Cancer Causes & Control

, Volume 19, Issue 10, pp 1051–1064

A summary measure of pro- and anti-oxidant exposures and risk of incident, sporadic, colorectal adenomas

Authors

    • Department of Epidemiology, Rollins School of Public HealthEmory University
  • Roberd M. Bostick
    • Department of Epidemiology, Rollins School of Public HealthEmory University
  • Chiranjeev Dash
    • Department of Epidemiology, Rollins School of Public HealthEmory University
  • Paul Terry
    • Department of Epidemiology, Rollins School of Public HealthEmory University
  • W. Dana Flanders
    • Department of Epidemiology, Rollins School of Public HealthEmory University
  • Jack Mandel
    • Department of Epidemiology, Rollins School of Public HealthEmory University
Original Paper

DOI: 10.1007/s10552-008-9169-y

Cite this article as:
Goodman, M., Bostick, R.M., Dash, C. et al. Cancer Causes Control (2008) 19: 1051. doi:10.1007/s10552-008-9169-y

Abstract

Despite compelling basic science evidence, the search for causal associations linking specific pro- and anti-oxidants to presumably oxidative stress-related neoplasms, such as colorectal adenoma, has produced inconsistent results. We developed an oxidative balance score (OBS) to characterize the pro-oxidant and anti-oxidant exposures of 2,305 participants in a case–control study of colorectal adenoma that used both endoscopy-confirmed and community controls. Twelve lifestyle medical and dietary factors with known pro- or anti-oxidant properties were considered. Each high anti-oxidant exposure and low pro-oxidant exposure was awarded one or two points depending on the level of exposure, and the points for each OBS component were summed. We observed a significant inverse association between OBS (continuous variable) and colorectal adenoma in the analyses with either community, or endoscopy controls (each p-trend < 0.01). When the OBS was treated as an ordinal variable and a score of ≤3 points was used as the referent category, in the analyses with the endoscopy controls the adjusted odds ratios for scores of 4–6, 7–9, 10–12, 13–15, and 16+, were 0.42, 0.32, 0.22, 0.20, and 0.19, respectively, with all 95% confidence intervals excluding 1.0. The corresponding analysis for community controls showed a similar trend. Our findings are in line with the basic science evidence supporting the role of oxidative stress in colorectal neoplasia.

Keywords

CancerColorectal adenomaOxidative stressScore

Introduction

The role of reactive oxygen species (ROS) as both initiators and promoters of colon carcinogenesis is supported by a considerable body of basic science literature [16] The degree and extent of ROS-induced damage (including damage to the colon epithelium) depends on the relative balance between pro- and anti-oxidant factors that modulate levels of potentially harmful ROS [7].

Although the pro-/anti-oxidant balance is determined to a large extent by endogenous enzymatic mechanisms, it is also affected by modifiable exogenous factors such as diet, medications, and lifestyle [8]. Tobacco smoke is considered a powerful exogenous pro-oxidant since high concentrations of ROS are present in both its tar and gas phases [9]. The direct increase in the oxidative burden of inhaled tobacco smoke can be further enhanced in the lungs through the secondary release of oxygen radicals from inflammatory cells [10]. Long-chain polyunsaturated fatty acids (PUFA) are well-documented direct contributors to oxidative stress through increased lipid peroxidation [11, 12]. Iron consumption is hypothesized to intensify oxidative stress in the colon by catalyzing the production of highly reactive hydroxyl radicals via the Haber-Weiss reaction [13]. Diet can also be an important source of anti-oxidants. In vitro data demonstrate that certain micronutrients, most notably fat-soluble carotenoids and tocopherols, act as scavengers of ROS and reduce mutagenicity in the Ames test [1416].

Another major contributor to oxidative stress is inflammation. Inflammatory cells respond to stimuli (e.g., microbial agents) by producing a variety of toxic compounds, such as superoxide, hydrogen peroxide, singlet oxygen, as well as nitric oxide, which can react further to form highly reactive peroxynitrite molecules [8]. These compounds directly interact with DNA in the host cells, or react with other cellular components such as lipids, initiating free radical chain reactions.

Several animal and in vitro studies demonstrated the effects of pro- and anti-oxidants on epithelial cell proliferation in colon tissue [1720]. For example, an anti-oxidant mixture primarily containing α-tocopherol reduced cell proliferation in the colon and rectum of mice [19]. Organic and inorganic selenium reduced colonic epithelial cell proliferation, while concomitantly reducing the incidence and multiplicity of colon adenocarcinomas in rats [20]. Several small trials in humans suggested that anti-oxidants can reduce colorectal epithelial cell proliferation [21, 22].

Despite compelling evidence from mechanistic studies, epidemiological results concerning the associations between specific determinants of oxidative stress and either colon cancer or its precursor, colorectal adenoma, have been inconsistent [2325]. Experience in other areas of chronic disease epidemiology shows that a combination of several risk factors may reveal an overall large increase in risk, even when associations with each individual factor are relatively weak and inconsistent [2628]. These findings suggest the potential value of epidemiologic approaches that account for a variety of factors acting (and interacting) along the same etiologic pathway.

Recently, we proposed the use of an oxidative stress score (OSS) as a measure of combined pro- and anti-oxidant exposures [29]. Using the data from two previously conducted studies, a population-based case–control study of prostate cancer, and a clinic-based case–control study of colorectal adenoma, we selected variables that are known to indicate pro- and anti-oxidant exposures. We then combined these variables into a single score, where higher values indicated a predominance of anti-oxidant exposures and lower values indicated a predominance of pro-oxidant exposures. In these analyses, we found a substantial decrease in risk associated with a high OSS for both prostate cancer and colorectal adenoma, lending support to our a priori hypothesis that persons with predominantly anti-oxidant exposures have lower risk than persons with predominantly pro-oxidant exposures. In the present study we attempt to confirm and extend our previous findings by applying a slightly modified approach in a larger case–control study of colorectal adenoma. In this paper we use the term ‘oxidative balance score’ (OBS) as proposed by van Hoydonck et al. [30] to reflect the hypothesized beneficial effect of higher score values.

