Fax: (313) 831-7806
Ethnic differences in survival among women with ovarian carcinoma
Article first published online: 15 MAR 2002
Copyright © 2002 American Cancer Society
Volume 94, Issue 6, pages 1886–1893, 15 March 2002
How to Cite
Barnholtz-Sloan, J. S., Tainsky, M. A., Abrams, J., Severson, R. K., Qureshi, F., Jacques, S. M., Levin, N. and Schwartz, A. G. (2002), Ethnic differences in survival among women with ovarian carcinoma. Cancer, 94: 1886–1893. doi: 10.1002/cncr.10415
- Issue published online: 15 MAR 2002
- Article first published online: 15 MAR 2002
- Manuscript Accepted: 2 NOV 2001
- Manuscript Revised: 1 NOV 2001
- Manuscript Received: 21 MAY 2001
- National Institutes of Health. Grant Number: N01-PC-65064
- ethnic differences;
- ovarian carcinoma;
- survival analysis;
- prognostic variables
Ovarian carcinoma is the leading cause of death among all female reproductive malignancies. There are substantial differences in age-adjusted incidence rates and survival rates between Caucasian women and African-American women. The objective of this study was to examine ethnic differences in survival after ovarian carcinoma in a population-based sample of women.
Thirteen thousand eighty-three patients (12285 Caucasian women and 798 African-American women) who were diagnosed with primary ovarian carcinoma from the population-based Surveillance, Epidemiology, and End Results (SEER) Program were used for analysis. Odds ratios were used to estimate the association between prognostic variables and ethnicity. Chi-square tests were used to determine the statistical significance of these associations (using two-sided P values). Univariable and multivariable Cox proportional hazards models were used to assess survival differences.
African-American women were significantly younger at the time of diagnosis, were more likely to be single, and were less likely to undergo site specific surgery compared with Caucasian women. In addition, the crude median survival for African-American women was nearly 1 year less than for Caucasian women (22 months vs. 32 months, respectively; P < 0.0001). African-American women were at a 30% increased risk of death from any cause when adjusting for all other prognostic variables that differed between the two ethnic groups.
African-American women who are diagnosed with ovarian carcinoma are at a significant increased risk of death from any cause compared with Caucasian women who are diagnosed with ovarian carcinoma. Cancer 2002;94:1886–93. © 2002 American Cancer Society.
More women die from ovarian carcinoma than from any other type of female reproductive malignancy. An estimated 23,100 new cases of ovarian carcinoma were diagnosed, and 14,000 deaths occurred from ovarian carcinoma in the United States in the year 2000.1 The 5-year overall survival rate for women with ovarian carcinoma is approximately 40%.2 Survival is related most directly to stage of disease at the time of diagnosis: The 5-year survival rate is 89% for women with Stage I disease, 57% for women with Stage II disease, 24% for women with Stage III disease, and 11% for women with Stage IV disease.3 Survival also varies with age at diagnosis; younger women (age < 45 years) have a 5-year survival rate of 70%, whereas women in the oldest age group (age > 75 years) have a 5-year survival rate of 20%.4 There also are substantial differences in age-adjusted ovarian carcinoma incidence rates between Caucasian women and African-American women in the United States, with Caucasian women at higher risk for the development of disease. Survival, however, is poorer among African-American women.5
Higher parity, oral contraceptive use, and breast feeding have been identified consistently as protective against ovarian carcinoma.6–8 In fact, it has been shown that these protective factors are similar for Caucasian women and African-American women.9 Less is known about predictors of survival. Older women have much poorer survival from ovarian carcinoma compared with younger women.10 To date, no one has used statistical modeling to study the relationship between ethnicity and survival from ovarian carcinoma between Caucasian women and African-American women, although other studies have examined ethnic differences in incidence and mortality rates.9, 11, 12 The objective of this study was to examine the effect of ethnicity on ovarian carcinoma survival using data from the population-based Surveillance, Epidemiology, and End Results (SEER) Program.
