Chronic Inflammation and Risk of Colorectal and Other Obesity-Related Cancers: The Health, Aging, and Body Composition Study

Abstract

Evidence of the association between chronic inflammation and the risk of colorectal cancer (CRC) and other obesity-related cancers (OBRC) remains inconsistent, possibly due to a paucity of studies examining repeated measures of inflammation. In the Health ABC prospective study of 2490 adults aged 70–79 years at baseline, we assessed whether circulating levels of three markers of systemic inflammation, IL-6, CRP and TNF-α, were associated with the risk of CRC and OBRC, a cluster including cancers of pancreas, prostate, breast and endometrium. Inflammatory markers were measured in stored fasting blood samples. While only baseline measures of TNF-α were available, IL-6 and CRP were additionally measured at Years 2, 4, 6 and 8. Multivariable Cox models were fit to determine whether tertiles and log-transformed baseline, updated and averaged measures of CRP and IL-6 and baseline measures of TNF-α were associated with the risk of incident cancer(s). During a median follow-up of 11.9 years, we observed 55 and 172 cases of CRC and OBRC, respectively. The hazard of CRC in the highest tertile of updated CRP was more than double that in the lowest tertile (HR = 2.29; 95% CI: 1.08–4.86). No significant associations were seen between colorectal cancer and IL-6 or TNF-α. Additionally, no significant associations were found between obesity-related cancers and the three inflammatory markers overall, but we observed a suggestion of effect modification by BMI and NSAID use. In summary, in this population, higher CRP levels were associated with increased risk of CRC, but not of OBRC. The findings provide new evidence that chronically-elevated levels of CRP, as reflected by repeated measures of this marker, may play a role in colorectal carcinogenesis in older adults.

INTRODUCTION

The evidence supporting a link between inflammation and the development of human cancers is growing ^{1, 2}. Chronic inflammation leading to cell hyperproliferation has been implicated in infection-associated cancers of the stomach, liver, bile duct, urinary bladder, as well as colorectal cancer ¹. Several chronic inflammatory conditions, such as Crohn’s disease, ulcerative colitis, chronic bronchitis and chronic pancreatitis, have been identified as important risk factors for colorectal, lung and pancreatic cancers, respectively ³. The link between chronic inflammation and cancer is further substantiated by the consistent inverse association between several cancer types and the use of anti-inflammatory agents ⁴.

A number of candidate markers of inflammation have been identified. Because C-reactive protein (CRP) is rapidly produced by hepatocytes primarily in response to the release of interleukin-6 (IL-6) in circulation by activated monocytes and other immune cells ⁵, it is a potential marker of chronic inflammation. Plasma CRP and IL-6 are both elevated in response to bacterial infections, as well as conditions resulting in tissue inflammation and injury ⁶. Inflammatory markers, such as tumor necrosis factor-α (TNF-α), also increase in response to infection and tissue damage and in active disease states ¹. High levels of TNF-α can lead to a chronic inflammatory state and may be implicated in tumor initiation and promotion ⁷.

The association of circulating C-reactive protein (CRP) and the risk of colorectal cancer has been found to be inconsistent across thirteen prospective studies ^8–20. Whereas five of these studies observed a significant positive association between baseline CRP and colorectal cancer ^{8, 9, 12, 13, 16}, the other eight have reported null ^{11, 14, 15, 17–20} or inverse ¹⁰ associations. In addition, a recent meta-analysis suggested that elevated CRP levels measured at one time were weakly associated with the risk of colorectal cancer and that the association was stronger in men than women ²¹. Moreover, another recent study reported a statistically-significant increase in the risk of death from obesity-related cancers (defined in that study as: breast, prostate, pancreas and endometrium) and death from colorectal cancer among participants with clinically-elevated CRP at baseline ²². Hitherto, evidence does not support a strong association of IL-6 ^{9, 18, 19, 23} and TNF-α ^{9, 23} with colorectal cancer risk, with the exception of one study that reported a significant effect of plasma soluble tumor necrosis factor receptor 2 (sNTFR-2), a surrogate for TNF-α ¹⁹.

An important limitation of studies to date has been that inflammation at a single time point has been used as a surrogate for persistent or chronic inflammation ^8–17. In fact, two recent reviews of inflammatory biomarkers in cancer research specifically cited the need for the assessment of repeated measures of these factors in future research ^{24, 25}. Thus, using data from a prospective cohort of adults aged 70–79 years at baseline, which collected inflammatory marker data at five time points over 8 eight years, and which had a relatively extensive list of covariate data updated over the follow-up period, we assessed whether circulating levels of three nonspecific serological markers of systemic inflammation, IL-6, CRP, and TNF-α, were associated with the risk of colorectal cancers. We also included obesity-related cancers (OBRC) as a secondary outcome.

METHODS

Study Population

Participants were enrolled in the Health, Aging, and Body Composition Study (Health ABC), a prospective cohort study of 3075 community-dwelling black and white men and women aged 70–79 years old. Between April 1997 and June 1998, participants were recruited from a random sample of Medicare beneficiaries and Medicare eligible residents in the areas surrounding Memphis, Tennessee, and Pittsburgh, Pennsylvania, USA. Eligibility criteria included: no reported difficulty walking one-quarter mile, climbing 10 steps without resting or performing activities of daily living; being free of known life-threatening illnesses, including no history of active cancer treatment in the prior 3 years; no plans to leave the study area for 3 years; and no active participation in a lifestyle intervention trial. All participants provided informed consent, and the protocols were approved by the institutional review boards of the two clinic sites and the data coordinating center. Participants were invited for in-person clinic visits every two years. After excluding subjects with (i) any prevalent cancer (n = 541) and (ii) those diagnosed with colorectal or obesity-related cancers during the first 2 years of follow-up and for whom elevated levels of inflammatory markers may be the result rather than a driver of cancer (n = 44), 2490 subjects were available for this analysis.

