The Dietary Inflammatory Index is Associated with Colorectal Cancer in the National Institutes of Health-American Association of Retired Persons Diet and Health Study

Abstract

Diet is a strong moderator of systemic inflammation, an established risk factor for colorectal cancer (CRC). The Dietary Inflammatory Index (DII) measures the inflammatory potential of individuals' diets. The association between the DII and incident CRC was examined using the National Institutes of Health-American Associations of Retired Persons Diet and Health Study individuals (n=489,422) aged 50-74 years at recruitment, starting between 1995-1996, and followed for a mean of 9.1(±2.9) years. Baseline data from a food frequency questionnaire were used to calculate the DII; higher scores are more pro-inflammatory and lower scores are more anti-inflammatory. First primary CRC diagnoses were identified through linkage to state cancer registries. Anatomic location and disease severity also were examined. Cox proportional hazards models estimated CRC hazard ratios (HRs) and 95% confidence intervals (95%CI) using quartile 1 as the referent. DII quartile 4 compared to 1 was associated with CRC risk among all subjects (HR=1.40; 95%CI=1.28-1.53, p-value for trend <0.01). Statistically significant associations also were observed for each anatomic site examined, for moderate and poorly differentiated tumours, and at each cancer stage among all subjects. Effects were similar when stratified by sex; however, results were statistically significant only in males. The only result reaching statistical significance in females was risk of moderately differentiated CRC tumours (DII quartile 4 vs. 1 HR=1.26; 95%CI=1.03-1.56). Overall, the DII was associated with CRC risk among all subjects. The DII may serve as a novel way to evaluate dietary risk for chronic disorders associated with inflammation, such as CRC.

Introduction

Inflammation is a normal part of the biological immune response, which is necessary for proper wound healing and combating infections⁽¹⁾. However, repeated insults and injuries (e.g., tobacco use, chronic infection, obesity, sleep disruption) can result in chronic systemic inflammation^{(1; 2; 3; 4; 5)}. Chronic inflammation is an underlying pathophysiological process that has been associated with numerous chronic disorders including cardiovascular disease, cancer, diabetes, stroke, and metabolic syndrome, as well as mortality^{(1; 6;7)}. Of all cancers, colorectal cancer (CRC) is the best described in terms of its association with inflammation. This is exemplified by epidemiologic evidence indicating increased rates of CRC among those with chronic inflammatory bowel disease⁽⁸⁾, and reduced risk of CRC with regular non-steroidal anti-inflammatory drug use⁽⁹⁾. Currently, CRC is the third most commonly diagnosed cancer among both men and women and the second most common cause of cancer death in the United States (US)⁽¹⁰⁾. Worldwide, CRC is the third and second most commonly diagnosed and the fourth and third most deadly cancers among men and women, respectively⁽¹¹⁾.

Diet is a strong moderator of chronic inflammation. Several dietary patterns consistently have been associated with systemic inflammation⁽¹²⁾. For example, Mediterranean diets (i.e., high in fruit and vegetables, fish, olive oil) have been associated with lower levels of systemic inflammation^{(12; 13)}; whereas, Western-style diets (i.e., high in fats, protein, simple carbohydrates, sweets) have typically been associated with increased systemic inflammation^{(12; 13)}. Many previous studies have found associations between dietary patterns and CRC risk. For example, several recent reviews or meta-analyses generally have indicated that ‘unhealthier’ diets (e.g., Western, meat-oriented) have been associated with increased CRC risk, whereas ‘healthier’ diets (e.g., Mediterranean, prudent vegetarian) have been associated with decreased CRC risk^{(14; 15; 16; 17; 18; 19)}.

Typically, dietary patterns or diet quality indices are derived using two methodologic techniques; a priori definitions based on dietary guidelines (e.g., Healthy Eating Index based on the Dietary Guidelines for Americans), or a posteriori analytic approaches (e.g., principal components analysis)^{(20; 21)}. A novel tool known as the Dietary Inflammatory Index (DII) has been developed to characterize diet on a continuum from maximally anti- to pro-inflammatory⁽²²⁾. Advantages of the DII over other dietary indices is that it is grounded in peer-reviewed literature focusing specifically on inflammation and it is standardized to dietary intake from numerous populations around the world. In addition, the DII can be estimated from a variety of diet assessment instruments (e.g., 24-hour recalls, 7-day dietary recalls, and food frequency questionnaires [FFQ]). The DII has been found to be associated with inflammatory cytokines including C-reactive protein and interleukin-6^{(23; 24;25)}, the glucose intolerance component of metabolic syndrome, increased odds of asthma and reduced forced expiratory volume in 1 second (FEV₁), shiftwork, CRC among women from the Iowa Women's Health Study, Women's Health Initiative, and a case-control study from Spain, and prostate and pancreatic cancers in Italy^{(24; 25; 26; 27; 28; 29; 30)}.