MATERIALS AND METHODS
The patients included 12,285 Caucasian women and 798 African-American women who were diagnosed with primary, malignant ovarian carcinoma between 1988 and 1997 whose information was attained from the population-based SEER Program data base for the nine standard SEER sites. Patients with tumors of borderline malignancy were excluded. SEER is a national cancer surveillance program that collects information from all incident cancer cases diagnosed in residents of nine standard geographic areas in the United States: Metropolitan Atlanta, Metropolitan Detroit, the San Francisco/Oakland Standard Metropolitan Statistical Area (SMSA), the Seattle/Puget Sound SMSA, and the states of Connecticut, New Mexico, Hawaii, Utah, and Iowa. Cases have been collected from these main SEER sites since 1973 (1974 for Seattle/Puget Sound and 1975 for Atlanta). All 11 SEER registries combined, 9 standard and 2 supplemental, represent approximately 14% of the United States population and are representative of national demographic characteristics (SEER website; www-seer.ims.nci.nih.gov).
Table 1 shows the distribution by ethnicity of the patients from the nine standard SEER registries used in this study. In addition to collecting demographic, treatment, and clinical information at the time of diagnosis, patients are followed annually for survival status.
First, differences in prognostic factors between the ethnic groups were estimated using odds ratios. Chi-square tests were used to assess statistical significance, generating two-sided P values (Table 2). Prognostic variables included age at diagnosis, tumor histology, tumor stage (American Joint Committee on Cancer [AJCC] staging), site specific surgery, lymph node involvement, grade of tumor, SEER registry site, and marital status at diagnosis. Although data have been collected since 1973 for most of the nine standard SEER sites, information on tumor staging prior to 1988 was classified into the following broad categories: in situ, localized, regional, and distant. Beginning in 1988, the more detailed AJCC TNM classification system has been used; thus, only women diagnosed since 1988 were included. The AJCC system takes into account information on the extent of disease variables: tumor size and tumor extension (T), regional lymph node involvement (N), and presence of distant metastasis (M). Ethnicity was categorized as Caucasian or African American. All other ethnicities had insufficient numbers for analysis and were excluded from this study. Patient age at diagnosis was categorized in 10-year age groups from age < 40 years to age ≥ 80 years and was also used as a continuous variable. Women who were single, widowed, or divorced at the time of diagnosis were considered unmarried.
|Variable of interest||African-American (%)||Caucasian (%)||Odds ratio (95%CI)||P valuea|
|No. of patients||798||12,285||—||—|
|Age category (yrs)|
|< 40||82 (10)||868 (7)||1.0||0.002|
|40–49||102 (13)||1629 (13)||0.7 (0.5–0.9)||—|
|50–59||145 (18)||2226 (18)||0.7 (0.5–0.9)||—|
|60–69||216 (27)||3121 (25)||0.7 (0.6–1.0)||—|
|70–79||180 (23)||2969 (24)||0.6 (0.5–0.8)||—|
|≥ 80||73 (9)||1472 (12)||0.5 (0.4–0.7)||—|
|Marital status at diagnosis|
|Married||253 (32)||6558 (53)||1.0||0.001|
|Single||505 (63)||5426 (44)||2.4 (2.1–2.8)||—|
|Missing||40 (5)||301 (3)||—||—|
|I||103 (13)||2060 (17)||1.0||0.04|
|II||32 (4)||619 (5)||1.0 (0.7–1.6)||—|
|III||126 (16)||1855 (15)||1.4 (1.0–1.8)||—|
|IV||351 (44)||4563 (37)||1.5 (1.2–1.9)||—|
|Unstaged||186 (23)||3188 (26)||1.2 (0.9–1.5)||—|
|Site specific surgery|
|Surgery||619 (78)||10232 (83)||1.0||0.001|
|No surgery||179 (22)||2041 (16)||1.4 (1.2–1.7)||—|
|Missing||0 (0)||12 (< 1)||—|
|Endometrioid||78 (10)||1440 (12)||1.0||0.10|
|Clear cell||31 (4)||579 (5)||1.0 (0.6–1.5)||—|
|Serous||506 (63)||7871 (64)||1.2 (0.9–1.5)||—|
|Mucinous||105 (13)||1299 (11)||1.5 (1.1–2.0)||—|
|Carcinosarcoma||27 (3)||346 (3)||1.4 (0.9–2.3)||—|
|Carcinoma, NOS||51 (6)||750 (6)||1.3 (0.9–1.8)||—|
|1||76 (10)||1060 (9)||1.0||0.99|
|2||96 (12)||2146 (17)||0.6 (0.5–0.9)||—|
