Race, Nutritional Status, and Survival From Breast Cancer Ralph J. Coates,* W. Scott Clark, I William Eley, Raymond S. Greenberg, Charles M. Huguley, Jr., Robert L. Brown
Numerous investigations examining national (1-3) and regional (4,5) US populations, as well as hospital case series (6), have found that white women with breast cancer have a longer survival time than do black women with breast cancer. Information collected from the National Cancer Instituted Surveillance, Epidemiology, and End Results (SEER) Program indicated that for the period 1979-1984, the 5-year relative survival rates in patients with breast cancer were 75.0% for white women and 62.2% for black women (3). This racial difference in survival was found among women aged 50 years and younger as well as among older women and persisted at roughly the same level from 1974-1976 through 1979-1984 (3). This racial difference in survival rates of patients with breast cancer has been attributed to many different factors. Black women are more likely to be diagnosed with cancer in a later stage (J). In addition, within specific stage classifications, white women survive longer than do black women (3), and racial differences in survival have been found to persist after analytic adjustment for stage (4,7,8). Some authors (5,9) have found most or all of the difference in survival rates to be eliminated when data are adjusted for socioeconomic status, but others (4,8) have found that substantial differences in survival rates remain. A number of other factors associated with poor prognosis have been found to be more prevalent among black patients than white patients. These include delays in diagnosis (10-12), larger tumors (6), less differentiated tumors (6,13), clinical symptoms at presentation indicating more severe disease (6,13), estrogen receptor-negative tumors (13-17), larger number of positive nodes (13), unfavor1684
Patients and Methods Women with breast cancer that was newly diagnosed between June 1975 and February 1979 were identified by a review of hospital tumor registry records at 14 private and public hospitals in Georgia. Of 2,641 identified subjects, 240 (9.1%) were excluded on the advice of their personal physicians, and an additional 309 (11.7%) declined to participate. After informed consent was obtained from the patients, trained oncology nurses under physician supervision collected interview and medical-record information and entered these data onto standard, precoded forms. Nurses interviewed patients in the hospitals to collect information on demographics, family history of cancer, medical history, and history of breast cancer detection, as well as on symptoms and their duration. Information concerning diagnosis, treatment, reha-
Received January 2, 1990; revised July 18, 1990; accepted August 22, 1990. Supported in part by Public Health Service contracts N01-CN55131 and N01CN55429 from the National Cancer Institute, National Institutes of Health, Department of Health and Human Services. R. J. Coates, W. S. Clark, R. S. Greenberg (Department of Epidemiology and Biostatistics), J. W. Eley (Winship Cancer Center), C M. Huguley, Jr. (Department of Medicine), R. L. Brown (Department of Surgery), Emory Unhwsity, Atlanta, Ga. We thank the patients, physicians, and nurse coordinators who participated in this study. 'Correspondence to: Ralph J. Coates, Ph.D., Department of Epidemiology and Biostatistics, Emory University, 1599 Clifton Rd., N.E., Atlanta, GA 30329.
Journal of the National Cancer Institute
Downloaded from http://jnci.oxfordjournals.org/ at Université Laval on March 3, 2015
The effects of nutritional status on differences in the survival of black and white women with breast cancer were studied in a cohort of 1,960 Georgia women diagnosed during 1975-1979. After data were adjusted for stage of disease, socioeconomic status, and other prognostic factors, poorer survival rates were shown in black women. Within each stage classification, lower levels of serum albumin and hemoglobin and higher relative body weight were more common among blacks and were independently associated with poorer survival. Among women with stage 3 disease, adjustment for these variables substantially reduced the excess mortality rate among blacks, suggesting that racial differences in survival may be partly explained by differences in nutritional status or extent of disease within stage. [J Natl Cancer Inst 82:1684-1692,1990]
able histologic type (13), less complete or aggressive treatment (14,18), and obesity (6). Various indicators of dietary or nutritional status have been found to be associated with survival of patients with breast cancer (6,19-30). These have included body size (6,19,21, 27-29); weight loss prior to treatment (22); dietary fat consumption (25,30); serum levels of cholesterol (23), albumin (20,26), and hemoglobin (20,26); and a number of hematologic measures, such as lymphocyte counts (20,26), which may be affected by nutritional status. Several of these and other indicators of nutritional status, including triceps skinfolds and delayed hypersensitivity tests (31-33), have been found to affect survival in patients with other cancers and in those having major surgery. To investigate further the potential of nutritional factorstoexplain differences in survival rates ofblack and white women with breast cancer, data from a large prospective cohort study of these patients were analyzed.
