In vitro evidence implicates oxidative stress in many adverse health conditions, including colorectal neoplasia. In human studies, however, oxidative stress is measured by imperfect biomarkers, which are inconsistently associated with health outcomes. Structural equation modeling (SEM) offers one possible solution by modeling a latent (unobserved) construct from multiple biomarkers. Our goal was to investigate the association of a latent oxidative stress variable with colorectal adenoma. Using SEM, we analyzed pooled data from two cross-sectional studies of colorectal adenoma (n = 526) that measured five plasma biomarkers of oxidative stress and inflammation that comprised the latent oxidative stress variable: F2-isoprostanes (FIP), fluorescent oxidation products (FOP), mitochondrial DNA (MtDNA) copy number, γ-tocopherol (Gtoc), and C-reactive protein (CRP). Higher levels of oxidative stress were associated with colorectal adenoma [OR = 3.23 per SD increase in oxidative stress; 95% confidence interval (CI), 1.28–8.18]. The latent variable estimate was considerably stronger than the associations of adenoma with the individual biomarkers, which were modest and mostly nonsignificant. Risk factors were associated with adenoma via the oxidative stress pathway, particularly overweight and obesity with an OR = 1.50; 95% CI, 1.10–2.81; and OR = 2.95; 95% CI, 1.28–12.45, respectively. Oxidative stress may be positively associated with colorectal adenoma, and important risk factors may act through this mechanism, but the cross-sectional design of the current study precludes observing the directionality of associations. The presence of an adenoma could affect levels of the circulating biomarkers; thus, we should be cautious of strong conclusions until the findings are replicated in a follow-up study. Cancer Prev Res; 11(1); 52–58. ©2017 AACR.

Defined as an imbalance between pro- and antioxidants (1), oxidative stress has been implicated in the initiation, promotion, and progression of carcinogenesis (2–4). Those in vitro findings, however, are not supported in the epidemiologic literature (5–7), and attempts to prevent cancer by antioxidant supplementation in human clinical trials did not yield the hypothesized results (8). This inconsistency may reflect insufficient knowledge on how to influence or measure oxidative stress in humans.

As oxidative stress is not directly observable in vivo, it is normally measured using biomarkers (9). Numerous biomarkers of oxidative stress have been studied (10), but because oxidative stress is complex and multifaceted, reliance on any single marker may be inadequate (11–13). Structural equation modeling (SEM) may offer a potential solution. SEM is a multivariate analytic technique for modeling theorized structural (causal) pathways, specified a priori, while incorporating observed variables as imperfect measures of unobservable (latent) constructs (14, 15). In our earlier analysis (16), five circulating biomarkers with hypothesized links to oxidative stress constructed an oxidative stress latent variable: F2-isoprostanes (FIP), a marker of lipid peroxidation; fluorescent oxidation products (FOP), a marker of nonspecific oxidation of large molecules; mitochondrial DNA copy number (MtDNA), a marker of oxidative cellular damage in close proximity to reactive oxygen species production (17); γ-tocopherol (Gtoc), a circulating marker more representative of its oxidative stress–influenced metabolism rather than nutritional intake (18); and C-reactive protein (CRP), a marker of acute inflammation accompanied by increased oxidation. In that analysis, we validated the oxidative stress latent variable in three ways: (i) demonstrating that the associations of the biomarkers with the latent variable followed expected directions; (ii) confirming hypothesized associations of the latent variable with pro- and antioxidant exposures; and (iii) showing that the associations of pro- and antioxidant exposures with the latent variable were stronger and more consistent than their corresponding associations with each of the five biomarkers considered individually (16).

In the current study, we assessed whether a similar latent oxidative stress variable was associated with newly diagnosed colorectal adenoma. We then compared the latent variable estimate to estimates of the five biomarkers modeled individually to demonstrate its novelty. For the analysis, we use pooled data from two cross-sectional studies of colorectal adenoma that collected information on demographic and lifestyle characteristics and measured plasma biomarkers of oxidative stress.

Study population

Markers of Adenomatous Polyps study I and II.

The Markers of Adenomatous Polyps I and II (MAP I and II) studies were conducted by the same principal investigator (R.M. Bostick) using almost identical protocols, described elsewhere (19). MAP I participants (n = 474) were recruited from gastroenterology practices in Winston-Salem and Charlotte, North Carolina, while MAP II (n = 233) participants were recruited from Consultants in Gastroenterology, a large, private clinic in Columbia, South Carolina. In both studies, participants were 30 to 74 years of age, with no prior history of colorectal neoplasms, scheduled for elective outpatient colonoscopy; the data were pooled for this analysis. For the analytic sample, 126 subjects were removed because of missing information on key variables (e.g., age, sex, race, etc.). Of the 581 remaining subjects, 91% were non-Hispanic whites; thus, we restricted the analysis to that sample (n = 526). Cases (n = 188) were defined as subjects newly diagnosed with a colon or rectal adenoma during colonoscopy, and noncases (n = 255) were defined as subjects found to be free of polyps. Participants with hyperplastic polyps or unknown outcome status (n = 83) contributed toward constructing the oxidative stress latent variable but not its association with colorectal adenoma.

Data collection

Before undergoing the colonoscopy, all consenting participants completed a mailed questionnaire collecting information on demographic characteristics, medical history, use of medications, and habits. The questionnaire included, but was not limited to, tobacco smoking and alcohol drinking history, physical activity, and use of aspirin and other NSAIDs. In addition, diet and supplement intakes were assessed through a modified 153-item Willet Food Frequency Questionnaire. Vitamin and mineral total intakes were calculated as dietary plus supplemental intakes.

All participants provided blood samples that were drawn into red-coated, prechilled Vacutainer tubes, plunged into ice, and protected from light exposure. Upon immediate delivery to the laboratory, the samples were centrifuged under refrigeration, after which plasma and serum were separately aliquoted into amber-colored cryopreservation vials. Air in the vials was displaced with inert gas, and the vials were sealed with O-ring screw caps then stored at −70°C.

