Associations of Novel Lifestyle- and Whole Foods-based Inflammation Scores with Incident Colorectal Cancer among Women

Abstract

Background:

Chronic inflammation, associated with lifestyle and dietary factors, may contribute to colorectal carcinogenesis. To address this, we investigated associations of previously validated, inflammation biomarker panel-weighted, novel, 4-component lifestyle (LIS) and 19-component predominately whole foods-based dietary (DIS) inflammation scores with incident colorectal cancer (CRC) in the prospective Iowa Women’s Health Study (IWHS; 1986–2012; n=34,254, of whom 1,632 developed CRC).

Methods:

We applied the published scores’ components’ weights, summed the weighted components to constitute the scores (higher scores reflect a higher balance of pro-inflammatory exposures), and investigated LIS- and DIS-CRC associations using multivariable Cox proportional hazards regression.

Results:

The multivariable-adjusted hazards ratios (HR) and their 95% confidence intervals (CI) for CRC among participants in the highest relative to the lowest LIS and DIS quintiles were 1.47 (1.26, 1.72; P_trend<0.01) and 1.07 (0.91, 1.25; (P_trend=0.22), respectively. The corresponding findings for distal colon cancers were HR 1.78 (1.29, 2.47) and HR 1.34 (0.98, 1.84), respectively. Among those in the highest relative to the lowest joint LIS/DIS quintile, the HR for CRC was 1.60 (95% CI 1.30, 1.98).

Conclusions:

Our results suggest that a more proinflammatory lifestyle, alone and jointly with a more proinflammatory diet, may be associated with higher CRC risk.

Introduction

Colorectal cancer (CRC) remains the second leading cause of cancer death among men and women combined in the United States (1). Inflammation contributes to colorectal carcinogenesis by altering DNA and promoting tumor proliferation and angiogenesis (2–5). Epidemiologic findings indicate that dietary and lifestyle exposures may be associated with CRC risk, likely in part through affecting chronic inflammation (6–8). Therefore, reducing inflammation through dietary and lifestyle interventions may reduce CRC risk.

The contributions of dietary components and lifestyle characteristics to inflammation and CRC risk may be more substantial collectively than individually. To address this, several dietary inflammation scores, including the dietary inflammatory index (DII) (9) and the empirical dietary inflammatory index (EDII) (10), recently renamed the empirical dietary inflammatory pattern (EDIP) (11), were developed. In previous studies, a higher (more pro-inflammatory) DII was associated with higher CRC risk (12–21), although in one cohort, the estimated association among women was modest and not statistically significant (19). A higher (more pro-inflammatory) EDIP was also associated with higher CRC risk in men and women, separately and combined (11). However, the DII and EDIP have several limitations. The DII is heavily focused on classical nutrients, and so may not account for many other known/unknown dietary constituents that directly or indirectly affect inflammation. The EDIP was developed using a primarily data-driven (rather than hypothesis-driven) approach in a demographically, occupationally homogenous population, thus potentially limiting its generalizability to other populations. Neither score addresses lifestyle.

To address these limitations, using food frequency and lifestyle questionnaire responses in a diverse United States population, we previously developed novel, hypothesis-based, inflammation biomarker panel-weighted dietary (DIS) and lifestyle (LIS) inflammation scores (22). The predominantly whole foods-based DIS and the LIS were more strongly associated with biomarkers of inflammation than were either the DII or EDIP in three external populations (22). Herein, we report an investigation of DIS- and LIS-incident CRC associations in the prospective Iowa Women’s Health Study (IWHS).

Methods

Study Population and Data Collection

Details of the IWHS were previously published (22). Briefly, the IWHS is a prospective cohort study of postmenopausal women, established in 1986. Prospective participants were 55 to 69-year-old women identified through the 1985 Iowa Department of Transportation’s current drivers list. Half of these women were randomly selected, of whom the 99,826 with valid Iowa mailing addresses were mailed a questionnaire. Of these prospective participants, 41,836 (42.6%) completed the questionnaire and were eligible for study enrollment. In addition to the original questionnaire, follow-up questionnaires were mailed to the participants in 1987, 1989, 1992, 1997, and 2004. The study was approved by the Minnesota Institutional Review Board (IRB), and the current analysis was approved by the Emory University IRB.

On the baseline questionnaire, participants provided detailed information on demographics, diet, lifestyle, medical and family history, self-measured anthropometrics (validated in the study population (23)), and other factors. Body mass index (BMI) was calculated as weight divided by height squared (kg/m²). Dietary and supplement intakes over the previous 12 months were collected through a 127-item Willett food frequency questionnaire (FFQ), for which the validity and reliability in the study population was previously reported (23). Physical activity was assessed via questions regarding participants’ usual moderate and vigorous activity frequencies, and categorized as low, medium, and high (24). Diet and physical activity were not comprehensively reassessed until 2004, at which time only 68.3% of participants remained alive (therefore, we used baseline exposure information for our primary analyses, and incorporated 2004 exposure data in one of two sensitivity analyses that supported our primary analyses). Information on aspirin and other nonsteroidal anti-inflammatory drug (NSAID) use was not collected until 1992 and was used in sensitivity analyses described further below.

Cancer diagnoses were collected through linkage with the State Health Registry of Iowa, a participant in the National Cancer Institute’s Surveillance, Epidemiology, and End Results Program. Ascertainment of cancer diagnoses was nearly 100% (25). CRC was defined as adenocarcinoma of the colon or rectum (International Classification of Diseases-Oncology-3 codes: C18.0–18.9, C19.9, and C20.9). Deaths were identified through the State Health Registry of Iowa and the National Death Index.

