The NUS Breast Cancer Registry has received approval from the NUS Institutional Review Board (NUS-IRB). The University Malaya Breast Cancer Registry was approved by the University Malaya's Ethical Committee. This study was approved by the respective ethical committees in both institutions. As the study relies on non-identifiable registry-based data, the need to obtain informed consent was waived.
We used data from the Singapore-Malaysia Breast Cancer Registry which currently encompasses 5,769 women. This multi-institutional breast cancer registry is a merger between the National University Hospital (NUH) Breast Cancer Registry, and the University Malaya Medical Center (UMMC) Breast Cancer Registry 
. The NUH is a tertiary university hospital in the city state of Singapore. Its breast cancer registry includes all consecutive 2,449 women diagnosed with breast cancer between 1990 and 2007. Information has been collected retrospectively between 1990 and 1995 and prospectively from 1995 to 2007. UMMC is an academic tertiary hospital situated in the relatively affluent part of Kuala Lumpur (Malaysia) and caters to a predominantly middle class urban population. The UMMC Breast Cancer Registry is a prospective database of 3,320 consecutive women who were newly diagnosed with breast cancer between 1993 and 2007.
In the current study, we included all 5,264 women from the three major ethnic groups in Malaysia and Singapore i.e. Chinese, Malays and Indians, who were diagnosed with invasive breast cancer. Women of other ethnic minorities (n
198) were excluded as were women with carcinoma in situ (n
The determinant of interest was ethnicity (Chinese, Malay, Indian), and the outcome was death from all causes. Data on patient characteristics included age at diagnosis, and center of treatment (NUH, UMMC). Patients were staged according to the 5th edition of TNM American Joint Committee on Cancer (AJCC) system if diagnosed before January 1st 2003, and according to the 6th edition of AJCC if diagnosed after this date. Variables on disease characteristics included pathologically determined tumor size (available as continuous value and categorized into 0·1–20.0 mm, 20·1–50·00 mm, >50·0 mm, unknown), pathologically determined lymph node involvement (available as absolute number of positive nodes and categorized into yes, no, unknown), distant metastasis (detected by means of computed tomography of the thorax, abdomen and pelvis, and bone scan in patients with clinical stage III breast cancer; yes, no, unknown), estrogen receptor (ER) status/progesterone receptor (PR) status (positive when >10% of tumor cells stained positive during immunohistochemical testing, negative, unknown), and tumor grade (Scarff-Bloom-Richardson classification; grade 1, grade 2, grade 3, unknown). Loco-regional treatment was classified as no loco-regional treatment, complete loco-regional treatment (i.e. mastectomy, or breast conserving surgery [BCS] followed by radiotherapy), and incomplete treatment (i.e. BCS only, or radiotherapy only). Administration of chemotherapy, and hormone therapy were categorized as yes and no.
Follow-up and outcome assessment
In both centers, patients were monitored via scheduled appointments in the specialist breast clinics. Data on mortality were obtained from the hospitals' medical records, as well as through active follow-up. In addition, vital status was verified through direct linkage with the National Registration Department in Malaysia. In this hospital based cancer registry, information on cause of death or cancer recurrence was not available for the majority of patients. Follow-up time was calculated as the interval between date of diagnosis and date of death, or date of last contact, whichever came first.
All categorical variables were described by proportions and compared using the Chi square test. Continuous variables were expressed in medians and compared using the Kruskal Wallis test. Overall survival was estimated using Kaplan-Meier analyses and compared by log-rank test.
Cox regression analysis was performed to estimate the relative risk for all-cause mortality expressed as hazard ratio (HR) between women of different ethnicities. Time at entry was date of diagnosis with breast cancer, and exit time was date of death (from all causes), date at last contact, or Nov, 2010 (linkage with national mortality registry in Malaysia). This model was adjusted for age at diagnosis, center, year of diagnosis, tumor size, lymph node involvement, metastases, estrogen and progesterone receptor status, tumor grade, loco-regional therapy, chemotherapy, and hormone therapy.
Within the subgroup of patients with pathologically confirmed tumor size and lymph node status (N
3,712), we studied possible effect modification by tumor size on the association between ethnicity and lymph node metastasis by including the interaction terms ‘ethnic groups multiplied by tumor size in mm’ into a logistic regression model with lymph node involvement as the outcome. Patients were subsequently grouped according to tumor size <20 mm, 20 to 50 mm, and >50 mm. Within each category, logistic regression analysis was used to determine the association between lymph node involvement (outcome) and ethnicity (predictor), adjusted for absolute tumor size (mm), tumor grade, ER status and PR status.
Two-tailed p-values below 0·05 and HRs with 95%CI which did not include 1·00 were considered as statistically significant. All analyses were performed using SAS version 9·1 (SAS Institute Inc, Cary, NC).