The abstracts (or titles for those listed prior to 1975) for these citations were then independently examined by 2 authors (P.B.B. and A.G.). The full text was retrieved if the abstract (1) mentioned both black and white patients; (2) made some comment regarding either similarity of treatment received or presented an analysis based on the treatment received; and (3) commented on survival. Agreement on these 3 criteria between the reviewers was 97%. Studies were then screened for potential overlapping cohorts. When encountered, we eliminated the smaller of the 2 overlapping cohorts; this occurred in 2 instances.^39- 42 Studies were then included in the analysis if they (1) included data for at least 10 black and 10 white patients; (2) specified what measures were undertaken to minimize loss to follow-up; (3) specified the cohort ascertainment method; (4) summarized survival of both blacks and whites using actuarial measures (eg, median survival, 5-year survival, Kaplan-Meier plot); (5) presented outcomes either within stage, adjusted for stage, or were based on cohorts with balanced-stage distributions; and (6) specified that blacks and whites in the study received similar treatment.

Our confidence in the extent to which treatments received by blacks and whites were comparable within each study was graded based on a hierarchy adapted from Roach and Alexander.⁴³ Grade 1 studies (the highest grade) analyzed black and white patients who were enrolled on a clinical trial or trials because participants in clinical trials are required to follow particular treatment protocols that are closely monitored. Grade 2 studies reported on the outcomes of black and white patients who had received a particular treatment (eg, all patients undergoing resection of stage I non–small cell lung cancer), allowing for within treatment group evaluation of survival. Grade 3 studies used statistical methods to control for treatment within a population of patients who received different treatments. The method could simply involve demonstration that similar proportions of blacks and whites in the study received the same treatments. Grade 4 studies presented outcomes for black and white patients in which the treatment was likely to be comparable (but was not observed to be so) because patients were all treated either within the same institution or within culturally similar institutions. Examples of the latter group include multiple sites within a single health maintenance organization or Veterans Affairs medical centers. We excluded multi-institutional studies in which such an obvious source of shared culture was not present and reports from state or multisite cancer registries in which treatment was not directly observed.

Two methods were required to calculate the hazard of death, each based on the assumption that survival-time distributions in the studies could be approximated by the exponential form S(t) = e^-λt (λ is the hazard; S(t) is the probability of surviving beyond time t).³⁸ In circumstances where the survival curves or statistics were comparable in terms of treatments received and stage of disease, we anchored the reported actuarial survival for each group at a landmark time point—5 years unless the authors identified alternatives as more representative—and solved for λ in the preceding equation.⁴⁴ If only the median (m) was reported in these situations, the hazard was calculated using λ = log(2)/m.

If the black and white cohorts were not comparable in terms of treatment, stage, or other relevant factors but a covariate-adjusted HR was reported, we derived the 2 hazards as follows. First, the hazard for whites was calculated using the reported actuarial estimate via the preceding equation. The hazard for blacks was then computed as the product of the covariate-adjusted HR × the hazard for whites.

For each cohort, from the overall hazard of death we subtracted the age, sex, and race appropriate hazard of death due to population mortality to determine the cancer-specific hazard of death. We determined population mortality rates from the National Center for Health Statistics 1997 decennial life tables, from which we subtracted death rates due to the cancer under study to avoid double counting.⁴⁵ The National Center for Health Statistics tables contain population estimates of annual death hazards (presented as probabilities) by race, sex, and age. The cancer hazards in the population were gleaned from the National Cancer Institute's 1994-1998 Surveillance, Epidemiology, and End Results (SEER) report, which classifies them similarly.⁴⁶ The hazard of population mortality for each racial group was taken to be the average of the hazards for each year of the study for a population matched to the race, sex distribution, and mean age of the cohort. When race-specific sex or age distributions were unavailable, we assumed that they were equal between the 2 groups.