The Molecular Epidemiology and Biomarker Research Laboratory at the University of Minnesota (Minneapolis, MN) performed all biomarker analyses. Gas chromatography–mass spectrometry was used to measure plasma FIP (20), using deuterium (4)-labeled 8-iso-prostaglandin F2-alpha as an internal standard with unlabeled, purified F2-isoprostane as a calibration standard. FOPs were measured in plasma samples using the modified Shimasaki method (21, 22). A mixed solution of plasma and ethanol/ether was centrifuged for 10 minutes at 3,000 rpm, and then 1.0 mL of supernatant was added to cuvettes for spectrofluorometric readings. Relative fluorescence intensity, in U/mL of plasma at 360/430 nm wavelengths, was calculated by a spectrofluorometer. Quinine sulfate diluted in 0.1 N H2SO4 was used for calibration. For approximately 22% of the participants, FOP was measured in serum because plasma samples were not available; this was not expected to affect the measurement because correlation between serum and plasma FOP concentrations was very high (r = 0.9; P < 0.001). High-sensitivity CRP was measured using latex-enhanced immunonephelometry on a Behring nephelometer II (BN-II) analyzer (inter-assay CV 4%; Behring Diagnostics). Gtoc was measured using high-performance liquid chromatography (23). MtDNA copy number was analyzed using qRT-PCR as described previously (24). Two primers were used, one for MtDNA and one for nuclear DNA. The ratio of MtDNA to nuclear DNA was determined by serial dilution of a healthy referent genomic DNA sample.

Structural equation model and statistical methods

As shown in Fig. 1, the baseline SEM model (Model 1) had one standardized continuous latent variable (here onward termed oxidative stress) composed of five continuous blood biomarkers thought to be consequences of oxidative stress (arrows emanating from oxidative stress): FIP, FOP, CRP, MtDNA, and Gtoc. The model included the main effect, oxidative stress to colorectal adenoma, and potential confounder variables (listed below) modeled as causes of oxidative stress and adenoma (arrows directed toward oxidative stress and adenoma). Because of distributional differences in MtDNA copy number between the MAP studies, we considered an alternative baseline SEM where the study variable directly affected the individual biomarkers, but not oxidative stress (Fig. 1B). This change may seem trivial, but the directionality of the arrows has implications for the model assumptions. In the first model, study-specific biomarker variability would manifest in the latent oxidative stress variable; in the second model, it would not, a feature we preferred. We considered two other models (Models 2 and 3) with additional potential confounders. Each model is summarized below:

  • Model 1: The baseline model included the following covariates: age (years), sex (women vs. men), body mass index (BMI; <25, 25–29.9, ≥30 kg/m2), smoking (never, former, current), alcohol drinking (never, former, current), regular aspirin use (≥once/week), regular NSAID use (≥once/week), and study (MAP I vs. MAP II); n = 526.

  • Model 2: The partially adjusted model included the covariates in the baseline model plus additional variables potentially associated with colorectal adenoma: reported physical activity (ordinal), total energy intake (ordinal), total dietary fiber intake (ordinal), plasma cholesterol levels (continuous, mg/dL), and a family history of colorectal cancer (yes/no); n = 457.

  • Model 3: The fully adjusted model included the covariates of Model 2 plus additional variables potentially associated with oxidative stress: total vitamin C intake (ordinal), and polyunsaturated fatty acid intake (PUFA; ordinal); n = 457. All ordinal variables were categorized as high, medium, or low using sex-specific tertile cut-off points among noncases.

Figure 1.

A, Baseline model of oxidative stress and colorectal adenoma in the combined MAP datasets. B, An alternative baseline model that allows a study sample variable to directly affect each biomarker so that study-specific variability in each biomarker will not manifest in the latent oxidative stress variable. Oxidative stress is derived from the shared covariance of the five biomarkers: FIP, FOP, CRP, Gtoc, MtDNA.

Figure 1.

A, Baseline model of oxidative stress and colorectal adenoma in the combined MAP datasets. B, An alternative baseline model that allows a study sample variable to directly affect each biomarker so that study-specific variability in each biomarker will not manifest in the latent oxidative stress variable. Oxidative stress is derived from the shared covariance of the five biomarkers: FIP, FOP, CRP, Gtoc, MtDNA.

Close modal

The primary result of these analyses is the model-estimated effect of oxidative stress on colorectal adenoma, interpreted as the change in odds of adenoma per SD change in oxidative stress. As a comparison, we estimated associations of the five biomarkers with adenoma, individually. In addition, we assessed the effects of the explanatory (exposure) variables on oxidative stress (γs), interpreted as the linear SD change in oxidative stress per unit change in exposure, and we evaluated the indirect effect of the exposures on adenoma mediated by oxidative stress (25, 26), interpreted as the change in odds of adenoma per unit change in exposure. Confidence intervals (CI) for the indirect effects were estimated by bootstrap resampling with 1,000 replications (27). We conducted a few sensitivity analyses: First, we ran the baseline model among never smokers over concerns about residual confounding; next, we removed CRP from the baseline model to examine what effect, if any, inflammation may contribute toward the latent variable.

All SEMs were analyzed using Mplus software version 7.2 (28). We chose a generalized linear logit SEM under maximum likelihood estimation, which provided ORs for adenoma but not traditional SEM fit statistics. In Supplementary Data, we report the results and fit statistics of a probit SEM using a weighted least squares estimator. A two-sided P < 0.05 indicated statistical significance.

Selected baseline characteristics of the analytic sample are in Table 1. The average age was 56.8 years, and the average BMI was 28.0 kg/m2. Women comprised 52% of all subjects, but only 41% of adenoma patients. Most of the subjects were never or former smokers (77%), and about half were current drinkers (51%).

Table 1.

