With Institutional Review Board approval, we established a database of all female patients diagnosed at our institution with invasive breast cancer between March 1998 and November 2006. Informed consent was waived because individual patient information was identified only by investigator-generated code numbers that are not linkable to patient identifiers. For each patient, we identified tumour grade, stage, level of ER, PR and HER2 expression (or gene amplification), patient age, BMI (using standard National Heart, Lung and Blood Institute categories) and self-identified racial group. Recurrent tumours were excluded. (Recurrence was defined by time elapsed since first tumour, interim treatment, immunophenotype, histology, metastases and clinical impression). Ten of 415 tumours (2%) were second primary or synchronous contralateral breast tumours and were included in this study.

We queried electronic medical records (EMR) and performed manual medical record review for all patients to ensure quality control. Tumour histology, grade, stage, ER, PR and HER2 expression were determined from the original pathology reports. Tumours had been diagnosed by experienced pathologists using standard criteria for histology and modified Scarff-Bloom-Richardson criteria for grade. ER and PR expression were determined using immunoperoxidase staining (Dako, Carpinteria, CA, USA) and quantified by image analysis (Biogenex, San Ramon, CA, USA) with values less than 5% categorised as negative. HER2 expression was determined by immunohistochemistry (Dako, Carpinteria, CA, USA). Tumours that showed 2+ Her2/neu immunohistochemistry staining based on HercepTest criteria were definitively assessed by fluorescence in situ hybridisation (FISH) (Vysis, Des Plaines, IL, USA). Tumours were designated as being HER2+ if they showed 3+ Her2/neu immunohistochemistry staining (based on HercepTest criteria) or if they were FISH positive 2728. Stage at diagnosis was coded according to the American Joint Commission on Cancer's Cancer Staging Manual 29.

Patient age at diagnosis was calculated using the dates of birth and diagnosis in the EMR. Patients were categorised as diagnosed at age 50 years or younger, when they are more likely to be premenopausal, or at age over 50 years, when they are more likely to be postmenopausal. BMI was calculated by weight (kg)/height squared (m²) using EMR data. In 97% of cases, weight and height were recorded within six months of diagnosis. If more than one weight or height value was available, we used the values closest to the date of diagnosis. Patients were placed into one of five BMI categories: under/normal weight (BMI <25), overweight (BMI 25 to <30), obesity I (BMI 30 to <35), obesity II (BMI 35 to <40) or obesity III (BMI ≥ 40).

Racial/ethnic group was determined by patient self-identification at the time of registration. Categories included: white/caucasian, black/African-American, hispanic, Asian/Pacific Islander, Middle Eastern and other. Due to sparse data, patients self-identified as Asian or Middle Eastern were included in the other category. To test the accuracy of the registration data, a manual review of the EMR was performed to determine provider-identified racial/ethnic group. A provider-identified racial/ethnic group was available for 402 of 415 patients and was concordant with the self-identified group in 91% of cases. When groups differed, the self-identified group was used. EMR data often included country of origin.

Because women born in the Caribbean constituted a large subgroup of our black population (n = 56, 27%), we compared this subgroup to the rest of the black population, which comprised patients who were provider-identified as African-American, African or black without further specification. Our Caribbean black patients were from countries with a majority pan-West African origin (Haiti, Jamaica, Trinidad, Tobago, Barbados and Montserrat). Patients from the Caribbean countries with a majority hispanic origin (Dominican Republic, Puerto Rico and Cuba) were classified as hispanic and not included in this black subgroup comparison.

For 56 of the 81 triple-negative tumours, residual paraffin-embedded tumour tissue was available and additional evaluation could be performed. For additional morphological assessment, tumours were classified as grade 2 or grade 3 ductal carcinoma not otherwise specified (NOS), medullary-like carcinoma or other (for example, grade 1 ductal NOS, lobular or micropapillary tumours). Additional immunohistochemistry was performed with antibodies to CK 5/6 (Biocare Medical, Concord, CA, USA), and EGFR (Dako, Carpinteria, CA, USA) using a streptavidin-biotin horseradish peroxidase detection kit (Biogenex, San Ramon, CA, USA). The expression of these markers was graded semi-quantitatively. Allred scoring 30 was used for CK5/6, with a score of 5 or more considered as positive. HercepTest scoring criteria 27 were used to grade EGFR expression, with a score of 2+ or more considered as positive 31.

Data on patient and tumour characteristics were entered into a Microsoft Excel worksheet (Redmond, WA, USA) and exported into the SAS (Cary, NC, USA) statistical package. We used chi-squares statistics based on contingency tables to test for homogeneity of proportions and multivariate logistic regression to determine associations between patient-tumour characteristics and triple-negative breast cancer. Odds ratios (OR) and 95% confidence intervals (CI) from the multivariate logistic regression analyses were adjusted for race, BMI and age (≤ 50 or > 50). We investigated whether there was an adjusted association between categorical variables (race and BMI group) and triple-negative tumour type by calculating twice the difference in the model log-likelihood with and without the categorical variable. This was distributed as chi-squared with degrees of freedom equal to the number of categories less one. Women with missing data were excluded from these analyses.

The demographic features of the 415 patients in our database are presented in Table 1. Some features are consistent with averages for the USA, but our population is unusual compared with most other reported cohorts because of its marked ethnic and racial diversity (Figure 1). In the present analysis, we grouped patients into four racial/ethnic categories: white (36%), black (43%), hispanic (10%) and other (11%). Within these broad categories, our patient population contained diverse subcategories. The median age at diagnosis of invasive breast cancer in our patients was 58 years, which is slightly younger than the US average age of 61 years, as reported in the National Cancer Institute's SEER Cancer Statistics Review 32. Of our patients, 29% were 50 years or younger and 71% were older than 50 years at diagnosis. These proportions are generally consistent with US averages 33. Notable in our population was the frequency of elevated BMI (only 23% of women fell into the under/normal weight category, while 30% were overweight, 27% were classified as obesity I, 13% as obesity II and 7% as obesity III).

