Association of cancer mortality with postdiagnosis overweight and obesity using body mass index

Although overweight and obesity increase cancer risk, it is still controversial with respect to cancer mortality. In the current study, we enrolled 2670 patients of 14 tumor types from the Cancer Genome Atlas (TCGA) project, to identify the prognostic role of overweight and obesity in cancer patients. After dividing the patients into different groups by the body mass index (BMI), we found significant lower mortality in the obesity group. In addition, we also treat BMI value as a binary categorical variable or continuous variable, respectively. We found significant lower mortality in the higher BMI group. Furthermore, when focusing on each tumor type, cervical cancer and bladder cancer showed lower mortality in the patients with higher BMI values. Taken together, our results demonstrate that postdiagnosis obesity might indicate a better prognosis in cancer patients. However, these findings should be interpreted cautiously because of small sample size.


INTRODUCTION
Prevalence of overweight and obesity has shown a significant increase during the last decades [1,2]. Besides higher risk of cardiovascular disease and diabetes, evidence suggests that the excess body weight is also associated with higher cancer risk [3,4]. However, it is still controversial with respect to cancer motality.
Body mass index (BMI) is a person's weight in kilograms divided by the square of height in meters. It is an attempt to quantify the amount of tissue mass (muscle, fat, and bone), and then categorize that individual as underweight, normal weight, overweight, or obese. By the analysis of more than 2.88 million individuals, Flegal et al. concluded that obesity was associated with higher mortality, while overweight was associated with lower mortality [5,6]. However, they did not specify the cancer population, and little conclusion could be drawn when focusing on cancer patients.
The cancer genome atlas (TCGA) project motivated large-scale coordinated cancer genomic efforts to obtain complete catalogs of the genomic alterations in cancer. Besides the comprehensive molecular profiling of each tumor, it also provides valuable clinical data [7]. In the present study, we used the survival data of over 2000 cancer patients (14 tumor types) from the TCGA project, to identify the roles of overweight and obesity in the prognosis of cancer.

Analyses of the cancer patients in the whole population
In this study, we applied the National Heart, Lung, and Blood Institute's BMI categories of underweight (BMI<18.5), normal weight (18.5 ≤ BMI < 25), overweight (25 ≤ BMI < 30), and obesity (BMI ≥ 30). Grade 1 obesity was defined as a BMI of 30 to less than 35; grade 2 obesity, a BMI of 35 to less than 40; and grade 3 obesity, a BMI of 40 or greater [5,6].
To gain insight into the prognostic role of obesity, we utilized the Kaplan-Meier curves stratified by different BMI groups. As shown in Figure 1A-1D, patients with grade 1, grade 2 and grade 3 obesity, showed significantly lower mortality (P < 0.05). We combined the grade 2 and grade 3 patients into one group, simply because the sample size is relatively small in grade 2 and grade 3. Although lower mortality was observed in overweight group, and higher mortality was observed in underweight group, they didn't reach statistical significance. In addition, cox regression analysis also confirmed the lower mortality in grade 1 obesity (HR=0.76, 95%CI, 0.58-0.99, P = 0.049), grade 2 and grade 3 obesity (HR=0.63, 95%CI, 0.46-0.85, P < 0.001), but not in the overweight group (HR=0.87, 95%CI, 0.70-1.08, P = 0.19), or in the underweight group (HR=1.50, 95%CI, 0.91-2.47, P = 0.115).
We then used the receiver operating characteristic (ROC) curves to determine the cutoff value of BMI. Thus, the BMI value of 31.2 (determined by the ROC curve) and 27.0 (median value)were used to explore the prognostic value of obesity, respectively. As shown in Figure 1E-1F, patients with higher BMI values showed lower mortality, no matter stratified by the ROC cutoff value ( Figure 1E) or the median value ( Figure 1F).

