VOLUME 27 䡠 NUMBER 13 䡠 MAY 1 2009

JOURNAL OF CLINICAL ONCOLOGY O R I G I N A L R E P O R T

Baseline C-Reactive Protein Is Associated With Incident

Cancer and Survival in Patients With Cancer
Kristine H. Allin, Stig E. Bojesen, and Børge G. Nordestgaard
From the Department of Clinical
Biochemistry, Herlev Hospital, and The A B S T R A C T
Copenhagen City Heart Study, Bispeb-
jerg Hospital, Copenhagen University Purpose
Hospital; and Faculty of Health We tested the hypothesis that baseline plasma levels of C-reactive protein (CRP) are associated
Sciences, University of Copenhagen, with risk of incident cancer in the general population and early death in patients with cancer.
Copenhagen, Denmark.
Patients and Methods
Submitted August 26, 2008; accepted A total of 10,408 individuals from the Danish general population who had CRP measured at
January 21, 2009; published online
baseline were observed for up to 16 years; 1,624 developed cancer, and of these, 998 patients
ahead of print at www.jco.org on
March 16, 2009.
died during follow-up. Follow-up was 100% complete. We excluded individuals with a cancer
diagnosis at baseline.
Supported by the Danish Heart Founda-
tion, the Danish Medical Research Results
Council, and the Research Council at Baseline CRP levels more than 3 versus less than 1 mg/L were associated with multifactorially
Herlev Hospital, Copenhagen University adjusted hazard ratios of 1.3 (95% CI, 1.0 to 1.6) for cancer of any type, 2.2 (95% CI, 1.0 to 4.6)
Hospital. for lung cancer, 1.9 (95% CI, 0.8 to 4.6) for colorectal cancer, and 0.7 (95% CI, 0.4 to 1.4) for breast
The funding sources are public or cancer. Corresponding hazard ratios for the highest versus the lowest quintile of baseline CRP
nonprofit organizations and had no role levels were 1.3 (95% CI, 1.0 to 1.6), 2.1 (95% CI, 1.2 to 3.8), 1.7 (95% CI, 0.8 to 3.2), and 0.9 (95%
in the design or conduct of the study, CI, 0.5 to 1.7), respectively. Multifactorially adjusted hazard ratios for early death in patients with
analyzing or interpreting the data, or
cancer were 1.8 (95% CI, 1.2 to 2.7) for CRP more than 3 versus less than 1 mg/L and 1.4 (95%
approving the submitted manuscript.
CI, 1.1 to 1.7) for the highest versus the lowest quintile. Elevated CRP levels were associated with
Authors’ disclosures of potential con-
early death in patients with cancer having localized disease, but not in those with metastases
flicts of interest and author contribu-
tions are found at the end of this
(interaction; P ⫽ .03).
article. Conclusion
Corresponding author: Børge G. Elevated levels of CRP in cancer-free individuals are associated with increased risk of cancer of any
Nordestgaard, MD, DMSc, Department type, of lung cancer, and possibly of colorectal cancer. Moreover, elevated levels of baseline
of Clinical Biochemistry, Herlev Hospi- CRP associate with early death after a diagnosis of any cancer, particularly in patients
tal, Copenhagen University Hospital, without metastases.
Herlev Ringvej 75, DK-2730 Herlev,
Denmark; e-mail: [email protected].
J Clin Oncol 27:2217-2224. © 2009 by American Society of Clinical Oncology
The Appendix is included in the
full-text version of this article,
available online at www.jco.org. is also associated with incident cancer.4-16 A limita-
It is not included in the PDF version INTRODUCTION
tion of the prospective studies is the use of a single
(via Adobe® Reader®).
C-reactive protein (CRP) is an acute-phase reactant CRP measurement at baseline, because lack of
© 2009 by American Society of Clinical
that is elevated during bacterial infection, inflamma- additional measurements precludes adjustment
Oncology
tory disease, trauma, myocardial infarction, surgery, for regression dilution bias17 and thereby tends
0732-183X/09/2713-2217/$20.00
and cancer1; CRP is produced in the liver in response to underestimate any association. Elevated levels
DOI: 10.1200/JCO.2008.19.8440
to elevated cytokine levels after an inflammatory of plasma CRP have also been associated with
stimulus.2 Two hypotheses have been proposed to poor survival in patients with cancer7,18-22; how-
explain the relationship between elevated CRP levels ever, in only two of these studies including few
and cancer.3 The first hypothesis states that elevated participants were CRP measured with a high-
CRP levels are a result of an underlying cancer or a sensitivity assay,7,19 allowing examination of asso-
premalignant state, whereas the second hypothesis ciations in the high-normal interval just greater
states that chronic inflammation and elevated CRP than 1 mg/L. Thus it is still unclear whether, and
might have a causal role in carcinogenesis. to which extent, CRP levels are associated with
Case-control studies have reported higher lev- incident cancer as well as early death in patients
els of CRP in patients with cancer compared with with cancer.
controls,3 whereas results from prospective studies We tested the hypothesis that baseline plasma
are conflicting, with some studies suggesting that levels of CRP in the general population are associ-
CRP is not merely a marker of prevalent cancer, but ated with risk of incident cancer, including the three

