C-reactive protein and colorectal cancer risk: A systematic review of prospective studies

Abstract

C-reactive protein is a sensitive but nonspecific systemic marker of inflammation. Several prospective studies have investigated the association of prediagnostic circulating C-reactive protein concentrations with the development of colorectal cancer, but the results have been inconsistent. We performed a systematic review of prospective studies of the association between prediagnostic measurements of circulating high-sensitivity C-reactive protein and development of invasive colorectal cancer. Authors of original studies were contacted to acquire uniform data. We combined relative risks (RR) for colorectal cancer associated with a one unit change in natural logarithm-transformed high-sensitivity C-reactive protein using inverse variance weighted random effects models. We identified eight eligible studies, which included 1,159 colorectal cancer cases and 37,986 controls. The summary RR per one unit change in natural log-transformed high-sensitivity C-reactive protein was 1.12 (95% confidence intervals [CI], 1.01–1.25) for colorectal cancer, 1.13 (95% CI, 1.00–1.27) for colon cancer, and 1.06 (95% CI, 0.86–1.30) for rectal cancer. The association was stronger in men (RR, 1.18; 95% CI, 1.04–1.34) compared to women (RR, 1.09; 95% CI, 0.93–1.27) but this difference was sensitive to the findings from a single study. Prediagnostic high-sensitivity C-reactive protein concentrations were weakly associated with an increased risk for colorectal cancer. More work is needed to understand the extent to which circulating high-sensitivity C-reactive protein or other blood inflammatory markers are related to colonic inflammation. © 2008 Wiley-Liss, Inc.

C-reactive protein (CRP) is a sensitive but nonspecific systemic marker of inflammation.1 CRP is produced mainly in the liver along with other acute-phase proteins in response to cytokines released by phagocytes during infection, trauma, surgery, burns, tissue infarction, advanced cancer and chronic inflammatory conditions.2–4 Several lines of evidence suggest that colorectal neoplasia may arise from colonic areas with chronic subclinical inflammation.5 Investigators have hypothesized that CRP may act as a biomarker for chronic low-grade intestinal inflammation and the subsequent development of colorectal cancer.

Several retrospective case–control studies have compared CRP concentrations between colorectal cancer patients and healthy controls, and have reported at least 10-fold higher concentrations in the cancer patients.6–9 These findings may be explained to some extent by reverse causality due to the host inflammatory response to existing advanced cancer among cases. Prospective studies, where CRP is measured long before cancer diagnosis in all participants, are thus required to answer whether CRP is associated with colorectal cancer incidence. Findings from such studies, however, have been inconsistent. The reasons underlying these heterogeneous findings need to be investigated, but no formal meta-analysis has been published. Therefore, we performed a systematic review and meta-regression analysis to summarize the findings and address the inconsistencies of the CRP and colorectal cancer incidence literature.

Material and methods

Study identification

We searched MEDLINE and EMBASE from 1966 through January 2007 to identify prospective epidemiologic studies of the association between prediagnostic measurements of circulating CRP and development of invasive colorectal cancer. In addition, we also performed a search of the Cochrane Systematic Review Database, and a manual review of references from relevant original or review articles. The search strategy used the Medical Subject Heading and text key words “(C-reactive protein or CRP) and (colorectal or colon or rectal) and (cancer or carcinoma or neoplasia or tumor or adenoma or neoplasm).” No language restrictions were imposed. We excluded articles that had no human or no original data, that did not have colorectal cancer or adenoma as an outcome, or that did not measure blood CRP concentrations. We also excluded prognostic studies.

Our search identified 399 articles potentially relevant for abstract review (Fig. 1). Abstracts were reviewed independently by 2 investigators (K.K.T. and C.B.). Differences were resolved by consensus. There were 37 articles retrieved for full-text review based on information in the abstracts. Of these, 9 studies were identified as relevant and were abstracted. Data abstraction was conducted independently by the same 2 investigators, and discrepancies were adjudicated. One study had 2 publications using data from overlapping study populations.10, 11 We only abstracted the results from the more recent publication, leading to 8 relevant studies to summarize.11–18 An additional search of MEDLINE and EMBASE through December 2007 did not identify more relevant studies.