For the majority of studies we determined the SE of the HR either based on the reported number of failures or by converting reported confidence intervals (CIs) using established methods.^38,44,47 In 23 (26%) of the cohorts, the data needed for these calculations were unavailable; therefore, we imputed an upper bound for the SE. This upper bound was calculated based on the assumption that exactly half of the total expected deaths at the landmark time point had contributed to the statistical power of the study, thus mirroring an analysis that would have been conducted if accrual were at a constant rate and the authors had analyzed the results when the first patient accrued had been followed up to the landmark (eg, the analysis occurred 5 years after the start of a study using a 5-year landmark). In reality, the results of most of these studies were likely analyzed at some later time point (after more failures) and so most estimates were probably associated with a smaller SE. Our assumption mitigates the influence of these cohorts on our overall pooled estimates, which is appropriate given our overall objectives. It was also assumed that the correction for other causes of death did not alter the SE of the log HR, because correction for population-based rates introduces virtually no additional random error.

For the meta-analyses, we used random effects models; our primary analysis was across all studies and our secondary analyses were by study grade and cancer type.⁴⁸ We assessed heterogeneity of the pooled HRs with the Q statistic.⁴⁸ We evaluated publication bias with a funnel plot and stratified and unstratified rank correlation tests.^49- 50 We conducted the meta-analyses using Stata version 7.0 (Statacorp, College Station, Tex). All P values were 2-sided and values less than .05 were considered significant.

Evaluation of the 891 citations yielded 157 full-text articles. A search of the reference lists of those articles yielded an additional 19 studies. From these articles, we identified 89 unique analyzable cohorts reported in 54 articles that met our inclusion criteria. In aggregate, these cohorts reported on the survival of 32 004 black patients (median, 108 patients per study) and 189 877 white patients (median, 467 patients per study) with cancer of 14 different organ sites (Table 1).

Across the 89 cohorts (Figure 1), blacks who received comparable treatment for similar stage cancer were at a 16% increased risk of death relative to whites (pooled HR, 1.16; 95% CI, 1.12-1.20; P<.001). After correction for differences in population mortality, the hazard of death was reduced to 1.07 (95% CI, 1.02-1.13; P = .01). The maximum influence on either of these estimates due to any particular cohort was 0.01, which we established by repeating our analyses removing 1 cohort at a time. These 2 results were robust across study grade, without evidence of heterogeneity.

In Figure 2, we present the results for overall survival and cancer-specific survival for particular cancer types, juxtaposed against the estimated proportion of cancer deaths due to each type in the black population.⁹⁷ For each of the 4 most common cancers (lung, colorectal, prostate, and breast, which collectively account for roughly half of all cancer deaths), blacks were at a significantly excess risk of death overall. However, after correction for population mortality, there was no statistical evidence of excess cancer-specific mortality in blacks with the first 3 cancer types; the excess mortality in blacks with breast cancer was only moderately reduced (HR, 1.22; 95% CI, 1.10-1.37; P<.001).

For each of the remaining cancer sites, the results are based on either single studies or aggregation of very few studies. For the majority, cancer-specific mortality appeared similar for the 2 racial groups. The exceptions were the doubling of risk for blacks with cancer of the uterine corpus (HR, 2.08; 95% CI, 1.34-3.21; P = .001) and the 70% increased risk of mortality for blacks with cancer of the bladder (HR, 1.68; 95% CI, 1.28-2.21; P<.001). The pooled cancer-specific HR for all cancers other than breast, uterine, and bladder was 1.02 (95% CI, 0.97-1.06; P = .49), without evidence of heterogeneity (Q, 8.21; P = .61).

We found no evidence of publication bias. We graphed the overall log HR for each cohort against its SE to form a funnel plot (Figure 3). In this figure, there is no systematic drift in the reported HR in relation to the SE of the study, constituting evidence against the hypothesis of publication bias. Statistical tests for detecting publication bias were also negative overall (P = .45), stratified by grade of study (P = .79), and stratified into studies of breast, uterine, or bladder cancer compared with other cancers (P = .40).

Disparities in treatment, rates of death due to other causes, and biological behavior of the cancer itself might each account for discrepancies in survival between blacks and whites with similar-stage cancer. We identified studies relevant to the bulk of cancer types and found that across these studies blacks were at a 16% increased risk of death. We then corrected this estimate for differences in underlying death rates and found that the pooled estimate of excess cancer-specific mortality for blacks was only 7%.