© 2009 by American Society of Clinical Oncology 2217

Downloaded from ascopubs.org by 157.46.111.191 on March 25, 2023 from 157.046.111.191

Copyright © 2023 American Society of Clinical Oncology. All rights reserved.
 Allin, Bojesen, and Nordestgaard

most common cancers (ie, lung, colorectal, and breast cancer), and we formation about covariates. Furthermore, we excluded participants with a
corrected for regression dilution bias. In addition, we tested the hy- diagnosis of any cancer (n ⫽ 772), lung cancer (n ⫽ 54), colorectal cancer
pothesis that baseline plasma levels of CRP are associated with early (n ⫽ 93), female breast cancer (n ⫽ 164), or other cancer (n ⫽ 581) before
study entry, from the analysis of the respective cancer subtypes. Follow-up
death in patients with cancer with any type of cancer.
time for each participant began at entry into the study (1991 to 1994 or
2001 to 2003) and ended at occurrence of cancer, death, emigration, or July
2007, whichever came first. During the study period, we had 100% follow-
PATIENTS AND METHODS up. The median and maximum follow-up periods were 13 and 16 years,
respectively (Appendix Fig A1, online only).
Diagnoses of invasive cancer from 1947 until July 2007 were obtained
Study Design from the Danish Cancer Registry,25,26 which identifies 97.8% of all cancers in
The Copenhagen City Heart Study is a prospective cohort study of the Denmark,27 as well as from the national Danish Patient Registry (Appendix
Danish general population, in which participants 20 years of age or older were Table A1, online only).
drawn randomly from the Copenhagen Central Person Register.23,24 It was Individuals with a diagnosis of cancer during the period of follow-up
initiated in 1976 to 1978, with follow-up examinations in 1981 to 1983, 1991 to (n ⫽ 1624) were observed from their date of diagnosis until death, emigration,
1994, and 2001 to 2003. Participants in the present study are from the 1991 to or July 2007, whichever came first. The median and maximum follow-up
1994 and/or 2001 to 2003 examination (additional information on patients periods were 2 and 15 years, respectively. Information about mortality was
and methods is described in the Appendix, online only). obtained from the national Danish Civil Registration System, which is
For the analysis of plasma CRP levels and incident cancer, we in- 100% complete.
cluded 10,520 individuals who had plasma CRP measured at the 1991 to
1994 and/or 2001 to 2003 examination. We excluded participants with Ethics
liver cirrhosis diagnosed before or during the study period from analysis The ethical committee of Copenhagen and Frederiksberg, Denmark,
(n ⫽ 112) because their ability to produce CRP must be expected to be approved the study (KF100.2039/91 and KF01-144/01). All participants gave
impaired, and 233 participants were excluded because of incomplete in- written informed consent.

Table 1. Baseline Characteristics of Participants in the Copenhagen City Heart Study 1991 to 1994 and/or 2001 to 2003 Examination
According to Plasma Levels of CRP at Study Entry
Plasma Levels of CRP, mg/L

⬍1 1-3 ⬎3
(n ⫽1,348) (n ⫽5,852) (n ⫽2,405)