Methods

Study population

The case–control study of incident, sporadic colorectal adenomas that provided data for the current analysis was conducted between April 1, 1991 and April 1, 1994 at the University of Minnesota in collaboration with Digestive Healthcare, a large multi-site gastroenterology practice. The protocol was approved by the Institutional Review Board of the University of Minnesota and all participating Digestive Healthcare sites [31].

The eligibility requirements for the study included residence in the Minneapolis-St. Paul metropolitan area, age between 30 and 74 years, sufficient command of the English language to respond to questionnaires; willingness to participate and ability to understand informed consent, no contraindications to endoscopy, no history of a previous colorectal adenoma; no known genetic syndromes associated with colonic neoplasia (e.g., familial polyposis coli or Gardner’s syndrome), and no history of ulcerative colitis or Crohn’s disease.

Potential participants were recruited over 36 months through the usual scheduling of endoscopies (sigmoidoscopies or colonoscopies) by practice staff. If a polyp was found on a sigmoidoscopy, all patients were referred for a colonoscopy. If, however, sigmoidoscopy was found to be completely normal, no referral for colonoscopy was deemed necessary. Some patients underwent colonoscopy without a preceding sigmoidoscopy. If polyps were confirmed or found for the first time during a colonoscopy, hospital pathologists supplied histologic information using a standardized study data collection form. The pathology slides, regardless of the original pathologist’s reading, were re-reviewed by a study index pathologist using criteria developed for the National Polyp Study [32]. Only the readings by the study index pathologist were used to assign case status.

The study included three categories of subjects: (1) cases were patients who underwent a colonoscopy (with or without previous sigmoidoscopy) and were found to have a pathology-confirmed, incident adenomatous polyp of the colon and/or rectum; (2) endoscopy controls were patients who underwent sigmoidoscopy or colonoscopy and were found to be free of any polyps of any type (e.g., adenomatous, hyperplastic); and (3) community controls were individuals selected from the Minnesota state drivers’ license registry and frequency matched to cases according to 5-year age intervals, sex, and zip code. The community controls were contacted by telephone and recruited using the same criteria previously established for the endoscopy patients. The participation rate was 68% among all endoscopy patients (cases and controls) and 65% among community controls. The final sample size included 574 adenoma cases, 1,227 endoscopy (707 colonoscopy and 520 sigmoidoscopy) controls, and 550 community controls.

Data collection

All eligible patients were asked to complete mailed medical and dietary questionnaires prior to their endoscopy visits. Dietary and nutritional supplement data were obtained using a previously validated, modified 153-item Willett Food Frequency Questionnaire (WFFQ) [33, 34], which included additional questions about vegetables, fruit, and other low-fat food items. This semi-quantitative instrument covered usual food intake by including questions about frequencies of use for various foods over the previous 12 months. The questionnaire also elicited information on vitamin and mineral supplement use. Dietary data were converted to nutrient intake measures using the dietary database developed in the Channing Laboratory [33]. As lycopene and lutein intake values were not available, the number of servings per week for foods rich in lycopene (tomatoes, tomato paste, and watermelon) and lutein (broccoli, peas, and spinach) was used instead [31].

The validity of WFFQ was assessed in several previously published studies, which compared self-reported carotene and vitamin E intakes to measured plasma carotenoid and α-tocopherol concentrations. The correlation coefficients for dietary carotenoids versus plasma α- and β-carotene were reported to be in the 0.3–0.5 range, whereas the corresponding correlations for vitamin E intake versus plasma α-tocopherol ranged from 0.4 to 0.5 [33, 35, 36]. Similar studies comparing WFFQ-derived intakes of fatty acids to the analyses of adipose tissue reported correlation coefficients ranging from 0.2 to 0.5 [37, 38]. Additional relevant data obtained from the questionnaires included demographics, family history, smoking habits, use of medications such as aspirin and other non-steroidal anti-inflammatory drugs (NSAIDs), as well as intakes of alcohol and dietary supplements.

Oxidative balance score

The oxidative balance score (OBS) was calculated by combining information from twelve factors selected a priori, which are summarized in Table 1. The twelve factors represent the relevant variables available in the dataset (other studies with different available data may necessarily require a slightly different set of factors, especially as new information about other relevant factors emerges).
Table 1

Oxidative balance score (OBS) assignment schemea

1. Smoking history

0 = Current smoker, 1 = former smoker, 2 = never smoker

2. Selenium supplement use

0 = Never, 1 = ≤2 years of use, 2 = 3+ years of use

3. Regular aspirin use

0 = Never, 1 = ≤2 years of use, 2 = 3+ years of use

4. Regular other NSAIDb use

0 = Never, 1 = ≤2 years of use, 2 = 3+ years of use

5. Alcohol consumption

0 = 8+ Drinks/week, 1 = <1 drink/week, 2 = 1–7 drinks/week

6. PUFAb intake

0 = High (3rd tertile), 1 = medium (2nd tertile), 2 = low (1st tertile)

7. Total (dietary and supplemental) iron intake

0 = High (3rd tertile), 1 = medium (2nd tertile), 2 = low (1st tertile)

8. Total (dietary and supplemental) vitamin C intake

0 = Low (1st tertile), 1 = medium (2nd tertile); 2 = high (3rd tertile)

9. Total (dietary and supplemental) carotene intakec

0 = Low (1st tertile), 1 = medium (2nd tertile); 2 = high (3rd tertile)

10. Total (dietary and supplemental) vitamin E intake

0 = Low (1st tertile), 1 = medium (2nd tertile); 2 = high (3rd tertile)

11. Intake of lutein-rich foods

0 = Low (1st tertile), 1 = medium (2nd tertile); 2 = high (3rd tertile)

12. Intake of lycopene-rich foods

0 = Low (1st tertile), 1 = medium (2nd tertile); 2 = high (3rd tertile)

aLow, medium and high categories correspond to sex-specific 1st, 2nd, and 3rd tertiles among community controls

bNSAID = Non-steroidal anti-inflammatory drug (not including aspirin); PUFA = Polyunsaturated fatty acid

cVariable ‘carotene’ represents total intake of plant-derived pro-vitamin A (primarily β-carotene)

Categorical variables were assigned scores from 0 to 2 based on exposure levels selected a priori. Smoking status was categorized as never (2 points), former (1 point), or current (zero points). For selenium supplements, NSAIDs and aspirin, zero points were assigned to participants who reported no regular use, one point to those who reported regular use for less than 3 years, and two points to those who reported regular use for 3 years or longer.