Exposure Assessment

Blood samples were collected by venipuncture in the morning, after an overnight fast. After processing, the specimens were aliquoted into cryovials, frozen at 70 °C and shipped to the Health, Aging, and Body Composition Core Laboratory at the University of Vermont.

Baseline (year 1997) levels of IL-6 (pg/mL), TNF-α (pg/mL) and CRP (ug/mL) were measured in frozen stored serum (IL-6 and CRP) or plasma (TNF-α). Baseline CRP levels were measured in duplicate by enzyme-linked immunosorbent assay (ELISA) based on purified protein and polyclonal anti-CRP antibodies (Calbiochem, San Diego, CA). Standardization was done using the WHO International Reference Standard with a sensitivity of 0.08 mg/L. The lower limit of detection for CRP was 0.007 mg/L. Baseline IL-6 and baseline TNF-α were determined by ELISA measured in duplicate using a high-sensitivity Quantikine colorimetric immunoassay kit from R&D Systems (Minneapolis, MN) with a detectable limit of 0.10 pg/mL and 0.18 pg/ml for IL-6 and TNF- α, respectively. Blind duplicate analyses (n = 150) for IL-6, CRP and TNF-α showed inter-assay coefficients of variation of 10.3%, 8.0% and 15.8%, respectively.

Only baseline measurements were available for TNF-α. However, CRP and IL-6 concentrations were additionally measured in the years 1998, 2000, 2002 and 2004. CRP was measured in EDTA-plasma by an automated chemiluminescent immunoassay system (IMMULITE, Diagnostic Products Corporation, Los Angeles) with a detectable limit of 0.1 μg/mL. CRP and IL-6 concentration measurements taken in the years 1998, 2000, 2002 and 2004, were analyzed at a different laboratory (Wake Forest University) than the baseline measurements (Core Laboratory at the University of Vermont). To account for possible laboratory differences, calibration was performed for both IL-6 and CRP in values from post-1997. This calibration was based on a set of 150 blind duplicate measurements obtained from both labs: one set of values was regressed on the other set. The regression function was then used to predict calibrated values on the rest of the samples. Calibrated values were used in our analyses. Measurements of IL-6 were obtained from a cell pack for the year 2004 and from serum for the years 1997, 1998, 2000 and 2002. To account for this difference in sample source, a second calibration was performed on IL-6 for the year 2004. This calibration was based on a set of 137 samples for the year 2002 that were derived from both serum and cell packs. In the analyses, participants were categorized in tertiles of inflammatory markers. Continuous natural log-transformed measures of inflammatory markers were also considered.

Endpoint Ascertainment

Participants or their proxies were contacted every 6 months, for up to 16 years, in person or by telephone and interviewed about any hospitalizations, major outpatient procedures and health conditions, including new cancer diagnoses. Incident cancers and the date of diagnosis were determined directly from hospital records or from the underlying cause of death from death certificates. The Health ABC Study Diagnosis and Disease Ascertainment Committee reviewed all hospital records, death certificates, informant interviews and autopsy data to adjudicate immediate and underlying causes of death. A panel of physicians verified diagnoses based on hospital records, interviews and death certificates. Medical records for overnight hospitalizations were reviewed at each site by local adjudicators, and diagnoses were confirmed by a review of imaging and biopsy reports. Adjudication of incident cancer events, excluding non-melanoma skin cancer, was completed through August 31, 2012. The primary outcome of interest was colorectal cancer. We additionally evaluated the association of our three inflammatory markers with obesity-related cancers as a secondary outcome.

Covariates

Age, education, race, gender, clinical site, physical activity (kilocalories/week), height (inches), abdominal circumference (cm), pack-years smoked and post-menopausal hormone replacement therapy use were considered as potential baseline confounders. Anthropometric characteristics were directly measured at clinical visits. Since calcium intake has previously shown an inverse association with colorectal cancer, quartiles of calcium consumption, obtained from the year 1998 clinical visit (the only occasion on which diet information was ascertained via a self-administered food frequency questionnaire), were also included in models of colorectal cancer incidence. In addition to the aforementioned factors, we also considered smoking history, alcohol consumption, body mass index (BMI), diabetes status, nonsteroidal anti-inflammatory drug (NSAID) and calcium supplement use as potential time-varying confounders. Smoking history was obtained from the baseline interview for former and never smokers and updated for the years 1998, 2001, 2004–2008 and 2011 for those that reported smoking at baseline and categorized as: never, former smoking of less than 20 pack-years, former smoking of more than 20 pack-years, current smoking of less than 20 pack-years and current smoking of more than 20 pack-years. BMI (kg/m²) was determined annually from measured height and weight through the year 2002 and biennially thereafter until the year 2007. Diabetes status was self-reported at baseline and updated annually until the year 2010 and categorized as no diabetes, previously-diagnosed diabetes or newly-diagnosed diabetes. NSAID use was reported at baseline and updated annually for the first three years of follow-up and for the years 2001, 2002, 2004, 2006 and 2007. A categorical variable for alcohol consumption (no consumption in the last year, less than once per week, 1–7 times per week, more than 1 per day), which was assessed at the baseline interview and then updated in 2007, was included in the analyses. Calcium supplement use, reported for the years 1997–1999, 2001 and 2002, was also considered. When a value for a time-varying confounder was missing, the last non-missing value was carried forward.