Figure 1. Flow diagram of the study selection process.

The quality of individual studies was assessed using the STROBE statement guidelines.19 Study quality was evaluated independently by 2 investigators (K.K.T. and E.A.P.) with disagreement resolved by consensus.

Statistical analysis

Corresponding authors from the 8 eligible studies were contacted to acquire uniform findings. We requested maximally adjusted relative risk (RR) estimates and 95% confidence intervals (CIs) for the association of CRP (per one unit change in mg/L), natural logarithm (ln) of CRP (per one unit change in ln mg/L), and quantiles of CRP (tertiles or quartiles in mg/L) with colorectal, colon, rectal, proximal colon, distal colon cancer and colorectal cancer in men and women. The study-specific dose-response slopes were approximately linear. Seven out of the 8 study authors provided additional information. Erlinger et al.12 provided additional results adjusted for smoking status, body mass index, use of aspirin and nonsteroidal anti-inflammatory drugs, and hormone use in women. Il'yasova et al.14 provided additional results for the colorectal cancer outcome. Ito et al.15 provided updated data with more cases and controls compared to the numbers in the published paper. Otani et al.16 reported results for both invasive and intramucosal colorectal cancer in their original study; invasive cancer results are included in this meta-analysis. For the Siemes et al. study17 that did not provide additional data, we computed the RR for colorectal cancer associated with each 1 mg/L increase in CRP.20, 21 This study17 provided estimates for sigmoid and nonsigmoid colon cancer in their article. These estimates were used to approximate distal and proximal colon cancer results, respectively. These estimates were combined to yield colon cancer estimates using the DerSimonian and Laird random effects model.22

We weighted the study-specific ln RR estimates by the inverse of their variance to calculate a summary estimate and its 95% CI using a random effects model.22 The primary analyses combined ln RR associated with one unit change in ln CRP. Alternative analyses were based on fixed effects models, on empirical Bayes models, on ln RR associated with one unit change in CRP (mg/L), and on ln RR estimates comparing the highest to the lowest category of CRP with similar findings (data not shown).

We performed sensitivity and subgroup analyses to investigate sources of heterogeneity. We repeated the random effects meta-analysis after omitting one study at a time. The association of circulating CRP on colorectal cancer risk was assessed by colorectal cancer site (colon, rectal, proximal colon, distal colon) and sex. We also assessed the influence on our findings of study location (Western vs. Japanese populations), study design (cohort vs. nested case–control), sample size (≥100 vs. <100 colorectal cancer cases), exclusion criteria [excluding patients with cardiovascular disease (CVD) at baseline vs. no], CRP assay methodology [enzyme-linked immunosorbent assay (ELISA) vs. immunoturbidimetric/immunonephelometric], and mean age at baseline (≥60 vs. <60 years). To test whether the summary estimates differed between strata of the latter characteristics, we conducted meta-regression analyses; we evaluated the coefficient for each characteristic using the Wald test.

Statistical heterogeneity among studies was assessed using the Q statistic. Publication bias was evaluated with the use of funnel plots,23 with the Begg rank correlation method,24 and with Egger's regression asymmetry test.23 All statistical analyses were performed with STATA version 9 (StataCorp, College Station, TX). All statistical tests were two-sided.

Results

Eight prospective studies met our inclusion criteria (Fig. 1, Tables I and II, appendix Table AI). Three studies were from US populations,11, 12, 14 2 from Japanese populations15, 16 and the remaining from European populations.13, 17, 18 The number of colorectal cancer cases varied between 41 and 250. Combined, these studies included 1,159 colorectal cancer cases and 37,986 controls. Five studies were nested case–control and 3 were cohort studies. All studies determined circulating CRP concentration before cancer diagnosis as a single measurement in time. All CRP assays were of high sensitivity; 2 studies14, 15 used an ELISA methodology to measure high-sensitivity CRP, 3 studies11, 13, 17 used an immunoturbidimetric assay, 2 studies12, 16 used an immunonephelometric assay, and one study18 used an automatic latex agglutination photometric assay. The assays were uniformly reliable. Colorectal cancer cases were always confirmed by medical or pathology reports. Maximum follow-up time was about 10 years in each study. The mean age of participants at baseline ranged from 53 to 73 years across studies. Appendix Table AI shows the findings reported in each of the 8 original publications. Most studies adjusted for colorectal cancer risk factors or stated that findings did not materially change after such adjustments. On the basis of the STROBE statement, the reporting of studies was uniformly very good (appendix Table AII).