The excess cancer-specific mortality that was observed appeared to be due to a modest increase in risk of death for blacks with breast cancer (9% of cancer deaths in blacks), a doubling of risk for blacks with uterine cancer (2%), and a 70% increase in risk for blacks with bladder cancer (1%). In contrast, there were no appreciable differences in cancer-specific survival in lung cancer (26%), colorectal cancer (11%), prostate cancer (9%), and 8 other cancers assessed in this study.

These findings should be considered in the context of the data on which they are based. Our search did not yield information on all types of cancer nor did our rules for categorizing treatment ensure that the cohorts received care of similar quality. In determining cancer severity, cancer stage either within strata or based on covariate adjustment was available in all studies but further detail on clinically important prognostic factors was absent. We also lacked information on either the mode of cancer detection (eg, screen detected vs symptom detected) or the types of staging evaluations undergone by these cohorts. In general, disparities between blacks and whites in these latter parameters tend to increase the survival of whites relative to blacks—the first through lead time and length time biases,⁹⁸ the second through stage migration.⁹⁹ Conceivably, the survival differences that were observed between blacks and whites with breast cancer (Figure 2), for example, reflected the impact of cancer being detected more often through screening mammography in whites (and thus earlier in its natural history) than in blacks.^100- 102

There are quantitative limitations as well. We modelled survival using mathematical models that likely do not mirror the observed relations precisely. We corrected for population mortality based on rates of death matched for age, sex, and race from the National Center for Health Statistics and the National Cancer Institute's 1994-1998 Surveillance, Epidemiology, and End Results; in fact, in these cohorts the actual rates of death due to other causes may be different. In studies set in relatively homogenous populations such as Veterans Affairs medical centers, the rates for blacks and whites may be more similar than they are in the United States as a whole. A parallel concern is that particular groups of patients with cancer, such as those with lung cancer, have a greater comorbid disease burden than the average population. Although this phenomenon will increase rates of death due to causes other than cancer, it may attenuate the differences in underlying death rates between blacks and whites.¹⁰³

Our methods, therefore, may have introduced measurement inaccuracies such that our reported CIs and accompanying P values do not capture the full degree of statistical uncertainty. The finding that differences in rates of death due to other causes reduced the HR for blacks relative to whites from 1.16 to 1.07 and eliminated the excess risk of mortality for blacks with colorectal, lung, and prostate cancer should be viewed in this context.

Finally, whether differences in death rates due to causes other than cancer constitute evidence of biological differences between blacks and whites was not addressed in this study. The death rates that were used in our calculations were derived from population statistics, which reflect the aggregate impact of differences in prevalence, severity, access to care, and, if present, biological differences. That there has been abundant documentation of disparities in care between blacks and whites for conditions such as heart disease and end-stage renal disease provides evidence that differences in treatment may be at least partially responsible for these disparities as well.¹⁰⁴ Similarly, this study focuses on subjects with a cancer diagnosis and therefore provides no insight into the extent to which biological differences may play a role in explaining racial differences in incidence rates, such as are seen in prostate cancer, gastric cancer, and uterine cancer.

Many researchers and physicians have concluded that poorer survival of blacks relative to whites after a cancer diagnosis reflects fundamental differences in the biology of the host or the attendant cancer or both. We did not observe the impact of these putative biological differences consistently in cohorts of comparably treated black and white patients with cancer of similar stage once we took into account differences in underlying death rates. We cannot be sure if our findings in breast cancer, uterine cancer, and bladder cancer constitute exceptions to this conclusion or reflect residual differences in treatment and disease severity that could not be identified through our study.

These findings suggest that if biological differences do exist, they are responsible for at most a very small fraction of all cancer deaths. Therefore, biological differences between blacks and whites cannot explain a meaningful share of the racial disparity in cancer survival observed in the United States. Numerous studies have demonstrated that blacks are less likely to receive optimal care for cancer than whites and are also more likely to be diagnosed initially at an advanced stage of disease. Finding workable remedies for these latter inequalities, as well as addressing the control of comorbid diseases, are important public health goals and should be an objective of future research.