Statistical Analysis

To evaluate the short- and long-term effects of inflammation on colorectal and obesity-related cancers risk, we evaluated the associations for CRP and IL-6 (1) at baseline, (2) using an updated exposure measure and (3) using a measure of average exposure, as is commonly done in analyses of repeated dietary data ²⁶. Baseline inflammatory markers were ascertained using levels of inflammatory markers measured in 1997 for all participants. To minimize the possibility of reverse causation, for participants who developed colorectal or obesity-related cancers, we only considered exposure measurements taken 2 or more years earlier than the cancer diagnosis. Therefore, cases that occurred within the first 2 years of follow-up were excluded from all analyses. If an exposure measurement was missing for the updated exposure analysis, the last non-missing value was carried forward. Values of inflammatory markers measured within 2 years before the cancer diagnosis of interest were set to the most recent previous measurement. Cancer outcomes for each time interval were then regressed onto the exposure values at the beginning of the interval. For example, among subjects that never developed colorectal or obesity-related cancer, the levels of inflammatory markers from 1997 were used to predict outcomes from 1997–1998 (the first interval); the levels from 1998 were used to predict outcomes from 1998–2000 (the second interval); the levels from 2000 were used to predict outcomes from 2000–2002 (the third interval); the levels from 2002 were used to predict outcomes from 2002–2004 (the fourth interval); and the levels from 2004 were used to predict outcomes from 2004 till the end of the follow-up (the fifth interval). For a hypothetical subject that developed colorectal cancer in 2003, the levels of inflammatory markers from 1997 were used to predict colorectal cancer from 1997–1998; the levels from 1998 were used to predict colorectal cancer from 1998–2000; and lastly, the levels from 2000 (not 2002) were used to predict colorectal cancer from 2000 till the end of the follow-up in 2003.

To reduce random within-person variation and to best represent the long-term effects of inflammation, we also computed current averages of the updated inflammatory markers. As described elsewhere ²⁶, in these analyses, baseline exposure was used to predict outcomes during the first time interval; the average of updated exposure levels for the first and second intervals was used to predict outcomes during the second time interval; the average of updated exposure levels for the first, second and third intervals was used to predict outcomes during the third interval, and so forth, for the remainder of the follow-up. Two-tailed p-values for linear trend tests across tertiles were computed by modeling the median value of each category as a continuous variable.

Cox proportional hazards models were used to estimate the hazard and corresponding 95% confidence intervals for the association between each exposure of interest and incident colorectal or obesity-related cancers. The proportional hazards assumption was assessed using Schoenfeld residuals. For each participant, the date of enrollment marked the beginning of the observation period. Person-months were accumulated until the analysis endpoint, the date of death or the date of last contact, whichever occurred first. Two models were considered for each outcome. The first model for both outcomes was adjusted for age, race, gender and clinical site. Fully-adjusted Cox models for obesity-related cancers were adjusted for age, race, gender, site, physical activity, body mass index, smoking, diabetes, NSAID use and alcohol consumption. Fully-adjusted Cox models of colorectal cancer were additionally adjusted for abdominal circumference, height, calcium consumption, calcium supplement use and post-menopausal hormone use. The simultaneous inclusion of abdominal circumference and BMI in fully-adjusted models for colorectal cancer is due to the differing risks between males and females with respect to colorectal cancer: while increased BMI is predictive of colorectal cancer among males, central adiposity is more predictive of colorectal cancer among women ^{27, 28, 39}.

Despite limited statistical power, we conducted analyses stratified by levels of obesity, NSAID use and gender to evaluate whether the association between plasma levels of inflammatory markers and cancer risk may exist in subgroups. Obesity was defined as a BMI of 30 kg/m² or greater. The Wald test was used to assess the statistical significance of corresponding interaction terms. All p-values were two-sided. All analyses were carried out using SAS software, Version 9.2 (SAS Institute, Inc., Cary, North Carolina), and R.

RESULTS

Our analytic cohort consisted of 2490 participants with a mean age of 73.5 years (SD = 2.9 years) whose median follow-up was 11.9 years (interquartile range (IQ): 6.5–14.5). During follow-up, we identified 55 incident cases of colorectal cancer and 172 other obesity-related cancers: 63 (36.6%) breast cancers, three endometrial cancers (1.7%), 21 pancreatic cancers (12.2%) and 85 (49.4%) prostate cancers. The distributions of selected demographic and clinical characteristics of the study population, overall and by tertile of inflammatory markers are presented in Table 1. Participants with higher CRP levels were more likely to have diabetes, higher mean BMI and abdominal circumference, and they were more likely to have smoked. Additionally, they were less likely to be male, white, to have completed postsecondary education or to engage in physical activity. Similar trends were observed in the distribution of the aforementioned characteristics across tertiles of IL-6 and TNF-α. Over 99% of the study group and obesity-related cancer cases, as well as 100% of the colorectal cancer cases had more than one measurement of CRP and IL-6.

Table 1.