Vol. 82, No. 21, November 7, 1990
indicated the time between the first medical consultation and diagnosis. Information on each of the demographic and clinical factors in this study, except histology (93.0%) and number of involved nodes (88.7%), was known for 95% or more of the subjects. There was no difference between blacks and whites in frequency of missing data for any of these variables except number of lymph nodes. Data on the number of involved lymph nodes were missing more frequently among blacks (21.3%) than among whites (8.0%), primarily because of the greater presence of advanced, inoperable disease among blacks. Estrogen receptor status could not be included in the analysis since it was available for only 30% of the participants. Selected characteristics of the present patient population were compared with those of all women with breast cancer registered by the population-based SEER Program in metropolitan Atlanta (4). The two patient groups were similar with respect to race, age, marital status, and tumor involvement of lymph nodes. We attempted to collect mortality data on all subjects for a period of at least 5 years. The primary sources of follow-up information were the 14 hospital tumor registries, each of which was approved by the American College of Surgeons. Supplemental follow-up information was also obtained from a population-based SEER Program registry, which included review of death certificates. Of 1,491 eligible white women and 469 eligible black women, 1,290 (86.5%) and 446 (95.1%), respectively, were followed into the 5th year after diagnosis. The associations between race and the nutritional status indicators and between race and other suspected prognostic factors in this study were evaluated with odds ratios and corresponding approximate 95% confidence intervals (56). In addition, stage- and race-specific median body mass indices, lymphocyte counts, and levels of serum albumin and hemoglobin were compared. To assess the relationship between suspected prognostic factors and survival from breast cancer, we performed univariate survival analyses, with death from breast cancer as the outcome, with the Kaplan-Meier product limit estimation procedure (37). The statistical significance of differences between survival curves was assessed with a generalized Wilcoxon test statistic (38). The Cox proportionalhazards regression approach (39) was used to evaluate the prognostic effect of race, with adjustments made for nutritional status variables and other potentially confounding prognostic factors. To assess whether the prognostic effect of the variables was uniform across levels, we categorized each variable and created binary indicator variables to estimate changes in rate ratios across levels of each variable. If changes in rate ratios were uniform, variables were used as ordinal measures. Otherwise, binary variables were used to estimate the rate ratio for each category relative to that of the referent category. We conducted trend tests by assigning an integer level to each ordered category and testing the significance of that integer variable in the model. In addition, we evaluated the potential for differential effects of the nutritional status indicators on survival among subgroups of patients defined by other prognostic variables by including interaction terms between the nutritional status indicators and other prognostic variables in the proportional-hazards model. Because the effects of race (4) and serum albumin levels (26) on survival ARTICLES 1685
Downloaded from http://jnci.oxfordjournals.org/ at Université Laval on March 3, 2015
bilitation, and follow-up was abstracted from hospital medical records. The case group for the present analysis consisted of 1,491 white women and 469 black women, excluding seven women of other races. Eighty women with carcinoma in situ were excluded. Because liver disease was likely to affect biochemical measures of nutritional status, 45 women were excluded with (a) a positive liver scan; (b) an alkaline phosphatase level of greater than 125 U/L, plus either an aspartate aminotransferase (AST) level of greater than 50 U/L or a serum total bilirubin level of greater than 1.2 mg/dL; or (c) an AST level of greater than 50 U/L and a serum total bilirubin level of greater than 1.2 mg/dL. Variables of nutritional status that were available for the present analysis included measured height (in meters) and weight (in kilograms), serum albumin (in grams per deciliter), hemoglobin (in grams per deciliter), and lymphocyte count (in cells per millimeter cubed). Weight and height were combined to create a body mass index (weight [in kilograms]/height [in meters] squared) (34). Information on height and weight was available for 98% of the study population. Pretreatment levels of serum albumin and hemoglobin and lymphocyte count, which were measured as part of the diagnostic evaluation, were available for 74% of the patients. Levels of serum albumin were more likely to be unknown for whites (26%) than for blacks (19%), while lymphocyte counts were more likely to be missing for blacks (19% vs. 8%). Weight loss, serum cholesterol, and data from immunocompetence tests were not collected and could not be included as nutritional status variables. A number of