Table I. Study Characteristics of the Eight Eligible Prospective Studies on C-Reactive Protein and Colorectal Cancer Risk

Study	Location/Population (Cohort name)	Design	No. of cases/controls	Exclusion criteria	Follow-up (years)1	CV% (CRP)	Outcome assessment
WHS, Women's Health Study; JACC, Japan Collaborative Cohort Study; HABC, Healthy Aging and Body Composition Cohort; ATBC, α-Tocopherol β-Carotene Cancer Prevention Study; JPHC, Japan Public Health Center-Based Prospective Study; EPIC, European Prospective Investigation into Cancer and Nutrition; CVD, cardiovascular disease; MI, myocardial infarction; CRP, C-reactive protein; CV%, coefficient of variation; C-C, case-control. 1 Maximum years of follow-up unless otherwise specified. 2 At CRP concentration of 0.15 mg/L. 3 Median. 4 Also excluded if difficulty walking ¼ mile, climbing 10 steps, or doing basic activities of daily living. 5 Mean. 6 Also excluded if users of vitamin A, E, or β-carotene supplements in excess of predefined doses, and anticoagulants. 7 Correlation coefficient for duplicated measures was 0.98.
Erlinger, 200412	USA/General population (CLUE II)	Nested C-C	172/342	Prior cancer	11	4.92	Registry, pathology reports
Zhang, 200511	USA/Female health professionals (WHS)	Cohort	169/27, 913	Prior cancer, CVD	10.13	3.3	Self-report, medical records
Ito, 200515	Japan/General population (JACC)	Nested C-C	160/433	Prior cancer	10	NR	Medical records
Il'yasova, 200514	USA/Medicare beneficiaries (HABC)	Cohort	41/2169	Prior cancer, life-threatening illness4	5.55	NR	Self-report, medical records
Gunter, 200613	Finland/Male smokers (ATBC)	Nested C-C	130/260	Prior cancer, serious illness6	14	8.0	Registry
Otani, 200616	Japan/General population (JPHC)	Nested C-C	250/500	Prior cancer	10	6.3	Registry, pathology reports
Trichopoulos, 200618	Greece/General population (Greek EPIC)	Nested C-C	48/96	Prior cancer, MI	11	2.7	Self-report, medical records
Siemes, 200617	Netherlands/General population (Rotterdam study)	Cohort	189/6273	Prior cancer, and CRP ≥ 10 mg/L	10.23	7	Medical records

Table II. Population Characteristics of the Eight Eligible Prospective Studies on C-Reactive Protein and Colorectal Cancer Risk

Study	Mean age (years)	% Male	% Current smoking	Mean BMI (kg/m^2)	% Hormone in females	% Aspirin	% NSAID	Mean Alcohol (g/d)	% CRC family history
BMI, body mass index; NSAID, non-steroidal anti-inflammatory drugs; CRC, colorectal cancer. 1 Cases. 2 Controls. 3 Hormone replacement therapy. 4 Oral contraceptives. 5 % in 60–69 age group: Cases, 43; Controls, 39. 6 % Current drinker: Cases, 48; Controls, 44. 7 Mean pack-years: Cases, 14; Controls, 12. 8 41% in > 65 age group (matched). 9 % Obese: Cases, 36; Controls, 35. 10 Aspirin and NSAIDs. 11 % > 10 g/d: Cases, 24; Controls, 28.
Erlinger, 200412	63 (matched)	45 (matched)	121, 122	271, 262	21, 102	61, 92	151, 202	NR	61, 42
Zhang, 200511	53	0	12	26	443, 704	12	NR	4	11
Ito, 200515	5	451, 442	211, 242	NR	NR	NR	NR	6	NR
Il'yasova, 200514	73	45	10	27	NR	NR	23	NR	NR
Gunter, 200613	561, 572	1001,1002	1001,1002	261, 262	NA	61, 202	NR	161, 132	31, 32
Otani, 200616	57 (matched)	52 (matched)	7	231, 232	NR	NR	NR	181, 132	21, 12
Trichopoulos, 200618	8	49 (matched)	261, 232	9	NR	221, 20210	221, 20210	11	NR
Siemes, 200617	70	40	22	26	NR	NR	NR	NR	NR