Characteristic No. % No. % No. % P

Sex, women 720 53.4 3,271 55.9 1,378 57.3 .07
Age at entry, years ⬍ .001
Median 44 58 63
Interquartile range 30-58 44-69 53-71
No. of cigarettes smoked per day ⬍ .001
Median 0 0 3
Interquartile range 0-9 0-15 0-18
Smoking ⬍ .001
Never 424 31.7 1,491 25.5 542 22.6
Former 442 33.0 1,675 28.7 618 25.8
Current 473 35.4 2,682 45.9 1,241 51.7
Alcohol consumption, g/wk .02
ⱕ 168/252ⴱ 1,184 87.8 5,004 85.5 2,027 84.3
⬎ 168/252ⴱ 167 12.4 848 14.5 378 15.7
Body mass index, kg/m2 ⬍ .001
⬍ 18.5 53 3.9 115 2.0 26 1.1
18.5-24.9 902 67.3 2,969 51.6 850 36.6
25-29.9 327 24.4 2,031 35.3 873 37.6
ⱖ 30.0 58 4.4 643 11.2 574 24.7
Current oral contraceptive
therapy among
premenopausal women 67 15.4 270 27.2 132 48.5 ⬍ .001
Postmenopausal status among
women 477 66.3 2,244 68.6 952 69.1 .39
Current hormone replacement
therapy among
postmenopausal women 55 19.3 456 20.0 268 24.2 .01

NOTE. All covariates (except number of women and age) are adjusted for age. P values were calculated using ␹ tests for categorical variables and Kruskal-Wallis
2

one-way analysis of variance tests for continuous variables.

Abbreviation: CRP, C-reactive protein.
ⴱ
Total of 168 g/wk for women and 252 g/wk for men; these cut-off values are from the recommendation concerning alcohol consumption from the Danish National
Board of Health.

2218 © 2009 by American Society of Clinical Oncology JOURNAL OF CLINICAL ONCOLOGY

Downloaded from ascopubs.org by 157.46.111.191 on March 25, 2023 from 157.046.111.191

Copyright © 2023 American Society of Clinical Oncology. All rights reserved.
 C-Reactive Protein, Cancer Risk, and Prognosis

Measurement of CRP trend tests. To estimate hazard ratios with 95% CIs for incident cancer, we
Plasma levels of CRP were determined with a high-sensitivity assay on used Cox proportional hazards regression. We used age as time scale, thus
8,778 participants from the 1991 to 1994 examination on plasma stored for 12 analyzing age at event using left truncation (ie, delayed entry). Hereby, age
to 15 years and on 5,931 participants from the 2001 to 2003 examination on differences are automatically adjusted for. We used a multifactorial Cox
fresh plasma samples using turbidimetry or nephelometry (Dako, Glostrup, regression model including sex and time-dependent covariates from the
Denmark, or Dade Behring, Deerfield, IL). 1991 to 1994 and 2001 to 2003 examinations: smoking (never, former, cur-
rent), smoking dose (cigarettes per day), alcohol consumption (women, ⱕ 168
Statistical Analysis
We analyzed data using the statistical software package STATA version or ⬎ 168 g per week; men, ⱕ 252 or ⬎ 252 g per week), body mass index
10.0 (StataCorp, College Station, TX). A two-sided P value less than .05 was (⬍ 18.5, 18.5 to 24.9, 25 to 29.9, or ⱖ 30.0 kg/m2), and, for women, also use of
considered statistically significant. A priori, we divided baseline CRP levels into oral contraceptive therapy, postmenopausal status, and use of hormone re-
quintiles as well as into three categories: low (⬍ 1.0 mg/L), average (1.0 to 3.0 placement therapy.
mg/L), or high (⬎ 3.0 mg/L). The CRP groups were coded 1, 2, 3, 4, and 5 To estimate hazard ratios with 95% CIs for early death of cancer, we
corresponding to the quintiles and 1, 2, and 3 corresponding to the categories used Cox proportional hazards regression analyzing time to event. We
for trend tests in log-rank and Cox statistics. used a Cox regression model including age at diagnosis, sex, cancer type
We plotted survival against follow-up time using the Kaplan-Meier (lung, colorectal, breast, or other cancer), cancer stage (localized, regional
method and tested differences between categories of CRP using log-rank metastases, or distant metastases), cancer histology (adenocarcinoma,

P-CRP (mg/L) Participants Events Adjusted multifactorially Cancer cases that occurred
within 2 years excluded
Any cancer