|3 and 4||322 (40)||5289 (43)||0.8 (0.7–1.1)||—|
|Missing||304 (38)||3790 (31)||—||—|
|Lymph node involvement|
|None||265 (33)||4699 (38)||1.0||< 0.001|
|Regional||113 (14)||1311 (11)||1.5 (1.2–1.9)||—|
|Distant||36 (5)||350 (3)||1.8 (1.2–2.6)||—|
|NOS||2 (< 1)||31 (< 1)||1.1 (0.1–4.5)||—|
|Missing||382 (48)||5894 (48)||—||—|
|SEER registry site|
|Hawaii||2 (< 1)||184 (2)||1.0||< 0.001|
|San Francisco/Oakland||162 (20)||1752 (14)||8.5 (2.3–71.4)||—|
|Connecticut||58 (7)||2064 (17)||2.6 (0.7–22.0)||—|
|Metro Detroit||338 (42)||2046 (17)||15.2 (4.1–127.0)||—|
|Iowa||8 (1)||1900 (15)||0.4 (0.1–3.8)||—|
|New Mexico||9 (1)||691 (6)||1.2 (0.2–11.5)||—|
|Seattle/Puget Sound||38 (5)||2024 (16)||1.7 (0.4–14.9)||—|
|Utah||4 (< 1)||743 (6)||0.5 (0.1–5.5)||—|
|Metro Atlanta||179 (22)||881 (7)||18.7 (5.0–156.7)||—|
In the SEER data base, the histologic type of tumor is coded using International Classification of Diseases for Oncology codes. For the purposes of this study, the World Health Organization histologic classification of ovarian tumors was used, and only surface epithelial-stromal tumors (carcinomas and carcinosarcomas) were considered, because there were insufficient numbers of nonsurface epithelial-stromal tumors. Many of the carcinomas could not be classified into a specific subtype, so they were classified together as carcinoma, not otherwise specified (NOS). The specific subtypes of surface epithelial-stromal tumors evaluated were serous, mucinous, endometrioid, clear cell, and carcinosarcoma.
Survival for all analyses was defined as the time from diagnosis until death from any cause. Patients with missing survival information were excluded from the survival analysis (n = 31 missing patients; n = 13,052 total surviving patients). Kaplan–Meier methods were used to estimate crude median survival by ethnicity, and the log-rank test was used to assess the statistical significance of the difference between these two groups (Fig. 1).13 Univariable Cox proportional hazards models14 were used to estimate hazard ratios (HR) and to assess the prognostic importance of each factor that differed significantly by ethnicity (using indicators for all variables) (Table 3) —age at diagnosis (using 10-year age categories), marital status at diagnosis, AJCC stage of disease, site specific surgery, histologic type, SEER registry site, and lymph node involvement—to calculate ethnicity-adjusted HRs and their 95% confidence intervals (95%CIs).
|Variable of interest||HR||95% CI|
|Age at diagnosis (yrs)|
|Marital status at diagnosis|
|Site specific surgery|
|Lymph node involvement|
|SEER registry site|
Multivariable Cox proportional hazards models were used to calculate HRs and their 95%CIs for the models involving multiple factors and their interactions to assess the importance of ethnicity as an independent predictor of survival after adjusting for the other factors that vary by ethnicity (Table 4). All data were analyzed using SAS (version 6.12; SAS, Inc., Cary, NC) and Splus (professional version 4.5; Insightful Corp., Seattle, WA) statistical software packages.
|Variable of interest||HR||95%CI|
African-American women had significantly shorter crude median survival compared with Caucasian women (22 months vs. 32 months, respectively; P < 0.0001) (Fig. 1). African-American women were younger than Caucasian women at the time of diagnosis (61 years vs. 63 years, respectively; P = 0.0008). The frequency within each category of the prognostic variables of interest is shown along with odds ratios for the ethnic differences in Table 2. Histologic tumor type and tumor grade did not vary significantly by ethnicity. Table 3 shows the HR estimates and 95%CIs after adjustment for ethnicity for the univariable Cox proportional hazards models for the variables of interest that differed between the ethnic groups (i.e., excluding histologic tumor type and tumor grade). These univariable models show that each of these variables, except for the SEER registry site, is an important predictor of survival after adjusting for ethnicity.