Comparison of baseline characteristics of the 2157 participants of the Health, Aging, and Body Composition (Health ABC) Study.

	Tertiles of Inflammatory Markers at Baseline
		CRP (ug/mL)		IL-6 (pg/mL)		TNF-α (pg/mL)
	Overall	T1	T3	T1	T3	T1	T3
Age at baseline, mean ± SD	73.5 ± 2.9	73.5 ± 2.9	73.3 ± 2.8	73.3 ± 2.7	73.5 ± 2.9	73.3 ± 2.9	73.8 ± 2.8
Gender, % male	46.8	53.7	38.5	41.7	49.8	43.5	50.4
Race, % white	57.2	65.1	49.7	62.9	51.2	50.6	63.3
Site, % Memphis	50.7	46.1	52.2	47.7	53.6	53.8	49.4
Education, %
Less than high school	25.9	22.4	26.4	23	28.5	27.6	26.9
High school	32.6	31.3	35	32.7	33.5	30	35.4
Post-secondary	41.3	46.1	38.2	44.3	37.7	42.2	37.5
Smoking status, %
Never	44.2	47.4	41.6	51.1	37.6	47.1	41
Former	44.8	42.6	45.2	42.4	46.9	40.3	48.7
Current	10.8	9.8	13.1	6.6	15.4	12.6	9.8
Pack-years, mean ± SD	18.4 ± 27.6	16.2 ± 25.4	19.6 ± 28.1	12.6 ± 20.9	23.7 ± 30.8	15.7 ± 24.6	22 ± 30.9
Body mass index (kg/m2), mean ± SD	27.4 ± 4.8	26.2 ± 4.1	28.9 ± 5.2	26 ± 3.9	28.3 ± 5.5	27 ± 5	27.8 ± 4.7
Abdominal circumference (cm), mean ± SD	99.4 ± 13.2	96.7 ± 13.5	102.5 ± 13.1	95.6 ± 11.1	101.9 ± 14.1	97.7 ± 12.9	101.1 ± 13
Height (inches), mean ± SD	65.4 ± 3.7	65.8 ± 3.7	64.8 ± 3.7	65.1 ± 3.7	65.5 ± 3.7	65.2 ± 3.7	65.5 ± 3.7
Alcohol consumption, %
No consumption in the past year	50.6	49	52	49.1	53.7	50.1	52.2
Less than once per week	20.5	20.9	21	21.2	19	21	21.2
1–7 times per week	21	22.3	19.5	24.5	18.2	21.3	18.9
More than 1 per day	7.5	7.3	7.3	5.3	8.6	7.4	7.4
Nutrient intake, mean ± SD
Dietary calcium intake (mg/day)	781.7 ± 400.1	783.2 ± 368.4	774.4 ± 433.5	782 ± 385.2	771.4 ± 398.3	799.5 ± 410.1	764.4 ± 407.3
Physical activity (kcal/wk), %
0–499.9	52.9	45.7	59.7	46	58.9	54.1	52.8
500–999.9	16	18	14.6	20.4	13.5	15.4	15.1
1,000–1,999.9	22.4	21.4	22.4	24.3	20.4	20.1	24.3
≥2,000	3.9	3	5.2	1.8	5.4	3.7	4.1
Medication use, %
NSAIDs	22.4	21.4	22.4	24.3	20.4	20.1	24.3
Corticosteroids	3.9	3	5.2	1.8	5.4	3.7	4.1
Statins	12.5	12.9	13.5	12	13.6	10.4	14.8
Calcium supplement	17.8	18.2	18.9	24.3	14.5	18.5	16.6
Postmenopausal hormonal therapy (in women), %	41.5	38.4	42.3	47.2	35.4	45.1	34.8
Diabetes, %	36.8	34.1	42.4	26.7	44.8	32.4	43.1
Baseline inflammatory markers, median ± SD	--	0.8 ± 0.6	5.7 ± 3.7	1.1 ± 0.3	4.1 ± 3.3	2.1 ± 0.4	4.5 ± 1.6

Colorectal Cancer

We computed the hazards ratios and the 95% confidence intervals for incident colorectal cancer using baseline, updated and average levels of inflammatory markers (Table 2). We found no association between baseline levels of CRP and the hazard of colorectal cancer. However, a one-unit increase in updated log-transformed CRP (ug/mL) was significantly associated with increased hazard of colorectal cancer in age, race, gender and site-adjusted (HR = 1.33; 95% CI: 1.03–1.73) and fully-adjusted (HR = 1.43; 95% CI: 1.09–1.87) models alike. Additionally, we found that, compared to the first tertile of updated CRP, the second (HR = 2.16; 95% CI: 1.05–4.43) and the third tertiles (HR = 2.29; 95% CI: 1.08–4.86) were significantly associated with increased hazard of colorectal cancer in fully-adjusted models (P_trend = 0.09). When average exposures were considered, log-transformed CRP (HR = 1.45; 95% CI: 1.02–2.05) and the second tertile of CRP (HR = 2.29; 95% CI: 1.15–4.52) were significantly associated with increased hazard of colorectal cancer in fully-adjusted models.

Table 2.

Hazards ratios (HR) and 95% CI of inflammatory markers for colorectal cancer.