Regular use was defined in the questionnaire as taking medication at least once a week. We used number of years rather than dose for anti-inflammatory medications and selenium supplements for three reasons: (1) more selenium supplement takers reported duration of use than dose; (2) for aspirin and NSAID use we had information on frequency of use, but not dose; and (3) recent evidence indicates that for NSAIDs and, even more so for aspirin, duration of use is at least as important as dose [3941]. The cutoff of 3 years was chosen in order to divide NSAID, aspirin, and selenium users into two groups of sufficient and reasonably comparable size. Among participants whose responses regarding selenium, aspirin, or other NSAIDs were incomplete, individuals who reported regular use of unspecified duration received one point and those who left the questions blank received zero points. The proportions of participants that left their responses blank were 2% for selenium, 19% for aspirin, and 19% for other NSAIDs. We did not use total (dietary and supplemental) selenium in the score because questionnaire-based selenium intake estimates are known to be inaccurate as the amount of selenium in the diet depends primarily on its content in the soil [42, 43].

Alcohol intake was categorized based on current or previous usual consumption as follows: eight or more drinks/week were assigned 0 points, 1–7 drinks/week were assigned 2 points, and one or fewer drinks/week were assigned 1 point. These categories were based on reports in the literature, which indicate that moderate alcohol drinkers have lower levels of oxidative stress and inflammation markers compared to non-drinkers and heavy drinkers [4447].

All continuous dietary variables were adjusted for total energy intake as proposed by Willett and Stampfer [48]. Variables reflecting pro-oxidant exposures, such as PUFA and saturated fat and iron intakes, were divided into high, medium, and low categories based on the sex-specific tertile values among the community controls. Participants whose exposure to a particular pro-oxidant was low (1st tertile) were assigned two points, and those whose exposure to the same pro-oxidant was medium (2nd tertile) or high (3rd tertile) received one or zero points, respectively. For anti-oxidant exposure variables (e.g., vitamin E and carotenoid intakes), two points were awarded for each high-level (3rd tertile) exposure, and one or zero points were awarded for each medium- (2nd tertile), or low-level (1st tertile) exposure. The decision to use tertiles in the analyses of continuous variables was based on the need to maintain consistency with categorical variables so that all factors under study had the same range of possible scores.

The points assigned for each of the twelve OBS components were then added to calculate the overall score. Thus, the possible OBS values ranged from a minimum of 0 points to a maximum of 22 points, where higher values indicate a predominance of anti-oxidant exposures, and lower values indicate a predominance of pro-oxidant exposures.

Compared to our previous publication [29] the current version of the OBS has three notable changes. We followed the recommendation of Mayne et al. that only polyunsaturated fatty acids rather than saturated fat should be included in the score [49]. Alcohol use was added to the score as suggested by an anonymous peer reviewer. Unlike our previous publication, we did not include beta-cryptoxanthin intake because it was not estimated in this study.

Statistical analyses

The associations between colorectal adenoma risk and OBS were examined using logistic regression models [50] to calculate odds ratios (ORs) and corresponding 95% confidence intervals (CIs). All models controlled for multiple a priori selected risk factors, which included age, race, sex, hormone replacement therapy (HRT), education, history of colorectal cancer in one or more first degree relatives, total energy intake, BMI, weekly metabolic equivalent task (MET) hours of physical activity, alcohol intake, and energy adjusted intakes of calcium, vitamin D, dietary fiber, red meat, and folic acid. Because hormone replacement therapy only applies to women, sex and HRT were combined in a single categorical variable that had separate values for men, women receiving HRT, and women not receiving HRT. In all models, OBS was treated as either a continuous or an ordinal variable with all categories representing an approximately equal interval. All models were examined for interaction based on likelihood ratio tests. Analyses were performed using SPSS 11.5 for Windows (LEAD Technologies, Inc., Chicago IL) and SAS 8.02 for Windows (SAS Institute, Inc., Cary, NC) statistical software packages.

To examine the impact of scoring assumptions, extreme observations and individual pro- and anti-oxidant exposures on the association between the OBS and colorectal adenoma risk, we conducted a series of sensitivity analyses through the use of alternative models and alternative case and control definitions. In these analyses we excluded participants who left questions about selenium, aspirin, and NSAID use blank, and we also used dose (for selenium) and frequency of use (for aspirin and other NSAIDs) instead of duration. We then evaluated the role of extreme observations by excluding participants who had the highest and lowest OBS values. We compared results of the analyses that used all endoscopy controls to the results for sigmoidoscopy and colonoscopy controls examined separately. Next, we used alternative subcategories of cases defined based on different adenoma characteristics, including location, size, multiplicity, and degree of atypia. We also removed each OBS component from the overall score individually as well as various groups of OBS components and instead adjusted for those components by including them as covariates in the model. The associations between colorectal adenoma and the OBS in these alternative models were then compared to the association between colorectal adenoma and the a priori-defined OBS in the original analyses. Finally, we calculated the adjusted ORs and 95% CIs for each individual OBS component to allow a comparison of these estimates with the overall OBS result.

Results

Participants with sufficient data to calculate the OBS included 564 (98%) of the 574 adenoma cases, 535 (97%) of the 550 community controls, and 1,204 (98%) of the 1,227 endoscopy controls. Among 564 cases, 391 (69%) had only one adenoma, and 173 (31%) had two or more adenomas. The size of the largest adenoma ranged from 1 mm to 80 mm with a mean and a median of 9 mm and 6 mm, respectively. With regards to location, just over 50% of all adenomas (394 of 742) were found in the distal colon and 16% were found in the rectum with the remainder located in proximal segments of the large intestine from the cecum through the splenic flexure. When only the largest adenoma was considered, the percentages of rectal, right colon (cecum through splenic flexure), and left colon (descending and sigmoid) lesions were 16%, 24%, and 60%, respectively.