Selected baseline characteristics of the analytic study sample from the non-Hispanic white subjects in the MAP I and II studies

Colon or rectal polyp (n = 188)Polyp free (n = 255)Hyperplastic polyp or unknowna (n = 83)Total (N = 526)
μ or n% or SDμ or n% or SDμ or n% or SDμ or n% or SD
Age, years 58.0 8.1 55.6 9.4 57.8 8.7 56.8 8.9 
Sex (% women) 77 41.0 163 63.9 35 42.2 275 52.3 
BMI, kg/m2 28.0 6.4 27.6 6.1 29.0 6.6 28.0 6.3 
Phys. act., MET-hours/week 186.6 138.4 189.2 130.7 208.6 124.5 191.3 132.5 
Smoker 
 Never 50 26.6 128 50.2 22 26.5 200 38.0 
 Former 78 41.5 95 37.3 32 38.6 205 39.0 
 Current 60 31.9 32 12.6 29 34.9 121 23.0 
Drinker 
 Never 43 22.9 87 34.1 23 27.7 153 29.1 
 Former 43 22.9 46 18.0 17 20.5 106 20.2 
 Current 102 54.3 122 47.8 43 51.8 267 50.8 
Aspirin useb 68 36.2 88 34.5 32 38.6 188 35.7 
NSAID useb 41 21.8 90 35.3 22 26.5 153 29.1 
Daily dietary intakes, questionnaire 
 Total energy, kcal 2,000 798 1,830 892 2,051 921 1,925 868 
 PUFA, gm 14.2 6.8 13.6 9.1 14.5 7.2 13.9 8.1 
 Totalc vitamin C, mg 268.7 347.7 268.7 308.9 254.1 282.3 266.4 319.0 
 Totalc fiber, gm 21.5 9.6 20.2 11.0 22.3 14.1 21.0 11.1 
Plasma/serum concentrations 
 Cholesterol, mg/dL 203.6 35.6 204.2 39.8 208.4 39.7 204.7 38.3 
 FIP, pg/mL 90.2 48.9 84.6 35.5 85.2 34.9 86.8 40.9 
 FOP, avg. std. ref. adj. 0.05 0.11 0.04 0.02 0.06 0.14 0.05 0.09 
 CRP, μg/mL 5.3 6.0 4.7 5.9 5.8 6.3 5.1 6.0 
 MtDNA, rel. to nucl. DNA 2.2 2.9 2.8 2.3 2.0 1.8 2.5 2.5 
 Gtoc, mg/dL 0.23 0.11 0.20 0.11 0.21 0.10 0.21 0.11 
Colon or rectal polyp (n = 188)Polyp free (n = 255)Hyperplastic polyp or unknowna (n = 83)Total (N = 526)
μ or n% or SDμ or n% or SDμ or n% or SDμ or n% or SD
Age, years 58.0 8.1 55.6 9.4 57.8 8.7 56.8 8.9 
Sex (% women) 77 41.0 163 63.9 35 42.2 275 52.3 
BMI, kg/m2 28.0 6.4 27.6 6.1 29.0 6.6 28.0 6.3 
Phys. act., MET-hours/week 186.6 138.4 189.2 130.7 208.6 124.5 191.3 132.5 
Smoker 
 Never 50 26.6 128 50.2 22 26.5 200 38.0 
 Former 78 41.5 95 37.3 32 38.6 205 39.0 
 Current 60 31.9 32 12.6 29 34.9 121 23.0 
Drinker 
 Never 43 22.9 87 34.1 23 27.7 153 29.1 
 Former 43 22.9 46 18.0 17 20.5 106 20.2 
 Current 102 54.3 122 47.8 43 51.8 267 50.8 
Aspirin useb 68 36.2 88 34.5 32 38.6 188 35.7 
NSAID useb 41 21.8 90 35.3 22 26.5 153 29.1 
Daily dietary intakes, questionnaire 
 Total energy, kcal 2,000 798 1,830 892 2,051 921 1,925 868 
 PUFA, gm 14.2 6.8 13.6 9.1 14.5 7.2 13.9 8.1 
 Totalc vitamin C, mg 268.7 347.7 268.7 308.9 254.1 282.3 266.4 319.0 
 Totalc fiber, gm 21.5 9.6 20.2 11.0 22.3 14.1 21.0 11.1 
Plasma/serum concentrations 
 Cholesterol, mg/dL 203.6 35.6 204.2 39.8 208.4 39.7 204.7 38.3 
 FIP, pg/mL 90.2 48.9 84.6 35.5 85.2 34.9 86.8 40.9 
 FOP, avg. std. ref. adj. 0.05 0.11 0.04 0.02 0.06 0.14 0.05 0.09 
 CRP, μg/mL 5.3 6.0 4.7 5.9 5.8 6.3 5.1 6.0 
 MtDNA, rel. to nucl. DNA 2.2 2.9 2.8 2.3 2.0 1.8 2.5 2.5 
 Gtoc, mg/dL 0.23 0.11 0.20 0.11 0.21 0.10 0.21 0.11 

aSubjects with hyperplastic polyps or otherwise unknown outcome status contributed information toward the construction of the oxidative stress latent variable but not its association with colorectal adenoma.

b≥Once/week.

cDietary + supplemental.

Higher oxidative stress was strongly associated with colorectal adenoma (Table 2). In the baseline model, the odds of adenoma increased approximately 2-fold for each SD increase in oxidative stress, with an estimated OR of 2.18 and a 95% CI from 1.22 to 3.92. The OR (95% CI) estimates for the partially and fully adjusted models were 3.03 (1.29–7.09) and 3.23 (1.28–8.18), respectively. In a sensitivity analysis among never smokers, the oxidative stress estimate did not materially change, although it was no longer statistically significant due to the smaller sample size, n = 200 (OR = 2.82; 95% CI, 0.82–9.64). Compared with oxidative stress, the associations of adenoma with each individual biomarker (per SD) were much weaker. At an OR = 1.37 (95% CI, 1.03–1.82), FIP was the only significant finding of the five biomarkers, although CRP (OR = 1.23; 95% CI, 0.97–1.57) and Gtoc (OR = 1.27; 95% CI, 0.99–1.63) were borderline. We observed the strongest point estimate for FOP, but it was very unstable and imprecise.