		Tertiles of Inflammatory Markers
	Log-transformed	T1	T2	T3	ptrend
CRP (ug/mL), Cases (PY*)	51 (23,073)	15 (7,943)	21 (7,726)	15 (7,404)
Baseline
Model 1	1.09 (0.84,1.40)	1.00	1.47 (0.76,2.86)	1.25 (0.60,2.59)	0.62
Model 2	1.17 (0.89,1.53)	1.00	1.61 (0.81,3.20)	1.52 (0.71,3.27)	0.32
Updated
Model 1	1.33 (1.03,1.73)	1.00	1.99 (0.99,4.01)	1.91 (0.93,3.94)	0.20
Model 2	1.43 (1.09,1.87)	1.00	2.16 (1.05,4.43)	2.29 (1.08,4.86)	0.09
Averaged
Model 1	1.29 (0.94,1.78)	1.00	2.09 (1.07,4.08)	1.73 (0.84,3.57)	0.28
Model 2	1.45 (1.02,2.05)	1.00	2.29 (1.15,4.52)	2.11 (0.98,4.53)	0.12
IL6 (pg/mL), Cases (PY)	55 (24,679)	14 (9,020)	28 (8,306)	13 (7,353)
Baseline
Model 1	1.12 (0.76,1.65)	1.00	2.06 (1.08,3.93)	1.10 (0.52,2.36)	0.81
Model 2	1.17 (0.78,1.77)	1.00	2.00 (1.03,3.88)	1.10 (0.50,2.44)	0.77
Updated
Model 1	1.22 (0.83,1.80)	1.00	1.40 (0.71,2.76)	1.42 (0.72,2.80)	0.46
Model 2	1.20 (0.81,1.79)	1.00	1.36 (0.68,2.72)	1.38 (0.68,2.79)	0.51
Averaged
Model 1	1.20 (0.77,1.85)	1.00	1.25 (0.65,2.40)	1.14 (0.58,2.23)	0.92
Model 2	1.20 (0.76,1.87)	1.00	1.26 (0.64,2.47)	1.15 (0.57,2.32)	0.93
TNF-α (pg/mL), Cases (PY)	50 (24,111)	20 (8,554)	14 (8,107)	16 (7,450)
Baseline
Model 1	0.97 (0.48,1.99)	1.00	0.74 (0.37,1.47)	0.87 (0.45,1.70)	0.72
Model 2	1.03 (0.50,2.14)	1.00	0.73 (0.37,1.47)	0.86 (0.43,1.70)	0.69

^*PY: person-years.

Model 1: adjusted for age in years (continuous), race (black/white), gender and site.

Model 2: adjusted for age in years (continuous), race (black/white), gender, site, body mass index (kg/m²), abdominal circumference, smoking (never, former <20 pack-years, former ≥20 pack-years, current <20 pack-years, current ≥20 pack-years), NSAID use (yes/no), height (inches), baseline physical activity (0, 0–499.9, 500–999.9, 1,000–1,499.9, 1,500–1,999.9, >2,000 Kcal/week), alcohol consumption (no consumption in the past year, <1 drink/week, 1–7 drinks/week, >1 drinks/day), quantile of calcium consumption (mg/day, reported at Year 2 visit), current diabetes, history of postmenopausal hormone use (yes/no) and calcium supplement use (yes/no).

The second tertile of baseline IL-6 was significantly associated with increased hazards of colorectal cancer in models adjusted for age, race, gender and site (HR = 2.06; 95% CI: 1.08–3.93), as well as in fully-adjusted models (HR = 2.00; 95% CI: 1.03–3.88; Table 2). Associations for the third tertile of baseline IL-6 did not reach statistical significance, possibly due to the small number of colorectal cases in this category (n = 13; P_trend = 0.77) or as a result of a non-linear pattern of outcome responses across tertiles of exposure. The hazards ratios for a one-unit increase in log-transformed IL-6 (pg/mL) and tertiles of updated and averaged IL-6 were elevated, but were not statistically significant.

We found no significant associations between baseline measures of TNF-α and the risk of incident colorectal cancer.

Obesity-Related Cancers

We found no significant associations between CRP, IL-6 and TNF-α and obesity-related cancers (Table 3). Despite the relatively small sample size, we evaluated whether the association of updated levels of CRP and IL-6 with obesity-related cancers varied within categories of obesity, NSAID use and gender (Table 4). Compared to the first tertile, the third tertile of updated CRP was associated with a statistically-significant increased hazard of obesity-related cancers (HR = 1.64; 95% CI: 1.06–2.52) among non-obese participants (BMI <30), but with an inverse (HR = 0.56; 95% CI: 0.27–1.17), non-statistically significant effect among obese participants (BMI ≥30; P_interaction = 0.11). The findings were similar for one-unit increases in log-transformed CRP (ug/mL), with hazards ratios of 1.21 (95% CI: 1.02–1.44) among non-obese and 0.76 (95% CI: 0.58–1.00) among obese participants (P_interaction = 0.04). We found no evidence of any modification of the effect of updated measures of IL-6 on the hazard of obesity-related cancers. Possible variations in the effect of baseline TNF-α were also evaluated (Table 4). Our findings suggest possible modification of the effect of baseline TNF-α by NSAID use (P_interaction = 0.09). While the hazards ratio for a one-unit increase in log-transformed baseline TNF-α (pg/mL) was suggestive of a positive association among NSAID users (HR = 1.12; 95% CI: 0.70–1.80), the effect was protective among non-users (HR = 0.60; 95% CI: 0.26–1.48; P_interaction = 0.26). A similar trend was observed among increasing tertiles of baseline TNF-α.