Scores

The development and validation of the DIS and LIS were previously published (22). Briefly, the scores’ components were determined a priori based on their hypothesized contributions to systemic inflammation and ease of reconstruction in major epidemiologic studies using different FFQs. The components’ weights were developed using baseline data from a diverse subset (n=639) of Reasons for Geographic and Racial Differences in Stroke Study (REGARDS) cohort participants (26, 27) on whom a panel of circulating inflammation biomarkers was measured (high sensitivity C-reactive protein [hsCRP], interleukin [IL]-6, IL-8, and IL-10). REGARDS is a national, on-going prospective cohort study of 30,239 White, Black, male, and female participants ≥45 years old from the United States’ 48 contiguous states, initiated January 2003-October 2007. Participants >74 years old and with hsCRP concentrations ≥10 mg/dL, missing responses to >10% of the FFQ items, implausible energy intakes (<500 and >4,500 kcal/day for women, and <800 and >5,000 kcal/day for men), missing lifestyle questionnaire data, and ≥2 comorbidities (diabetes mellitus, heart disease, cirrhosis, chronic kidney disease) were excluded from analysis. The DIS components were based on Block 98 FFQ (28–30) responses, and the LIS components on lifestyle questionnaire responses and interviewer-measured anthropometrics.

The DIS and LIS components’ weights were calculated according to their multivariable-adjusted strengths of associations with an inflammation biomarker score comprised of the aforementioned biomarkers in the REGARDS subset. The biomarker score for each participant was created by transforming each biomarker by the natural logarithm (ln), then standardizing the ln-transformed biomarker values to a mean of 0 and a standard deviation of 1.0, and then summing the values. Then, the associations of the individual DIS and LIS components with the biomarker score were estimated using multivariable linear regression models, and the β-coefficient for each component-biomarker score association was taken as the component’s weight. The estimated DIS and LIS associations with inflammation biomarkers were similar across sex and race (22).

Calculating the DIS and LIS in the IWHS

The components of the 19-component DIS and 4-component LIS in the IWHS are summarized in Table 1. For inclusion in the 18 whole-foods group components of the DIS, we disaggregated mixed foods into their components using the “My Pyramid Equivalents Database” (31), and then added the disaggregated components to their respective DIS food groups. For the nineteenth DIS component, we calculated a supplement score by ranking supplemental micronutrient intakes into tertiles, which we assigned values of 0–2; then multiplied the values by +1 or −1 for a micronutrient’s hypothesized anti- or pro-inflammatory properties, respectively; and then summed the values for each participant. Then, each participant’s DIS was created by ln-transforming the components’ values, then standardizing each to a mean of 0 and a standard deviation of 1.0 based on the baseline distribution of intake among all participants, then multiplying the component’s value by its respective weight (see Table 1), and then summing the weighted components. A higher score indicated a more pro-inflammatory diet.

Table 1.

Components, descriptions, and weights of the components of the dietary (DIS) and lifestyle (LIS) inflammation scores in the Iowa Women’s Health Study.

Components	Descriptions	Weightsa
DIS components
Leafy greens and cruciferous vegetables	Broccoli, cabbage or coleslaw, cauliflower, Brussels sprouts, cooked or raw spinach, kale, mustard or chard greens, iceberg lettuce, romaine lettuce, endive, parsley, kohlrabi, and watercress	−0.14
Tomatoes	Tomatoes, tomato juice, and tomato sauce	−0.78
Apples and berries	Apple, apple juice or cider, strawberries, blueberries, apple sauce, fresh blackberries, fresh raspberries, and quince	−0.65
Deep yellow or orange vegetables and fruit	Cantaloupe, peach, carrots, carrot juice, persimmons, and figs	−0.57
Other fruits and real fruit juices	Artichoke, Crenshaw melon, fresh coconut, fresh currants, dates, grapefruit, honeydew, apricot juice, grapefruit juice, mango juice, orange juice, other fruit juices, papaya juice, fresh pineapple or pineapple juice, prune juice, kiwi fruit, lemons, limes, mangos, nectarines, oranges, olives, papayas, tangerines, and watermelon	−0.16
Other vegetables	Asparagus, beets, celery, celery juice, corn, daikon radish, eggplant, garlic, horseradish, Jerusalem artichokes, mixed vegetables, mushrooms, okra, parsnips, green or chili peppers, rutabaga, rhubarb, scallions, yellow squash, zucchini, or summer squash, turnips, and V8 juice	−0.16
Legumes	Beans, fava beans, string beans, peas, peapods, alfalfa sprouts, and bean sprouts	−0.04
Fish	Canned tuna fish, dark meat fish, and other fish	−0.08
Poultry	Chicken with and without skin	−0.45
Red and organ meats	Hamburger, liver, beef, pork, and lamb as a main dish or stew	0.02
Processed meats	Bacon, hotdogs, and other processed meats	0.68
Added sugars	Sweetened carbonated beverages, non-carbonated fruit drinks, candy bars, candy without chocolate, chocolate bars or pieces, dried fruits (apple, banana, papayas, peaches, pineapple, and mixed dried fruit), fruit cocktail, honey, jams, jellies, preserves, prunes, pudding, raisins or grapes, canned cherries, sweet pickles, and syrup	0.56
High-fat dairy	Cream, ice cream, sour cream, cream cheese, other high-fat cheese, cream sauce, sherbet or ice milk, whole milk, and yogurt	−0.14
Low-fat dairy	Low-fat cottage or ricotta cheese, and skim or low-fat milk	−0.12
Coffee and tea	Coffee (decaffeinated and caffeinated) and tea	−0.25
Nuts	Nuts, peanut butter, seeds, and water chestnuts	−0.44
Fats	Butter, gravy, margarine, and mayonnaise or other creamy dressing	0.31
Refined grains and starchy vegetables	Dark or white bread, brownies, home-baked or ready-made cakes, cold or other cooked breakfast cereal, cooked oatmeal, crackers, home-baked or ready-made cookies, doughnuts, granola bars or other granola, English muffin, bagels, rolls, muffins or biscuits, pancakes or waffles, pasta, home-baked or ready-made pastries, homemade or ready-made pie, popcorn, potatoes, French-fried potatoes, potato chips, brown rice, and yams	0.72
Supplement scoreb	Ranked score of supplements, including: β-carotene, B-complex vitamins, calcium, copper, folic acid, iron, magnesium, selenium, zinc, and vitamins A, C, D, and E	−0.80
LIS components c
Heavy drinker	Heavy (> 7 drinks/wk) vs. non-drinker	0.30
Moderate drinker	Moderate (1 – 7 drinks/wk) vs. non-drinker	−0.66
Moderately physically active	Vigorous activity 1 time/week and moderate activity 1 time/week, or moderate activity 2–4 times/week	−0.18
Heavily physically active	Vigorous activity ≥ 2 times/week or moderate activity ≥ 4 times/week	−0.41
Current smoker	Currently smoked tobacco at baseline vs. did not currently smoke tobacco	0.50
Overweight BMI	Overweight BMI (25 – 29.99 kg/m2) vs. normal/underweight BMI (< 25 kg/m2)	0.89
Obese BMI	Obese BMI (≥ 30 kg/m2) vs. normal/underweight BMI (< 25 kg/m2)	1.57

Abbreviations: BMI, body mass index; DIS, dietary inflammation score; hsCRP, high sensitivity C-reactive protein; IL, interleukin; LIS, lifestyle inflammation score; FFQ, food frequency questionnaire; NSAID, nonsteroidal anti-inflammatory drug; REGARDS, Reasons for Geographical and Racial Differences in Stroke study.