The odds that an African-American woman would be older were 30–50% less than the odds that a Caucasian woman would be older as age at diagnosis increased from 40 years to ≥ 80 years (P = 0.002). The univariable models for age, after adjusting for ethnicity, showed that, as women got older, they had a significantly increased risk of death from any cause compared with younger women. The HR estimates increased from 1.8 (95% CI, 1.5–2.1) to 9.9 (95% CI, 8.5–11.4) as age at diagnosis increased from 40–49 years to ≥ 80 years compared with women who were diagnosed at age < 40 years.
The odds that an African-American woman would be single were 2.4 times greater than the odds that a Caucasian woman would be single, with a higher proportion of unmarried African-American women (63%) compared with 44% of Caucasian women (P = 0.0001). Unmarried women have a 50% increased risk of death from any cause compared with married women (HR, 1.5; 95% CI, 1.4–1.5).
The distribution of AJCC staging differed between the two ethnic groups (P = 0.001). The proportion of women with unstaged disease was 26% for Caucasian women compared with 23% for African-American women. There was no difference in the odds that a woman with Stage II, Stage III, or unstaged disease would be African-American compared with the odds that she would be Caucasian. The odds that an African-American woman would have Stage IV disease were 50% greater compared with Caucasian women. Women with a more advanced stage of disease had a higher risk of death from any cause compared with women who had Stage I disease. The HR estimates increased from 2.4 (95% CI, 2.0–2.8) to 10.5 (95% CI, 9.3–11.8) as the stage of disease became more severe (Stage II to Stage IV disease compared with Stage I disease). Women who had unstaged disease had an HR of 6.8 (95% CI, 6.0–7.7) compared with women who had Stage I disease, most likely due to the fact that this group of women is highly heterogeneous in terms of their other prognostic factors for survival.
Fewer African-American women than Caucasian women underwent surgery (P = 0.001). The odds that an African-American woman would not undergo surgery were 40% greater than the odds that a Caucasian woman would not undergo surgery. In both ethnic groups, there were more women with Stage IV disease or unstaged disease that did not undergo surgery. In addition, the distribution of surgery by stage was similar between the two ethnic groups (results not shown). Women who did not undergo surgery had a significantly increased risk of death from any cause compared with women who underwent surgery (HR, 4.5; 95% CI, 4.2–4.7).
Higher proportions of African-American women had regional and distant lymph node involvement compared with Caucasian women (P = 0.001). The odds that an African-American woman would have regional lymph node involvement or distant lymph node involvement were increased by 50% or 80%, respectively, compared with Caucasian women. Women with regional lymph node involvement (HR, 1.2; 95% CI, 1.1–1.3), distant lymph node involvement (HR, 1.7; 95% CI, 1.5–1.9), or NOS lymph node involvement (HR, 2.7; 95% CI, 1.8–3.9) were at an increased risk of death from any cause compared with women with no lymph node involvement.
The highest proportions of African-American women came from the Metro Detroit (42%), Metro Atlanta (22%), and San Francisco/Oakland (20%) registries (Tables 1 and 2). Although there were significant differences by ethnicity within the registry sites (P < 0.0001), women from all sites, compared with the women from Hawaii, were not at any increased risk for death from any cause after they were diagnosed with ovarian carcinoma (i.e., the 95% CIs of all HRs in Table 3 for SEER registry site included 1.0).
Histologic Grade 3 and 4 tumors were combined, because there were relatively few women with Grade 4 tumors and because survival was essentially the same in the two categories. The distribution of grade did not differ between the ethnic groups (P = 0.99). The odds that an African-American woman would have a Grade 2 tumor or a Grade 3–4 tumor were reduced by 40% and 20%, respectively, compared with Caucasian women. Because grade did not differ by ethnicity, univariable Cox models were not fit for this variable.