As shown in Table 2, colorectal adenoma patients were similar to both control groups with respect to race, physical activity, and intakes of calcium and dietary fiber, but significantly different with respect to sex distributions, alcohol and red meat consumption, folate intake, multivitamin use, and BMI. The age distribution and education levels among cases were similar to those of the community controls, but were significantly different from those of the endoscopy controls. By contrast, the differences in the percentage of persons with a family history of colorectal cancer in a first degree relative were significant when comparing cases to community (p < 0.01) but not endoscopy (p = 0.07) controls.
Table 2

Selected demographic, clinical, lifestyle, and dietary characteristics of study participants

Variablesa

Cases (n = 564)

Endoscopy controls (n = 1,204)

p-valuesc

Community controls (n = 535)

p-valuesc

Age, years

59 (54–64)

53 (47–59)

<0.01

59 (54–63)

0.65

Sex

    Males

348 (61.7%)

468 (38.9%)

<0.01

295 (55.1%)

0.03

    Females

216 (38.3%)

736 (61.1%)

 

240 (44.9%)

 

Females on HRT

    Yes

83 (14.7%)

371 (30.8)

<0.01

106 (19.8%)

0.25

    No

133 (23.6%)

365 (30.3%)

 

134 (25.1%)

 

Race

    White

551 (97.7%)

1,168 (97.0%)

0.26

520 (97.2%)

0.56

    Non-white

12 (2.3%)

36 (3.0%)

 

15 (2.8%)

 

Education

    High school or less

210 (37.2%)

368 (30.6%)

<0.01

197 (36.8%)

0.90

    More than high school (13+ years)

354 (62.8%)

836 (69.4%)

 

338 (63.2%)

 

1st Degree relatives with CRCd

    Yes

91 (16.3%)

239 (19.9%)

0.07

37 (6.9%)

<0.01

    No

469 (83.8%)

959 (80.1%)

 

497 (93.1%)

 

Multivitamin use

    Yes

127 (22.5%)

420 (34.8%)

<0.001

164 (30.7%)

<0.01

    No

437 (77.5%)

784 (65.2%)

 

371 (69.3%)

 

Energy intake, kcal/days

1,940 (1,698–2,299)

1,897 (1,632–2,194)

0.04

1,979 (1,686–2,288)

0.82

Physical activity, METd h/week

24.4 (15.2–38.7)

25.7 (16.9–36.4)

0.99

25.0 (16.8–41.1)

0.61

Calcium intake, mg/dayb

856.5 (723.4–1,042.2)

883.9 (738.1–1,113.2)

0.23

891.7 (735.5–1,141.0)

0.29

Vitamin D intake, IU/dayb

247.9 (184.2–358.3)

275.1 (200.0–399.6)

0.08

281.5 (196.3–414.7)

0.05

Red meat intake servings/weekb

4.3 (3.3–5.4)

3.9 (3.1–5.1)

0.03

3.9 (3.1–5.1)

0.17

Fiber intake g/dayb

20.9 (18.3–23.8)

21.3 (18.6–24.2)

0.52

21.2 (18.9–24.2)

0.39

Folate intake, mcg/dayb

330.6 (284.8–398.0)

358.0 (298.8–467.4)

<0.01

352.5 (296.4–433.1)

0.02

BMId, kg/m2

26.7 (25.1–28.8)

25.9 (24.1–27.8)

<0.01

26.4 (24.6–28.2)

0.04

aContinuous variables are expressed as median (inter-tertile range), categorical variables are expressed as number (percent)

bAll dietary variables are adjusted for total energy intake using the residual method of Willett and Stampfer [48], and all nutrients are expressed as dietary + (if applicable) supplemental intake

cCompare cases to each control group separately, based on Mann-Whitney test for continuous variables and Fisher’s exact test (excluding missing data) for categorical variables

dCRC = Colorectal cancer, BMI = Body mass index, MET = Metabolic equivalent task

An evaluation of the study participants with respect to individual OBS components (Table 3) demonstrated that cases were more likely to drink alcohol and be current or former smokers than were either the endoscopy or community controls (p-values < 0.01); they were also significantly less likely than endoscopy controls to report regular NSAID use (p < 0.01). Compared to cases, community controls had significantly higher intakes of vitamin E (p < 0.05) and lycopene (p < 0.01), whereas endoscopy controls had a significantly lower intake of PUFA (p < 0.01) and significantly higher intakes of total vitamin C (p < 0.01), vitamin E (p < 0.01), and carotene (p < 0.05).
Table 3

Distribution of individual oxidative balance score (OBS) components among cases and controls

Variablesa

Cases (n = 564)

Endoscopy controls (n = 1,204)

p-valuesc

Community controls (n = 535)

p-valuesc

Smoking

    Never

183 (32.4%)

558 (46.4%)

<0.01

236 (44.1%)

<0.01

    Former

264 (46.8%)

489 (40.6%)

216 (40.4%)

    Current

117 (20.7%)

157 (13.0%)

83 (15.5%)

Selenium supplement use

    None

548 (97.2%)

1,171 (97.2%)

0.99

520 (97.2%)

0.14

    1–2 years

12 (2.1%)

25 (2.1%)

6 (1.1%)

    3+ years

4 (0.7%)

8 (0.7%)

9 (1.7%)

Aspirin use

    None

405 (71.8%)

822 (68.4%)

0.09

377 (70.5%)

0.87

    1–2 years

97 (17.2%)

203 (16.8%)

98 (18.3%)

    3+ years

62 (11.0%)

179 (14.8%)

60 (11.2%)

Other (non-aspirin) NSAID# use

    None

491 (87.1%)

919 (76.4%)

<0.01

439 (82.1%)

0.07

    1–2 years

55 (9.8%)

164 (13.6%)

71 (13.3%)

    3+ years

18 (3.2%)

121 (10.0%)

25 (4.7%)

Alcohol intake

    <1 drink/week

246 (43.6%)

683 (56.7%)

<0.01

280 (52.3%)

<0.01

    1–7 drinks/week

114 (20.2%)

258 (21.4%)

116 (21.7%)

    8+ drinks/week

204 (36.2%)

263 (21.8%)

139 (26.0%)

PUFA intake, gm/dayb

14.3 (13.0–15.5)

12.1 (10.8–13.4)

<0.01

13.8 (12.6–15.2)

0.28

Iron intake, mg/dayb

14.5 (13.4–16.2)

14.6 (12.9–17.3)

0.17

15.4 (13.4–18.1)

0.11

Vitamin C intake, mg/dayb

149.1 (117.1–194.9)