Tumour characteristics are presented in Table 1. Of the 415 invasive breast cancers in our database, 81 (20%) were triple negative, 297 (72%) were ER and/or PR positive and 53 (13%) were HER2 positive. Overall, our patients had fewer HER2-positive tumours and more triple-negative tumours than in some series, although our proportions are consistent with those in racially/ethnically heterogeneous cohorts 343536. In addition, our patients presented at a somewhat later stage than average for the USA: 47% of our tumours were lymph node-positive; the USA average is 39% 32. However, the presence of 96 tumours with unconfirmed node status may have affected these proportions.

We characterised all available triple-negative tumours (56 of 81 (69%)) with additional morphological classification and immunohistochemical staining for CK5/6 and EGFR to determine what proportion of the triple-negative tumours had the basal-like phenotype. We found that 19 of 56 (34%) tumours were medullary-like, 14 of 56 (25%) were grade 2 and 18 of 56 (32%) were grade 3 ductal carcinoma NOS and 5 of 56 (9%) were of other histologies. Of 56 tumours, 38 (68%) were CK5/6 and/or EGFR positive. There was a significant association between CK5/6 expression and EGFR expression with a Spearman correlation coefficient of 0.34 (p = 0.01). Taken together, these studies suggest that our triple-negative tumours include a high proportion of basal-like tumours.

We examined associations between race, BMI, age (and presumed menopausal status), tumour grade, ER expression, PR expression, HER2 expression and nodal involvement. We confirmed previously noted associations between patient and tumour variables, for example, positive associations between markers of poor prognosis (e.g., grade and stage). Of particular relevance, we found in bivariate analyses that obesity was associated with race (p = 0.01) 1537 and that triple-negative status was associated with race (p = 0.0002) 63435363839.

Table 1 shows the associations of race/ethnicity with tumour prognostic markers and other clinicopathological features. We noted two associations pertinent to triple-negative tumours and BMI. First, 30% of tumours in black women were triple negative, compared with 11 to 13% of tumours in other women. Second, 55% of black women were obese, compared with 36 to 45% of other women. There was no substantial dependence on race/ethnicity categories other than black (Table 1): the results of our analyses did not substantially depend on more finely-divided race categories, as determined by visual examination of the associations and their intervals with the reference category limited to whites versus the reference category defined as non-blacks. Therefore, we combined white, hispanic and other race/ethnicity categories into a single category 'non-black', which serves as our reference group, to address two questions.

Black women were both more likely than other women to be obese and to have triple-negative tumours, so we asked whether obese black women had a higher proportion of triple-negative tumours than other obese women. As shown in Table 2, stratifying the dataset to black vs. non-black women, we found that 29% of obese black women had triple-negative tumours compared with 8.6% of obese non-black women (OR = 4.3: 95 CI = 1.8 to 10; p = 0.0004). (Using whites as the reference category, the OR = 4.2 and 95% CI = 1.6 to 13). Similarly, 31% of non-obese black women had triple-negative tumours compared with 15% of non-obese non-black women (OR = 2.7, 95% CI = 1.4 to 5.3; p = 0.003). (Using whites as the reference category, the OR = 2.5 and 95% CI = 1.2 to 5.4). These two ORs were not significantly different from one another (p = 0.41), suggesting that among black women, BMI does not appear to be associated with triple-negative status.

Next, we examined associations of age at diagnosis with tumour characteristics within black and non-black women (Table 3). In contrast to previous reports 640, we did not find a strong association between triple-negative tumour status and younger age when we considered all patients (24% triple-negative tumours in women aged ≤ 50 years compared with 18% triple-negative women aged > 50 years; p = 0.22). Adjustment for more finely divided age categories made no difference in the estimates of association. We next considered black vs. non-black women. We found that the proportions of triple-negative tumours were similar in younger and older black women: 31% of black women aged 50 years or younger and 29% of black women aged over 50 years had triple-negative tumours (p = 0.76). In contrast, we found a marginal association between triple-negative tumours and age in non-black women: 17% of non-black women aged 50 years or younger had triple-negative tumours, compared with 10% of non-black women aged over 50 years (p = 0.11).

To further characterise the possible relations between race/ethnicity, BMI and triple-negative breast cancer, we performed multiple logistic regression analyses. Table 4 shows the adjusted OR and 95% CI from the multiple logistic regression analyses for patient characteristics of triple-negative breast cancers compared with other types of breast cancer. The odds of having a triple-negative tumour were three-fold higher (95% CI = 1.6 to 5.4) in black women as compared with white women. Mutually adjusting for race/ethnicity, BMI and age (age ≤ 50 years vs. > 50 years, as a surrogate for menopausal status), there remained a strong association between race/ethnicity and triple-negative tumours (p = 0.0001). After adjusting for race/ethnicity and age, we noted decreasing proportions of triple-negative tumours with increasing category of BMI (p = 0.08).

Because women of Caribbean origin constituted a large proportion of our black population (n = 56, 27%), we compared this subgroup to the rest of the black population. Controlling for age (≤ 50 years vs > 50 years), no significant differences were seen between Caribbean black women and other black women in the proportion of triple-negative tumours, BMI, grade or node involvement (see Additional Data File 1).