The proportions of histologic tumor types did not vary significantly by ethnicity (P = 0.10). The majority of tumors from the women in either ethnic group were of the serous histologic type followed by the mucinous and endometrioid histologic types (Table 2). The odds that an African-American woman would have any of the histologic tumor types was not increased or decreased significantly compared with Caucasian women (i.e., all 95%CIs included 1.0); therefore, the univariable Cox models also were not fit for this variable.
The univariable model for ethnicity showed that African-American women who were diagnosed with ovarian carcinoma were at a 30% increased risk of death from any cause (crude HR for ethnicity, 1.3; 95% CI, 1.2–1.4) compared with Caucasian women who were diagnosed with ovarian carcinoma. Taken together, the results show that there were differences between the ethnic groups for each of these variables and that, the older the woman and the more advanced the disease (i.e., higher stage and more lymph node involvement), the more pronounced the difference between the ethnicities.
The final prognostic survival model, as shown in Table 4, included ethnicity, age at diagnosis (using 10-year age categories), marital status at diagnosis, AJCC stage of disease, and site specific surgery. The histologic type of tumor and tumor grade were not used because neither varied by ethnicity, and information on tumor grade was missing for approximately 33% of patients. The lymph node involvement variable is used in the staging of patients, so it also was not included in the model. SEER registry site was not used in the model, because it was not an important predictor of survival in the univariable Cox models after adjustment for ethnicity. There was no difference in the HR estimates for the model with age as a categoric variable and with age as a continuous variable (continuous model not shown); hence, the categoric variable was used. The models showed that, after adjusting for all variables that showed significant differences between the two ethnic groups and that were important predictors of survival, African-American women, after they were diagnosed with ovarian carcinoma, were at a 30% increased risk of death from any cause compared with Caucasian women who had the same diagnosis.
Interactions between age and marital status, age and stage, age and ethnicity, ethnicity and marital status, ethnicity and stage, and ethnicity and surgery were tested in the multivariable Cox models. The only interaction that had any significant components was age and stage of disease. Only 3 of 20 interactions that were fitted were significant, and, after applying a Bonferroni correction to the P values for multiple comparisons (which would make the significance level P = 0.003), only 1 interaction remained significant: women age 80+ and Stage IV disease. The global likelihood ratio test showed no difference between the interaction model and the model with no interaction terms (−2LOGL = 124549.08 vs. −2LOGL = 124623.11; P > 0.05); thus, the interaction between age and stage also was not significant.
African-American women who are diagnosed with ovarian carcinoma have a 30% increased risk of dying from any cause compared with Caucasian women who are diagnosed with ovarian carcinoma, before and after adjusting for age at diagnosis (10-year categories), marital status, AJCC stage of disease, and site specific surgery. Because the estimated risk of death for African-American women compared with Caucasian women did not change after adjustment for the variables that were shown to differ by ethnicity, additional variables should be investigated that are not available in the SEER data base, such as socioeconomic status (SES), course of treatment, and accessibility to care, to help explain the ethnic difference in survival.
These data support the notion that African-American women have poorer survival compared with Caucasian women.5 There is a more pronounced ethnic difference for older women with more advanced disease. When models were fit for each stage separately, after adjusting for all important prognostic variables, only women with Stage IV disease and unstaged disease had a significant increased risk of death from any cause of 30% and 30%, respectively. It must be noted that we had a total of > 13,000 patients in this study, and that allowed us enough power to detect very small, statistically significant differences between the ethnic groups.