186.6 (145.0–248.3)

<0.01

157.8 (124.5–217.3)

0.17

Total carotene intake, IU/dayb,d

6,598 (5,216–8,528)

8,227 (6,286–10,934)

0.04

6,903 (5,655–9,213)

0.16

Vitamin E intake, mg/dayb

9.0 (7.9–11.3)

11.5 (8.9–17.9)

<0.01

10.0 (8.5–12.7)

0.02

Lutein-rich foods, servings/weekb

5.0 (3.7–6.5)

6.1 (4.5–8.2)

0.12

5.4 (3.9–7.2)

0.13

Lycopene-rich foods, servings/weekb

1.4 (0.9–2.0)

1.5 (1.0–2.4)

0.77

1.8 (1.2–2.9)

<0.01

aContinuous variables are expressed as median (inter-tertile range), categorical variables are expressed as number (percent)

bAll dietary components of OBS are adjusted for total energy intake using the residual method of Willett and Stampfer [48] and all nutrients are expressed as dietary + (if applicable) supplemental intake

cCompare cases to each control group separately, based on Mann-Whitney test for continuous variables and Fisher’s exact test for categorical variables

dVariable ‘carotene’ represents intake of plant-derived pro-vitamin A (primarily β-carotene)

The correlations among individual dietary nutrients included in the OBS are presented in Table 4. The strongest correlations were observed between intakes of vitamin C and vitamin E (r = 0.52) and between intakes of lutein and carotene (r = 0.47). Conversely, the weakest correlations were observed between iron and PUFA (r = −0.04) and between lycopene and iron (r = −0.01). All other correlation coefficients ranged between −0.05 and 0.37.
Table 4

Correlations among individual nutrients included in the oxidative balance score (OBS)

Nutrientsa

Correlation coefficients

Iron

Vitamin E

Carotene

Vitamin C

Lycopeneb

Luteinb

PUFA

−0.04

−0.05

−0.10

−0.15

−0.06

−0.05

Iron

 

0.20

0.15

0.23

−0.01

0.13

Vitamin E

  

0.27

0.52

0.03

0.10

Carotenec

   

0.37

0.21

0.47

Vitamin C

    

0.13

0.20

Lycopeneb

     

0.31

aAll dietary variables are adjusted for total energy intake using the residual method of Willett and Stampfer [48], and all nutrients are expressed as dietary + (if applicable) supplemental intake

bAs lycopene and lutein intake values were not available, the number of servings per week for foods rich in lycopene and lutein was used instead

cVariable ‘carotene’ represents total intake of plant-derived pro-vitamin A (primarily β-carotene)

The OBS values ranged from 2 to 18 among cases, 1 to 19 among endoscopy controls, and 2 to 18 among community controls. There was an inverse association between OBS (continuous variable) and colorectal adenoma in both the community control (OR = 0.92; 95% CI 0.88–0.97; p-trend = 0.001) and endoscopy control (OR = 0.90; 95% CI: 0.87–0.94; p-trend < 0.001) analyses. When the OBS was treated as a categorical variable and a score of 3 points or less was used as the referent category, in the analyses with the endoscopy controls the ORs for scores of 4–6, 7–9, 10–12, 13–15, and 16–18, were, respectively, 0.42, 0.32, 0.22, 0.20, and 0.19 (p-trend < 0.001). In the corresponding analysis with the community controls, the ORs were 0.49, 0.37, 0.27, 0.28, and 0.24 (p-trend 0.002). Similar analyses for each 6-point increase in the OBS yielded ORs of 0.61 (95% CI: 0.45–0.83) and 0.47 (0.31–0.71) in the analysis with endoscopy controls; and 0.63 (0.44–0.90) and 0.55 (0.35–0.89) in the analysis with community controls (Table 5).
Table 5

Association of oxidative balance score (OBS) with incident, sporadic colorectal adenoma

OBSa

Cases (n = 564)

Endoscopy controls (n = 1,204)

OR (95% CI)b

Community controls (n = 535)

OR (95% CI)b

For every 3-point OBS increase

For every 6-point OBS increase

For every 3-point OBS increase

For every 6-point OBS increase

1

0

1

  

0

  

2

5

2

1.0

 

1

1.0

 

3

11

7

 

1.0

3

 

1.0

4

21

24

 

9

 

5

37

58

0.42 (0.17–1.02)

 

31

0.49 (0.15–1.57)

 

6

53

78

  

37

  

7

42

114

  

52

  

8

64

116

0.32 (0.13–0.78)

 

42

0.37 (0.12–1.15)

 

9

78

118

 

0.61 (0.45–0.83)

65

 

0.63 (0.44–0.90)

10

58

128

 

71

 

11

54

147

0.22 (0.09–0.54)

 

51

0.27 (0.09–0.87)

 

12

46

135

  

61

  

13

45

103

  

40

  

14

28

81

0.20 (0.08–0.50)

 

31

0.28 (0.08–0.91)

 

15

7

49

 

0.47 (0.31–0.71)

23

 

0.55 (0.35–0.89)

16

13

23

 

10

 

17

1

11

0.19 (0.06–0.57)

 

7

0.24 (0.06–0.94)

 

18

1

7

  

1

  

19

0

2

  

0

  
   

p-trend < 0.001

p-trend < 0.001

 

p-trend = 0.002

p-trend = 0.015

aAll dietary components of OBS are adjusted for total energy intake using the residual method of Willett and Stampfer [48], and all nutrients are expressed as dietary + (if applicable) supplemental intake

bOdds ratio (ninety-five percent confidence interval) adjusted for age, sex, hormone replacement therapy, race, education, family history of colorectal cancer in a first degree relative, total daily energy intake, BMI, alcohol consumption, and intakes of calcium, vitamin D, folic acid, red meat, multivitamin supplements, and dietary fiber