^aWeights are β−estimates from multivariable linear regression models estimating associations of the dietary and lifestyle components with a summary inflammation biomarker score (a sum of standardized circulating hsCRP, IL-6, IL-8, and IL-10 concentrations [the latter with a negative sign] in a subset (n = 639) of the REGARDS cohort. Participants had ≤ 1 chronic disease, were < 75 years old, had plausible energy intake (500 – 6,000 kcal/day), answered ≥ 90% of FFQ questions, had hsCRP concentrations < 10 mg/L, and had non-outlying values for other inflammation biomarker concentrations. All weights are adjusted for sex, hormone replacement therapy (among women), race (Black or White), education (less than high school or high school graduate vs. some college or more), self-reported regular use of aspirin, NSAID, or lipid lowering medication (≥ 2 times/wk), region of residence in United States (Stroke Belt, Buckle, Other), season of baseline interview (Spring, Summer, Fall, or Winter), comorbidity (history of cancer, heart disease, diabetes, or chronic kidney disease), age (continuous), and total energy intake (kcal/day), and all dietary and lifestyle components in the DIS and LIS.

^bAll supplemental micronutrients from individual supplements and multivitamins were ranked into tertiles of intake and assigned values of 0, 1, or 2 for predominantly anti-inflammatory supplements (β-carotene, B-complex vitamins, calcium, folic acid, magnesium, selenium, zinc, and vitamins A, C, D, and E) and 0, −1, or −2 for predominantly pro-inflammatory supplements (copper and iron) as described in the text.

^cAll lifestyle components were dummy variables, coded as 1 for the non-referent category and 0 for the referent category.

The LIS comprised alcohol consumption, physical activity, smoking status, and BMI. Alcohol consumption was categorized as none, moderate (>0 to ≤7 drinks/week), and heavy (>7 drinks/week). Physical activity was categorized as heavy (defined as vigorous activity twice a week or moderate activity >4 times/week), moderate (vigorous activity once a week and moderate activity once a week, or moderate activity 2–4 times/week), or low. Baseline smoking status was categorized as “current” or “former and never” (former and never combined since former was not considered to be contributing to current inflammation). Baseline BMI was categorized as normal/underweight (<25 kg/m²; inclusion/exclusion of underweight with normal weight made no difference in the development/validation study or the present study), overweight (25–29.99 kg/m²), or obese (≥30 kg/m²). Each LIS component’s value was multiplied by its respective weight (see Table 1), and the weighted values were summed. A higher score indicated a more pro-inflammatory lifestyle (22).

Statistical Analyses

Participants with a history of cancer (other than nonmelanoma skin cancer) at baseline (n=3,830), who skipped >30 FFQ questionnaire items (n=2,499), had self-reported implausible energy intakes (<600 or >5,000 kcal/day; n=286), or were missing data on any LIS component (n=967) were excluded from analysis, leaving an analytic cohort of 34,254. Follow-up time was calculated as the time between the date of baseline questionnaire completion and the date of a first CRC diagnosis, the date of moving from Iowa, the date of death, or the end of the last follow up (12/31/2012), whichever was first.

Selected characteristics of participants, by score quintiles, were summarized and compared using the χ² test for categorical variables and analysis of variance for continuous variables (ln-transformed to meet the normality assumption, when indicated). Associations of the various baseline scores with incident CRC were estimated using multivariable Cox proportional hazards regression to calculate hazards ratios (HR) and their 95% confidence intervals (CI). The scores were analyzed as both continuous and categorical variables based on the distributions of all participants’ scores at baseline. Trend tests were calculated using the median value of each LIS or DIS quintile. Correlation between the DIS and LIS was assessed using a Spearman correlation coefficient.

Based on previous literature and biological plausibility, the following variables were considered potential confounders and included in all fully-adjusted models: age (years; continuous), education (< high school, high school, > high school and < college, ≥ college), family history of CRC in a first-degree relative (yes/no), hormone replacement therapy use (current, past, never), comorbidity score (includes sum of yes/no for diabetes, heart disease, and cirrhosis), and total energy intake (kcal/day; continuous). For the DIS model, we also included smoking status (current, past, never smoker), alcohol use (servings/week; continuous), physical activity (low, medium, high), and BMI (kg/m²; continuous). For the LIS model, we also included former smoker (yes/no), since it is not included in the LIS but has been associated with higher CRC risk; and an equally-weighted dietary inflammation score, to account for the dietary components’ inflammation plus other potential colorectal carcinogenic-related effects. Prior to conducting Cox proportional hazards regression, the proportional hazards assumption was assessed by calculating Schoenfeld residuals. No variables violated the proportional hazards assumption. These analyses were also repeated by CRC subsite, including the proximal colon (cecum through transverse colon), distal colon (splenic flexure through sigmoid colon), and rectum. To test for heterogeneity based on CRC subsite, we conducted a case-only multivariable logistic regression analysis with CRC subsite as the dependent variable, and took the P-values for the continuous DIS and LIS to be the P_{heterogeneities}.

A joint/combined (cross-classification) analysis was conducted to assess potential interaction between the DIS and LIS. The reference group was participants in the first quintile of both scores. We entered a categorical DIS x LIS interaction term into the multivariable joint/combined model to calculate a P_interaction.