<1 1,331 / 1,311 105 / 85 P = .0 6 P = .5 7

1 to 3 5,754 / 5,615 1,029 / 890

>3 2,318 / 2,222 458 / 362

Lung cancer

<1 1,385 /1,365 11 / 9 P = .008 P = .06

1 to 3 6,194 /6,055 151 /135

>3 2,542 /2,446 93 / 75

Fig 1. Hazard ratios for cancer inci-

Colorectal cancer dence by plasma levels of C-reactive
protein (P-CRP) in categories. Multifacto-
<1 1,379 / 1,359 9/9 P = .10 P = .72 rial adjustment was for age, sex, smoking,
alcohol consumption, body mass index,
1 to 3 6,176 / 6,037 122 / 108 and, for women, also oral contraceptive
therapy, menopausal status, and hormone
>3 2,527 / 2,431 60 / 45 replacement therapy. P values for trend
tests examine whether increasing levels
of CRP are associated with increasing
hazard ratios.
Breast cancer

<1 733 / 724 20 / 18 P = .4 0 P = .2 2

1 to 3 3,418 / 3,343 135 / 120

>3 1,410 / 1,364 52 / 43

Other cancer

<1 1,348 / 1,328 73 / 55 P = .52 P = .95

1 to 3 5,864 / 5,725 699 / 595

>3 2,382 / 2,286 288 / 227

1 3 5 1 3 5
Hazard Ratio (95% confidence interval)

www.jco.org © 2009 by American Society of Clinical Oncology 2219

Downloaded from ascopubs.org by 157.46.111.191 on March 25, 2023 from 157.046.111.191

Copyright © 2023 American Society of Clinical Oncology. All rights reserved.
 Allin, Bojesen, and Nordestgaard

P-CRP (mg/L) Participants Events Adjusted multifactorially Cancer cases that occurred
within 2 years excluded
Quintiles

Any cancer
1. (< 1.10) 1,877 / 1,829 168 / 134 P = .0 2 P = .2 1

2. (1.10 to 1.41) 1,884 / 1,830 311 / 266

3. (1.42 to 1.98) 1,881 / 1,830 361 / 311
4. (1.99 to 3.56) 1,881 / 1,830 389 / 337
5. (≥ 3.57) 1,880 / 1,829 363 / 289
Lung cancer
1. (< 1.11) 2,024 / 1,974 20 / 16 P = .00 1 P = .0 1

2. (1.11 to 1.43) 2,024 / 1,972 36 / 34

3. (1.44 to 2.00) 2,024 / 1,974 57 / 47
4. (2.01 to 3.64) 2,025 / 1,973 66 / 62
5. (≥ 3.65) 2,024 / 1,973 76 / 60
Fig 2. Hazard ratios for cancer inci-
Colorectal cancer
dence by plasma levels of C-reactive
1. (< 1.11) 2,019 / 1,964 17 / 13 P = .33 P = .8 7 protein (P-CRP) in quintiles. Multifactorial
adjustment was for age, sex, smoking,
2. (1.11 to 1.43) 2,014 / 1,971 41 / 39
alcohol consumption, body mass index,
3. (1.44 to 2.00) 2,016 / 1,961 44 / 41 and, for women, also oral contraceptive
therapy, menopausal status, and hormone
4. (2.01 to 3.62) 2,017 / 1,966 44 / 36
replacement therapy. P values for trend
5. (≥ 3.63) 2,016 / 1,965 45 / 33 tests examine whether increasing levels
Breast cancer of CRP are associated with increasing
hazard ratios.
1. (< 1.12) 1,112 / 1,086 30 / 27 P = .7 9 P = .6 7
2. (1.12 to 1.43) 1,110 / 1,086 44 / 37
3. (1.44 to 2.00) 1,115 / 1,087 47 / 42
4. (2.01 to 3.67) 1,111 / 1,086 46 / 40
5. (≥ 3.68) 1,113 / 1,086 40 / 35
Other cancer
1. (< 1.10) 1,919 / 1,867 119 / 91 P = .51 P = .8 7
2. (1.10 to 1.42) 1,920 / 1,868 221 / 184
3. (1.43 to 1.99) 1,917 / 1,869 234 / 202
4. (2.00 to 3.59) 1,918 / 1,867 264 / 225
5. (≥ 3.60) 1,920 / 1,868 222 / 175