Table 6 presents the sensitivity analyses in which the observed association between colorectal adenoma and the a priori-defined OBS (treated as a continuous variable) was compared to the results of various alternative models. In the analyses that excluded participants who left aspirin, NSAID, and selenium supplement questions blank, and in the analyses that used dose (for selenium) and frequency of use (for aspirin and other NSAIDs) instead of duration, the ORs remained statistically significant and the point estimate changed by no more than two percent. The results for sigmoidoscopy controls (OR = 0.85; 95% CI: 0.81–0.90) were stronger than those for colonoscopy controls (OR = 0.94; 95% CI: 0.90–0.99). Compared to all adenoma cases combined, the association with OBS was more pronounced for those with multiple adenoma and for those with adenoma characterized by moderate/severe atypia, but not for those with large (10+ mm) lesions. Inverse associations according to the site of the largest adenoma tended to move toward the null when progressing from the right colon to the left colon to the rectum. In the analyses in which one or more OBS components were removed, most OR estimates remained within 5% of the original result. When all carotenoids (carotene, lycopene, and lutein) were removed from the OBS, the OR changed from 0.90 to 0.82 in the analysis that used the endoscopy controls, but remained unchanged in the analysis that used the community controls. After the OBS was divided into separate pro-oxidant and anti-oxidant scores the direct association of the pro-oxidants with adenoma appeared to be stronger (Table 6).
Table 6

Sensitivity analyses to evaluate the impact of incomplete reports, alternative exposure definitions, extreme OBS values, different endoscopy procedures, different categories of cases, individual OBS components, and combinations of OBS components on study results

Model

Cases versus endoscopy controls

Cases versus community controls

ORa

95% CI

ORa

95% CI

Original model (reference)

0.90

0.87–0.94

0.92

0.88–0.97

Excluding those who left aspirin, other NSAID and selenium questions blankb

0.92

0.87–0.96

0.92

0.87–0.97

Replacing selenium, other NSAID and aspirin duration of use with dosec

0.92

0.88–0.96

0.94

0.90–0.99

Restricted OBS range 4–15

0.91

0.87–0.95

0.93

0.88–0.98

Colonoscopy controls only

0.94

0.90–0.99

  

Sigmoidoscopy controls only

0.85

0.81–0.90

  

Right sided colon adenoma(cecum through splenic flexure) cases

0.86

0.80–0.92

0.88

0.82–0.95

Left sided colon adenoma (descending through sigmoid) cases

0.92

0.87–0.97

0.94

0.89–0.99

Rectal adenoma cases

0.95

0.87–1.04

0.95

0.87–1.05

Small adenoma (<10 mm) cases

0.90

0.85–0.95

0.92

0.87–0.98

Large adenoma (10+ mm) cases

0.91

0.85–0.97

0.93

0.87–1.00

Single adenoma cases only

0.92

0.88–0.97

0.94

0.89–0.99

Two or more adenoma cases

0.87

0.81–0.93

0.88

0.82–0.95

Mild atypia cases

0.93

0.88–0.98

0.95

0.90–1.01

Moderate/severe atypia cases

0.88

0.83–0.92

0.90

0.84–0.95

OBS excluding smoking controlled for smoking

0.92

0.88–0.97

0.94

0.90–0.99

OBS excluding selenium controlled for selenium

0.90

0.86–0.94

0.92

0.88–0.97

OBS excluding aspirin controlled for aspirin

0.91

0.87–0.96

0.92

0.87–0.97

OBS excluding non-aspirin NSAID controlled for non-aspirin NSAID

0.92

0.88–0.96

0.93

0.89–0.98

OBS excluding alcohol controlled for alcohol

0.91

0.87–0.95

0.93

0.89–0.98

OBS excluding PUFA controlled for PUFA

0.91

0.87–0.95

0.91

0.87–0.96

OBS excluding iron controlled for iron

0.90

0.86–0.94

0.92

0.88–0.97

OBS excluding vitamin E controlled for vitamin E

0.90

0.86–0.94

0.92

0.87–0.97

OBS excluding vitamin C controlled for vitamin C

0.91

0.87–0.96

0.90

0.85–0.96

OBS excluding carotene controlled for carotene

0.88

0.84–0.92

0.91

0.86–0.97

OBS excluding lutein controlled for lutein

0.88

0.84–0.92

0.93

0.88–0.98

OBS excluding lycopene controlled for lycopene

0.87

0.83–0.91

0.93

0.88–0.98

OBS excluding all carotenoids controlled for lycopene, carotene & lutein

0.82

0.78–0.87

0.92

0.86–0.98

OBS excluding drugs controlled for drugs (aspirin and other NSAIDs)

0.93

0.89–0.98

0.93

0.88–0.98

OBS excluding alcohol and smoking controlled for alcohol & smoking

0.93

0.89–0.97

0.95

0.90–1.00

OBS limited to antioxidants only, controlled for pro-oxidants

0.95

0.91–1.00

0.94

0.88–0.99

OBS limited to pro-oxidants only, controlled for antioxidants

0.84

0.77–0.90

0.89

0.81–0.97

aOR represents a change in odds for each additional OBS point. All results are adjusted for age, sex, hormone replacement therapy, race, education, family history of colorectal cancer in a first degree relative, total daily energy intake, BMI, alcohol consumption, and intakes of calcium, vitamin D, folic acid, red meat, and dietary fiber, and multivitamins

bCurrently assigned 0 points

cSelenium dose categorized as none (0 points), <80 mcg/days (1 point), and 80+ mcg/days (2 points), aspirin and NSAID doses categorized as none (0 points), less 7 pills/week (1 point); 7+ pills/week

Adjusted ORs and 95% CIs for each individual OBS component are shown in Table 7. Smoking and alcohol consumption were significantly associated with colorectal adenoma regardless of the control group, although, in the case of alcohol, the observed trend was not in the direction that was expected based on our a priori considerations. The most pronounced association was observed for prolonged (3+ years) NSAID use relative to no NSAID use (OR = 0.27; 95% CI: 0.15–0.47) in the analyses comparing cases to endoscopy controls; whereas the corresponding result in the analyses comparing cases to community controls was not statistically significant (OR = 0.52; 95% CI: 0.27–1.01). Similarly, the OR for prolonged (3+ years) aspirin use was significant in the endoscopy, but not in the community control analyses. The odds ratios for all other individual OBS components were not significantly different from the null (Table 7).
Table 7

Associations between individual oxidative balance score (OBS) components and colorectal adenoma

Variablesa

Cases versus endoscopy controls

Cases versus community controls

OR (95% CI)b

OR (95% CI)b

Smoking:

Former (1 point) relative to current (0 points)