Table 2.

Model estimated effects on colorectal adenoma for one SD increase in the oxidative stress latent variable, or the respective individual biomarker

ORs (95% CI)P
SEMs with a latent oxidative stress variable comprised of five biomarkers 
  Baseline modela 2.18 (1.22–3.92) 0.009 
  Partially adjusted modelb 3.03 (1.29–7.09) 0.011 
  Fully adjusted modelc 3.23 (1.28–8.18) 0.013 
Biomarkers modeled individually using the baseline model covariatesa 
  FIP 1.37 (1.03–1.82) 0.031 
  FOP 1.67 (0.70–3.96) 0.249 
  CRP 1.23 (0.97–1.57) 0.094 
  Gtoc 1.27 (0.99–1.63) 0.056 
  MtDNA copy number 1.14 (0.84–1.55) 0.412 
ORs (95% CI)P
SEMs with a latent oxidative stress variable comprised of five biomarkers 
  Baseline modela 2.18 (1.22–3.92) 0.009 
  Partially adjusted modelb 3.03 (1.29–7.09) 0.011 
  Fully adjusted modelc 3.23 (1.28–8.18) 0.013 
Biomarkers modeled individually using the baseline model covariatesa 
  FIP 1.37 (1.03–1.82) 0.031 
  FOP 1.67 (0.70–3.96) 0.249 
  CRP 1.23 (0.97–1.57) 0.094 
  Gtoc 1.27 (0.99–1.63) 0.056 
  MtDNA copy number 1.14 (0.84–1.55) 0.412 

Abbreviation: ORs = standardized odds ratio.

aAdjusted for age, sex, BMI, smoking, alcohol consumption, aspirin, and nonsteroidal anti-inflammatory use.

bFurther adjusted for physical activity, total fiber intake, total energy intake, and plasma cholesterol.

cFurther adjusted for total vitamin C intake, and PUFA intake.

The covariance between FIP, Gtoc, and CRP characterized the latent variable in the baseline and alternative SEM (Table 3); their standardized loading factors were similar in both models. FIP contributed the most to the oxidative stress latent variable (λs = 0.70; 95% CI, 0.61–0.79), followed by Gtoc (λs = 0.51; 95% CI, 0.41–0.61), and CRP (λs = 0.45; 95% CI, 0.35–0.54). The MtDNA loading factor depended entirely on its distributional differences between the two studies: baseline model (λs = −0.26; 95% CI, −0.41 to −0.12); alternative SEM in which the study variable directly affected the biomarkers (λs = <0.01; 95% CI, −0.10–0.10); we achieved the same result if study was removed from the model (data not shown). In neither model did FOP meaningfully contribute to oxidative stress. The loading factor results did not change with additional confounders. In the CRP-excluded sensitivity analysis, the factor loadings for FIP, FOP, MtDNA, and Gtoc did not materially change, and neither did the oxidative stress association with colorectal adenoma (OR = 1.96; 95% CI, 1.13–3.38). In Supplementary Table S1, we provide model fit statistics from a probit model analysis of the baseline models.

Table 3.

Indicator variables and their standardized loading factors from oxidative stress

λs (95% CI)P
 Baseline model 
 FIP 0.64 (0.54–0.74) <0.001 
 FOP 0.07 (−0.05–0.18) 0.231 
 CRP 0.46 (0.37–0.55) <0.001 
 MtDNA −0.26 (−0.41 to −0.12) <0.001 
 Gtoc 0.55 (0.46–0.65) <0.001 
 Alternative model 
 FIP 0.70 (0.61–0.79) <0.001 
 FOP 0.04 (−0.08–0.16) 0.488 
 CRP 0.45 (0.35–0.54) <0.001 
 MtDNA <0.01 (−0.10–0.10) 0.994 
 Gtoc 0.51 (0.41–0.61) <0.001 
λs (95% CI)P
 Baseline model 
 FIP 0.64 (0.54–0.74) <0.001 
 FOP 0.07 (−0.05–0.18) 0.231 
 CRP 0.46 (0.37–0.55) <0.001 
 MtDNA −0.26 (−0.41 to −0.12) <0.001 
 Gtoc 0.55 (0.46–0.65) <0.001 
 Alternative model 
 FIP 0.70 (0.61–0.79) <0.001 
 FOP 0.04 (−0.08–0.16) 0.488 
 CRP 0.45 (0.35–0.54) <0.001 
 MtDNA <0.01 (−0.10–0.10) 0.994 
 Gtoc 0.51 (0.41–0.61) <0.001 

Abbreviation: λs = standardized loading factor.

Several explanatory variables were associated with oxidative stress, and some of these were associated with adenoma via the oxidative stress pathway. From the adjusted models, higher amounts of physical activity, dietary fiber, and total vitamin C intakes were estimated to lower oxidative stress (Table 4). Conversely, higher serum total cholesterol concentrations, and greater PUFA and total energy intakes were estimated to increase oxidative stress. Obesity, smoking, alcohol consumption, and regular aspirin use were all associated with oxidative stress (Table 4). The estimated oxidative stress pathway effects on adenoma for overweight and obesity were 1.50 (95% CI, 1.10–2.81) and 2.95 (95% CI, 1.28–12.45), respectively. The estimates for current smoking, current drinking, and regular aspirin use were borderline statistically significant, while those for former smoking, former drinking, and regular NSAID use did not appreciably differ from the null value (Table 5).

Table 4.