1 3 5 1 3 5

Hazard ratio (95% confidence interval)

squamous carcinoma, other carcinoma, or other histology), and time from Any Cancer
blood sampling to diagnosis. Increasing levels of CRP by quintiles were associated with in-
Because approximately 41% of the participants had CRP measured at creasing risk of incident cancer, whereas a similar trend was seen for
both the 1991 to 1994 and 2001 to 2003 examination, we were able to correct
hazard ratios for regression dilution bias with a nonparametric method.17 CRP by categories: P for trend ⫽ .02 and .06 (Figs 1 and 2). Multifac-
Spearman’s rank correlation coefficient between the two CRP measurements torially adjusted hazard ratios were 1.3 (95% CI, 1.0 to 1.6) for CRP
was 0.5. more than 3 versus less than 1 mg/L and 1.3 (95% CI, 1.0 to 1.6) for the
highest versus the lowest quintile.
RESULTS
Lung Cancer
Baseline characteristics of the participants by plasma levels of CRP at Increasing levels of CRP, by categories or quintiles, were associ-
study entry are listed in Table 1. Levels of CRP were correlated with ated with increasing risk of incident lung cancer: P for trend ⫽ .008
several of these characteristics. Median values of CRP were 1.7 mg/L and .001, respectively (Figs 1 and 2). Multifactorially adjusted hazard
among all participants, and 1.9, 2.3, 1.8, and 1.7 mg/L among individ- ratios were 2.2 (95% CI, 1.0 to 4.6) for CRP more than 3 versus less
uals who developed any cancer, lung cancer, colorectal cancer, or than 1 mg/L and 2.1 (95% CI, 1.2 to 3.8) for the highest versus the
breast cancer, respectively, during the period of follow-up. lowest quintile.

2220 © 2009 by American Society of Clinical Oncology JOURNAL OF CLINICAL ONCOLOGY

Downloaded from ascopubs.org by 157.46.111.191 on March 25, 2023 from 157.046.111.191

Copyright © 2023 American Society of Clinical Oncology. All rights reserved.
 C-Reactive Protein, Cancer Risk, and Prognosis

Colorectal Cancer (Figs 1 and 2). Multifactorially adjusted hazard ratios were 0.7 (95%
Levels of CRP, by categories or quintiles, were not associated CI, 0.4 to 1.4) for CRP more than 3 versus less than 1 mg/L and 0.9
statistically significant with risk of incident colorectal cancer: P for (95% CI, 0.5 to 1.7) for the highest versus the lowest quintile. For CRP
trend ⫽ .10 and .33, respectively (Figs 1 and 2). Multifactorially more than 3 versus less than 1 mg/L and for the highest versus the
adjusted hazard ratios were 1.9 (95% CI, 0.8 to 4.6) for CRP lowest quintile, we had 90% statistical power to exclude a hazard
more than 3 versus less than 1 mg/L and 1.7 (95% CI, 0.8 to 3.2) ratio of ⱖ 1.9 and ⱖ 1.9, respectively.
for the highest versus the lowest quintile. For CRP more than 3
versus less than 1 mg/L and for the highest versus the lowest Sensitivity Analyses
quintile, we had 90% statistical power to exclude a hazard ratio of After exclusion of individuals with CRP more than 10 mg/L, as an
ⱖ 2.3 and ⱖ 2.1, respectively. indication of overt inflammation at baseline, the results were similar to
those in Figures 1 and 2. To eliminate an effect of occult cancer on CRP
Breast Cancer levels, we also conducted additional analyses excluding incident cases
Levels of CRP, by categories or quintiles, were not associated with of cancer diagnosed within 2 years after blood sampling, and hazard
risk of incident breast cancer: P for trend ⫽ .40 and .79, respectively ratios were attenuated (Figs 1 and 2). Mean time from blood sampling

P-CRP Participants Events Localized Metastases

(mg/L)

Any cancer

<1 1,269 / 1,246 43 / 20 P = .4 7 P = .0 0 3

1 to 3 5,149 / 5,004 424 / 279

>3 2,028 / 2,009 168 / 149

Lung cancer

<1 1,371 / 1,377 4/3 P = .63 P = .04

1 to 3 5,961 / 6,121 30 / 78

>3 2,028 / 2,493 19 / 44

Fig 3. Hazard ratios for cancer inci-

Colorectal cancer dence by plasma levels of C-reactive
protein (P-CRP) stratified for cancer stage.
<1 1,368 / 1,371 6/1 P = .83 P = .02 Multifactorial adjustment was for age,
70 sex, smoking, alcohol consumption, body
1 to 3 5,970 / 6,094 48 / 40 mass index, and, for women, also oral
118 contraceptive therapy, menopausal status
>3 2,426 / 2,489 20 / 22 and hormone replacement therapy. P val-
ues for trend tests examine whether in-
creasing levels of CRP are associated with
increasing hazard ratios.
Breast cancer