0.55 (0.40–0.75)

0.84 (0.58–1.20)

Never (2 points) relative to current (0 points)

0.43 (0.31–0.60)

0.54 (0.37–0.78)

Selenium supplement use:

1–2 Years (1 point) relative to never (0 points)

1.51 (0.71–3.22)

1.98 (0.72–5.47)

3+ Years (2 points) relative to none (0 points)

1.95 (0.53–7.16)

0.56 (0.16–1.92)

Aspirin use:

1–2 Years (1 point) relative to never (0 points)

0.80 (0.60–1.08)

0.95 (0.68–1.32)

3+ Years (2 points) relative to never (0 points)

0.61 (0.44–0.86)

0.93 (0.62–1.38)

Other NSAID use:

1–2 Years (1 point) relative to never (0 points)

0.74 (0.52–1.05)

0.72 (0.49–1.07)

3+ Years (2 points) relative to never (0 points)

0.27 (0.15–0.47)

0.52 (0.27–1.01)

Alcohol intake:

<1 Drink/week (1 point) relative to 8+ drinks/week (0 points)

0.60 (0.46–0.78)

0.61 (0.45–0.82)

1–7 Drinks per week (2 points) relative to 8+ drinks/week (0 points)

0.69 (0.51–0.95)

0.66 (0.46–0.94)

PUFA intake:

2nd Tertile (1 point) relative to 3rd tertile (0 points)

0.83 (0.63–1.10)

0.97 (0.71–1.33)

1st Tertile (2 points) relative to 3rd tertile (0 points)

0.80 (0.61–1.04)

1.15 (0.84–1.56)

Iron intake:

2nd Tertile (1 point)relative to 3rd tertile (0 points)

1.04 (0.78–1.38)

1.25 (0.89–1.75)

1st Tertile (2 points) relative to 3rd tertile (0 points)

1.00 (0.73–1.38)

1.03 (0.71–1.49)

Vitamin C intake:

2nd Tertile (1 point) relative to 1st tertile (0 points)

0.84 (0.63–1.13)

1.02 (0.74–1.41)

3rd Tertile (2 points) relative to 1st tertile (0 points)

0.62 (0.45–0.86)

0.94 (0.65–1.37)

Total carotene intake:

2nd Tertile (1 point) relative to 1st tertile (0 points)

0.91 (0.69–1.20)

0.82 (0.60–1.11)

3rd Tertile (2 points) relative to 1st tertile (0 points)

0.91 (0.67–1.23)

0.85 (0.60–1.21)

Vitamin E intake:

2nd Tertile (1 point) relative to 1st tertile (0 points)

1.05 (0.78–1.40)

0.97 (0.70–1.35)

3rd Tertile (2 points) relative to 1st tertile (0 points)

0.99 (0.70–1.42)

1.10 (0.72–1.68)

Lutein-rich foods intake:

2nd Tertile (1 point) relative to 1st tertile (0 points)

1.15 (0.88–1.50)

0.95 (0.70–1.29)

3rd Tertile (2 points) relative to 1st tertile (0 points)

1.07 (0.78–1.46)

0.75 (0.52–1.07)

Lycopene-rich foods intake:

2nd Tertile (1 point) relative to 1st tertile (0 points)

1.21 (0.93–1.58)

0.99 (0.73–1.35)

3rd Tertile (2 points) relative to 1st tertile (0 points)

1.19 (0.90–1.59)

0.75 (0.54–1.03)

aAll dietary components of OBS are adjusted for total energy intake using the residual method of Willett and Stampfer [48], and all nutrients are expressed as dietary + (if applicable) supplemental intake

bOdds ratio (ninety-five percent confidence interval) adjusted for age, sex, hormone replacement therapy, race, education, family history of colorectal cancer in a first degree relative, total daily energy intake, BMI, alcohol consumption, and intakes of calcium, vitamin D, folic acid, red meat, multivitamin supplements, and dietary fiber

Discussion

Despite the strong biological rationale and basic science evidence, the search for causal associations linking specific pro- and anti-oxidants to presumably oxidative stress-related neoplasms and precursor lesions, such as colorectal adenoma, has not produced encouraging results in epidemiologic studies [2325, 5166]. Our analyses suggest that one possible explanation for the failure to detect consistent associations with pro- and anti-oxidant factors studied individually is the multifactorial nature by which oxidative stress, inflammation, diet, and other factors act and interact in affecting risk of colorectal neoplasia. By way of illustration, when we conducted “traditional” analyses to evaluate associations between individual OBS components and risk for colorectal adenoma we found multiple statistically significant (at the 0.05 level) two-way interactions in each of several analyses, but which interaction terms were statistically significant depended on the model that was used. If one were also to consider higher order interactions and multi-collinearity among all of the individual OBS components, the proper application and interpretation of traditional models becomes even more problematic.

Ours is not the first attempt to evaluate the effect of multiple determinants of oxidative stress simultaneously. For example, Wright and colleagues constructed a dietary anti-oxidant index and evaluated its ability to predict lung cancer risk within the Alpha-Tocopherol, Beta-Carotene Cancer Prevention Study cohort [67]. In multivariate proportional hazards models, the relative risks for lung cancer according to increasing quintiles of the anti-oxidant index were 1.00 (referent), 1.00 (95% CI: 0.87, 1.14), 0.91 (0.79, 1.05), 0.79 (0.68, 0.92), and 0.84 (0.72, 0.98). In a Swedish population-based case–control study of esophageal cancer, Terry and colleagues created an anti-oxidant index by summing the quartile scores of vitamin C, alpha tocopherol, and beta-carotene. There was a monotonic 20–30% decrease in esophageal cancer risk with each increasing anti-oxidant index quartile [68].

Although conceptually similar, the studies by Wright et al. and Terry et al. are different from ours in several aspects. First, in contrast to our study and that by Terry et al., Wright and colleagues used a data-driven principal components analysis to develop their composite anti-oxidant index. Although this is a potentially useful methodology, the inherent variation in exposure patterns across populations, as well as subjective decisions in the principal components analysis (e.g., those regarding the selection and grouping of exposure variables and the selection, rotation, and interpretation of principal components, or “factors”) limits the comparability of results across such studies.