Model estimated effects of explanatory variables on the linear oxidative stress latent variable; γs estimates are per SD of oxidative stress

Baseline model
γs (95% CI)P
Age, per year −0.001 (−0.013–0.012) 0.913 
Women vs. men 0.80 (0.57–1.02) <0.001 
BMI (kg/m2)a 
 <25 Referent  
 25–29.9 0.53 (0.28–0.78) <0.001 
 ≥30 1.40 (1.16–1.61) <0.001 
Smokinga 
 Never Referent  
 Former 0.01 (−0.25–0.23) 0.915 
 Current 0.29 (0.00–0.53) 0.048 
Alcohol drinkinga 
 Never Referent  
 Former −0.30 (−0.63 to −0.02) 0.074 
 Current −0.38 (−0.65 to −0.15) 0.006 
Aspirin useb −0.34 (−0.57 to −0.15) 0.003 
NSAID useb −0.05 (−0.29 to 0.15) 0.669 
 Partially adjusted model 
Physical activityc −0.12 (−0.24–0.01) 0.060 
Total fiber intakec −0.39 (−0.55 to −0.23) <0.001 
Total energy intakec 0.19 (0.04–0.35) 0.013 
Serum total cholesterol, mg/dl 0.009 (0.006–0.011) <0.001 
Family hist. of CRC −0.13 (−0.37–0.10) 0.266 
 Fully adjusted model 
 Totald vitamin C intakec −0.32 (−0.45 to −0.18) <0.001 
 PUFA intakec 0.13 (−0.03–0.29) 0.106 
Baseline model
γs (95% CI)P
Age, per year −0.001 (−0.013–0.012) 0.913 
Women vs. men 0.80 (0.57–1.02) <0.001 
BMI (kg/m2)a 
 <25 Referent  
 25–29.9 0.53 (0.28–0.78) <0.001 
 ≥30 1.40 (1.16–1.61) <0.001 
Smokinga 
 Never Referent  
 Former 0.01 (−0.25–0.23) 0.915 
 Current 0.29 (0.00–0.53) 0.048 
Alcohol drinkinga 
 Never Referent  
 Former −0.30 (−0.63 to −0.02) 0.074 
 Current −0.38 (−0.65 to −0.15) 0.006 
Aspirin useb −0.34 (−0.57 to −0.15) 0.003 
NSAID useb −0.05 (−0.29 to 0.15) 0.669 
 Partially adjusted model 
Physical activityc −0.12 (−0.24–0.01) 0.060 
Total fiber intakec −0.39 (−0.55 to −0.23) <0.001 
Total energy intakec 0.19 (0.04–0.35) 0.013 
Serum total cholesterol, mg/dl 0.009 (0.006–0.011) <0.001 
Family hist. of CRC −0.13 (−0.37–0.10) 0.266 
 Fully adjusted model 
 Totald vitamin C intakec −0.32 (−0.45 to −0.18) <0.001 
 PUFA intakec 0.13 (−0.03–0.29) 0.106 

Abbreviation: γs, standardized effect estimate.

aP value is a test of trend.

b≥Once/week.

cOrdinal tertile variable.

dTotal = dietary + supplemental.

Table 5.

Model estimated indirect effects for adenoma risk factors via the oxidative stress pathway

OR (95% CI)b
BMI (kg/m2
 <25 Referent 
 25–29.9 1.50 (1.10–2.63) 
 ≥30 2.95 (1.26–11.69) 
Smoking 
 Never Referent 
 Former 1.01 (0.79–1.32) 
 Current 1.24 (1.01–2.00) 
Alcohol drinking 
 Never Referent 
 Former 0.80 (0.48–1.04) 
 Current 0.75 (0.46–0.97) 
Aspirin usea 0.77 (0.48–0.97) 
NSAID usea 0.96 (0.73–1.19) 
OR (95% CI)b
BMI (kg/m2
 <25 Referent 
 25–29.9 1.50 (1.10–2.63) 
 ≥30 2.95 (1.26–11.69) 
Smoking 
 Never Referent 
 Former 1.01 (0.79–1.32) 
 Current 1.24 (1.01–2.00) 
Alcohol drinking 
 Never Referent 
 Former 0.80 (0.48–1.04) 
 Current 0.75 (0.46–0.97) 
Aspirin usea 0.77 (0.48–0.97) 
NSAID usea 0.96 (0.73–1.19) 

a≥Once/week.

bObtained by bootstrap resampling with 1,000 replications.

In a previous study, we presented evidence that a latent variable comprised of FIP, CRP, and Gtoc, but not FOP and MtDNA copy number, may adequately represent the construct of oxidative stress (16). In the current study, we found that a similar oxidative stress latent variable was strongly associated with colorectal adenoma risk. When modeled individually, the biomarker–adenoma associations were weaker and nonsignificant, demonstrating the utility of the latent variable. Numerous pro- and antioxidant risk factors predicted oxidative stress. Of those, overweight and obesity were associated with adenoma via the oxidative stress pathway, although we must be cautious because the cross-sectional study design precludes observing directionality. To our knowledge, this is the first study to use SEM with latent variables to investigate how oxidative stress may be related to a health outcome.

Directly comparing our results with the current literature is difficult as most researchers study biomarkers individually. In previous publications using the same dataset as ours, Kong and colleagues and Thyagarajan and colleagues reported associations of colorectal adenoma with log-transformed biomarkers of FIP, FOP, CRP, and MtDNA copy number ranging from 1.02 to 1.38 (29, 30), similar to the individual biomarker results we report. With different data, Siamakpour-Reihani and colleagues reported associations of four urinary isoprostanes with adenoma ranging from 0.88 to 1.16 (31). That prospective study (n = 425) measured FIP in urine approximately 10 years before colonoscopy, while we measured FIP in plasma shortly before colonoscopy; it is believed that plasma samples are generally preferred to urine (32). One case–control study (n = 695) reported nonsignificant higher levels of Gtoc in persons with adenoma relative to sigmoidoscopy-confirmed controls (33). Unlike Gtoc and FIP, CRP has been frequently studied with colorectal neoplasia. A meta-analysis of colorectal cancer found a higher risk (RR = 1.12; 95% CI, 1.01–1.25) for one log-transformed mg/L unit change in CRP (34). Four studies have focused on the CRP-adenoma association: Two case–control studies report nonsignificant, but modest, positive associations (OR = 1.45) for the highest study-specific level of CRP compared with the lowest (29, 35); two prospective studies found a null result and a nonsignificant protective result (OR = 0.61; refs. 36, 37).