Table 3.

Hazards ratios (HR) and 95% CI of inflammatory markers for obesity-related cancers ^*.

		Tertiles of Inflammatory Markers
	Log-transformed	T1	T2	T3	ptrend
CRP (ug/mL), Cases (PY *)	156 (23,047)	48 (7,935)	51 (7,643)	57 (7,469)
Baseline
Model 1	1.03 (0.89,1.18)	1.00	1.12 (0.76,1.67)	1.32 (0.89,1.96)	0.16
Model 2	1.01 (0.88,1.17)	1.00	1.11 (0.75,1.65)	1.30 (0.87,1.94)	0.19
Updated
Model 1	1.08 (0.93,1.24)	1.00	0.86 (0.58,1.27)	1.26 (0.87,1.82)	0.10
Model 2	1.07 (0.92,1.24)	1.00	0.83 (0.56,1.24)	1.22 (0.83,1.78)	0.14
Averaged
Model 1	1.01 (0.86,1.20)	1.00	0.69 (0.46,1.03)	1.29 (0.91,1.83)	0.05
Model 2	0.98 (0.82,1.17)	1.00	0.66 (0.44,1.00)	1.22 (0.85,1.77)	0.11
IL6 (pg/mL), Cases (PY)	159 (24,646)	51 (9,009)	67 (8,318)	41 (7,319)
Baseline
Model 1	0.98 (0.78,1.23)	1.00	1.37 (0.95,1.97)	0.92 (0.61,1.40)	0.45
Model 2	0.94 (0.74,1.20)	1.00	1.34 (0.92,1.94)	0.90 (0.58,1.38)	0.37
Updated
Model 1	0.94 (0.74,1.19)	1.00	1.15 (0.79,1.66)	0.92 (0.62,1.36)	0.37
Model 2	0.90 (0.71,1.16)	1.00	1.08 (0.74,1.57)	0.87 (0.58,1.30)	0.26
Averaged
Model 1	0.85 (0.66,1.10)	1.00	1.09 (0.77,1.56)	0.80 (0.54,1.17)	0.15
Model 2	0.78 (0.59,1.02)	1.00	1.03 (0.72,1.49)	0.72 (0.48,1.07)	0.06
TNF-α (pg/mL), Cases (PY)	155 (24,079)	59 (8,564)	54 (8,108)	42 (7,408)
Baseline
Model 1	0.99 (0.66,1.48)	1.00	1.00 (0.69,1.45)	0.87 (0.59,1.31)	0.51
Model 2	0.99 (0.65,1.49)	1.00	0.97 (0.67,1.41)	0.86 (0.57,1.30)	0.48

^*Obesity-related cancers include cancers of breast, endometrium, pancreas and prostate. *PY: person-years.

Model 1: adjusted for age in years (continuous), race (black/white), gender and site.

Model 2: adjusted for age in years (continuous), race (black/white), gender, site, body mass index (kg/m²), smoking (never, former <20 pack-years, former ≥20 pack-years, current <20 pack-years, current ≥20 pack-years), current diabetes, NSAID use (yes/no), baseline physical activity (0, 0–499.9, 500–999.9, 1,000–1,499.9, 1,500–1,999.9, >2,000 Kcal/week) and alcohol consumption (no consumption in the past year, <1 drink/week, 1–7 drinks/week, >1 drinks/day).

Table 4.

Hazards ratios (HR) and 95% CI of inflammatory markers for obesity-related cancers by obesity, NSAID use and gender.