High-sensitivity CRP concentration was statistically significantly associated with a small increased risk of colorectal (8 studies; per one unit change in ln mg/L; RR, 1.12; 95% CI, 1.01–1.25; p_{heterogeneity}, 0.05) and colon (7 studies; RR, 1.13; 95% CI, 1.00–1.27; p_{heterogeneity}, 0.16) cancer, but not with rectal (7 studies; RR, 1.06; 95% CI, 0.86–1.30; p_{heterogeneity}, 0.02), proximal colon (5 studies; RR, 1.19; 95% CI, 0.90–1.56; p_{heterogeneity}, <0.01), or distal (5 studies; RR, 1.05; 95% CI, 0.90–1.22; p_{heterogeneity}, 0.33) colon cancer (Figs. 2–4). There was some indication of heterogeneity across studies. In sensitivity analysis, exclusion of individual studies did not modify the estimates substantially except for one study. After excluding the Zhang et al. study, the findings for colorectal (7 studies; RR, 1.15; 95% CI, 1.07–1.24; p_{heterogeneity}, 0.44), and colon cancer (6 studies; RR, 1.18; 95% CI, 1.06–1.32; p_{heterogeneity}, 0.59) were stronger and less heterogeneous; for proximal colon cancer the findings were statistically significant (4 studies; RR, 1.32; 95% CI, 1.09–1.61; p_{heterogeneity}, 0.24). Funnel plots and other statistical tests did not provide evidence of substantial publication bias (data not shown).

Figure 2. Relative risk for colorectal cancer (CRC) associated with a 1 unit increase in ln-transformed C-reactive protein concentration (p_{heterogeneity}, 0.05).

Figure 3. Relative risk for colon (p_{heterogeneity}, 0.16) and rectal (p_{heterogeneity}, 0.02) cancer associated with a 1 unit increase in ln-transformed C-reactive protein concentration.

Figure 4. Relative risk for proximal (p_{heterogeneity}, <0.01) and distal (p_{heterogeneity}, 0.33) colon cancer associated with a 1 unit increase in ln-transformed C-reactive protein concentration.

In meta-analyses performed separately by sex (Fig. 5), ln high-sensitivity CRP was not associated with colorectal cancer in women, but a significant positive association was observed in men. After excluding the Zhang et al. study, the findings in women (4 studies; RR, 1.17; 95% CI, 1.03–1.34; p_{heterogeneity}, 0.69) were similar to those in men. The associations of ln high-sensitivity CRP with colorectal cancer risk were stronger in nested case–control compared to cohort studies (from meta-regression model: p, 0.08), and in studies that did not exclude participants with history of CVD at baseline (from meta-regression model: p, 0.01). The associations did not differ by study location, possibility of residual confounding, age, sample size, and high-sensitivity CRP assay methodology (appendix Table AIII).

Figure 5. Relative risk for colorectal cancer (CRC) in men (p_{heterogeneity}, 0.79) and women (p_{heterogeneity}, 0.08) associated with a 1 unit increase in ln-transformed C-reactive protein concentration.

Discussion

In this meta-analysis, we identified a positive but weak association between pre-diagnostic circulating high-sensitivity CRP concentrations and colorectal and colon cancer risk. The association was in the same direction but weaker and/or not statistically significant for rectal, proximal colon, or distal colon cancer. There was some evidence that the association was stronger in men than in women, although this difference was sensitive to the findings from a single study. Our results confirm and extend with additional data and with use of meta-analytic and meta-regression techniques the findings of a previous systematic review on high-sensitivity CRP and cancer.25 Two published studies and 1 preliminary report in abstract form did not observe statistically significant positive associations between high-sensitivity CRP concentration and colorectal adenomas.26–28