Analyses of the cancer patients in each tumor type
Our above data showed that obesity might indicate a lower mortality in the whole population of cancer patients. Next we sought to find out the specific associations in different cancer types.
Unexpectedly, no significant associations were found in any of the tumor type by the log-rank test (data not shown) or cox regression analysis (Figure 2A-2C). However, when we set the BMI value as a binary categorical variable, patients with higher BMI values (cutoff by the ROC curve, not by the median value) of BLCA and CESC showed significantly lower mortality ( Figure 2D-2E). On the other hand, when we set the BMI as a continuous variable, patients with higher BMI values of BLCA and CESC also showed significantly lower mortality ( Figure 2F).
This relationship was also verified by the cox regression analysis, which demonstrated that higher BMI value was correlated with lower mortality in CESC and BLCA (P < 0.05, Table 3). In addition, cox regression analysis also showed that some other variables, such as age and tumor stage, were also correlated with mortality in some cancer types (Table 3).

DISCUSSION
In this study, we systemically report that postdiagnosis obesity might indicate a better prognosis in cancer patients. The association of obesity and cancer risk has been extensively explored. The mechanisms included insulin resistance and resultant chronic hyperinsulinaemia, increased bioavailability of steroid hormones, and localized chronic inflammation [8][9][10][11][12]. However, it is still controversial with respect to cancer motality.
Calle et al. reported that obesity was a negative prognosis factor for cancer by analyzing the largest cohort of patients in the US [13]. Reeves et al. provided similar www.impactjournals.com/oncotarget results through the analysis in the UK [14]. These studies assessed the BMI before cancer development, leading to the popular perspective that prediangosis obesity indicated poor prognosis [15,16]. However, discrepancies emerged when Schlesinger et al. indicated a decreased mortality risk among overweight colorectal cancer survivors (HR=0.79, 95%CI, 0.71-0.88, P < 0.05). We realized that they assessed the BMI after cancer development, which is consistent with our study (HR=0.64, 95%CI, 0.51-0.81, P < 0.001) [17].
Our results supported the survival advantage conferred by obesity where energy balance is likely to be negative [18,19]. It's not hard to imagine that cachexia might indicate a poor prognosis [20]. However, when the patient suffers obesity before the development of cancer, side effects such as diabetes and cardiovascular diseases might have more impact on survial, which explains the results by Calle et al. and Reeves et al. In our study, besides the whole cancer population, only cervical cancer and bladder cancer showed lower mortality in patients with higher BMI values. We speculate that, the sample size is relatively large (>250) in cervical cancer and bladder cancer in our study, which might improve the statistical power in these patients. On the other hand, due to some hidden mechanisms, maybe the negative energy balance is more likely to affect the survival in these cancer types.
There are some limitations in our studies. Firstly, these findings should be interpreted cautiously because of small sample size. Secondly, we don't have the postdiagnosis weight loss data, which is a better evidence to demonstrate the impact of large energy store on cancer mortality. In conclusion, our data demonstrate that postdiagnosis obesity might indicate a lower cancer mortality.

Patient samples
Clinical and survival data for 2670 cancer patients from 14 tumor types were obtained from the TCGA data portal (https://tcga-data.nci.nih.gov/tcga/) on July 24, 2015. The study was approved by the TCGA project manager and the Institutional Review Board of the Xi'an Jiaotong University.

Clinical and follow-up data collection
Patient clinical data, including age, gender, height, weight and tumor stage were collected. Follow-up data such as vital status and survival time were also recoreded. We only included the patients with available BMI data.

Statistical analysis
All data were analyzed by the R version 3.2.0 software and the SPSS version 19.0 software (SPSS, Chicago, IL, United States). Survival curves were constructed by the Kaplan-Meier method and compared by the log-rank test. The BMI cutoff points were determined by the receiver operating characteristic (ROC) curve analysis. Cox regression analysis was performed to evaluate the prognostic significance of the variables. Results are given as mean ± SD. P < 0.05 was considered statistically significant.