With respect to CRC, no study has examined the DII in a large follow-up cohort of both men and women. The National Institutes of Health American Association of Retired Persons (NIH-AARP) Diet and Health Study provides an excellent opportunity to examine the relationship between the DII and CRC incidence in a large population (≈500,000) of aging (50-74 years of age at baseline) US adults followed-up for ≈10 years. Specifically, this study tested the hypothesis that those with more pro-inflammatory DII scores would have greater risk of developing CRC compared to those with lower scores. Additionally, this study explored the relationship between the DII and CRC severity (i.e., tumour stage, grade, lymph node involvement) and location, as well as effect modification by sex.

Materials and Methods

Study Population

This analysis utilized data from the NIH-AARP Diet and Health Study, which has been described previously⁽³¹⁾. Briefly, about 3.5 million AARP members living in California, Florida, Louisiana, New Jersey, North Carolina, Pennsylvania and two metropolitan areas (Atlanta, Georgia and Detroit, Michigan) aged 50 to 74 years were mailed a self-administered questionnaire. The questionnaires were completed between 1995 and 1997 and included information on demographic characteristics, medical history, and diet. Exposure data only from this initial baseline questionnaire were utilized for this analysis. Exclusions were applied to the baseline cohort and included those who used a proxy for questionnaire completion (n=15,760); those with self-reported prostate (n=10,640), breast (n=10,875), colon (n=4,584), or other (n=23,219) cancer; those with self-reported end-stage renal disease (n=997); those with prevalent CRC (n=202) or any other cancer (n=1,697) not self-reported; death certificate-only confirmation of CRC (n=272) or any other cancer (n=2,187); and those with an FFQ-derived daily caloric intake <500 (n=3,563) or >6000 (n=2,718) kcal/day. This study was approved by the National Cancer Institute Special Studies Institutional Review Board and this analysis was approved by the University of South Carolina Institutional Review Board.

Dietary Inflammatory Index

The baseline questionnaire (administered between 1995 and 1996) included an FFQ, which obtained self-reported frequency and portion size information on 124 food items⁽³¹⁾. Data from the FFQ were linked to the United States Department of Agriculture's (USDA) 1994-1996 Continuous Survey of Food Intake by Individuals in order to estimate nutrient, foods, and food group intake. Various micro and macronutrients, as well as several individual food items (collectively termed “food parameters”), were used to calculate the DII. These food parameters included calories; carbohydrates; protein; total fat; unsaturated, monounsaturated, and polyunsaturated fat; trans-fat; alcohol; fibre; cholesterol; vitamins B1, B2, B6, B12, A, C, D, and E; iron; magnesium; zinc; selenium; folate; beta-carotene; anthocyanidins; flavan-3-ols; flavones; flavonols; flavanones; caffeine; green peppers; and tea. To calculate flavonoid classes, FFQ-derived daily gram intake of fruits and vegetables were linked to the USDA's Database for Flavonoid Content from Selected Foods (Release 3.1, December 2013) by matching foods with the USDA's 5-digit nutrient database number. Once linked, the content levels for each flavonoid class were applied to each fruit and vegetable and were summed to provide a total value for each flavonoid class.

The development and validation of the DII have been described previously^{(22; 23)}. In short, the food parameters were assigned scores based on research summarizing findings from 1,943 articles published through 2010 describing the relationship between the 45 possible food parameters and inflammation. DII calculation is linked to a regionally representative world database constructed by the authors that provided a mean and standard deviation for each food parameters. This world database included food consumption from 11 populations around the world (i.e., United States, United Kingdom, Bahrain, Mexico, Australia, South Korea, Taiwan, India, New Zealand, Japan, and Denmark). More detail on the world database can be found elsewhere⁽²³⁾. The “standard mean” was subtracted from the actual food parameter value and divided by its standard deviation. This z-score was then converted to a percentile (in order to minimize the effect of outliers or right-skewing) and centred by doubling the value and subtracting 1. The product of each food parameter z-score and adjusted article score was calculated and summed across all food parameters to create the overall DII score, which were then converted to equally distributed quartiles (quartile ranges: 1 = from -7.33 to -0.59; 2 = -0.58 to 1.36; 3 = 1.37 to 3.24; 4 = 3.25 to 6.97). The greater the DII score the more pro-inflammatory the diet, while lower values are more anti-inflammatory. These values lie within the theoretical limits of -9 to +8⁽²³⁾.

Follow-up and Colorectal Cancer Diagnoses

Follow-up began at the return of the baseline questionnaire (between 1995 and 1996) and continued until diagnosis of first cancer, movement out of cancer registry catchment areas, death, or December 31, 2006, whichever came first. Incident CRCs were identified through linkage of the NIH-AARP cohort data with cancer registries of the 8 states listed above, plus Arizona and Texas. The case ascertainment protocol has been described previously; linkage validity has been found to identify about 90% of all cancer cases⁽³²⁾. The cancer of interest was the first primary CRC. Information on disease anatomic location and severity also was obtained. Anatomic location was defined as ascending colon or cecum, transverse colon or flexures (i.e., hepatic and splenic), descending or sigmoid colon, and rectum or rectosigmoid; grade was defined as well differentiated, moderately differentiated, or poorly or undifferentiated; lymph node involvement was defined as 0 or >0; stage was defined as in situ or local (combined due to small sample size among in situ), regional, or distant.