To assess whether associations differed by categories of selected participant characteristics, we conducted separate analyses within each category of: age (≤/> median age of 61 years), education (≤ high school/> high school), and HRT use (current or past/never) for the DIS and LIS, and smoking status (current and past/never), physical activity (low/medium and high), and BMI (≤25 kg/m²/>25 kg/m²) for the DIS only. We calculated a P_interaction by including a stratification factor x continuous score interaction term in the multivariable Cox proportional hazards regression models.

To assess the associations’ sensitivity to various considerations, we repeated the analyses with several variations. Since comprehensive data on diet and physical activity during follow up were not collected until 2004 and some participants could have changed their exposures somewhat during follow up, we assessed DIS- and LIS-CRC associations 5, 10, 15, 20, and 25 years from baseline. To further address this issue, we incorporated exposure data from the 2004 follow-up questionnaire two ways: among those who were not censored prior to 2004, we conducted analyses using (i) the means of their baseline (1986) and 2004 follow-up DIS and LIS, and (ii) only their 2004 DIS and LIS. Since NSAID use was not ascertained until 1992, we repeated the analyses with 1992 as the baseline, and assessed whether or not including regular (once a week or more) aspirin and/or other NSAID use as a covariate affected the estimated associations. In other sensitivity analyses, we excluded participants who died or were diagnosed with CRC within 1 or 2 years of follow-up (to rule out a substantial reverse causality effect during early follow up); censored participants upon reaching the age of 75 or 80 (to rule out a substantial attenuating effect of chance due to aging); and used equally-weighted (22) rather than weighted DIS and LIS (to assess potential differences in associations primarily related to inflammation versus all mechanisms combined). We also assessed contributions of the individual DIS and LIS components to the DIS and LIS, respectively, by removing each component one at a time and adding it as a covariate in the multivariable Cox proportional hazards regression models. We calculated Spearman’s correlation coefficients to assess the correlation of each removed DIS and LIS component with the remaining DIS or LIS. Finally, we calculated the 18-component EDIP described by Tabung et al. (10) (see Supplemental Table 1 footnote ‘b’) and a DII, as described by Shivappa et al. (9), based on the 33 components available from the IWHS Willett FFQ (see Supplemental Table 2 footnote ‘b’), and assessed EDIP- and DII-CRC associations.

We conducted all analyses using SAS statistical software, version 9.4 (SAS Institute, Cary, NC). We considered 2-sided P-values ≤0.05 or 95% CIs that excluded 1.0 statistically significant.

Results

Selected participants’ baseline characteristics by DIS and LIS quintiles are summarized in Table 2. Participants in the higher relative to the lower DIS and LIS quintiles tended to be less formally educated, less likely to take HRT, less physically active, and less likely to take a multivitamin/mineral supplement, and had lower mean total calcium intakes. Participants in the higher DIS quintiles were also more likely to be current or past smokers, and had lower mean dietary fiber and higher mean saturated fat intakes. Participants in the higher LIS quintiles were also more likely to have diabetes or heart disease, and had higher mean BMIs. In 1992, among those who completed the 1992 follow-up questionnaire (n = 27,891), those in the higher DIS and those in the lower LIS quintiles were more likely to regularly take a non-aspirin NSAID.

Table 2.

Selected baseline characteristics of participants according to quintiles of the dietary (DIS) and lifestyle (LIS) inflammation scores in the Iowa Women’s Health Study (n = 34,254), 1986 – 2012

	Dietary inflammation score (DIS)a quintiles				Lifestyle inflammation score (LIS)b quintiles
Characteristicsc	1 (n = 6,850)	3 (n = 6,851)	5 (n = 6,851)	P d	1 (n = 6,798)	3 (n = 5,796)	5 (n = 6,909)	P d
Age (years)	61.4 ± 4.2	61.6 ± 4.2	61.4 ± 4.2	0.11	61.4 ± 4.2	61.5 ± 4.2	61.4 ± 4.2	0.09
White race (%)	98.2	98.4	97.8	0.04	98.8	98.2	97.6	<0.01
> High school education (%)	49.9	40.3	28.6	<0.01	48.4	39.9	33.6	<0.01
Family history of CRCe (%)	3.1	3.2	2.8	0.71	3.4	3.1	3.0	0.52
Current or past smoker (%)	34.4	32.8	37.0	<0.01	31.9	45.5	38.4	<0.01
Alcohol intake (g/day)	4.0 ± 8.8	3.8 ± 8.9	3.7 ± 9.6	0.33	3.3 ± 3.7	5.6 ± 11.1	4.1 ± 11.5	<0.01
Current or past HRT use (%)	43.3	38.6	34.9	<0.01	42.1	39.7	34.5	<0.01
Take an NSAIDf (%)	26.4	24.1	22.1	<0.01	19.5	24.5	30.9	<0.01
Have comorbidityg (%)	16.2	13.8	13.5	<0.01	9.6	13.2	22.0	<0.01
High physical activityh (%)	38.6	24.0	14.0	<0.01	51.1	40.7	14.1	<0.01
Body mass index (kg/m2)	26.8 ± 4.9	26.8 ± 4.9	27.0 ± 5.3	<0.01	22.7 ± 1.7	26.0 ± 3.4	33.1 ± 4.8	<0.01
Take multivitamin (%)	59.0	30.6	12.2	<0.01	37.1	34.5	28.5	<0.01
Total energy intake (kcal/day)	1,817 ± 614	1,782 ± 606	1,826 ± 625	<0.01	1,803 ± 583	1,811 ± 625	1,796 ± 622	0.09
Total fat intake (%kcal/day)	31.4 ± 5.6	34.1 ± 5.3	36.2 ± 5.8	<0.01	33.2 ± 5.7	33.9 ± 6.0	34.6 ± 5.8	<0.01
Saturated fat intake (g/1,000 kcal/day)	12.1 ± 2.7	13.3 ± 2.6	14.2 ± 3.0	<0.01	12.8 ± 2.8	13.2 ± 3.0	13.5 ± 2.9	<0.01
Protein intake (%kcal/day)	19.7 ± 3.4	18.0 ± 2.9	16.6 ± 3.0	<0.01	18.0 ± 3.1	18.0 ± 3.4	18.4 ± 3.4	<0.01
Carbohydrates intake (%kcal/day)	50.0 ± 7.9	48.7 ± 7.4	47.7 ± 7.9	<0.01	50.0 ± 7.4	48.1 ± 7.9	47.5 ± 8.0	<0.01
Dietary fiber intake (g/1,000 kcal/day)	24.2 ± 9.3	19.4 ± 7.1	16.2 ± 6.6	<0.01	11.8 ± 3.4	11.0 ± 3.3	10.8 ± 3.2	<0.01
Total calcium intakei (mg/1,000 kcal/day)	789 ± 371	630 ± 318	493 ± 262	<0.01	688 ± 342	635 ± 350	599 ± 321	<0.01

Abbreviations: CRC, colorectal cancer; HRT, hormone replacement therapy; NSAID, nonsteroidal anti-inflammatory drug.

