Racial/ethnic disparities in weight or BMI change in adulthood and pancreatic cancer incidence: The multiethnic cohort

Abstract Introduction Compared to non‐Hispanic Whites, Japanese Americans, Native Hawaiians, and African Americans have higher incidences of pancreatic cancer (PCa) that are not entirely explained by rates of obesity but may be explained by weight changes throughout adulthood. Methods The multiethnic cohort is a population‐based prospective cohort study that has followed 155,308 participants since its establishment between 1993 and 1996. A total of 1,328 incident cases with invasive PCa were identified through 2015. We conducted separate multivariable Cox proportional hazards models for self‐reported weight‐change and BMI‐change (age 21 to cohort entry) to determine the association with PCa risk, adjusting for potential confounders including weight or BMI at age 21. Results The mean age at cohort entry was 59.3 years (SD 8.9). An increased risk of PCa was associated with: 1) weight (HR per10 lbs = 1.06; 95% CI = 1.03–1.09) or BMI (HR per kg/m2 = 1.04; 95% CI = 1.02–1.05) at age 21; and 2) weight (HR per 10 lbs = 1.03; 95% CI = 1.01–1.05) or BMI (HR = 1.02; 95% CI = 1.00–1.03) at cohort entry. We found increased risk of PCa between weight (HR per 10 lbs = 1.03; 95% CI = 1.01–1.05) and BMI (HR per 5 kg/m2 = 1.08; 95% CI = 1.01–1.15) change from age 21 to baseline. There were significant interactions between race/ethnicity and weight (p = 0.008) or BMI (p = 0.03) at baseline, and weight (p = 0.02) or BMI (p = 0.02) change. Weight and BMI change through adulthood significantly increased the risk of PCa for Japanese Americans and Latinos, but not for African American, White, or Hawaiian participants. Conclusion Our findings indicate that weight or BMI gain has a significant and independent impact on PCa risk, specifically among Latinos and Japanese Americans.

and potential preventive measures. 3 Evidence suggests that risk factors for PCa include cigarette smoking, non-O ABO blood group, chronic pancreatitis, long-term diabetes mellitus, obesity, uncommon high-penetrance germline mutations, and common low-risk single nucleotide polymorphisms. [4][5][6][7][8][9][10][11][12][13] Epidemiologic evidence from observational studies, including meta-analysis and pooled analysis, have consistently shown body mass index (BMI) in older adulthood to be a risk factor for PCa in a dose-dependent manner. [14][15][16][17] When BMI in adolescence or early adulthood was examined, studies have also shown it to be associated with PCa. 16,[18][19][20][21][22][23] However, it remains unclear how adult weight and BMI gain influence PCa risk. While past studies of these variables have been more limited with few PCa cases, most have shown a nonsignificant increased risk between greater weight or BMI change from adolescence/early adulthood to older adulthood and PCa. 19,24,25 There are also striking racial/ethnic differences in PCa incidence in the United States that are not explained by rates of obesity. 26 Native Hawaiians (RR: 1.60, 95% CI: 1.30-1.98), Japanese Americans (1.33, 1.15-1.54), and African Americans (1.20, 1.01-1.42) have a 60%, 33%, and 20% higher risk of developing PCa compared to non-Hispanic Whites. While increased weight or BMI has been associated with risk of PCa within different populations, no individual study has compared the association between weight or BMI and risk of PCa among different racial/ethnic US populations. 23,27,28 To further understand the relationship between weight or BMI and risk of PCa, the association between multiple weight or BMI measurements in adulthood and risk of PCa were examined in a diverse multiethnic cohort (MEC). The goal of this study was to examine the relationship between early adulthood weight and weight change (early adulthood and later in life) and incidence of PCa in the MEC, overall, and by sex and race/ethnicity.

| Study design and participants
A prospective cohort analysis was conducted among patients enrolled in the MEC. The MEC was established in 1993-1996 to investigate cancer etiology. It is comprised of over 215,000 participants between the ages of 45 and 75 at cohort entry, who were recruited from Los Angeles County and Hawaii. The five main ethnic groups represented in the MEC are White, African American, Latino American, Japanese American, and Native Hawaiian. 29 All participants completed a self-administered epidemiological baseline questionnaire, which included information on demographics, medical conditions, family history of cancer, and lifestyle factors.
Individuals were excluded from this study if they were not of the five main race/ethnicity groups (N = 13,987), had missing BMI at baseline (N = 2,237), missing BMI at age 21 (N = 11,604), missing smoking status (N = 1,734), missing education status (N = 713), missing diabetes status (N = 2), missing alcohol intake (N = 6,948), missing vigorous physical activity (N = 7,166), invalid entry and exit dates (N = 6), had prevalent cancer at baseline (N = 14,914), had a reported BMI at baseline outside of 15-50 kg/m 2 or had a BMI at age 21 outside of 15-50 kg/m 2 . In total, there were 155,308 eligible MEC participants for this analysis.