Although it is conventional to assess oxidative stress with individual biomarkers, we considered it an unobservable construct from which we could model a latent variable using the shared covariance of multiple imperfect biomarkers. In theory, the individual biomarkers can be weakly associated with adenoma, but the latent variable would result in a stronger and more robust association, something we observed in this study. In a recent prospective study (n = 2,520), Aleksandrova and colleagues combined a reactive oxygen metabolite biomarker with CRP (ROM/CRP) through principal component analysis, and although only two biomarkers created the “factor,” the authors found that the highest quartile of ROM/CRP compared with the lowest increased colorectal cancer risk by 70% (38). This association, however, did not persist after excluding cases diagnosed within the first 2 years of follow-up, leading the authors to suggest that the ROM biomarker was either a late cause or early consequence of colorectal cancer (39). The current study suggests that oxidative stress may be associated earlier in the colorectal neoplasia natural history, as a cause and/or consequence of adenoma. If oxidative stress is more a consequence rather than a cause of colorectal neoplasia, it may still be useful as an early predictor.

We found that higher BMI was positively associated with colorectal adenoma via the oxidative stress pathway. Obesity is an established risk factor for adenoma (40, 41) and oxidative stress (42). It affects inflammatory cytokines, weakens antioxidant defenses, and elevates blood lipid and glucose levels, resulting in higher oxidative stress (43). Our inverse pathway findings for regular aspirin use, and current alcohol drinking were modest, along with the positive finding for current smoking. Aspirin is inversely associated with colorectal neoplasms and has reduced adenoma recurrence in randomized controlled trials (44), presumably by suppressing oxidative stress and inflammation (45). Research shows smoking increases oxidative stress (46) and is associated with higher adenoma risk (47, 48). The estimated inverse indirect effect of current drinking on adenoma may require further exploration. Although the association of alcohol consumption with oxidative stress is mixed, alcohol consumption is associated with higher adenoma risk (49). Interestingly, for both alcohol drinking and smoking, the model estimated influential, and harmful, direct effects on adenoma, suggesting that drinking and smoking may act through additional mechanisms.

An important limitation of the current study is that the biomarkers were measured from a single blood draw shortly before colonoscopy. The design cannot establish directionality of the oxidative stress–adenoma association, although the application of SEM assumes that exposure precedes the outcome. There is evidence suggesting that colorectal cancer may affect reactive oxygen metabolites (39), and despite colorectal adenomas being more benign tissues than carcinomas, we cannot rule out that adenoma growth had some effect on the circulating biomarkers. Despite this limitation, the analysis demonstrates that a latent variable approach consisting of multiple biomarkers may produce stronger associations with a health outcome than individual biomarkers. We cannot be certain that the latent variable in this study exclusively, or wholly, reflects the construct of oxidative stress, which may require additional biomarkers. As oxidative stress is both a cause and consequence of inflammation (2), it is possible the latent variable represents some combination of these two interrelated processes. We did, however, observe no material difference when we excluded CRP from the analysis, and NSAID use in the current study was associated with CRP and colorectal adenoma but not via the oxidative stress pathway. Finally, we note that separation of direct and indirect effects is difficult and requires strong assumptions (50). Although we attempted to adjust for as many confounders available to us, we suggest caution interpreting the magnitudes of the pathway associations.

Future studies of oxidative stress should consider the limitations noted for the current study. As colorectal neoplasms tend to grow slowly, only a prospective design, preferably with multiyear follow-up, can assure the proper temporal relation between blood biomarkers and adenoma incidence. Although we validated the oxidative stress latent variable in a previous study, further use of SEM with more oxidative stress biomarkers could help clarify this biologic phenomenon and its role in disease etiology and progression. The SEM methodology can be applied in other areas of research where it is necessary to combine different but imperfect measurements to describe a complex biologic phenomenon.

No potential conflicts of interest were disclosed.

Conception and design: R.C. Eldridge, M. Goodman, R.M. Bostick, M. Gross, W.D. Flanders

Development of methodology: R.C. Eldridge, W.D. Flanders

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Goodman, R.M. Bostick, M. Gross, B. Thyagarajan

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): R.C. Eldridge, W.D. Flanders

Writing, review, and/or revision of the manuscript: R.C. Eldridge, M. Goodman, R.M. Bostick, V. Fedirko, B. Thyagarajan, W.D. Flanders