^aFor score construction, see text and Table 1; a higher score indicates a more pro-inflammatory diet.

^bFor score construction, see text and Table 1; a higher score indicates a more pro-inflammatory lifestyle.

^cContinuous variables presented as mean ± standard deviation, and categorical variables as percentages.

^dP values from the χ² test for categorical variables and one-way analysis of variance (ANOVA) for continuous variables.

^eHistory of a first degree relative with colorectal cancer.

^fRegularly take a non-aspirin NSAID once a week or more. NSAID use was not collected until the 1992 follow up, so calculations are based on the 1992 follow up population (n = 27,891); the quintile sample sizes for the DIS were 1 (n = 5,567), 3 (n = 5,590), and 5 (n = 5,576), and for the LIS they were 1 (n = 5,513), 3 (n = 5,797), and 5 (n = 5,375).

^gSelf-reported history of diabetes mellitus, heart disease, and/or cirrhosis.

^hPhysical activity level derived from two questions regarding the frequency of moderate and vigorous physical activity (24), and categorized as high (vigorous activity twice a week or moderate activity > 4 times/week), medium (vigorous activity once a week plus moderate activity once a week, or moderate activity 2 – 4 times/week), and low.

^ITotal = diet + supplements.

DIS- and LIS-incident CRC associations are presented in Table 3. Adjustment for known and suspected risk factors had minimal effect on the risk estimates. In the multivariable-adjusted analysis, CRC risk was statistically significantly 16% higher per 1-point increase in the LIS. When analyzed by quintiles, there was a statistically significant trend for increasing CRC risk with an increasing LIS, and participants in the highest relative to the lowest quintile had statistically significant 47% higher risk. In the multivariable-adjusted analysis, CRC risk was estimated to be non-statistically significantly 1% higher per 1-point increase in the DIS. When analyzed by quintiles, there was a pattern of increasing risk with an increasing DIS, and those in the fifth relative to the first quintile had an estimated 7% higher risk, although these findings were not statistically significant.

Table 3.

Associations^a of the dietary inflammation score (DIS) and the lifestyle inflammation score (LIS) with incident colorectal cancer; the Iowa Women’s Health Study (n = 34,254), 1986 – 2012.

	Inflammation scores
Score variable	Dietaryb				Lifestylec
	Minimally-adjusted modeld		Fully-adjusted modele		Minimally-adjusted modeld		Fully-adjusted modelf
	HR	95% CI	HR	95% CI	HR	95% CI	HR	95% CI
Continuous	1.02	1.00, 1.04	1.01	0.99–1.03	1.19	1.12–1.26	1.16	1.09–1.24
Quintiles
1	1.00	Referent	1.00	Referent	1.00	Referent	1.00	Referent
2	0.95	0.81, 1.11	0.94	0.80, 1.10	1.20	1.02, 1.40	1.18	1.01, 1.38
3	1.19	1.03, 1.38	1.16	1.00, 1.35	1.28	1.08, 1.50	1.25	1.06, 1.47
4	1.08	0.93, 1.26	1.04	0.89, 1.21	1.30	1.11, 1.51	1.26	1.08, 1.48
5	1.13	0.97, 1.32	1.07	0.91, 1.25	1.54	1.32, 1.80	1.47	1.26, 1.72
P trend	0.04		0.22		<0.01		<0.01

Abbreviations: CI, confidence interval; DIS, dietary inflammation score; HR, hazards ratio; LIS, lifestyle inflammation score.

^aHRs and 95% CIs from Cox proportional hazards models.

^bFor score construction, see text and Table 1; a higher score indicates a more pro-inflammatory diet.

^cIncludes smoking, physical activity, alcohol use, and body mass index; for score construction, see text; a higher score indicates a more pro-inflammatory lifestyle

^dCovariates included age (years; continuous) and total energy intake (kcal/day; continuous).

^eCovariates for DIS model included age (years; continuous), education (< high school, high school, > high school and < college, ≥ college), family history of colorectal cancer in a first-degree relative (yes/no), smoking status (current, past, never smoker), alcohol use (servings/week; continuous), comorbidity score (includes sum of yes/no for diabetes, heart disease, and cirrhosis), hormone replacement therapy use (current, past, never), physical activity (low, medium, high), body mass index (weight [kg]/height [m]²; continuous), and total energy intake (kcal/day; continuous).

^fCovariates for LIS model included age (years; continuous), education (< high school, high school, > high school and < college, ≥ college), family history of colorectal cancer in a first-degree relative (yes/no), comorbidity score (includes sum of yes/no for diabetes, heart disease, or cirrhosis), hormone replacement therapy use (current, past, never use), total energy intake (kcal/day; continuous), former smoker (yes/no), and unweighted dietary inflammation score (DIS).

The joint/combined associations of the DIS and LIS with CRC risk are presented in Table 4. There were patterns of increasing risk with an increasing DIS among those in the lower LIS quintile, and of increasing risk with an increasing LIS among those in the lower DIS quintile. CRC risk was highest (statistically significantly 60% higher) among those in the highest relative to the lowest joint DIS/LIS quintile (P_interaction=0.23). The Spearman correlation coefficient for correlation of the DIS and LIS was 0.11.

Table 4.

Joint/Combined Associations^a of the dietary (DIS) and lifestyle (LIS) inflammation scores^b with incident colorectal cancer; the Iowa Women’s Health Study (n = 34,254), 1986 – 2012.