Statistical Analyses

All analyses were performed using SAS version 9.3^® (Cary, North Carolina, USA). Descriptive analyses included frequencies or means and standard deviations for population characteristics at baseline among all subjects and stratified by sex. Differences by sex were determined using chi-square or t-tests. Possible confounders included age at baseline; body mass index (BMI=kg/m²); family history of CRC or any cancer; self-reported gallbladder disease, diabetes, or any circulatory disorder; smoking status; physical activity (frequency of ≥20 minute bouts of exercise per week in the past 12 months that caused increases in breathing or heart rate, or working up a sweat); race; education; marital status; census-based annual household income; and perceived health. Model variable selections began with a series of bi-variable Cox proportional hazards regressions (i.e., the DII + covariate). If a covariate had a p-value of ≤0.20, it was added to the full model. Backward elimination procedures were then used to develop the final models, which included all covariates that, when removed, led to a 10% change in the hazard ratio (HR) of the DII; statistically significant (p<0.05) covariates also were included in the final models. Confounders for which adjustments were made in the various models are located in Table 1, which displays whether confounders were categorical or continuous in nature. Final model selections for each analysis can be found in the footnotes of Tables 2 and 3. Smoking status, age, and BMI were included in every model. Cox proportional hazards regression was used to estimate CRC HRs and 95%CIs for DII quartiles 2, 3 and 4, compared to 1; the comparison of interest was between quartiles 1 and 4. The proportional hazards assumption was tested using methods derived from the cumulative sums of Martingale residuals. Proportional hazards assumptions were met for the DII; however, several covariates among the models (see footnotes in Tables 2 and 3) did not meet this assumption. The STRATA statement in the PHREG procedure in SAS was used for these covariates. In addition to examining CRC, each CRC anatomic location and disease severity category was analysed as outcomes. For sensitivity analyses, CRC cases diagnosed within 3 years of enrolment date were excluded.

Table 1. Baseline Population Characteristics by Sex.

Characteristic	All Subjects (n=489,442)	Males (n=292,118)	Females (n=197,324)
Age
Mean (standard deviation)	62.0 (5.4)	62.1 (5.4)	61.8 (5.4)
Race
European-American	446,705 (92%)	270,371 (94%)	176,334 (91%)
Other	36,396 (8%)	18,385 (6%)	18,011 (9%)
Education
≤High School	124,577 (26%)	62,416 (22%)	62,161 (33%)
Vocational School	48,179 (10%)	27,127 (10%)	21,052 (11%)
Some College	113,468 (24%)	64,699 (23%)	48,769 (26%)
College Graduate	92,258 (19%)	63,129 (22%)	29,129 (15%)
Graduate School	96,831 (20%)	67,081 (24%)	29,750 (16%)
Marital Status
Married or Living with Partner	336,000 (69%)	248,385 (86%)	87,615 (45%)
Widowed	53,269 (11%)	8,989 (3%)	44,280 (23%)
Divorced, Separated, Never Married	96,419 (20%)	32,870 (11%)	63,549 (33%)
Household Income (per $10,000)*
Mean (standard deviation)	5.38 (2.36)	5.54 (2.42)	51.5 (2.25)
Smoking Status
Never	172,077 (37%)	85,235 (30%)	86,842 (46%)
Former	240,716 (51%)	165,127 (59%)	75,589 (40%)
Current	58,172 (12%)	30,292 (11%)	27,880 (15%)
Physical Activity Level†
Never/Rarely	87,432 (18%)	43,461 (15%)	43,971 (23%)
1-3 Times Per Month	66,211 (14%)	38,096 (13%)	28,115 (14%)
1-2 Times Per Week	105,295 (22%)	63,907 (22%)	41,388 (21%)
3-4 Times Per Week	131,174 (27%)	81,864 (28%)	49,310 (25%)
≥5 Times Per Week	94,174 (19%)	62,175 (22%)	31,999 (16%)
Self-Reported Perceived Health
Excellent	84,305 (17%)	51,004 (18%)	33,301 (17%)
Very Good	172,629 (36%)	104,053 (36%)	68,576 (35%)
Good	166,734 (35%)	99,406 (34%)	67,328 (35%)
Fair or Poor	58,611 (12%)	33,877 (12%)	24,734 (13%)
Self-Reported Diabetes
Yes	43,863 (9%)	29,448 (10%)	14,415 (7%)
No	445,579 (91%)	262,670 (90%)	182,909 (93%)
Self-Reported Polyps
Yes	45,099 (9%)	32,203 (11%)	12,896 (7%)
No	444,343 (91%)	259,915 (89%)	184,428 (93%)
Self-Reported Cardiovascular Disease
Yes	73,525 (15%)	54,817 (19%)	18,708 (9%)
No	415,917 (85%)	237,301 (81%)	178,616 (91%)
Family History of Cancer
Yes	226,110 (49%)	139,593 (50%)	86,517 (46%)
No	238,032 (51%)	137,134 (50%)	100,898 (54%)
Family History of Colorectal Cancer
Yes	42,612 (9%)	24,046 (9%)	18,566 (10%)
No	421,530 (91%)	252,681 (91%)	168,849 (90%)
Body Mass Index (kg/m2)
Mean (standard deviation)	27.0 (4.8)	27.3 (4.2)	26.8 (5.6)
Dietary Inflammatory Index
Mean (standard deviation)	1.27 (2.47)	1.06 (2.40)	1.58 (2.54)