| Independent study variables
The baseline epidemiological questionnaire assessed weight and height at cohort entry and asked respondents to recall weight at age 21. Body mass index (BMI) was calculated as weight (in kilograms) divided by height 2 (in meters). In order to account for variations in the time that elapsed from age 21 to age at cohort entry, variables called weight change and weight change rate from age 21 to cohort entry and BMI change and BMI change rate from age 21 to cohort entry were created. Weight change from age 21 to cohort entry was created by subtracting weight at age 21 from weight at cohort entry. Weight change rate from age 21 to cohort entry was created by dividing the weight change from age 21 to cohort entry by the number of years elapsed between age 21 to cohort entry, which represents the average weight (lbs) change per year. In a similar manner, BMI change from age 21 to cohort entry was created by subtracting BMI at age 21 from BMI at cohort entry. BMI change rate from age 21 to cohort entry was created by dividing the BMI change from age 21 to cohort entry by the number of years elapsed between age 21 to cohort entry, which represents the average BMI (kg/ m 2 ) gain per year. Weight and BMI were modelled in four different ways: a continuous variable at age 21, a continuous variable at baseline, a continuous variable for change from age 21 to entry into MEC cohort, a continuous variable for change rate from age 21 to entry into MEC cohort and categories of weight/BMI change, similar to other studies. 30

| Covariates
We included covariates from data that were available from the baseline survey for age at cohort entry, sex (male, female, in overall, and race/ethnicity stratified models), race (African American, Japanese American, Latino, White, Native Hawaiian, in overall and sex stratified models), education (high school graduate or less, some college or technical school, college graduate, graduate and professional school), smoking status (never, former, current), alcohol intake (0, <12 g/day, ≥12 g/day), vigorous physical activity (hours/ day), Health Eating Index 2010 (quartiles), diabetes at cohort entry (yes, no), and family history of pancreatic cancer (yes, no). The baseline dietary questionnaire was used to develop a Health Eating Index 2010 that captures the key nutrient and food group recommendations of the 2010 Dietary Guidelines and is used to assess the diet quality of the US population. 32

| Statistical analysis
Descriptive distributions were examined in the overall cohort, as well as stratified by weight change and BMI change from age 21 to cohort entry categories for descriptive purposes (>−10 lbs, −10 to +10 lbs, +10 to +30 lbs, +30 to +50 lbs, ≥50 lbs and <0 kg/m 2 , 0 to 5 kg/m 2 , 5 to 10 kg/m 2 , ≥15 kg/m 2 , respectively). The chi-square test was used to compare categorical variables and the analysis of variance (ANOVA) test was used to compare continuous variables. multivariable cox proportional hazards (PH) analysis was used to calculate hazard ratios (HRs) and their 95% confidence intervals (95% CIs) as independent predictors of pancreatic cancer diagnosis. Models were adjusted for all potential confounders previously listed. Weight at age 21 (continuous) was adjusted for in the weight at baseline and weight change models. BMI at age 21 (continuous) was adjusted for in the BMI at baseline and BMI change models. The analyses were stratified by sex and race/ethnicity to assess whether the influence of absolute weight or BMI and weight or BMI change on pancreatic cancer risk varied by sex or racial/ethnic groups. We tested for an interaction by including a multiplicative variable to the regression model for the BMI and weight categories and race/ethnicity or sex. We also tested for significant interactions between smoking status and the weight or BMI variables using product terms. All statistical analysis was performed using SAS version 9.4 (North Carolina) and reported p-values are based on two-sided tests (summary alpha = 0.05). Values of p < 0.05 were used to define statistical significance. Smoking and diabetes status were found to violate the PH assumption in all models and were corrected by adding these variables to the STRATA command in SAS.