demographic, medical, and clinical characteristics found to predict survival in published reports were included in the present analysis. These included age (in yr), race (black or white), marital status (married or not married), education (yr), menopausal status at diagnosis (premenopausal or postmenopausal), prior history of cancer (yes or no), family history ofbreast cancer (yes or no), history ofbreast self-examination (done or not done), number of lymph nodes involved, and treatment (yes or no) for each of the following: surgery, radiation therapy, hormone therapy, and chemotherapy. Information on the presence of nipple bleeding or discharge (yes or no) and skin ulceration (yes or no) at the time of diagnosis was also available. The occupational status of the patient^ household was indicated by the occupation (professional, nonprofessional salaried, hourly wage, or not employed) of the patient or her spouse, whichever was higher. Since the pathologic classification of stage was considered more definitive, the present analysis was based on pathologic rather than clinical stage (35). Ordinal categories of stage were coded as follows: 1 = localized tumor 2 cm or less in diameter with histologically negative movable lymph nodes (referent); 2 = localized tumor 2-5 cm or any tumor 5 cm or smaller with spread to movable axillary lymph nodes; 3 = any tumor greater than 5 cm in diameter, involving fixation to skin or chest wall, or involvingfixationof axillary lymph nodes; and 4 = disease outside the involved breast and homolateral axillary nodes. Histology was classified as lobular, ductal, or other. A patient time-to-care variable was created by calculating the time between the recognition of symptoms and the first medical consultation. A system time-to-diagnosis variable
have been found to vary by stage of disease, stage-specific analyses were conducted. A list-wise deletion of records was used in the bivariate analysis, so that the maximum number of observations with complete data was used in any given analysis (40). Multivariable analysis was conducted by use of a sample of participants for whom data were available for all variables used in the final analysis.
Results
Table 1. Prognostic demographic and medical characteristics of black and white patients with breast cancer
Characteristic
Level
Whites 1,491)* No.
Age at diagnosis (yr)
Marital status Education (yr)
Household occupational status}:
Menopausal status at diagnosis Prior cancer Family history of breast cancer
«49 50-64 5*65 Married Not married *13 12 9-11 24.5) had the worst survival rates. Women in the lower two categories of serum albumin levels tended to have poorer survival rates than did women with higher levels. Uniform trends of increased survival by albumin level, however, were present only among women with stage 1 disease and in the combined analysis of all stages. Women with higher hemoglobin levels fared better than did those with lower levels, both within each stage and overall. Although an inverse relationship between survival time and lymphocyte count was apparent when all stages were combined, it was not statistically significant. Further analyses of the relationship between nutritional status and survival time revealed no racial differences in survival patterns according to body mass index or levels of serum albumin or hemoglobin. An additional analysis was conducted to compare nutritional status indicators for women lost to follow-up before the 5th year with those either identified as deceased prior to the 5th year or followed alive into the 5th year. There were no statistically significant differences in nutritional status indicators according to followup status for either black or white women. Because of the small Vol. 82, No. 21, November 7, 1990
numbers of premenopausal women and of women who had received treatment other than surgery, treatment-specific and menopausal status-specific analyses were not performed. In constructing multivariate models, we observed that for stage 1 and stage 2 diseases, the black-to-white hazard rate ratios were affected by the exclusion of records lacking complete information on all variables. The rate ratios for the total sample were 1.90 for stage 1 (n = 458) and 1.22 for stage 2 (« = 1,057). For the sample with complete information, the rate ratios were 1.26 for stage 1 (n = 278) and 1.05 for stage 2 (n = 649). These alterations in the hazard rate ratios were perceived to be too great to permit valid inferences from multivariable analysis of stage 1 or stage 2 disease. Among women with stage 3 disease, exclusion of 154 women with missing data resulted in a somewhat reduced rate ratio (from 1.86 to 1.61) (table 7). Adjustment for demographic, medical, and clinical characteristics reduced this black-white difference in rates (rate ratio = 1.54). Further adjustment for lymphocytes had little effect on the rate ratio (1.52), but adjustment for body mass index or levels of serum albumin or hemoglobin reduced the ratio to approximately 1.40-1.45. Simultaneous adjustment for all nutritional status variables reduced the rate ratio to 1.14. Although the adjustment for each ARTICLES 1687
Downloaded from http://jnci.oxfordjournals.org/ at Université Laval on March 3, 2015
2-6
Whites (« = 1,491)* % No.