Frequencies not equalling column frequencies are due to missing data. Strata frequencies not equalling 100% are due to rounding.

^*Income is based on United States census-derived median household income in American dollars.

^†Refers to frequency of at least 20 minute bouts of physical activity per week.

Table 2. Any colorectal cancer and location-specific hazard ratios among quartiles of the Dietary Inflammatory Index stratified by sex.

DII Quartile	All Subjects		Males		Females
	Person-Years (Diagnoses)	Adjusted HRs (95%CI)	Person-Years (Diagnoses)	Adjusted HRs (95%CI)	Person-Years (Diagnoses)	Adjusted HRs (95%CI)
Colorectal Cancer vs. CRC-Free Participants
1	1,002,822 (1,497)	1.0 (Reference)	648,012 (1,055)	1.0 (Reference)	354,810 (442)	1.0 (Reference)
2	1,018,090 (1,549)	1.13 (1.05-1.22)	649,860 (1,156)	1.18 (1.08-1.29)	368,230 (393)	0.90 (0.78-1.04)
3	1,012,643 (1,594)	1.27 (1.17-1.38)	585,457 (1,065)	1.28 (1.16-1.41)	427,186 (529)	1.08 (0.93-1.25)
4	1,002,557 (1,585)	1.40 (1.28-1.53)	499,344 (955)	1.44 (1.29-1.61)	503,212 (630)	1.12 (0.95-1.31)
Location: Ascending/Cecum vs. CRC-Free Participants
1	1,037,510 (500)	1.0 (Reference)	665,525 (333)	1.0 (Reference)	371,986 (167)	1.0 (Reference)
2	1,047,094 (505)	1.07 (0.94-1.21)	665,193 (352)	1.11 (0.95-1.30)	381,900 (153)	0.96 (0.76-1.20)
3	1,041,325 (519)	1.16 (1.01-1.33)	599,147 (308)	1.15 (0.94-1.33)	442,178 (211)	1.19 (0.94-1.51)
4	1,036,250 (536)	1.27 (1.09-1.49)	513,597 (286)	1.27 (1.04-1.54)	522,653 (250)	1.26 (0.98-1.63)
Location: Transverse/Hepatic and Splenic Flexure vs. CRC-Free Participants
1	1,035,754 (177)	1.0 (Reference)	664,329 (112)	1.0 (Reference)	371,426 (65)	1.0 (Reference)
2	1,045,464 (210)	1.32 (1.07-1.63)	664,108 (158)	1.53 (1.19-1.98)	381,355 (52)	0.88 (0.59-1.29)
3	1,039,668 (210)	1.46 (1.16-1.83)	598,287 (149)	1.70 (1.29-2.24)	441,381 (61)	0.96 (0.64-1.43)
4	1,034,389 (205)	1.58 (1.23-2.03)	512,700 (122)	1.74 (1.27-2.39)	521,690 (83)	1.19 (0.78-1.83)
Location: Descending/Sigmoid vs. CRC-Free Participants
1	992,781 (389)	1.0 (Reference)	641,569 (292)	1.0 (Reference)	351,212 (97)	1.0 (Reference)
2	1,007,762 (390)	1.14 (0.99-1.32)	643,298 (299)	1.14 (0.96-1.35)	364,464 (91)	0.99 (0.73-1.34)
3	1,002,438 (409)	1.35 (1.15-1.58)	579,173 (285)	1.31 (1.09-1.57)	423,265 (124)	1.23 (0.90-1.67)
4	993,310 (426)	1.61 (1.35-1.91)	494,476 (276)	1.62 (1.31-1.99)	498,833 (150)	1.33 (0.95-1.86)
Location: Rectum/Rectosigmoid vs. CRC-Free Participants
1	1,006,594 (403)	1.0 (Reference)	648,632 (288)	1.0 (Reference)	357,962 (115)	1.0 (Reference)
2	1,019,916 (428)	1.20 (1.04-1.39)	649,361 (333)	1.30 (1.10-1.53)	370,554 (95)	0.78 (0.59-1.04)
3	1,014,919 (433)	1.35 (1.16-1.58)	585,064 (303)	1.43 (1.19-1.71)	429,855 (130)	0.91 (0.68-1.22)
4	1,005,736 (416)	1.45 (1.22-1.73)	499,474 (260)	1.57 (1.27-1.93)	506,261 (156)	0.91 (0.67-1.25)