Despite the weak strength of the association between high-sensitivity CRP and colorectal cancer, evidence is mounting for a role of inflammation in colorectal neoplasia from studies using several different markers of inflammation and multiple designs. Animal models and mechanistic studies implicate inflammation as a key predisposing factor to colorectal neoplasia.29–31 Dysplasia and cancer developed in healthy mice or rats when colitis was induced by exposure to repeated cycles of dextran sulfate sodium (DSS).32, 33 DSS is a synthetic, sulfated polysaccharide that disrupts the colonic mucosal barrier function, resulting in acute and chronic colitis. Certain bacteria (e.g., bifidobacteria) and agents with demulcent (e.g., polyethylene glycol), anti-inflammatory (e.g., nonsteroidal anti-inflammatory drugs [NSAIDs]), and anti-oxidant (e.g., folic acid) properties diminish the growth and subsequent development of colorectal neoplasia.29, 30 In epidemiologic studies, blood levels of fibrinogen, albumin, white blood cell count and immune response-related genes were associated with colorectal neoplasia.34–39 Moreover, patients with inflammatory bowel diseases are at increased risk for developing colitis-associated colorectal cancer.31, 40 Observational studies and human intervention trials have indicated that the regular administration of NSAIDs confers a 30–50% reduction in colorectal cancer risk or adenoma recurrence.41–44

Given that high-sensitivity CRP is a circulating marker for inflammation, the lack of a consistent association with colorectal neoplasia raises questions on its use as a marker of the inflammatory process associated with colorectal cancer. Persistent production of CRP reflects the signaling of the innate immune system. Chronic uncontrolled low-grade colonic inflammation may be responsible for the development and progression of colorectal neoplasia.5 Thus, high-sensitivity CRP will be associated with colorectal cancer development and/or progression only to the extent that high-sensitivity CRP correlates well with colonic inflammation.

Several factors should be considered in the interpretation of our findings. First, meta-analyses of observational studies are particularly vulnerable to biases and confounding inherent in the original studies. The temporal association between high-sensitivity CRP and colorectal cancer is supported by the prospective design of the 8 eligible studies, and by the confirmation of the study findings after excluding colorectal cancer cases that occurred within one to 5 years from baseline. Most studies in this meta-analysis adjusted for colorectal cancer risk factors. However, residual confounding may still be present especially due to the use of body mass index as a surrogate for obesity. Studies of the association between functional CRP gene polymorphisms and colorectal neoplasia may be used in the future to overcome residual confounding using the Mendelian randomization principles.45 Second, heterogeneity may be introduced because of methodologic and demographic differences among studies. We used appropriate well-motivated inclusion criteria to maximize homogeneity, and investigated in detail potential sources of heterogeneity to assess their contributions to our findings. Only high-quality prospective studies of invasive colorectal cancer were included. Heterogeneity tests and meta-regression models, sensitivity and subgroup analyses were performed. We identified 4 sources of between-study heterogeneity: the inclusion/exclusion of the Zhang et al. study,11 the differential association by sex, the study design, and the exclusion of subjects with a history of CVD at baseline. The Zhang et al. study was performed within the Women's Health Study (randomized clinical trial of low-dose aspirin and vitamin E), which recruited female health professionals, and found an almost significant inverse association between high-sensitivity CRP and colorectal cancer. This study appeared to be responsible for most of the heterogeneity in our meta-analysis, as the remaining 3 variables (study design, sex distribution and exclusion of subjects with CVD) are probably not true sources of heterogeneity, but simply correlates of the inclusion/exclusion of the Zhang et al. study. Additional large prospective studies are needed to verify the association of high-sensitivity CRP with colorectal cancer in women.

Several lessons can be gleaned from this review and should be considered in future epidemiologic studies. Prospective study design is needed to elucidate the role of high-sensitivity CRP in colorectal carcinogenesis. Because the presence of malignant disease might itself affect CRP circulating concentrations, measurements from retrospective case–control studies or from cohort studies where case diagnosis is close in time to blood draw might reflect tumor marker status rather than true risk assessment. Whether circulating concentration of high-sensitivity CRP truly reflects colonic inflammation and/or translates into biological activity is unclear, emphasizing the need for more research to explore the correlation of high-sensitivity CRP with colonic inflammation.

In our meta-analysis, we identified only a weak positive association between circulating high-sensitivity CRP concentrations and colorectal cancer risk. More work is needed to understand the extent to which circulating high-sensitivity CRP or other blood inflammatory markers are related to colonic inflammation.