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): V. Fedirko

This work was supported by the Laney Graduate School, Emory University, R01 CA66539, R01 CA116795, the Fullerton Foundation, and Georgia Cancer Coalition.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1.
Sies
H
. 
Oxidative stress: oxidants and antioxidants
.
Exp Physiol
1997
;
82
:
291
5
.
2.
Reuter
S
,
Gupta
SC
,
Chaturvedi
MM
,
Aggarwal
BB
. 
Oxidative stress, inflammation, and cancer: how are they linked?
Free Radic Biol Med
2010
;
49
:
1603
16
.
3.
Poulsen
HE
,
Prieme
H
,
Loft
S
. 
Role of oxidative DNA damage in cancer initiation and promotion
.
Eur J Cancer Prev
1998
;
7
:
9
16
.
4.
Klaunig
JE
,
Kamendulis
LM
. 
The role of oxidative stress in carcinogenesis
.
Annu Rev Pharmacol Toxicol
2004
;
44
:
239
67
.
5.
Papaioannou
D
,
Cooper
KL
,
Carroll
C
,
Hind
D
,
Squires
H
,
Tappenden
P
, et al
Antioxidants in the chemoprevention of colorectal cancer and colorectal adenomas in the general population: a systematic review and meta-analysis
.
Colorectal Dis
2011
;
13
:
1085
99
.
6.
Garland
M
,
Willett
WC
,
Manson
JE
,
Hunter
DJ
. 
Antioxidant micronutrients and breast cancer
.
J Am Coll Nutr
1993
;
12
:
400
11
.
7.
Byers
T
,
Perry
G
. 
Dietary carotenes, vitamin C, and vitamin E as protective antioxidants in human cancers
.
Annu Rev Nutr
1992
;
12
:
139
59
.
8.
Goodman
M
,
Bostick
RM
,
Kucuk
O
,
Jones
DP
. 
Clinical trials of antioxidants as cancer prevention agents: past, present, and future
.
Free Radic Biol Med
2011
;
51
:
1068
84
.
9.
Kohen
R
,
Nyska
A
. 
Oxidation of biological systems: oxidative stress phenomena, antioxidants, redox reactions, and methods for their quantification
.
Toxicol Pathol
2002
;
30
:
620
50
.
10.
Jones
DP
. 
Redefining oxidative stress
.
Antioxid Redox Signal
2006
;
8
:
1865
79
.
11.
Maiese
K
. 
Marking the onset of oxidative stress: biomarkers and novel strategies
.
Oxid Med Cell Longev
2009
;
2
:
1
.
12.
Dalle-Donne
I
,
Rossi
R
,
Colombo
R
,
Giustarini
D
,
Milzani
A
. 
Biomarkers of oxidative damage in human disease
.
Clin Chem
2006
;
52
:
601
23
.
13.
Valko
M
,
Leibfritz
D
,
Moncol
J
,
Cronin
MT
,
Mazur
M
,
Telser
J
. 
Free radicals and antioxidants in normal physiological functions and human disease
.
Int J Biochem Cell Biol
2007
;
39
:
44
84
.
14.
Bollen
KA
.
Structural equations with latent variables
.
New York, NY
:
Wiley
; 
1989
.
15.
Kline
RB
.
Principles and practice of structural equation modeling
. 2nd ed.
New York, NY
:
Guilford Press
; 
2005
.
16.
Eldridge
RC
,
Flanders
WD
,
Bostick
RM
,
Fedirko
V
,
Gross
M
,
Thyagarajan
B
, et al
Using multiple biomarkers and determinants to obtain a better measurement of oxidative stress: a latent variable structural equation model approach
.
Biomarkers
2017
;
22
:
517
24
.
17.
Lee
H-C
,
Yin
P-H
,
Lu
C-Y
,
Chi
C-W
,
Wei
Y-H
. 
Increase of mitochondria and mitochondrial DNA in response to oxidative stress in human cells
.
Biochem J
2000
;
348
:
425
32
.
18.
Cooney
RV
,
Franke
AA
,
Wilkens
LR
,
Gill
J
,
Kolonel
LN
. 
Elevated plasma gamma-tocopherol and decreased alpha-tocopherol in men are associated with inflammatory markers and decreased plasma 25-OH vitamin D
.
Nutr Cancer
2008
;
60
Suppl 1
:
21
9
.
19.
Fedirko
V
,
Bostick
RM
,
Goodman
M
,
Flanders
WD
,
Gross
MD
. 
Blood 25-hydroxyvitamin D3 concentrations and incident sporadic colorectal adenoma risk: a pooled case-control study
.
Am J Epidemiol
2010
;
172
:
489
500
.
20.
Morrow
JD
,
Roberts
LJ
. 
Mass spectrometric quantification of F2-isoprostanes in biological fluids and tissues as measure of oxidant stress
.
Methods Enzymol
1999
;
300
:
3
12
.
21.
Shimasaki
H
. 
Assay of fluorescent lipid peroxidation products
.
Methods Enzymol
1994
;
233
:
338
46
.
22.
Wu
T
,
Rifai
N
,
Roberts
LJ
 II
,
Willett
WC
,
Rimm
EB
. 
Stability of measurements of biomarkers of oxidative stress in blood over 36 hours
.
Cancer Epidemiol Biomarkers Prev
2004
;
13
:
1399
402
.
23.
De Leenheer
AP
,
De Bevere
VO
,
De Ruyter
MG
,
Claeys
AE
. 
Simultaneous determination of retinol and alpha-tocopherol in human serum by high-performance liquid chromatography
.
J Chromatogr
1979
;
162
:
408
13
.
24.
Thyagarajan
B
,
Wang
R
,
Barcelo
H
,
Koh
WP
,
Yuan
JM
. 
Mitochondrial copy number is associated with colorectal cancer risk
.
Cancer Epidemiol Biomarkers Prev
2012
;
21
:
1574
81
.
25.
Muthén
B
,
Asparouhov
T
. 
Causal effects in mediation modeling: an introduction with applications to latent variables
.
Struct Equ Modeling
2014
;
22
:
12
23
.
26.
Vanderweele
T
,
Vansteelandt
S
. 
Conceptual issues concerning mediation, interventions and composition
.
Stat Interface
2009
;
2
:
457
68
.
27.
MacKinnon
DP
,
Lockwood
CM
,
Williams
J
. 