Abbreviations: DII = dietary inflammatory index; HRs = hazard ratios; 95%CI = 95% confidence interval; BMI = body mass index (kg/m²). DII quartile ranges: 1 = from -7.33 to -0.59; 2 = -0.58 to 1.36; 3 = 1.37 to 3.24; 4 = 3.25 to 6.97. Adjustments: All models adjusted for age, smoking status, BMI, self-reported diabetes, and energy intake. Additional adjustments included: Colorectal Cancer = physical activity (frequency of ≥20 minutes bouts in the past 12 months, marital status, education, and age (STRATA statement); Ascending/Cecum = age (STRATA statement); Transverse/Hepatic and Splenic Flexures = race, and age; Descending/Sigmoid = marital status, education, perceived health, census-based income, and age (STRATA statement); Rectum/Rectosigmoid = self-reported polyps, education, age, and census-based income.

Table 3. Tumour grade, node involvement, and stage-specific hazard ratios among quartiles of the Dietary Inflammatory Index stratified by sex.

DII Quartile	All Subjects		Males		Females
	Person-Years (Diagnoses)	Adjusted HRs (95%CI)	Person-Years (Diagnoses)	Adjusted HRs (95%CI)	Person-Years (Diagnoses)	Adjusted HRs (95%CI)
Grade: Well Differentiated vs. CRC-Free Participants
1	1,006,471 (174)	1.0 (Reference)	648,425 (120)	1.0 (Reference)	358,046 (54)	1.0 (Reference)
2	1,019,975 (176)	1.09 (0.87-1.35)	649,135 (128)	1.14 (0.88-1.48)	370,840 (48)	0.85 (0.56-1.28)
3	1,014,996 (185)	1.22 (0.97-1.55)	584,965 (130)	1.34 (1.01-1.78)	430,031 (55)	0.84 (0.55-1.30)
4	1,005,875 (178)	1.27 (0.98-1.66)	499,454 (109)	1.39 (1.00-1.92)	506,421 (69)	0.90 (0.56-1.43)
Grade: Moderately Differentiated vs. CRC-Free Participants
1	1,016,766 (897)	1.0 (Reference)	655,309 (630)	1.0 (Reference)	361,457 (267)	1.0 (Reference)
2	1,029,339 (944)	1.17 (1.06-1.29)	656,628 (710)	1.23 (1.10-1.37)	372,711 (234)	0.93 (0.77-1.12)
3	1,022,274 (951)	1.30 (1.18-1.45)	590,695 (647)	1.32 (1.17-1.50)	431,579 (304)	1.10 (0.91-1.34)
4	1,015,748 (990)	1.52 (1.35-1.70)	505,507 (602)	1.54 (1.34-1.78)	510,240 (388)	1.26 (1.03-1.56)
Grade: Poorly or Undifferentiated vs. CRC-Free Participants
1	997,100 (233)	1.0 (Reference)	643,476 (153)	1.0 (Reference)	353,624 (80)	1.0 (Reference)
2	1,012,453 (248)	1.16 (0.96-1.40)	645,130 (174)	1.25 (0.99-1.57)	367,323 (74)	0.95 (0.68-1.33)
3	1,007,065 (264)	1.34 (1.10-1.64)	580,925 (173)	1.49 (1.17-1.91)	426,140 (91)	1.06 (0.75-1.49)
4	997,383 (259)	1.45 (1.16-1.82)	495,276 (141)	1.56 (1.17-2.07)	502,108 (118)	1.21 (0.83-1.75)
Nodes: 0 vs. CRC-Free Participants
1	1,018,250 (521)	1.00 (Referent)	655,707 (365)	1.00 (Referent)	362,543 (156)	1.00 (Referent)
2	1,031,856 (575)	1.22 (1.08-1.38)	657,084 (428)	1.26 (1.09-1.46)	374,771 (147)	1.01 (0.79-1.28)
3	1,024,593 (568)	1.33 (1.16-1.52)	591,077 (394)	1.37 (1.17-1.61)	433,516 (174)	1.10 (0.85-1.41)
4	1,017,286 (565)	1.48 (1.27-1.72)	505,250 (335)	1.47 (1.22-1.77)	512,036 (230)	1.31 (1.00-1.72)
Nodes: 1+ vs. CRC-Free Participants
1	1,007,548 (345)	1.00 (Referent)	649,057 (226)	1.00 (Referent)	358,491 (119)	1.00 (Referent)
2	1,021,030 (341)	1.07 (0.92-1.25)	649,913 (246)	1.19 (0.99-1.44)	371,118 (95)	0.80 (0.60-1.06)
3	1,016,174 (384)	1.31 (1.11-1.54)	585,704 (252)	1.45 (1.18-1.77)	430,470 (132)	0.99 (0.74-1.32)
4	1,007,138 (376)	1.41 (1.17-1.70)	500,085 (212)	1.55 (1.22-1.96)	507,053 (164)	1.08 (0.79-1.48)
Stage: In situ or Local vs. CRC-Free Participants
1	987,376 (460)	1.00 (Referent)	638,922 (338)	1.00 (Referent)	348,455 (122)	1.00 (Referent)
2	1,003,693 (453)	1.09 (0.95-1.25)	640,792 (338)	1.09 (0.93-1.28)	362,901 (115)	0.98 (0.75-1.28)
3	997,948 (454)	1.21 (1.05-1.41)	577,070 (310)	1.19 (1.99-1.42)	420,878 (144)	1.12 (0.84-1.48)
4	987,406 (412)	1.25 (1.06-1.48)	491,487 (255)	1.25 (1.02-1.54)	495,919 (157)	1.08 (0.79-1.47)
Stage: Regional vs. CRC-Free Participants
1	1,006,666 (439)	1.00 (Referent)	648,644 (310)	1.00 (Referent)	358,022 (129)	1.00 (Referent)
2	1,019,921 (437)	1.10 (0.96-1.27)	649,219 (317)	1.11 (0.94-1.31)	370,702 (120)	0.97 (0.75-1.26)
3	1,014,949 (440)	1.21 (1.04-1.41)	585,081 (310)	1.28 (1.07-1.52)	429,868 (130)	0.95 (0.72-1.28)
4	1,006,027 (471)	1.43 (1.21-1.68)	499,588 (270)	1.39 (1.13-1.70)	506,439 (201)	1.30 (0.97-1.74)
Stage: Distant vs. CRC-Free Participants
1	1,026,672 (153)	1.00 (Referent)	659,737 (106)	1.00 (Referent)	366,935 (47)	1.00 (Referent)
2	1,037,903 (168)	1.21 (0.96-1.52)	659,893 (126)	1.24 (0.95-1.63)	378,010 (42)	1.01 (0.65-1.57)
3	1,030,994 (182)	1.42 (1.11-1.81)	593,880 (111)	1.25 (0.93-1.70)	437,114 (71)	1.61 (1.05-2.48)
4	1,024,598 (189)	1.60 (1.22-2.10)	508,534 (113)	1.54 (1.10-2.15)	516,064 (76)	1.60 (0.99-2.58)