| DISCUSSION
The relationships between weight or BMI at age 21 and at cohort entry and change from age 21 to cohort entry and risk of PCa were examined in the MEC, overall, and by sex and race/ethnicity. Overall, after controlling for potential confounding variables, our study shows a positive association between weight and BMI variables and risk of PCa. To our knowledge, this is the largest prospective cohort to examine weight or BMI change throughout adulthood and risk of PCa, the first study to examine racial/ethnic-specific weight or BMI variables in a diverse racial/ethnic population, and the largest study to examine weight or BMI change variables among large cohorts of Japanese Americans, Native Hawaiians, and Latinos.
In our study, we did not observe significant differences in PCa risk by weight or BMI variables by sex. However, we did find that weight and BMI changes, change rate, and change categories were more strongly associated with PCa risk for men than for women. Nothings and colleagues 14 examined the relationship between BMI in older adults and risk of PCa in an earlier analysis of the MEC study with follow-up to 2002 and 237 cases and found that while the interaction between sex and BMI was not significant (p = 0.09), in men obesity (BMI≥30 kg/m 2 ) was associated with an increased risk of PCa (HR = 1.51; 95% CI = 1.02-2.26), but in women it was associated with a reduced risk (HR = 0.65; 95% CI = 0.43-0.99). 14 With 12 additional years of follow-up and 1,091 more incident PCa cases in our current study, the association between weight or BMI variables and risk of PCa remained generally stronger in men than women, especially for weight or BMI change rate, but a reduced risk of PCa for women was no longer apparent. While several other studies have also revealed stronger relationships between weight or BMI variables in men compared to women, sex is not considered as an established modifier of the relationship between weight or BMI and risk of PCa, and the mechanism of this possible modification has not yet been explored. 23,27,28,33,34 When we stratified our analysis by race/ethnicity, weight or BMI variables were positively associated with PCa risk among Japanese Americans and Latinos, but there was no consistent association for African Americans, Whites, or Native Hawaiians. The relative distribution of body fat is known to differ by race/ethnicity, with Latinos storing the highest amount as trunk fat and Japanese Americans storing the highest amounts as visceral fat, which may partially explain our differing findings by race/ethnicity. 35 While is it unknown if the location of adiposity modifies the association between total adiposity and risk of PCa, several studies highlight the importance of exploring this potential modifier. For example, recent studies have shown that visceral fat is more strongly associated with risk of cancers such as colorectal and breast, compared to total adiposity. 36 To our knowledge, our study is the first to examine weight or BMI change and risk of PCa in a large MEC study. Only one other large study in African Americans and one in Japanese have examined the relationship between absolute BMI in adulthood and PCa. A large pooled analysis of various African-American studies that included 29,306 MEC subjects (140 PCa deaths), found a significant positive association between older adult BMI and death from PCa overall (P trend = 0.03), but no significant association between older adult BMI and PCa death when limited to MEC subjects (HR = 1.11, 95% CI = 0.95-1.30). 27 A large pooled analysis of Japanese in Japan showed an association between BMI in older adults and risk of PCa in females (HR per 1 kg/m 2 = 1.02, 95% CI = 1.00-1.05), but not in males (HR per 1 kg/m 2 = 0.97, 95% CI = 0.94-1.01). 23 However, in our MEC study, we did not find a significant interaction between sex and BMI but we did find that BMI change from adulthood to cohort entry was significantly associated with PCa risk for males but not females. Further, we found that BMI and BMI change and weight and weight change was more strongly significantly associated with PCa risk among Japanese Americans than the other racial/ethnic groups included in our study.
The strengths of our analysis study include the large ethnically and racially diverse study population. There are some limitations to consider in interpreting these findings. First, weight at age 21 and baseline were self-reported by subjects (not directly measured) and therefore subject to misclassification bias if participants did not accurately report weight at age 21. Second, although our cohort was large, the highest BMI change category (≥15 kg/m 2 ), did not have enough subjects so the power to detect an association with risk of PCa is insufficient, especially when stratified by sex and race/ ethnicity. Third, despite the fact that we controlled for a number of known and suspected confounders such as smoking history, history of diabetes, and family history of pancreatic cancer, there could still be residual or unmeasured confounding that we could not account for. Finally, there is a possibility of reverse causality due to disease-related weight change, however, we conducted a sensitivity analysis excluding 118 patients diagnosed within three years of cohort entry and our findings remain unchanged.
In summary, our findings suggest that weight or BMI at age 21 and baseline and weight or BMI change are important risk factors for PCa. While there were no differences in weight or BMI variables when stratified by sex, there were significant disparities in weight or BMI variables when stratified by race/ethnicity, with only Japanese Americans and Latinos having positive associations between BMI or weight variables and risk of PCa.