Background/Objective: Obesity predicts a higher incidence of postmenopausal breast cancer, but body fat distribution may provide information beyond body mass index (BMI), in particular among women with non-White ancestry. Therefore, we evaluated the association of body shape phenotypes previously identified in European populations with breast cancer among women with African American, Native Hawaiian, Japanese American, Latino, and White ancestry in the Multiethnic Cohort (MEC).
Methods: The MEC is a prospective study of >215,000 residents of Hawaii and California recruited in 1993-1996. This analysis examined a subset of 40,787 women who responded to a follow-up survey in 2003-2008 (QX3) when their mean age was 70.9±8.6 years. Demographic information, lifestyle factors, height, weight, waist and hip circumferences were self-reported by questionnaire. Using these four anthropometric measures plus BMI and waist-to-hip ratio (WHR), we identified three principal components (PC) to represent body fat distribution derived through PC analysis. To estimate breast cancer risk associated with quintiles of PCs, Cox proportional hazards regression was applied to obtain hazard ratios (HRs) and 95% confidence intervals (95%CI) with follow-up time (from QX3 completion to breast cancer diagnosis, death, or end of 2019) as underlying time metric while adjusting for age, ethnicity, hormone therapy, family history of breast cancer, smoking, and parity, plus BMI in a separate model. Stratified analysis by ethnicity were performed to assess possible differences across groups.
Results: During 11.1±3.6 years of follow-up, 2,069 women were diagnosed with breast cancer. While the loadings of the six anthropometric measures were generally comparable across ethnic groups for PC1, representing general obesity, and PC3, interpreted as a measure of abdominal obesity, clear inter-ethnic variability was detected for PC2 loadings, which reflects a tall/lean figure, with a similar configuration for African American, Latina, and Native Hawaiian women, while Japanese American and White women showed an almost reversed pattern. PC1 was positively associated with breast cancer risk (HR for Q5 vs. Q1 1.97; 95%CI 1.70-2.30); this association remained significant after adding BMI to the model (HR for Q5 vs. Q1 1.55; 95%CI 1.16-2.07). In contrast, PC2 (HR 1.05; 95%CI 0.91-1.21) and PC3 (HR 0.92; 95%CI 0.80-1.06) showed no significant relation with breast cancer. The association of  PC1 with breast cancer varied considerably by ethnic group. The risk estimated for Q5 vs. Q1 was strongest among Japanese Americans (HR 2.61; 95%CI 1.97-3.45) and Native Hawaiians (HR 2.41; 95%CI 1.25-4.63), while the values for Latinas (HR 1.79; 95%CI 1.15-2.80) and Whites (HR 1.57; 95%CI 1.18-2.08) were intermediate; no association was detected among African Americans. PC2 and PC3 were not associated with incidence in any group. HRs for BMI were lower than PC1 among all women (HR 1.89; 95%CI 1.62-2.21) and Japanese American women (HR 2.45; 95%CI 1.88-3.21) but higher in White women (HR 1.71; 95%CI 1.30-2.25).
Conclusion: Our findings suggest that PC1, which captures general adiposity, only adds limited incremental information beyond BMI to the prediction of breast cancer incidence. However, the utility of these PCs may vary by ethnicity with the strongest associations among Japanese American and Native Hawaiian women.