	Log-Transformed Inflammatory Markers		Tertiles of Inflammatory Markers
		Pinteraction	T1	T2	T3	Pinteraction
CRP (ug/mL), updated
Obesity
BMI <30, Cases (PY *)	119 (18,808)	0.04	40 (7,128)	31 (6,319)	48 (5,361)	0.11
RR (95% CI)	1.21 (1.02,1.44)		1.00	0.84 (0.52,1.34)	1.64 (1.06,2.52)
BMI ≥30, Cases (PY)	45 (6,598)		13 (1,351)	15 (2,151)	17 (3,096)
RR (95% CI)	0.76 (0.58,1.00)		1.00	0.73 (0.35,1.56)	0.56 (0.27,1.17)
NSAID Use
Yes, Cases (PY)	130 (20,773)	0.85	41 (6,978)	39 (6,939)	50 (6,855)	0.60
RR (95% CI)	1.08 (0.92,1.28)		1.00	0.93 (0.59,1.44)	1.24 (0.80,1.90)
No, Cases (PY)	34 (4,633)		12 (1,501)	7 (1,530)	15 (1,602)
RR (95% CI)	1.03 (0.75,1.42)		1.00	0.56 (0.21,1.44)	1.20 (0.53,2.69)
Gender
Men, Cases (PY)	93 (10,897)	0.29	34 (4,042)	27 (3,849)	32 (3,006)	0.24
RR (95% CI)	0.99 (0.81,1.20)		1.00	0.75 (0.45,1.25)	1.12 (0.67,1.85)
Women, Cases (PY)	71 (14,509)		19 (4,438)	19 (4,620)	33 (5,451)
RR (95% CI)	1.14 (0.91,1.43)		1.00	0.90 (0.47,1.71)	1.29 (0.72,2.33)
IL-6 (pg/mL), updated
Obesity
BMI <30, Cases (PY)	116 (18,948)	0.22	40 (7,177)	43 (6,154)	33 (5,617)	0.21
RR (95% CI)	0.99 (0.75,1.31)		1.00	1.14 (0.73,1.76)	0.97 (0.61,1.56)
BMI ≥30, Cases (PY)	49 (6,679)		12 (1,394)	19 (2,422)	18 (2,863)
RR (95% CI)	0.67 (0.40,1.14)		1.00	0.79 (0.37,1.68)	0.61 (0.28,1.31)
NSAID Use
Yes, Cases (PY)	132 (20,947)	0.96	40 (6,850)	48 (7,069)	44 (7,028)	0.35
RR (95% CI)	0.90 (0.68,1.19)		1.00	1.05 (0.69,1.62)	0.94 (0.60,1.47)
No, Cases (PY)	33 (4,680)		12 (1,721)	14 (1,508)	7 (1,452)
RR (95% CI)	1.00 (0.59,1.67)		1.00	1.22 (0.56,2.70)	0.60 (0.23,1.58)
Gender
Men, Cases (PY)	94 (11,010)	0.70	27 (3,226)	39 (3,876)	28 (3,908)	0.53
RR (95% CI)	0.90 (0.65,1.25)		1.00	1.18 (0.71,1.94)	0.84 (0.49,1.45)
Women, Cases (PY)	71 (14,617)		25 (5,344)	23 (4,700)	23 (4,572)
RR (95% CI)	0.89 (0.61,1.30)		1.00	0.94 (0.52,1.67)	0.93 (0.51,1.70)
TNF-α (pg/mL), baseline
Obesity
BMI <30, Cases (PY)	109 (17,830)	0.64	44 (6,340)	33 (6,049)	32 (5,441)	0.64
RR (95% CI)	1.01 (0.62,1.65)		1.00	0.79 (0.50,1.26)	0.90 (0.56,1.43)
BMI ≥30, Cases (PY)	46 (6,249)		15 (2,223)	21 (2,058)	10 (1,967)
RR (95% CI)	1.00 (0.46,2.17)		1.00	1.62 (0.82,3.20)	0.80 (0.35,1.83)
NSAID Use
Yes, Cases (PY)	123 (19,653)	0.26	43 (7,075)	45 (6,628)	35 (5,950)	0.09
RR (95% CI)	1.12 (0.70,1.80)		1.00	1.13 (0.74,1.73)	1.00 (0.63,1.59)
No, Cases (PY)	32 (4,427)		16 (1,489)	9 (1,479)	7 (1,458)
RR (95% CI)	0.62 (0.26,1.48)		1.00	0.60 (0.26,1.39)	0.51 (0.20,1.31)
Gender
Men, Cases (PY)	85 (10,363)	0.96	29 (3,374)	34 (3,468)	22 (3,521)	0.78
RR (95% CI)	0.91 (0.52,1.60)		1.00	1.21 (0.73,2.01)	0.84 (0.47,1.49)
Women, Cases (PY)	70 (13,717)		30 (5,190)	20 (4,640)	20 (3,887)
RR (95% CI)	1.05 (0.56,1.97)		1.00	0.67 (0.37,1.19)	0.92 (0.51,1.66)

^*PY: person-years.

Models were adjusted for age in years (continuous), race (black/white), gender, site, body mass index (kg/m²), smoking (never, former <20 pack-years, former ≥20 pack-years, current <20 pack-years, current ≥20 pack-years), current diabetes, NSAID use (yes/no), baseline physical activity (0, 0–499.9, 500–999.9, 1,000–1,499.9, 1500–1,999.9, >2,000 Kcal/week) and alcohol consumption (no consumption in the past year, <1 drink/week, 1–7 drinks/week, >1 drinks/day).

DISCUSSION

We found that participants in the highest tertile of updated and averaged CRP had a more than two-fold higher risk of colorectal cancer compared to those in the lowest tertile. Participants in the second tertile of baseline IL-6 also had a two-fold higher risk of colorectal cancer compared to those in the lowest tertile. We found no association between updated or averaged measures of IL-6 or baseline TNF-α with incident colorectal cancer. Although we found no significant associations between CRP, IL-6 and TNF-α and obesity-related cancers, overall, we found possible evidence for effect modification by gender, BMI and NSAID use, but had very limited statistical power for these analyses.

Whereas several prospective studies reported a significant association between CRP and colorectal cancer ^{8, 9, 12, 13, 16}, others have generally reported null findings ^{11, 14, 15, 17–20} or inverse associations ¹⁰. The interpretation of those findings is limited, because all utilized inflammation marker levels which were measured at a single point in time. Specifically, the temporal instability of markers can lead to measurement error, reflected as low long-term intraclass correlation, which can result in marked attenuation of risk estimates ²⁴. In addition, only a few studies accounted for important potential confounders, such as dietary factors and physical activity ^8–17. One of the studies ²⁹ adjusted for body mass index as a general marker of obesity, but not for measures of abdominal adiposity. Adipose tissue produces pro-inflammatory cytokines that induce hepatic CRP secretion ^{30, 31}. CRP concentrations are, in turn, associated with abdominal obesity, insulin resistance and dyslipidemia ^{30, 31}, which have been suggested to be associated with colorectal cancer risk ^32–35. Therefore, abdominal adiposity may be the most etiologically relevant measure to take into account when studying inflammation. In addition to having repeated measures of the exposure, the present analysis extends previous work by including a number of important time-varying confounders that could be related to inflammation, such as abdominal adiposity.