	LIS quintiles
	1			2		3		4		5
	n	HR (95% CI)	n	HR (95% CI)	n	HR (95% CI)	n	HR (95% CI)	n	HR (95% CI)
DIS quintiles
1	1,727	1.00 (ref)	1,459	1.19 (1.01, 1.39)	1,208	1.26 (1.06, 1.48)	1,226	1.27 (1.09, 1.48)	1,230	1.48 (1.26, 1.73)
2	1,460	0.94 (0.80, 1.10)	1,509	1.12 (0.90, 1.39)	1,194	1.18 (0.94, 1.48)	1,383	1.19 (0.96, 1.48)	1,305	1.39 (1.12, 1.73)
3	1,389	1.17 (1.01, 1.36)	1,519	1.39 (1.12, 1.72)	1,155	1.47 (1.18, 1.83)	1,462	1.48 (1.20, 1.83)	1,326	1.73 (1.39, 2.14)
4	1,209	1.05 (0.90, 1.22)	1,485	1.24 (1.00, 1.54)	1,103	1.32 (1.05, 1.65)	1,575	1.33 (1.07, 1.65)	1,479	1.55 (1.25, 1.92)
5	1,013	1.09 (0.93, 1.27)	1,425	1.29 (1.04, 1.60)	1,136	1.36 (1.09, 1.70)	1,708	1.38 (1.11, 1.71)	1,569	1.60 (1.30, 1.98)

Abbreviations: CI, confidence interval; DIS, dietary inflammation score; HR, hazards ratio; LIS, lifestyle inflammation score; ref, reference.

^aHRs and 95% CIs from Cox proportional hazards models; covariates included age (years; continuous), education (< high school, high school, > high school and < college, ≥ college), family history of colorectal cancer in a first-degree relative (yes/no), comorbidity score (includes sum of yes/no for diabetes, heart disease, and cirrhosis), hormone replacement therapy use (current, past, never), former smoker (yes/no), and total energy intake (kcal/day; continuous).

^bFor construction of scores, see text and Table 1; a higher score indicates a more pro-inflammatory diet or lifestyle.

Multivariable-adjusted associations of the DIS and LIS with incident CRC, by colorectal sites, are presented in Table 5. Both scores were more strongly directly associated with distal than with proximal colon cancers. For the DIS-distal colon cancer association, risk was borderline statistically significantly 4% higher per 1-point increase in the DIS; when analyzed by quintiles, there was a statistically significant trend for increasing risk with an increasing DIS, and those in the highest relative to the lowest DIS quintile had statistically significant 34% higher risk. However, the DIS-proximal colon cancer association was close to null and not statistically significant. For the LIS, there were statistically significant trends for increasing risk for both distal and proximal colon cancers with an increasing LIS; those in the highest relative to the lowest LIS quintile had statistically significant 78% and 43% higher risk for distal colon and proximal colon cancers, respectively. The associations of both scores with rectal cancer were close to null.

Table 5.

Adjusted Associations^a of dietary (DIS) and lifestyle (LIS) inflammation scores with incident colorectal cancer according to colorectal cancer site; the Iowa Women’s Health Study (n = 34,254), 1986 – 2012.

	Dietary inflammation scoreb		Lifestyle inflammation scorec
Colorectal site	Adjusted HR	95% CI	Adjusted HR	95% CI
Proximal colond (n = 890)
Continuous	1.00	0.97, 1.03	1.16	1.06, 1.26
Quintiles
1	1.00	Referent	1.00	Referent
2	0.85	0.69, 1.06	1.10	0.89, 1.36
3	1.09	0.89, 1.33	1.16	0.92, 1.46
4	0.97	0.78, 1.19	1.21	0.98, 1.50
5	0.95	0.77, 1.18	1.43	1.16, 1.77
Ptrend	0.96		<0.01
Pheterogeneity	0.01		0.21
Distal colone (n = 422)
Continuous	1.04	1.00, 1.09	1.24	1.09, 1.41
Quintiles
1	1.00	Referent	1.00	Referent
2	0.90	0.64, 1.25	1.39	1.00, 1.94
3	1.40	1.03, 1.90	1.68	1.20, 2.35
4	1.10	0.80, 1.52	1.49	1.07, 2.06
5	1.34	0.98, 1.84	1.78	1.29, 2.47
Ptrend	0.03		<0.01
Pheterogeneity	0.27		0.05
Rectum (n = 236)
Continuous	1.00	0.95, 1.06	1.01	0.85, 1.19
Quintiles
1	1.00	Referent	1.00	Referent
2	1.16	0.77, 1.75	1.16	0.78, 1.73
3	0.95	0.61, 1.47	1.11	0.72, 1.70
4	1.33	0.89, 2.00	1.06	0.70, 1.59
5	1.18	0.77, 1.80	1.10	0.72, 1.67
Ptrend	0.31		0.83
Pheterogeneity	ref		ref

Abbreviations: CI, confidence interval; HR, hazards ratio; ref, referent.

^aHRs and 95% CIs from Cox proportional hazards models.

^bFor score construction, see text and Table 1; a higher score indicates a more proinflammatory diet; model covariates included age (years; continuous), education (< high school, high school, > high school and < college, ≥ college), family history of colorectal cancer in a first-degree relative (yes/no), smoking status (current, past, never smoker), alcohol use (servings/week; continuous), comorbidity score (sum of yes/no for diabetes, heart disease, and cirrhosis), hormone replacement therapy use (current, past, never), physical activity (low, medium, high), body mass index (weight [kg]/height [m]²; continuous), former smoker (yes/no), and total energy intake (kcal/day; continuous).

^cIncludes current smoking (yes/no), physical activity, alcohol use, and body mass index; for score construction, see text; a higher score indicates a more proinflammatory lifestyle; model covariates included age (years; continuous), education (< high school, high school, > high school and < college, ≥ college), family history of colorectal cancer in a first-degree relative (yes/no), comorbidity score (includes sum of yes/no for diabetes, heart disease, and chronic colitis), hormone replacement therapy use (current, past, never use), former smoker (yes/no), total energy intake (kcal/day; continuous), and unweighted dietary inflammation score (DIS).

^dIncludes cecum, ascending colon, hepatic flexure, and transverse colon.