Table 3. Comparison of nutritional status of black and white patients with breast cancer
287 557 623 152 168 292 199 287 69 921 457 299 650 205 211
«20.5 20.6-24.5 *24.6
Numerous investigations examining national (1-3) and regional (4,5) US populations, as well as hospital case series (6), have found that white women with breast cancer have a longer survival time than do black women with breast cancer. Information collected from the National Cancer Instituted Surveillance, Epidemiology, and End Results (SEER) Program indicated that for the period 1979-1984, the 5-year relative survival rates in patients with breast cancer were 75.0% for white women and 62.2% for black women (3). This racial difference in survival was found among women aged 50 years and younger as well as among older women and persisted at roughly the same level from 1974-1976 through 1979-1984 (3). This racial difference in survival rates of patients with breast cancer has been attributed to many different factors. Black women are more likely to be diagnosed with cancer in a later stage (J). In addition, within specific stage classifications, white women survive longer than do black women (3), and racial differences in survival have been found to persist after analytic adjustment for stage (4,7,8). Some authors (5,9) have found most or all of the difference in survival rates to be eliminated when data are adjusted for socioeconomic status, but others (4,8) have found that substantial differences in survival rates remain. A number of other factors associated with poor prognosis have been found to be more prevalent among black patients than white patients. These include delays in diagnosis (10-12), larger tumors (6), less differentiated tumors (6,13), clinical symptoms at presentation indicating more severe disease (6,13), estrogen receptor-negative tumors (13-17), larger number of positive nodes (13), unfavor1684
Patients and Methods Women with breast cancer that was newly diagnosed between June 1975 and February 1979 were identified by a review of hospital tumor registry records at 14 private and public hospitals in Georgia. Of 2,641 identified subjects, 240 (9.1%) were excluded on the advice of their personal physicians, and an additional 309 (11.7%) declined to participate. After informed consent was obtained from the patients, trained oncology nurses under physician supervision collected interview and medical-record information and entered these data onto standard, precoded forms. Nurses interviewed patients in the hospitals to collect information on demographics, family history of cancer, medical history, and history of breast cancer detection, as well as on symptoms and their duration. Information concerning diagnosis, treatment, reha-
Received January 2, 1990; revised July 18, 1990; accepted August 22, 1990. Supported in part by Public Health Service contracts N01-CN55131 and N01CN55429 from the National Cancer Institute, National Institutes of Health, Department of Health and Human Services. R. J. Coates, W. S. Clark, R. S. Greenberg (Department of Epidemiology and Biostatistics), J. W. Eley (Winship Cancer Center), C M. Huguley, Jr. (Department of Medicine), R. L. Brown (Department of Surgery), Emory Unhwsity, Atlanta, Ga. We thank the patients, physicians, and nurse coordinators who participated in this study. 'Correspondence to: Ralph J. Coates, Ph.D., Department of Epidemiology and Biostatistics, Emory University, 1599 Clifton Rd., N.E., Atlanta, GA 30329.
Journal of the National Cancer Institute
Downloaded from http://jnci.oxfordjournals.org/ at Université Laval on March 3, 2015
The effects of nutritional status on differences in the survival of black and white women with breast cancer were studied in a cohort of 1,960 Georgia women diagnosed during 1975-1979. After data were adjusted for stage of disease, socioeconomic status, and other prognostic factors, poorer survival rates were shown in black women. Within each stage classification, lower levels of serum albumin and hemoglobin and higher relative body weight were more common among blacks and were independently associated with poorer survival. Among women with stage 3 disease, adjustment for these variables substantially reduced the excess mortality rate among blacks, suggesting that racial differences in survival may be partly explained by differences in nutritional status or extent of disease within stage. [J Natl Cancer Inst 82:1684-1692,1990]
able histologic type (13), less complete or aggressive treatment (14,18), and obesity (6). Various indicators of dietary or nutritional status have been found to be associated with survival of patients with breast cancer (6,19-30). These have included body size (6,19,21, 27-29); weight loss prior to treatment (22); dietary fat consumption (25,30); serum levels of cholesterol (23), albumin (20,26), and hemoglobin (20,26); and a number of hematologic measures, such as lymphocyte counts (20,26), which may be affected by nutritional status. Several of these and other indicators of nutritional status, including triceps skinfolds and delayed hypersensitivity tests (31-33), have been found to affect survival in patients with other cancers and in those having major surgery. To investigate further the potential of nutritional factorstoexplain differences in survival rates ofblack and white women with breast cancer, data from a large prospective cohort study of these patients were analyzed.