Abbreviations: DII = dietary inflammatory index; HRs = hazard ratios; 95%CI = 95% confidence interval; BMI = body mass index (kg/m²). DII quartile ranges: 1 = from -7.33 to -0.59; 2 = -0.58 to 1.36; 3 = 1.37 to 3.24; 4 = 3.25 to 6.97. Adjustments: All models adjusted for smoking status, BMI, and energy intake. Grade: Well-Differentiated = education and age (STRATA statement); Grade: Moderately Differentiated = Self-reported diabetes and polyps, physical activity , marital status, perceived health, age, , and census-based income (STRATA statement); Grade: Poorly or Undifferentiated = Self-reported polyps, race, education, and age. Nodes: 0 = self-reported diabetes and circulatory disorders, physical activity, race, and age (STRATA statement); Nodes: 1+ = self-reported diabetes and polyps, education, and age,; Stage: in situ or local: self-reported diabetes, marital status, race, with age, census-based income, education and physical activity in STRATA statement; Stage: Regional = self-reported diabetes and polyps, education, age, and census-based income; Stage: Distant = self-reported diabetes and polyps, physical activity, and age .

Results

This analysis included a total of 489,442 participants at baseline with a mean follow-up of 9.1 ± 2.9 years per participant contributing a total of 4,451,383 accumulated person-years of observation. There were 6,944 incident first primary CRC diagnoses (67% were in males). Most (63%) were localized or regional tumours (as opposed to distant); and 25% were unknown. A total of 342,870 participants had complete follow-up; 74,754 were diagnosed with a cancer other than colorectal; 33,795 died; and 31,079 moved out of the cancer registry areas. For a graphical representation of study follow-up and censorship, see Supplementary Figure 1. Overall, participants (mean baseline age: 62.0 ± 5.4 years) included in this analysis were predominantly (92%) European-American, somewhat well educated (63% with at least some college), married or living with a partner (69%), and overweight (mean BMI: 27.0 ± 4.8kg/m²) with an average household income around $54,000. Nearly 50% of participants reported a family history of any cancer, with 9% specifically reporting a family history of CRC. The mean DII was 1.27 ± 2.47, which was higher for females than males (1.58 vs. 1.06, respectively, p<0.01). This was partially due to the fact that males had higher absolute intake amounts of many anti-inflammatory components of the DII (see Supplemental Table 1). Additionally, each covariate presented in Table 1 statistically significantly (p<0.01) differed between males and females. Supplemental Table 2 further stratifies these covariates by DII quartiles among males and females.