^eIncludes splenic flexure, descending colon, and sigmoid colon.

We found no consistent, clear patterns of differences in associations of the DIS with incident CRC according to age, education, HRT use, smoking status, physical activity, or BMI, or of the LIS with CRC according to education (Supplement Table 3). The LIS-CRC association tended to be stronger among those who were younger (≤ the median age of 61 years), and those who did not take HRT; however, the 95% CIs for the estimated HRs for the corresponding quintiles across strata overlapped substantially and no P_interaction was statistically significant.

In sensitivity analyses, censoring participants after 5, 10, 15, 20, and 25 years of follow up (Supplement Table 4) yielded similar results, and incorporating 2004 exposure information two different ways did not materially affect our estimated DIS- or LIS-CRC associations (Supplement Table 5). When we used 1992 as baseline, inclusion/exclusion of regular aspirin and/or other NSAID use as model covariates did not materially affect the estimated associations (Supplement Table 6). Neither excluding those who died or were diagnosed with CRC within 1 or 2 years of follow-up (Supplement Table 7), nor censoring participants upon reaching the ages of 75 or 80 years (Supplement Table 8), materially affected our estimated DIS- or LIS-CRC associations. The findings for the equally-weighted and weighted LIS were similar to each other, and those for the equally-weighted DIS were slightly stronger than those for the weighted DIS (Supplement Table 9). We also found that removing any individual DIS or LIS component did not substantially affect the DIS- and LIS-CRC associations (Supplement Table 10). Whereas among those in the highest relative to the lowest DIS quintile, the estimated multivariable-adjusted HR for the DIS-CRC association was 1.07 (95% CI 0.91, 1.25) (Table 3), and for the EDII- and DII-CRC associations the corresponding HRs (95% CIs) were 1.04 (0.88, 1.22) (Supplement Table 1) and 1.16 (0.99, 1.37) (Supplement Table 2), respectively.

Discussion

Our findings suggest that a more proinflammatory lifestyle, alone and jointly with a more proinflammatory diet, may be associated with higher risk for CRC, especially distal colon cancers, in women.

There is strong biological and epidemiological support for a role of inflammation in colorectal carcinogenesis. Expression of proinflammatory cyclooxygenase-2 (COX-2) was repeatedly observed to be progressively elevated in colorectal carcinogenesis (32), and COX-2 inhibitors, such as aspirin and other NSAIDs, are consistently reported to be inversely associated with CRC risk (7, 33–35). For example, in a pooled analysis of four randomized controlled trials (n=14,033), aspirin statistically significantly reduced 20-year CRC risk by 24% (36). In addition, in a systematic review and meta-analysis of 13 case-control and 10 cohort studies, non-aspirin NSAID use was statistically significantly associated with 19% lower CRC risk among women (37). Finally, inflammatory bowel disease is strongly associated with CRC risk—risk that is progressively higher with the greater the extent of bowel involved (38).

As summarized previously (22), there is strong plausibility that the components of the DIS and LIS affect inflammation and CRC risk. Whereas many past studies on dietary components and inflammation focused on selected dietary nutrients, these nutrients are consumed as parts of whole foods and beverages comprising numerous constituents that may be acting separately and/or interacting with each other to collectively affect inflammation (39, 40). Substantial evidence supports that components of the LIS may be strongly associated with inflammation (41–47). We previously assessed associations of the DIS and LIS with circulating inflammation biomarker concentrations in the entire REGARDS cohort (excluding the development population; n=14,210) (22). In that study, participants in the highest relative to the lowest DIS, LIS, DII, and EDIP quintiles had statistically significant 1.66-, 4.29-, 1.56-, and 1.32-fold higher odds of a clinically high hsCRP concentration (>3 mg/dL), respectively (all P_trends<0.001). Furthermore, those in the highest relative to the lowest joint DIS/LIS quintile had statistically significant 7.26-fold higher odds of a high hsCRP concentration. Similar findings were noted in two other validation populations. Those results support 1) that dietary and lifestyle exposures collectively—especially in interaction with one another—contribute substantially to systemic inflammation, and 2) the use of our novel DIS and LIS.

Our LIS components—obesity, smoking, heavy alcohol intake, and physical activity—individually have been consistently associated with CRC risk (the first three directly and the latter inversely) (48–52), and they have also been collectively associated with risk (53, 54). In the IWHS, a higher relative to a lower “evolutionary-concordance” lifestyle score, which includes smoking, BMI, and physical activity, was associated with 34% lower CRC risk (53). However, that score’s components were not weighted according to associations with inflammation biomarkers since that score was intended to reflect all mechanisms related to CRC risk, whereas our LIS components were weighted so that they would primarily reflect the components’ contributions to inflammation as the mechanism of interest.

The components of our DIS, individually, have generally been less consistently associated with CRC risk (55); however, emerging evidence suggests that the components collectively may be more substantially associated with risk. Several dietary patterns that encompass many of these dietary components, including the Mediterranean and evolutionary concordance diet (rich in fruits, vegetables, eggs, nuts, lean meats, and calcium; very low in sodium; excludes grains, dairy products, and refined fats and sugar) patterns (53) and the Healthy Eating Index (56), have been inversely associated with colorectal neoplasms. Importantly, in the IWHS, there appeared to be synergism between the evolutionary concordance diet and lifestyle pattern scores in relation to CRC risk, such that those who were most concordant with both the evolutionary concordance diet and lifestyle patterns were at lowest risk (53). The DII was directly associated with CRC risk in 10 of 10 studies (12–21), including the IWHS (17), and the EDIP was directly associated with CRC risk in two of two cohorts (11, 57). The DIS and LIS, separately and jointly, were directly associated with CRC risk in the prospective NIH-AARP Diet and Health Study (n=453,465), associations that were stronger among men (58).