It is possible that the association of chronic inflammation with colorectal and obesity-related cancers is confounded by unmeasured lifestyle factors over time, such as physical activity and postmenopausal hormone use in women. The lack of specific data on such factors and the lack of consistent updates on inflammatory markers until the end of study follow-up are limitations of this study. Moreover, inflammation can be both the cause and the consequence of adverse lifestyle factors, such as physical inactivity. Thus, such factors may be important components of the causal pathway between inflammation and cancer. Non-differential misclassification of exposure due to potential batch effects is also possible. Exposure categories were defined separately for each year to minimize any potential bias introduced by differences in sample source, laboratory protocols or procedures and time in frozen storage. Furthermore, inflammatory markers are known to be correlated due to functional redundancy ³⁶. Because we evaluated only three inflammatory markers, our findings provide limited insight into the correlation structures that may exist across inflammation markers and key inflammatory pathways involved in carcinogenesis. For example, although TNF-α stimulates prostate cancer angiogenesis, it may inhibit neovascularization, induce apoptosis and stimulate tumor immunity ³⁷. The opposing effect of TNF-α on microvessels of neoplasms ad inflammatory reaction, which depend on its local tissue concentration ³⁸, may explain our inability to detect a significant association between levels of circulating TNF-α and colorectal cancer. Similarly, variable effects of IL-6 on the proliferation of cancer cells have been reported ³⁹.

The pattern of the relationships between the inflammation markers and incident colorectal cancer supports the possible role of CRP as a systemic marker of underlying colonic inflammation, including a direct marker of neoplastic lesions ²². Although the findings from this study population of older adults may not easily generalize to younger populations, in a study of a representative sample of the U.S. population of adults aged 50 and older, participants with clinically-raised CRP at baseline exhibited a greater than two-fold risk of colorectal cancer death compared to those with undetected levels, suggesting that elevated CRP levels may exert an adverse effect on colorectal cancer across the spectrum of age ²². Additionally, it should be noted that the evidence supporting an etiological role for obesity in prostate carcinogenesis has been increasingly equivocal. Individual studies have found no association ^40–46 or modest increases in the risk of prostate cancer among obese men ^47–53; however, a recent meta-analysis of 17 cohort studies reported that while obesity was not associated with overall prostate cancer incidence (RR = 1.00; 95% CI: 0.95–1.06), it was significantly associated with increased risk of aggressive prostate cancer (RR = 1.24; 95% CI: 1.15–1.33) ⁵⁴.

To our knowledge, this is the first study to date to evaluate repeated measures of inflammatory markers in relation to the risk of colorectal and obesity-related cancers. Other strengths of this study include a racially- and geographically-diverse prospective cohort design that reduces the likelihood of reverse causality, while also permitting the evaluation of whether elevated levels of inflammatory markers serve as contributors to carcinogenesis or simply as a manifestation of subclinical tumor-related inflammation ²⁵. We also excluded exposure measurements occurring less than two years prior to cancer diagnosis to safeguard against temporally-biased associations. Moreover, biomarkers were measured in blood, resulting in more accurate assays than urine biomarker measurements that have been used in other studies ²⁵. The risk of outcome misclassification in our study is small in light of the thorough adjudication process of incident diseases. Since multiple factors, including smoking, obesity, physical inactivity, persistent or transient infections, may lead to increased levels of inflammatory markers and attendant cancer risk, our ability to adjust for many of the aforementioned time-varying confounders is another significant strength of this study. Similar to a recent study of colorectal cancer in the European Prospective Investigations into Cancer and Nutrition ⁵⁵, a strength of this study is the simultaneous inclusion of BMI and abdominal circumference in multivariable models due to gender differences in risk factors for colorectal cancer. While BMI has been found to predict colorectal cancer among men ^{27, 28}, evidence remains inconsistent for women ²⁷, and central adiposity may be a more informative measure of risk than BMI for women ^{27, 29, 56}.

In conclusion, we found that elevated CRP levels, measured over a period of eight years, were associated with significantly increased risk of colorectal cancer. In contrast, IL-6 and TNF-α were not major predictors of colorectal cancer risk. The inflammatory markers were not associated with other obesity-related cancers, although there was a suggestion of effect modification by BMI and NSAID use. Our observations, combined with those of other investigators, suggest that chronically-elevated levels of inflammatory markers are likely associated with an increased risk of colorectal cancer. Larger studies with repeated measures of inflammatory markers, more specific methods of quantifying inflammation and a better understanding of how circulating markers of inflammation operate at the tissue level will clarify the possible causal role of inflammation in carcinogenesis.

Impact/Novelty.

Given the temporal instability and high intra-person variability of markers of chronic inflammation, it is important to assess repeated measures of these markers when examining their association with cancer outcomes. To our knowledge, this analysis is the first to evaluate the relationship between repeated measures of inflammatory markers and incident cancer. Our findings add to the evidence that CRP may play a role in colorectal cancer carcinogenesis among older adults.

Abbreviations

Health ABC Study

Health, Aging and Body Composition Study

vs

versus

CRP

C-reactive protein

IL-6

interleukin-6

TNF-α

tumor necrosis factor-α

BMI

body mass index

HR

hazards ratio

95% CI

95% confidence interval

kcal/wk

kilocalories per week

kg

kilograms

cm

centimeters

ICD

International Classification of Diseases

IQ

interquartile range