Individuals in DII quartile 4 were 1.40 (95%CI=1.28-1.53) times more likely to develop CRC compared to quartile 1 (Table 2). Similar results were observed for tumours located in ascending colon or cecum, transverse colon or flexures, descending colon or sigmoid, and rectum or rectosigmoid. The results among all subjects were primarily driven by results among males. HRs were between 27% and 74% greater among males in DII quartile 4 compared to quartile 1 for all CRC and CRC at different anatomical locations; all were statistically significant. Non-statistically significant HRs for females in DII quartile 4 were 12%, 26%, and 33% greater compared to quartile 1 for all CRC, ascending or cecum tumours, and descending or sigmoid tumours, respectively (Table 2). Among all subjects, when the DII was analysed continuously, a one-unit increase was associated with an increase in all CRC (HR=1.06, 95%CI=1.05-1.08), ascending or cecum (HR=1.05, 95%CI=1.02-1.07), transverse or flexures (HR=1.06, 95%CI=1.02-1.10), descending or sigmoid (HR=1.08, 95%CI=1.05-1.11), and rectum or rectosigmoid tumours (HR=1.08, 95%CI=1.05-1.10) among all subjects (data not tabulated).

Table 3 displays HRs for disease severity markers. DII quartile 4, compared to 1, was statistically significantly associated with moderately and poorly differentiated tumours; tumours with and without lymph node involvement; and local, regional, and distant tumours. The same was true among males. Females in DII quartile 4, compared to 1, were more likely to develop moderately differentiated tumours (HR=1.26, 95%CI=1.03-1.56). Several other HRs among females were elevated, but did not achieve statistical significance (e.g., distant tumours HR=1.60, 95%CI=0.99-2.58). Among all subjects, a one-unit increase in the DII was associated with an increase in well- (HR=1.05, 95%CI=1.01-1.09), moderate- (HR=1.07, 95%CI=1.05-1.09), and poorly- (HR=1.07, 95%CI=1.03-1.10) differentiated tumours; tumours with lymph node involvement (HR=1.07, 95%CI=1.05-1.10) and without lymph node involvement (HR=1.06, 95%CI=1.03-1.09); and for localized (HR=1.05, 95%CI=1.01-1.09), regional (HR=1.05, 95%CI=1.01-1.09), and distant tumours (HR=1.05, 95%CI=1.01-1.09) among all subjects. The p-value for trend was statistically significant for all outcomes presented in Tables 2 and 3 for all subjects and males specifically. However, for females, only any CRC, ascending colon or cecum, descending colon or sigmoid, moderately differentiated tumours, tumours with no lymph node involvement, and distant tumours had significant trend p-values (data not tabulated).

Results did not differ after additional adjustment for hormone use among women (data not shown). After excluding cases (n=1,984) that were diagnosed within 3-years of enrolment the HR for ascending or cecum (HR=1.21, 95%CI=0.96-1.52) or local (HR=1.21, 95%CI=0.95-1.54) tumours among males and tumours with no node involvement among females (HR=1.26, 95%CI=0.94-1.70) for DII quartile 4 were attenuated and became non-significant. However, the HR for DII quartile 4, compared to 1, for distant tumours became significant for females (HR=1.85, 95%CI=1.04-3.26). An additional post-hoc sensitivity analysis, using sex-specific DII quartiles, produced a marginally significant HR for DII quartile 4, compared to 1 for all CRC combined (HR=1.18, 95%CI=1.00-1.38).

Discussion

Higher (i.e., more pro-inflammatory) DII scores were associated with increased risk of any CRC; CRC at each anatomic site examined; for moderately and poorly differentiated tumours; tumours with and without lymph node involvement; and local, regional, and distant tumours among all subjects and males. The direction of effect among females was similar; however, HRs were only statistically significant for moderately differentiated tumours and tumours with no lymph node involvement. Previously, individual food groups, micronutrients, and macronutrients have been associated with increases or decreases in CRC risk including red and processed meat, fibre, vitamin D, animal fat, and selenium^{(33; 34)}. For example, excessive alcohol consumption has been shown to increase CRC by 8-52%, which is generally stronger among males than females⁽³³⁾; excessive or no alcohol consumption also has been associated with increased inflammation compared to moderate consumption⁽³⁵⁾. However, analysis of individual dietary factors does not allow one to take into account the complicated interactions or high intercorrelations between dietary factors. Additionally, the effect of any single nutrient may be too small to detect or may be confounded by dietary habits and patterns ^{(20; 36)}. Compared to other dietary indices, the DII was designed based on a specific biologic mechanisms (i.e., inflammation) and was standardized to dietary intake from numerous populations around the world⁽²³⁾.

