Of the 67,947 subjects, 0.8% were underweight, 34.8 were normal, 43% were overweight, 16.7% were moderately obese (obese I) and 4.7% were highly obese (obese II/III) (). Compared to normal weight subjects, underweight subjects were more likely to be women, young (<40 years) or old (≥70 years), current smokers, never alcohol drinkers, non-hypertensives, high fruit consumers (≥3 servings/day), and do no or little exercise. In contrast, compared to normal weight subjects, overweight or obese subjects were more likely to be men, middle-aged (40–69), lower educated (high school or less), former smokers, hypertensive, diabetic, low fruit and vegetable consumers, not take vitamin supplements, and do no or little exercise.
Distribution of Study Population by BMI categories and Selected Characteristics
We found a statisitcally significant positive linear trend for BMI and colon cancer among men (HR 1.05, 95% CI 1.02–1.09; p-trend=0.005) adjusting for race, education, and family history of colon cancer (). Overweight men had a 1.26-fold (95% CI=0.86–1.86), obese I men a 1.88-fold (95% CI =1.23–2.91), and obese II/III men a 2.03-fold 95% CI=1.05–3.93) risk compared to normal weight men. For rectal cancer, obese II/III men had a 3.21-fold (95% CI=1.34–7.71) risk compared to normal weight men adjusting for meat consumption, but the test of trend was not statistically significant. We also observed increased, but not statistically significant, risks of stomach and bladder cancers with higher BMI. In contrast, the risks of lung and oral cancers decreased with higher BMI. Stratifying by smoking status, we found a statistically significant test of trend for lung cancer among ever smokers (HR=0.93, 95% CI=0.88–0.97, p-trend=0.001) adjusting for state, race, vegetable consumption, exercise, and pack-years of cigarette smoking. There were only 21 men with lung cancer who never smoked, thus we could not completely calculate risk estimates for this group or the test of interaction between smoking status and BMI. In order to rule out the possibility of weight loss due to preclinical disease, we restricted the cases of lung cancer among ever smokers to those diagnosed 2 or 5 years after enrollment and found that the inverse association with BMI remained (2 years: HR=0.94, 95% CI=0.89–0.99, p-trend=0.01; 5 years: HR=0.93; 95% CI=0.88–0.99, p-trend=0.02).
Hazard Rate Ratio (HR) and 95% Confidence Intervals (CI) for cancers in relation to BMI at enrollment by sex
Among women, we found statistically significant positive linear trend for BMI and all cancer sites combined (HR=1.02, 95% CI=1.01–1.04; p-trend=0.001) adjusting for smoking status, hypertension, taking vitamin supplements, and parity; and specifically for postmenopausal breast cancer (HR=1.03, 95% CI =1.01–1.06; p-trend=0.02) adjusting for diabetes, vitamin supplement, parity, and family history of breast cancer. For breast cancer, overweight postmenopausal women had a 1.22-fold (95% CI 0.93–1.60), obese I a 1.62-fold (95% CI 1.17–2.24), and obese II/III a 1.07-fold (95% CI 0.61–1.88) risk compared to normal weight postmenopausal women. To explore the null association for breast cancer in obese II/III women, we compared the frequency of the factors in between the obese I and II postmenopausal women using the chi-square test, and found that only alcohol consumption was significantly different (p=0.02), with 50% of the obese II versus 26% of the obese I women ever drinking alcohol. Stratifying by alcohol consumption, the linear trend for BMI and breast cancer among postmenopausal women was statistically significant among those who never drank alcohol (HR=1.05, 95% CI=1.02–1.08, p-trend=0.003), but not among those who drank alcohol (HR=1.00, 95% CI=0.95–1.04, p-trend=0.85) (p-interaction=0.01). In addition to breast cancer, women with higher BMI had increased, but not statistically significant, risks of pancreatic and kidney cancers, and melanoma. BMI was not associated with colon cancer in women (HR=0.99, 95% CI 0.95–1.04). Also, the risk of lung cancer decreased with increasing BMI, but the test of trend was not statistically significant for all women, ever smokers, or those diagnosed 2 or 5 years after enrollment.
For the interaction between pesticide use and BMI on colon cancer risk in men, we evaluated 22 pesticides that were used by at least 10% of the male study population and for which there were at least 5 cancer cases in each BMI group, as well as any organochlorine or organophophate insecticide (). Of these 22 pesticides, carbofuran, metolachlor and alachlor had statistically significant (p≤0.05) modifying effects. Specifically, BMI was significantly associated with colon cancer among male users of carbofuran (HR=1.10, 95% CI 1.04–1.17, p-trend=0.002), but not among non-users of carbofuran (HR=1.02, 95% CI 0.97–1.06, p-trend=0.52). Similarly, BMI was significantly associated with colon cancer among male users of metolachlor (HR=1.09, 95% CI 1.04–1.15, p-trend=0.001) and alachlor (HR=1.08, 95% CI 1.03–1.13, p=trend=0.002), but not among non-users of each of these herbicides (No metolachlor: HR=1.01, 95% CI 0.96–1.06, p-trend=0.70; No alachlor: HR=1.01, 95% CI 0.95–1.06, p-trend=0.87). To explore possible confounding by use of multiple pesticides, we examined the correlation between never/ever use of carbofuran, metolachlor and alachlor among male participants and found that they were not highly correlated (metolachlor and alachlor r=0.31, metolachlor and carbofuran=0.18, and alachlor and carbofuran=0.26). We also examined whether each of these pesticides still had statistically significant modifying effects when we further adjusted for use of the other two pesticides, and found that the test of interaction was still statistically significant (<0.05). In addition to these three pesticides, we also found that men who used DDT, fonofos, malathion, carbaryl, permethrin, atrazine, cyanazine, trifluralin, EPTC, imazethpyr, and glyphosate had significant associations between BMI and colon cancer, but non-users of these pesticides did not, and the tests of interaction were not statistically significant. Furthermore, we also observed that men who did not use terbufos, chlorpyrifos, metribuzin, pendimethalin, chlorimuron-ethyl, and methyl bromide had significant associations between BMI and colon cancer, but users of these pesticides did not, and tests of interaction were not statistically significant.
Hazard Rate Ratio (HR) and 95% Confidence Intervals (CI) for Colon Cancer in relation to BMI at enrollment by ever/never use of pesticides among men
We evaluated the interaction between pesticide use and BMI on postmenopausal breast cancer for four pesticides used by at least 10% of the women in the study population and for which there were at least 5 cancer cases in each BMI group, as well as any organochlorine or organophosphate insecticide. There was no evidence of a significant interaction between pesticide use and BMI on breast cancer risk among postmenopausal women; however, women who did not use malathion, diazinon, carbaryl or any organochlorine insecticide or any organophosphate insecticide had significant associations between BMI and breast cancer, but users did not. The significant associations among non-users of these pesticides remained statistically significant among women who did not drink alcohol, but not among women who drank alcohol.
We also examined the interactions between pesticide use and BMI on the risks of lung, bladder, prostate, and rectal cancers, and NHL and melanoma in men, as well as colon cancer in women. For lung cancer among male ever smokers, we found that the inverse effect of BMI was statistically significant among non-users of carbofuran (HR=0.87, 95% CI=0.82–0.92, p-trend=<0.0001), but not among users (HR=1.01, 95% CI=0.93–1.09; p-trend=0.77) (p-interaction=0.02). We did not find any other statistically significant modifying effects on any of the other associations examined.