Acknowledgements

We greatly thank the authors of the original 8 studies, who kindly and promptly replied to our requests for additional information. Our C-reactive protein and colorectal neoplasia research was funded by the American Institute for Cancer Research and the Maryland Cigarette Restitution Fund at Johns Hopkins University. Konstantinos K. Tsilidis was supported by a Fulbright grant and a scholarship from the Hellenic State Scholarships Foundation.

Appendix

Table AI. Published Findings Extracted from the Eight Eligible Prospective Studies on C-Reactive Protein and Colorectal Cancer Risk

Study	Median CRP (mg/L)	RR CRC (95% CI)1	RR colon (95% CI)1	RR rectum (95% CI)1	RR proximal colon (95% CI)1	RR distal colon (95% CI)1	Potential confounders assessed	Matching variables
CRP, C-reactive protein; CRC, colorectal cancer; RR, relative risk; BMI, body mass index; NSAIDs, non steroidal anti-inflammatory drugs. 1 Relative risk estimates as reported in each study of invasive cancer comparing top vs. bottom quantile of CRP unless otherwise specified. 2 Cases. 3 Controls. 4 Additional adjustments for BMI, smoking, hormone, and aspirin/NSAID use did not change the reported findings. 5 Per one unit change in ln-C-reactive protein. 6 Additional adjustments for adiposity, smoking, physical activity, education, medical conditions at baseline and NSAID use did not change reported findings. 7 Per one standard deviation change in C-reactive protein. 8 Nonsigmoid colon. 9 Sigmoid colon.
Erlinger, 200412	2.442, 1.943	2.00 (1.16–3.46)	2.55 (1.34–4.88)	NR	NR	NR	None4	Age, sex, race, blood draw date, time since last meal
Zhang, 200511	2.02	0.66 (0.43–1.03)	NR	0.44 (0.18–1.08)	0.44 (0.22–0.89)	1.27 (0.59–2.73)	Age, treatment assignment, hormones, BMI, CRC family history, physical activity, smoking, aspirin, alcohol, hormones, menopausal status, multivitamins, polyp history	–
Ito,200515	0.432, 0.453	1.18 (0.68–2.06)	1.42 (0.73–2.74)	0.90 (0.3–2.3)	NR	NR	Smoking, alcohol, BMI	Age, sex, institution
Il'yasova, 200514	1.69	1.44 (1.03–2.02)5	NR	NR	NR	NR	Age, sex, race, study site6	–
Gunter, 200613	3.42, 2.63	2.9 (1.4–6.0)	1.8 (0.7–4.4)	7.8 (2.2–28.1)	NR	NR	Age, BMI, aspirin, smoking intensity and duration	Age, blood draw date, intervention group
Otani,200616	0.522, 0.453	NR	1.2 (0.64–2.4)	NR	1.3 (0.56–3.2)	0.86 (0.29–2.6)	Smoking, BMI, alcohol, physical activity, CRC family history	Age, sex, blood draw date, time since last meal, study area
Trichopoulos, 200618	NR	1.17 (0.93–1.46)7	NR	NR	NR	NR	Age, sex, BMI, smoking, alcohol, NSAIDs, blood storage duration	Age, gender, enrollment date
Siemes, 200617	NR	0.93 (0.63–1.37)	NR	0.38 (0.16–0.92)	2.07 (1.1–3.9)8	0.62 (0.31–1.24)9	Age, sex, smoking, NSAIDs, cholesterol, physical activity, hormones, fruit, selenium, energy	–

Table AII. Quality Assessment of the Eight Eligible Prospective Studies on C-Reactive Protein and Colorectal Cancer Risk Using the Strobe Statement Guidelines