Vol. 82, No. 21, November 7, 1990
indicated the time between the first medical consultation and diagnosis. Information on each of the demographic and clinical factors in this study, except histology (93.0%) and number of involved nodes (88.7%), was known for 95% or more of the subjects. There was no difference between blacks and whites in frequency of missing data for any of these variables except number of lymph nodes. Data on the number of involved lymph nodes were missing more frequently among blacks (21.3%) than among whites (8.0%), primarily because of the greater presence of advanced, inoperable disease among blacks. Estrogen receptor status could not be included in the analysis since it was available for only 30% of the participants. Selected characteristics of the present patient population were compared with those of all women with breast cancer registered by the population-based SEER Program in metropolitan Atlanta (4). The two patient groups were similar with respect to race, age, marital status, and tumor involvement of lymph nodes. We attempted to collect mortality data on all subjects for a period of at least 5 years. The primary sources of follow-up information were the 14 hospital tumor registries, each of which was approved by the American College of Surgeons. Supplemental follow-up information was also obtained from a population-based SEER Program registry, which included review of death certificates. Of 1,491 eligible white women and 469 eligible black women, 1,290 (86.5%) and 446 (95.1%), respectively, were followed into the 5th year after diagnosis. The associations between race and the nutritional status indicators and between race and other suspected prognostic factors in this study were evaluated with odds ratios and corresponding approximate 95% confidence intervals (56). In addition, stage- and race-specific median body mass indices, lymphocyte counts, and levels of serum albumin and hemoglobin were compared. To assess the relationship between suspected prognostic factors and survival from breast cancer, we performed univariate survival analyses, with death from breast cancer as the outcome, with the Kaplan-Meier product limit estimation procedure (37). The statistical significance of differences between survival curves was assessed with a generalized Wilcoxon test statistic (38). The Cox proportionalhazards regression approach (39) was used to evaluate the prognostic effect of race, with adjustments made for nutritional status variables and other potentially confounding prognostic factors. To assess whether the prognostic effect of the variables was uniform across levels, we categorized each variable and created binary indicator variables to estimate changes in rate ratios across levels of each variable. If changes in rate ratios were uniform, variables were used as ordinal measures. Otherwise, binary variables were used to estimate the rate ratio for each category relative to that of the referent category. We conducted trend tests by assigning an integer level to each ordered category and testing the significance of that integer variable in the model. In addition, we evaluated the potential for differential effects of the nutritional status indicators on survival among subgroups of patients defined by other prognostic variables by including interaction terms between the nutritional status indicators and other prognostic variables in the proportional-hazards model. Because the effects of race (4) and serum albumin levels (26) on survival ARTICLES 1685
Downloaded from http://jnci.oxfordjournals.org/ at Université Laval on March 3, 2015
bilitation, and follow-up was abstracted from hospital medical records. The case group for the present analysis consisted of 1,491 white women and 469 black women, excluding seven women of other races. Eighty women with carcinoma in situ were excluded. Because liver disease was likely to affect biochemical measures of nutritional status, 45 women were excluded with (a) a positive liver scan; (b) an alkaline phosphatase level of greater than 125 U/L, plus either an aspartate aminotransferase (AST) level of greater than 50 U/L or a serum total bilirubin level of greater than 1.2 mg/dL; or (c) an AST level of greater than 50 U/L and a serum total bilirubin level of greater than 1.2 mg/dL. Variables of nutritional status that were available for the present analysis included measured height (in meters) and weight (in kilograms), serum albumin (in grams per deciliter), hemoglobin (in grams per deciliter), and lymphocyte count (in cells per millimeter cubed). Weight and height were combined to create a body mass index (weight [in kilograms]/height [in meters] squared) (34). Information on height and weight was available for 98% of the study population. Pretreatment levels of serum albumin and hemoglobin and lymphocyte count, which were measured as part of the diagnostic evaluation, were available for 74% of the patients. Levels of serum albumin were more likely to be unknown for whites (26%) than for blacks (19%), while lymphocyte counts were more likely to be missing for blacks (19% vs. 8%). Weight loss, serum cholesterol, and data from immunocompetence tests were not collected and could not be included as nutritional status variables. A number of demographic, medical, and clinical characteristics found to predict survival in published