Confidence limits for the indirect effect: distribution of the product and resampling methods
.
Multivariate Behav Res
2004
;
39
:
99
128
.
28.
Muthen
LK
,
Muthen
BO
.
Mplus User's Guide 7th edition
. Seventh ed.
Los Angeles, CA
:
Muthen & Muthen
; 
1998–2012
.
29.
Kong
SY
,
Bostick
RM
,
Flanders
WD
,
McClellan
WM
,
Thyagarajan
B
,
Gross
MD
, et al
Oxidative balance score, colorectal adenoma, and markers of oxidative stress and inflammation
.
Cancer Epidemiol Biomarkers Prev
2014
;
23
:
545
54
.
30.
Thyagarajan
B
,
Guan
W
,
Fedirko
V
,
Barcelo
H
,
Tu
H
,
Gross
M
, et al
No association between mitochondrial DNA copy number and colorectal adenomas
.
Mol Carcinog
2016
;
55
:
1290
6
.
31.
Siamakpour-Reihani
S
,
Scarbrough
PM
,
Wang
F
,
Spasojevic
I
,
Base
K
,
Sedjo
R
, et al
Systemic markers of oxidative status and colorectal adenomatous polyps
.
Ann Epidemiol
2012
;
22
:
587
91
.
32.
Halliwell
B
,
Lee
CY
. 
Using isoprostanes as biomarkers of oxidative stress: some rarely considered issues
.
Antioxid Redox Signal
2010
;
13
:
145
56
.
33.
Ingles
SA
,
Bird
CL
,
Shikany
JM
,
Frankl
HD
,
Lee
ER
,
Haile
RW
. 
Plasma tocopherol and prevalence of colorectal adenomas in a multiethnic population
.
Cancer Res
1998
;
58
:
661
6
.
34.
Tsilidis
KK
,
Branchini
C
,
Guallar
E
,
Helzlsouer
KJ
,
Erlinger
TP
,
Platz
EA
. 
C-reactive protein and colorectal cancer risk: a systematic review of prospective studies
.
Int J Cancer
2008
;
123
:
1133
40
.
35.
Kim
S
,
Keku
TO
,
Martin
C
,
Galanko
J
,
Woosley
JT
,
Schroeder
JC
, et al
Circulating levels of inflammatory cytokines and risk of colorectal adenomas
.
Cancer Res
2008
;
68
:
323
8
.
36.
Tsilidis
KK
,
Erlinger
TP
,
Rifai
N
,
Hoffman
S
,
Hoffman-Bolton
J
,
Helzlsouer
KJ
, et al
C-reactive protein and colorectal adenoma in the CLUE II cohort
.
Cancer Causes Control
2008
;
19
:
559
67
.
37.
Ognjanovic
S
,
Yamamoto
J
,
Saltzman
B
,
Franke
A
,
Ognjanovic
M
,
Yokochi
L
, et al
Serum CRP and IL-6, genetic variants and risk of colorectal adenoma in a multiethnic population
.
Cancer Causes Control
2010
;
21
:
1131
8
.
38.
Aleksandrova
K
,
Jenab
M
,
Bueno-de-Mesquita
HB
,
Fedirko
V
,
Kaaks
R
,
Lukanova
A
, et al
Biomarker patterns of inflammatory and metabolic pathways are associated with risk of colorectal cancer: results from the European Prospective Investigation into Cancer and Nutrition (EPIC)
.
Eur J Epidemiol
2014
;
29
:
261
75
.
39.
Leufkens
AM
,
van Duijnhoven
FJB
,
Woudt
SHS
,
Siersema
PD
,
Jenab
M
,
Jansen
EHJM
, et al
Biomarkers of oxidative stress and risk of developing colorectal cancer: a cohort-nested case-control study in the European Prospective Investigation Into Cancer and Nutrition
.
Am J Epidemiol
2012
;
175
:
653
63
.
40.
Giovannucci
E
,
Ascherio
A
,
Rimm
EB
,
Colditz
GA
,
Stampfer
MJ
,
Willett
WC
. 
Physical activity, obesity, and risk for colon cancer and adenoma in men
.
Ann Intern Med
1995
;
122
:
327
34
.
41.
Giovannucci
E
,
Colditz
G
,
Stampfer
M
,
Willett
W
. 
Physical activity, obesity, and risk of colorectal adenoma in women (United States)
.
Cancer Causes Control
1996
;
7
:
253
63
.
42.
Vincent
HK
,
Taylor
AG
. 
Biomarkers and potential mechanisms of obesity-induced oxidant stress in humans
.
Int J Obes
2006
;
30
:
400
18
.
43.
Khandekar
MJ
,
Cohen
P
,
Spiegelman
BM
. 
Molecular mechanisms of cancer development in obesity
.
Nat Rev Cancer
2011
;
11
:
886
95
.
44.
Cole
BF
,
Logan
RF
,
Halabi
S
,
Benamouzig
R
,
Sandler
RS
,
Grainge
MJ
, et al
Aspirin for the chemoprevention of colorectal adenomas: meta-analysis of the randomized trials
.
J Natl Cancer Inst
2009
;
101
:
256
66
.
45.
Podhaisky
H-P
,
Abate
A
,
Polte
T
,
Oberle
S
,
Schröder
H
. 
Aspirin protects endothelial cells from oxidative stress – possible synergism with vitamin E
.
FEBS Lett
1997
;
417
:
349
51
.
46.
van der Vaart
H
,
Postma
DS
,
Timens
W
,
ten Hacken
NH
. 
Acute effects of cigarette smoke on inflammation and oxidative stress: a review
.
Thorax
2004
;
59
:
713
21
.
47.
Giovannucci
E
,
Rimm
EB
,
Stampfer
MJ
,
Colditz
GA
,
Ascherio
A
,
Kearney
J
, et al
A prospective study of cigarette smoking and risk of colorectal adenoma and colorectal cancer in U.S. men
.
J Natl Cancer Inst
1994
;
86
:
183
91
.
48.
Giovannucci
E
,
Colditz
GA
,
Stampfer
MJ
,
Hunter
D
,
Rosner
BA
,
Willett
WC
, et al
A prospective study of cigarette smoking and risk of colorectal adenoma and colorectal cancer in U.S. women
.
J Natl Cancer Inst
1994
;
86
:
192
9
.
49.
Haggar
FA
,
Boushey
RP
. 
Colorectal cancer epidemiology: incidence, mortality, survival, and risk factors
.
Clin Colon Rectal Surg
2009
;
22
:
191
7
.
50.
Robins
JM
,
Greenland
S
. 
Identifiability and exchangeability for direct and indirect effects
.
Epidemiology
1992
;
3
:
143
55
.