Previous studies of dietary indices and CRC are generally consistent with the current findings. Studies using a posteriori or a priori methods for describing dietary patterns have typically found that ‘healthier’ diet patterns (e.g., high in fruits and vegetables, fish, poultry, and whole grains) are associated with lower CRC risk; whereas ‘less healthy’ diets (e.g., high in red or processed meat, refined grains, sweets) are associated with increased CRC risk, including some specific to the NIH-AARP Diet and Health Study^{(14; 15;16; 18; 37; 38; 39; 40)}. However, some studies have found no association between various dietary patterns and CRC risk^{(41; 42;43)}. Not surprisingly, these ‘healthier’ dietary patterns also are typically associated with lower levels of inflammation⁽¹²⁾.

CRC risks among men tended to be elevated with higher DII scores; whereas there was less consistent evidence of elevated risk observed among women in the current study. This is somewhat similar to previous studies that have found no association between various dietary patterns and CRC among women^{(37; 39; 44; 45)}. In the current study, use of sex-specific DII quartile cut-points did not change the overall interpretation of results. Social desirability has been shown to influence dietary reports and the bias is expressed much more strongly among women than men^{(46; 47)}; social desirability was unmeasured in the NIH-AARP Diet and Health Study. Previously, the DII has been shown to be associated with CRC among women in the Iowa Women's Health Study (HR for DII quintiles: Q5 vs Q1=1.20; 95%CI=1.01-1.43)⁽²⁷⁾ and in the Women's Health Initiative (HR for DII quintiles: Q5 vs Q1=1.22; 95%CI=1.05-1.43)⁽³⁰⁾. We observed a similar magnitude of effect; however, results did not achieve statistical significance. Differences in cohort characteristics and available data on food parameters comprising the DII may, at least partially, explain the differences in results among women in the NIH-AARP Diet and Health Study compared to other studies.

Several reports have investigated the effect of dietary patterns on CRC risk for different anatomic locations (e.g., distal or proximal colon, rectum)^{(38; 39; 41; 42; 48; 49)}. Margalhaes and colleagues recently published a meta-analysis and found elevated risks for the proximal (RR=1.11, 95%CI=0.93-1.32) and distal (RR=1.32, 95%CI=0.99-1.77) colon for Western-type diets. However, these did not achieve statistical significance. The authors concluded that, overall, there were no differences in CRC risk by anatomic location⁽¹⁵⁾. This is somewhat consistent with the current findings.

This was one of the first studies to examine the association between a dietary index and severity of CRC. Elevated CRC risk by disease severity seemed to be restricted to males. Except for local tumours, DII quartile 4 conferred between 25% and 60% greater risk compared to quartile 1 for all CRC regardless of disease severity among males. Interestingly, the HR for DII quartile 4 compared to quartile 1 increased as tumour stage increased. If confirmed in other studies, these results may indicate that more virulent cancers may be associated with greater dietary inflammatory potential.

Numerous pathways exist through which dietary patterns influence CRC risk. Pro-inflammatory diets can increase insulin resistance by increasing systemic inflammation^{(13; 50)} which, in turn, could increase levels of insulin, triglycerides, and non-esterified fatty acids^{(51; 52)}. These factors could then promote excessive proliferation of colonic epithelial cells and potentially expose them to reactive oxygen species^{(51; 52)}. Diets high in red and processed meats can be high in N-nitroso compounds, which could damage DNA^{(33; 53)}. Diets high in fruits and vegetables (more anti-inflammatory) contain antioxidants and micronutrients with antitumour capabilities, as well as fibre which can decrease transit time for food in the digestive tract⁽³³⁾.

This study had several weaknesses. A small number of CRC cases by some anatomic locations or by disease severity may have limited the ability to detect statistically significant associations in women. A measure of social desirability was not obtained and there may be other unmeasured factors that differ by sex that influence either self-report measures, exposure to CRC risk factors, or both. The longitudinal nature of the NIH-AARP Diet and Health Study is a strength in that diet was assessed prior to disease diagnosis; however, only baseline diet assessment was used in this analysis. Therefore, changes in dietary patterns could not be examined. Also, the FFQ has been shown to be subject to both random and systematic errors^{(46; 47)}. Despite its weaknesses, the NIH-AARP Diet and Health Study is a large (n≈500,000) well-established follow-up cohort with a strong record of publication. This was one of the first studies to examine both CRC location and disease severity by levels of a dietary index. Additionally, this is the first report of the association between the DII and CRC among males. The use of the DII has several unique advantages over other dietary measures and was designed specifically in reference to inflammation⁽²³⁾; a strong risk factor for CRC⁽⁸⁾.

In conclusion, this study found that the novel DII predicted CRC incidence among NIH-AARP Diet and Health Study participants. As noted by Fung and colleagues, no dietary indices or patterns have been developed specifically for CRC prevention⁽¹⁴⁾. The DII was designed based on peer-reviewed literature on diet and inflammation, an established risk factor for CRC. Future research should test whether changing the inflammatory potential of diet can reduce chronic inflammation and the risk of CRC. The utility of the DII may be able to extend to clinical settings to address inflammatory potency of one's diet and possibly reduce future risk of chronic inflammatory-related disease.