Of the 2,888 breast cancer patients in the study cohort, 1,575 (54%) were non-Hispanic White, 1,106 (39%) Black, 126 (4%) Asian, and 81 (3%) Hispanic. Baseline demographic and clinical characteristics by racial/ethnic groups are summarized in Table 1. Compared to non-Hispanic White, Black women were older, more likely to smoke currently, receive Medicare or Medicaid benefits, and have comorbidities. As expected, Black women were more likely to have triple-negative breast cancer. The distribution of tumor stage was similar among the groups. Black women had the highest BMI at diagnosis (mean 31.5 kg/m2) with 30.9% being overweight and 51.8% obese, followed by Hispanic, and non-Hispanic Whites, while Asian Americans had the lowest BMI (Figure 2A).
As shown in Figure 2B-E, Black patients had significant weight loss after 6 months, 18 months, 30 months, and 48 months from diagnosis (0.84 kg/m2 reduction), while body weight did not change substantively in other racial/ethnic groups. Patients younger than 50 gained weight after diagnosis, while older patients lost weight. Patients with at least one comorbidity lost substantial weight after diagnosis. In the multivariable analysis using a mixed-effects linear model, we found that Black patients had higher BMI at diagnosis but had significant weight loss after cancer diagnosis (0.47 kg/m2/year reduction) compared with Whites (Table 2). Older patients or patients with comorbidities had higher BMI at diagnosis but had significantly more weight loss after diagnosis, compared with younger patients or those without comorbidities, respectively. Patients diagnosed with different tumor stages had similar baseline weight but those with advanced stage had larger weight loss in a long run. At baseline, patients with or without chemotherapy had similar body weight; at months 6 and 18, patients receiving chemotherapy had more weight loss than patients without chemotherapy, but at month 48 after diagnosis, patients receiving chemotherapy had slightly less weight loss. We also found ex-smokers at diagnosis had higher baseline BMI and lost more weight during follow-up.
After a median follow-up time of 6.4 years, 387 patients died (190 deaths due to breast cancer) and 265 patients had a breast cancer recurrence. We modeled the shape of BMI changes according to months before death or censoring stratified by vital status in both White and Black women (Figure 3). Among breast cancer survivors, BMI was almost constant during the follow-up years in both White and Black women. In contrast, for patients who died, there were large BMI losses starting about 30 months before death with the rapid BMI reduction observed in both Whites and Blacks (3.2 and 3.8 kg/m2 reduction, respectively). These BMI reductions were equivalent to 8.8 and 10.3 kg absolute weight loss for an average person in our cohort. Using a consensus definition of cancer cachexia,24 about 45% of the patients who were deceased had cachexia (40% in Whites and 48% in Blacks). The median time from recurrence to death was 24 months and 13 months for Whites and Blacks who were deceased, respectively. For patients who had distant metastasis and died later, 19% had cachexia before the recurrence.
Being obese or overweight at diagnosis was significantly associated with higher risk of all-cause mortality, breast cancer specific mortality, and disease recurrence in the univariate analysis. However, after adjusting for age, race, Carlson comorbidity index, tumor stage, histologic grade, estrogen receptor, progesterone receptor, Her2, radiotherapy, hormonal therapy, and chemotherapy, BMI at baseline was no longer statistically associated with survival outcomes (Table 3).
Table 4 shows that after adjustment for multiple clinical and pathological factors, compared to stable weight, BMI loss (>0.5 kg/m2/year) was associated with a higher risk of all-cause mortality (hazard ratio [HR]=2.60, 95% CI 1.88-3.59), breast cancer specific mortality (HR=3.05, 95% CI 1.91-4.86), and disease-free survival (HR=2.12, 95% CI 1.52-2.96). BMI gain (>0.5 kg/m2/year) was also associated with worse survival outcomes compared with stable weight (HR=1.60, 95% CI 1.10-2.33 for overall survival; HR=1.73, 95% CI 1.04-2.87 for breast cancer-specific survival, and HR=1.54, 95% CI 1.06-2.24 for disease-free survival). A slope of 0.5 kg/m2/year is equivalent to 3% weight change over 18 months for women with BMI of 25 kg/m2 (i.e. 5’5 height and 150 pounds weight). There was a U-shaped relationship between change in BMI and overall survival (Figure 4, p-for-nonlinearity <0.001). The larger the weight loss, the higher the risk of all-cause mortality, and BMI gain was also associated with gradual increased risk of death. Similar U-shaped relationships were also observed for breast cancer-specific survival and disease-free survival.
In addition, we found that age was strongly associated with overall survival and disease-free survival, and moderately associated with breast cancer-specific survival (Table 4). There was a monotonic strong association between higher comorbidity index with worse outcomes in overall survival and disease-free survival, but no association between comorbidity and breast cancer-specific survival. Tumor stage was strongly associated with all three survival outcomes, in particular with breast cancer-specific survival (HR=5.92, 95% CI 3.35-10.48, comparing stage 3 vs. stage 1). Black women had about a 2-fold increased risk of the three survival outcomes compared with White women (Table 4). Consistent with a previous clinical trial-based study,25 we also found that the racial difference in overall survival was stronger among non-obese patients (HR=2.22, 95% CI 1.48-3.33) than among obese patients (HR=1.28, 95% CI 0.82-1.98).
We also included recurrence status as a time-varying covariate in the Cox model of overall survival, and found that relationship between weight change and overall survival remained: compared to stable weight, BMI loss (>0.5 kg/m2/year) was associated with a higher risk of all-cause mortality (HR=2.55, 95% CI 1.84-3.53); BMI gain (>0.5 kg/m2/year) was also associated with worse survival compared with stable weight (HR=1.62, 95% CI 1.10-2.39). As expected, recurrence status strongly predicted overall survival (HR=13.09, 95% CI 9.76-17.56).
In the stratified analysis of changing slope of BMI and survival outcomes, we found the U-shaped relationships were generally preserved for White and Black patients (Table 5): both weight gain and loss after diagnosis were associated with increased risk of the three outcomes. Although the effects were seemingly stronger in Blacks than in Whites, the tests for interaction were not statistically significant. The effects of BMI change on survival outcomes seemed to be limited to those women who were overweight/obese at diagnosis, but the test for interaction were not significant. BMI loss was consistently associated with worse survival outcomes regardless of the age at diagnosis.
We further examined whether the slope of BMI change within the first two years after diagnosis and BMI change from diagnosis to 18 months can predict survival outcomes (Table 6). We found that weight loss within the first 2 years of diagnosis can predict increased risk of the three survival outcomes after adjustment for other prognostic factors. Using changing slope as a measure of weight loss was more powerful in risk prediction than change between two time points, reflecting more data being efficiently utilized.