Second, both the Wright et al. and Terry et al. studies included only dietary anti-oxidant exposures in their summary index, whereas we used a wider range of variables thought to affect oxidative balance. It is important to note, however, that Wright et al. observed the strongest inverse association between the index and lung cancer among subjects in the upper quartiles of heme iron (a pro-oxidant) intake, which was not apparent among men in the lower quartiles of heme iron intake. The index also appeared to be most strongly associated with risk among men consuming higher quantities of fish polyunsaturated fatty acids, with 25–30 percent reductions in the risk of lung cancer for the highest anti-oxidant index quintile versus the lowest [67].

Finally, in their ordinal category analyses both Terry et al. and Wright et al. separated the anti-oxidant index distribution according to quartiles or quintiles, an approach that divides participants into more or less equal groups, but could mask important associations at the extremes. In our study, we used categories that reflected equal score intervals, but not necessarily equal numbers of participants. We based this decision on experience with scores in other health disciplines, such as urology or neonatology, which typically shows that most individuals fall within a very narrow range of values, whereas clinically important scores apply to a relatively small proportion of the population [69, 70].

Our study most closely follows the method previously described by Van Hoydonck et al. In this Belgian mortality study the authors constructed an oxidative balance score to summarize the combined intake of vitamin C, beta-carotene, and iron among smokers. The risk of all-cause and total cancer mortality was primarily elevated in the high-score group, which allowed the authors to conclude that the association between oxidative stress related exposures and mortality may follow a threshold pattern rather than a linear trend [30].

There is an overlap between dietary factors used in our study and those considered in studies of dietary patterns, which generally found inverse associations between healthy diets and both colorectal adenoma and colorectal cancer [26, 7175]. Thus, factors included in the OBS may simply act as surrogate indicators of healthier diets. Conversely, one could argue that the so-called ‘prudent’ diets are those rich in anti-oxidants and low in pro-oxidants. One should also keep in mind that the association between the OBS and colorectal adenoma risk in our data is attributable at least in part to non-dietary factors such as smoking, aspirin, and other NSAIDs.

The OSB approach undoubtedly requires further discussion and refinement. A general problem with characterizing complex processes by using unidimensional scores is that such scores often rely on the subjective assignment of points that may be susceptible to substantial inter-observer and intra-observer variability [7678]. In our proposed scoring approach this is unlikely to be a problem, as we use a priori selected cutoffs as opposed to subjective judgment. There may, however, be disagreement as to whether the use of tertile cut points among controls for constructing the score is appropriate, particularly for exposures that produce a threshold effect and in circumstances where controls may not be representative of the general population [79, 80]. Another likely limitation of the scoring method used in this analysis is the implicit assumption that all pro- and anti-oxidant exposures produce effects of similar magnitude, and all equally deserve to be included in the score. For instance, it may be reasonable to assume that the impact of non-aspirin NSAIDs is higher than that of aspirin, and that lycopene may be a more potent prostate anti-oxidant than, say, beta-carotene or lutein. Further, our proposed method may not provide the degree of biologic insight as would some other approaches, such as one based on direct incorporation of biochemical pathway information [81].

Although our results are in keeping with the stated a priori hypothesis and (judging from the sensitivity analyses) are reasonably robust, we emphasize the need for caution in interpreting these findings because of the potential problems with case–control studies such as ours [8285]. The relatively low response rate raises concerns about selection bias, whereas reliance on dietary recall may leave room for exposure misclassification. Additional exposure misclassification may occur due to less than optimal characterization of some OBS components. For example, if the data allowed, we would prefer to categorize intakes of long-chain PUFA instead of all PUFA, individual tocopherols instead of vitamin E, and heme iron instead of all forms of iron. We also feel that a more comprehensive evaluation of the effects of aspirin and other NSAIDs should consider cumulative dose rather than duration of use.

Each of the control groups used in this study has its own strengths and weaknesses. As described previously, the primary advantage of the colonoscopy control group is that it minimizes outcome misclassification in contrast to the community control group, which likely included individuals with undiagnosed adenomatous polyps [31]. On the other hand, colonoscopy controls in this study (which was conducted before the widespread use of colonoscopies for screening purposes) represent a highly selected group of individuals, nearly all of whom either had symptoms that prompted colonoscopy, or a strong family history of colorectal cancer that put them in a high-risk group. Previous comparisons between cases and colonoscopy controls showed an inverse association between adenoma and a family history of colorectal cancer in a first degree relative, because individuals with a negative family history were more likely to have clinical signs such as hemtochezia, melena, or iron deficiency anemia [86]. The main advantage of the sigmoidoscopy control group compared to the colonoscopy control group is that sigmoidoscopy controls were mostly asymptomatic, and thus more representative of the underlying population. Although sigmoidoscopy controls may still be subject to outcome misclassification due to missed proximal adenomas, the resulting bias is expected to be less pronounced than that with community controls. Another approach would have been to weight controls inversely in proportion to their probability of selection [87], but we did not adopt it here because of the lack of data about the appropriate weights and because of the robustness of our results with respect to the choice of control groups.

Despite the above concerns, it is important to point out that we have obtained similar results from two previous studies [29]. Both these previous studies found consistent, statistically significant associations between combined exposure to pro- and anti-oxidants and risk of developing oxidative stress-related neoplasms. Given that all our analyses so far were based on data from case–control studies, a reasonable next step would be to confirm the observed association using data from prospective studies.

Future studies should also take into consideration endogenous components of the oxidative stress pathway, such as polymorphic variants and phenotypic expression of genes encoding enzymes responsible for cytoplasmic and mitochondrial anti-ROS defense [88], as well as genes regulating repair of oxidative stress-induced DNA damage [89]. Additional exogenous factors that warrant further investigation and potential incorporation into the OBS may include physical exercise and intakes of other dietary pro-oxidants and anti-oxidants such as copper, cadmium, and polyphenols [90]. Finally, one needs to consider that some presumed anti-oxidants may lose their activity and even exhibit pro-oxidant properties at high doses or at high partial pressures of oxygen [91]. Nonetheless, based on the results of this study, and those of previous studies, the OBS appears to be a useful way to characterize the pro-oxidant/anti-oxidant balance status of participants in epidemiologic studies.

Copyright information

© Springer Science+Business Media B.V. 2008