Our data support previous studies that have shown that survival is related most directly to stage of disease and age at diagnosis.3, 4 Because older African-American women with more advanced disease showed the highest increase risk for death from any cause after they were diagnosed with ovarian carcinoma, we speculate that this may be attributable to a difference in treatment and care and/or social support. These data show that more Caucasian women underwent surgery, which is the most efficient treatment for patients with ovarian carcinoma, and more Caucasian women were married. Women who are married typically have a better familial/social support system and possibly better access to care and treatment options. It is also possible that the women in these two ethnic groups had different sized tumors with different histologic types. For women with Stage III disease, there was a significant difference (P = 0.03) in the histologic type of tumor between African-American women and Caucasian women, with more African-American women having clear cell and carcinosarcoma tumors. For women with Stage IV tumors (P = 0.004) and unstaged tumors (P = 0.01), there was a significant difference in the size of tumors between African-American women and Caucasian women, with more African-American women having larger sized tumors (> 100 mm), although it must be noted that information on tumor size was missing for 69% of the total study population. Notable differences between the ethnicities in Stage III tumors and in the size of tumor for women with Stage IV and unstaged tumors may explain the small yet significant increased risk of death for African-American women. Had there been significant differences between the two groups for all age categories, all tumor sizes, and all histologic types in all stages, we would have expected a greater increased risk of death from any cause for African-American women compared with Caucasian women.
This study has many strengths. It is well known that the SEER data are of very high quality. The information collected on each individual is monitored meticulously for its accuracy. Because this study was population based, generalizability of the results to the greater United States population can be made. We had a very large sample size, giving us enough power to detect even the smallest of increased risks of death after diagnosis with ovarian carcinoma. We constructed models using only 50% of the total data (≈7500 patients) at a time and 25% of the total data (≈4000 patients) at a time. The factors that were significant in all of these models were the same as the significant factors in the models that included all 13083 observations. The estimates changed slightly but still were highly significant in all models.
Possible reasons why the risk of death for African-American women compared with Caucasian women did not change from the crude, unadjusted model to the adjusted model are 1) it was found that disease stage violated the proportional hazards assumption of the Cox model, and 2) there may have been a difference in the length of follow-up between the ethnic groups. To examine the proportional hazards assumption for disease stage, we fit a stratified model by stage and found that the results were almost identical to the results from an unstratified model. To assess differences in the length of follow-up, we estimated follow-up among women who were alive at the end of the study period. Figure 2 shows the results of this analysis, indicating that, until approximately 3 years after diagnosis (36 months), follow-up is the same regardless of ethnicity. The median follow-up did differ for African-American women compared with Caucasian women (54 months vs. 58 months, respectively; P = 0.05). We believe that the 30% increased risk of death from any cause for African-American women compared with Caucasian women, after they are diagnosed with ovarian carcinoma, is real, even with the slight differences in the length of follow-up.
There also were limitations to this study. We did not have SES information or insurance information (accessibility to care) on our patients, because SEER does not collect these variables. Although SEER does collect zip code of residence for all patients; we did not use zip code information to impute SES, because such imputation is likely to result in nondifferential misclassification and is more likely to misclassify minority populations.15 We realize that both SES and insurance coverage are possible confounders for the modeling of survival after diagnosis with ovarian carcinoma. Marital status may be used as a surrogate for insurance coverage, because women who are married tend to have better insurance coverage and a better social support system in place.16–18 Although we had information on chemotherapy and cause of death, these kinds of variables preclude a meaningful interpretation. In the SEER data base, both of these variables are subject to misclassification bias. Cause of death information is gathered from death certificates, which are known to be incorrect in many cases, and general treatment information is gathered only from hospital records. Thus, if a woman had chemotherapy treatment anywhere but at a participating hospital, the SEER registry would not have that information.
In summary, we conclude that, compared with Caucasian women, African-American women are at a 30% increased risk of death from any cause after they are diagnosed with ovarian carcinoma, after adjustment for all variables that differ significantly between these two ethnic groups. The crude HR was identical to the HR that resulted after adjusting for the prognostic factors. However, further work needs to be done to understand all of the factors associated with the observed differences in survival and focusing on other factors that are not included in the SEER data base.
- 1American Cancer Society. Cancer facts and figures. Atlanta: American Cancer Society, 2000.
- 5RiesLAG, EisnerMP, KosaryCL, et al., editors. EER cancer statistics review, 1973–1997. Bethesda: National Cancer Institute, 2000.
- 14Regression models and life-tables. J R Stat Soc B. 1972; 30: 248–275..