None of the point estimates for the associations of the DIS, EDII, and DII with CRC among those in the highest relative to the lowest score quintiles was statistically significant in our analyses. However, our estimated direct DIS-CRC association was modestly stronger than our estimated EDII-CRC association, but modestly weaker than the DII-CRC association. We note, however, that in the IWHS after 24 years of follow up (vs. after 26 years of follow up in the present analyses), the DII-CRC association for those in the highest relative to the lowest DII quintile was reported to be HR 1.20 (95% CI 1.01, 1.43) (17). The reason(s) for the stronger estimated DII-CRC association is unclear. The reported DII analysis was done after two years less follow up, but perhaps more importantly, the model covariates did not include family history of CRC in a first degree relative and physical activity, both important CRC risk factors. When we updated the DII-CRC analysis with the full 26 years of follow up, and added family history of CRC in a first degree relative and physical activity as covariates to make the model covariates comparable to those for the DIS- and EDII-CRC association models, the HR for those in the highest relative to the lowest DII quintile became more attenuated and was not statistically significant. Perhaps most important is that regardless of the three different methods to estimate dietary contributions to systemic inflammation, we estimated all to be modestly, directly associated with CRC. Although the DIS was more strongly associated with circulating biomarkers of inflammation than were the DII and the EDII among both men and women in three populations, in the present study, among those in the highest relative to the lowest score quintiles, none of the HRs for the associations of the scores with CRC were statistically significant, and thus not definitively different from one another. We also note that in the IWHS population, the estimated associations of the Mediterranean and evolutionary concordance diet pattern scores with CRC were null (HRs [95% CIs] for those in the fifth relative to the first quintiles were 1.01 [0.86, 1.18] and 1.01 [0.85, 1.19], respectively) (53). As we pointed out in the latter paper, the close-to-null dietary pattern-CRC associations in the IWHS may be because the diets across the study participants were more homogeneous than those in other study populations.

Another consideration concerning the modest estimated associations of the three dietary inflammation indices with CRC is that dietary inflammation and other dietary scores tend to be more modestly associated with CRC among women than they are among men. As examples, in the prospective NIH-AARP Diet and Health Study, the direct associations of both the DII (19) and the DIS (58) with CRC were stronger among men. Also, several previous studies found no or weaker associations of different dietary patterns with CRC risk among women (59–61). The reason(s) for the associations’ differences by sex are unclear, but could be methodological (e.g., more biased reporting of dietary intakes among women (62)) and/or involve unidentified biological mechanisms. In the present study, we found that the LIS, alone and jointly with the DIS, was more strongly, directly associated with CRC risk in the IWHS than was the DIS alone, suggesting that lifestyle-related inflammation may play a bigger role in CRC risk than diet-related inflammation in women.

CRC pathogenesis may differ somewhat by colorectal site (63). The distal and proximal colons have different embryological origins, the fecal stream becomes progressively less fluid moving proximally to distally, and the microbiome may differ along the colon’s length (63, 64). Cancers related to DNA mismatch repair (65) and p16 tumor suppression gene inactivation (66) are more common in the proximal colon. Importantly, COX-2 expression was found to be greater in the distal colon (67), suggesting a possible stronger link between inflammation and distal CRC. These observations suggest that risk factors for, and etiologies of, CRCs at different anatomical subsites may differ somewhat. However, the findings from previous studies that examined CRC risk by anatomical subsite were inconsistent (37, 68, 69). We found that both the DIS and LIS were more strongly, directly associated with distal than with proximal colon and rectal cancers, suggesting that inflammation may increase risk for distal more than for proximal colon or rectal cancers, at least in women. Further studies are needed to assess whether this finding can be consistently replicated in other study populations, and to elucidate possible differences in effects of environmental risk factors on colorectal carcinogenesis at different anatomical subsites.

Strengths of our study include the novel DIS and LIS, which reflect diet and lifestyle contributions to inflammation, and in turn to CRC risk, and address several limitations of the DII and EDIP. The DII is primarily nutrient based, which may under-account for numerous whole food constituents that may affect inflammation. Although the EDIP is whole foods-based, it is a primarily data-driven score that was developed in a relatively demographically, occupationally homogeneous population (nurses in the Nurses’ Health Study), potentially limiting its generalizability to other populations. Also, our LIS, which is the first reported lifestyle inflammation score, was more strongly associated with CRC than was any of the dietary inflammation scores, and, furthermore, appeared to interact with diet to be most strongly associated with risk. Other study strengths included the large sample size and number of cases, which allowed for stratified analyses; the prospective design; accurate and complete data on cancer diagnoses; comprehensive data on potential confounding variables; the use of a validated dietary assessment instrument; and robustness to multiple sensitivity analyses.

Our study also had several limitations. Important exposure data were collected only in 1986 (baseline) and 2004, and some participants’ exposures may have changed during follow up. However, such measurement error in prospective cohort studies is considered non-differential (because participants at baseline do not know their eventual outcomes), which tends to attenuate associations. Also, a previous study on FFQ long-term reproducibility found that 60–70% of participants remained in the same or adjacent food group quintiles after a mean 68.8 (SD 4.1) months of follow-up (70), suggesting that dietary intake estimates from a single FFQ may be good estimates of long-term dietary exposures. Importantly, in our sensitivity analyses, the estimated inflammation score-CRC associations, for both the DIS and LIS, were similar across 5-year follow-up intervals, and including 2004 exposure data two different ways negligibly affected our estimated inflammation score-CRC associations. We did not collect data on aspirin and other NSAID use until six years after baseline; however, when we used 1992 rather than 1986 as baseline, adjustment for aspirin/NSAID use did not meaningfully affect our findings. FFQs have known limitations, such as recall error and limited detail on food preparation (71); however, these types of measurement error in prospective cohort studies are considered non-differential, likely attenuating true associations. Also, we lacked data on CRC screening, which also may have resulted in attenuated observed associations. This is because no matter how high risk someone’s diet or lifestyle may be, if via CRC screening their adenomas (the immediate precursor to most CRCs (72, 73)) are detected and removed, they are unlikely to develop CRC. So, in a sense, these patients are ‘misclassified’, thus attenuating what the associations may have been had there been no screening. Finally, our study population included only women in Iowa, 99% of whom were White, thus possibly limiting the generalizability of our findings.

In conclusion, our findings, taken together with previous literature, support that a more pro-inflammatory lifestyle, alone and jointly with a more pro-inflammatory diet, may be directly associated with CRC risk, especially distal colon cancer risk, in women. Our study supports the use of the DIS and LIS in further research on inflammation and CRC.