Item/recommendation	Erlinger	Zhang	Ito	Il'yasova	Gunter	Otani	Trichopoulos	Siemes
NA, not applicable; RR, relative risk.
Title & abstract
Indicate study design	√	√	√	√	√	√	√	√
Provide informative summary	√	√	√	√	√	√	√	√
Introduction
Explain scientific background and rationale	√	√	√	√	√	√	√	√
State specific objectives/hypotheses	√	√	√	√	√	√	√	√
Methods
Present key elements of study design	√	√	√	√	√	√	√	√
Describe setting, location, and relevant dates	√	√	√	√	√	√	√	√
Give eligibility criteria, selection/follow-up methods	√	√	√	√	√	√	√	√
Give matching criteria	√	NA	√	NA	√	√	√	NA
Define outcomes, exposures, and confounders	√	√	√	√	√	√	√	√
Give assessment methods for all variables	√	√	√	√	√	√	√	√
Describe efforts to assess potential sources of bias	√	√	No	√	√	√	√	√
Explain how the study size was arrived at	√	√	√	√	√	√	√	√
Explain how variables were handled in analyses	√	√	√	√	√	√	√	√
Describe all statistical methods	√	√	√	√	√	√	√	√
Describe methods to examine subgroups/interaction	√	√	NA	√	√	√	√	√
Explain how missing data were addressed	√	No	No	No	No	No	No	√
Explain how loss to follow-up was addressed	NA	√	NA	√	NA	NA	NA	√
Explain how matching was addressed	√	NA	√	NA	√	√	No	√
Describe any sensitivity analyses	√	√	No	√	√	√	√	√
Results
Report participant numbers at each study stage	√	√	√	√	√	√	√	√
Give reasons for nonparticipation at each stage	√	√	√	√	√	√	√	√
Consider using a flow diagram	No	No	No	No	No	No	No	√
Give study population characteristics	√	√	√	√	√	√	√	√
Indicate % missing for all variables	No	No	No	No	No	No	No	No
Summarize follow-up time	√	√	√	√	√	√	√	√
Report numbers of outcome/exposure categories	√	√	√	√	√	√	√	√
Give unadjusted/adjusted estimates & precision	√	√	√	√	√	√	√	√
Give category bounds for categorized continuous variables	√	√	√	√	√	√	√	√
Consider translating RR estimates to absolute risk	No	No	No	No	No	No	No	No
Report other analyses (e.g., subgroups, sensitivity analyses)	√	√	No	√	√	√	√	√
Discussion
Summarize key results	√	√	√	√	√	√	√	√
Discuss study limitations	√	√	√	√	√	√	√	√
Give a cautious overall interpretation of results	√	√	√	√	√	√	√	√
Discuss generalizability of study results	No	√	No	√	√	√	No	No
Other information
Give source of funding and role of funders	√	√	No	√	No	√	√	√

Table AIII. Relative Risk for Colorectal Cancer Associated with A One Unit Increase in Ln-Transformed C-Reactive Protein Concentrations in Subgroups

Outcome measures	No. studies	RR estimate (95% CI)	p value for heterogeneity
CVD, cardiovascular diseases; CRC, colorectal cancer; IN, immunonephelometric assay; IT, immunoturbidimetric assay; CRP, C-reactive protein.
Study location
Non-USA	513, 15–18	1.12 (1.03–1.22)	0.57
Western	611–14, 17, 18	1.15 (0.98–1.34)	0.01
Japan	215, 16	1.11 (0.99–1.24)	> 0.95
Study design
Nested case-control	512, 13, 15, 16, 18	1.17 (1.07–1.28)	0.64
Cohort	311, 14, 17	1.05 (0.85–1.31)	0.03
Exclusion criteria
Exclude CVD at baseline	211, 18	0.98 (0.75–1.27)	0.17
Do not exclude CVD at baseline	612–17	1.16 (1.06–1.26)	0.33
Possible residual confounding
Yes	215, 18	1.12 (0.97–1.30)	0.74
No	611–14, 16, 17	1.13 (0.99–1.29)	0.015
Age at baseline
≥ 60 years	512, 14, 15, 17, 18	1.15 (1.04–1.26)	0.38
< 60 years	311, 13, 16	1.08 (0.87–1.34)	0.01
Sample size
< 100 CRC cases	214, 18	1.32 (1.03–1.70)	0.46
≥ 100 CRC cases	611–13, 15–17	1.10 (0.98–1.22)	0.04
CRP assay methodology
ELISA	214, 15	1.21 (0.95–1.54)	0.17
IN/IT	511–13, 16, 17	1.10 (0.96–1.26)	0.02