reports were included in the present analysis. These included age (in yr), race (black or white), marital status (married or not married), education (yr), menopausal status at diagnosis (premenopausal or postmenopausal), prior history of cancer (yes or no), family history ofbreast cancer (yes or no), history ofbreast self-examination (done or not done), number of lymph nodes involved, and treatment (yes or no) for each of the following: surgery, radiation therapy, hormone therapy, and chemotherapy. Information on the presence of nipple bleeding or discharge (yes or no) and skin ulceration (yes or no) at the time of diagnosis was also available. The occupational status of the patient^ household was indicated by the occupation (professional, nonprofessional salaried, hourly wage, or not employed) of the patient or her spouse, whichever was higher. Since the pathologic classification of stage was considered more definitive, the present analysis was based on pathologic rather than clinical stage (35). Ordinal categories of stage were coded as follows: 1 = localized tumor 2 cm or less in diameter with histologically negative movable lymph nodes (referent); 2 = localized tumor 2-5 cm or any tumor 5 cm or smaller with spread to movable axillary lymph nodes; 3 = any tumor greater than 5 cm in diameter, involving fixation to skin or chest wall, or involvingfixationof axillary lymph nodes; and 4 = disease outside the involved breast and homolateral axillary nodes. Histology was classified as lobular, ductal, or other. A patient time-to-care variable was created by calculating the time between the recognition of symptoms and the first medical consultation. A system time-to-diagnosis variable
have been found to vary by stage of disease, stage-specific analyses were conducted. A list-wise deletion of records was used in the bivariate analysis, so that the maximum number of observations with complete data was used in any given analysis (40). Multivariable analysis was conducted by use of a sample of participants for whom data were available for all variables used in the final analysis.
Results
Table 1. Prognostic demographic and medical characteristics of black and white patients with breast cancer
Characteristic
Level
Whites 1,491)* No.
Age at diagnosis (yr)
Marital status Education (yr)
Household occupational status}:
Menopausal status at diagnosis Prior cancer Family history of breast cancer
«49 50-64 5*65 Married Not married *13 12 9-11 24.5) had the worst survival rates. Women in the lower two categories of serum albumin levels tended to have poorer survival rates than did women with higher levels. Uniform trends of increased survival by albumin level, however, were present only among women with stage 1 disease and in the combined analysis of all stages. Women with higher hemoglobin levels fared better than did those with lower levels, both within each stage and overall. Although an inverse relationship between survival time and lymphocyte count was apparent when all stages were combined, it was not statistically significant. Further analyses of the relationship between nutritional status and survival time revealed no racial differences in survival patterns according to body mass index or levels of serum albumin or hemoglobin. An additional analysis was conducted to compare nutritional status indicators for women lost to follow-up before the 5th year with those either identified as deceased prior to the 5th year or followed alive into the 5th year. There were no statistically significant differences in nutritional status indicators according to followup status for either black or white women. Because of the small Vol. 82, No. 21, November 7, 1990
numbers of premenopausal women and of women who had received treatment other than surgery, treatment-specific and menopausal status-specific analyses were not performed. In constructing multivariate models, we observed that for stage 1 and stage 2 diseases, the black-to-white hazard rate ratios were affected by the exclusion of records lacking complete information on all variables. The rate ratios for the total sample were 1.90 for stage 1 (n = 458) and 1.22 for stage 2 (« = 1,057). For the sample with complete information, the rate ratios were 1.26 for stage 1 (n = 278) and 1.05 for stage 2 (n = 649). These alterations in the hazard rate ratios were perceived to be too great to permit valid inferences from multivariable analysis of stage 1 or stage 2 disease. Among women with stage 3 disease, exclusion of 154 women with missing data resulted in a somewhat reduced rate ratio (from 1.86 to 1.61) (table 7). Adjustment for demographic, medical, and clinical characteristics reduced this black-white difference in rates (rate ratio = 1.54). Further adjustment for lymphocytes had little effect on the rate ratio (1.52), but adjustment for body mass index or levels of serum albumin or hemoglobin reduced the ratio to approximately 1.40-1.45. Simultaneous adjustment for all nutritional status variables reduced the rate ratio to 1.14. Although the adjustment for each ARTICLES 1687
Downloaded from http://jnci.oxfordjournals.org/ at Université Laval on March 3, 2015
2-6
Whites (« = 1,491)* % No.
Table 3. Comparison of nutritional status of black and white patients with breast cancer
287 557 623 152 168 292 199 287 69 921 457 299 650 205 211
«20.5 20.6-24.5 *24.6