<1 717 / 717 6/4 P = .7 9 P = .9 6

1 to 3 3,303 / 3,314 68 / 31

>3 1,357 / 1,374 21 / 16

Other cancer

<1 1,305 / 1,288 32 / 13 P = .55 P = .14

1 to 3 5,425 / 5,322 310 / 157

>3 2,186 / 2,171 123 / 77

1 3 1 3 6 9 15

Hazard ratio (95% confidence interval)

www.jco.org © 2009 by American Society of Clinical Oncology 2221

Downloaded from ascopubs.org by 157.46.111.191 on March 25, 2023 from 157.046.111.191

Copyright © 2023 American Society of Clinical Oncology. All rights reserved.
 Allin, Bojesen, and Nordestgaard

to a diagnosis of cancer was 6.3 years for all participants and 5.8 years localized cancer, squamous carcinoma, and less than 2 years as well as
for individuals with CRP more than 3 mg/L. more than 8 years from blood sampling to diagnosis, but not in the
To examine whether CRP is associated with cancer of different other stratified groups (Table 2). However, tests of interaction were
stages at diagnosis, we stratified for localized cancer and metastases all nonsignificant, except for cancer stage (P ⫽ .03.), suggesting
(Fig 3). Increasing levels of CRP, by categories, were associated with that elevated CRP levels are associated with early death irrespective
increasing risk of incident any, lung, and colorectal cancer with me- of cancer type and histology and time from blood sampling to
tastases: P for trend ⫽ .003, .04, and .02, respectively. Corresponding diagnosis, but that elevated CRP levels are associated with early
multifactorially adjusted hazard ratios were 2.3 (95% CI, 1.3 to 4.0), death in patients with cancer having localized disease, but not in
4.4 (95% CI, 1.1 to 18), and 10.4 (95% CI, 0.9 to 118), respectively, for those with metastases.
CRP more than 3 versus less than 1 mg/L.
Appendix Table A1 (online only) shows the specific distribution
(including International Classification of Diseases codes) of the full DISCUSSION
spectrum of malignancies that occurred within the cohort. CRP levels
were not associated with nonmelanoma skin cancer (n ⫽ 297), pros-
tate cancer (n ⫽ 106), cancer of the bladder and excretory urinary tract In this prospective cohort study that included approximately 10,000
(n ⫽ 100), melanoma (n ⫽ 60), pancreas cancer (n ⫽ 55), or cancer of individuals observed for up to 16 years, we found that increasing levels
the ovary and female genital organs (n ⫽ 54). Cancer types with fewer of CRP were associated with increasing risk of incident cancer of any
than 50 incident cases were not examined as a result of limited statis- type, of lung cancer, and possibly of colorectal cancer, but CRP levels
tical power. were not associated with breast cancer. Furthermore, we found that
elevated levels of CRP were associated with early death after a
Survival After a Diagnosis of Cancer diagnosis of cancer.
Survival after a diagnosis of cancer, unadjusted for potential What is the biologic underlying mechanism of the association
confounders, decreased with increasing levels of plasma CRP at study between levels of CRP and risk of cancer? Several findings support the
entry: log-rank trend test P ⬍ .001 (Fig 4). Multifactorially adjusted hypothesis that elevated levels of CRP are a marker of occult cancer.
hazard ratios were 1.8 (95% CI, 1.2 to 2.7) for CRP more than 3 versus First, tumor growth can cause tissue inflammation around the tumor
less than 1 mg/L and 1.4 (95% CI, 1.1 to 1.7) for the highest versus the and thus increase plasma levels of CRP.3 Second, tumor cells are
lowest quintile: P for trend ⫽ .002 and .002. known to produce various cytokines and chemokines that attract
When we stratified for cancer type, stage, and histology, and for leukocytes, and some cancerous cells have been shown to express CRP
time from blood sampling to diagnosis, elevated levels of CRP were and secrete interleukin-6 and interleukin-8, which stimulate CRP
associated with early death separately in patients with other cancer, production in the liver.3,28 Finally, it is possible that CRP is a part of a
host immune response to the tumor.3 However, there is increasing
evidence that chronic inflammation, of which CRP is a marker, is a
causal factor in several malignancies.28,29 It is evident that inflam-
100 Log-rank trend, P < .001
matory cells act as tumor promoters, producing an attractive envi-
ronment for tumor growth, inducing DNA damage, promoting
angiogenesis, and favoring neoplastic spread and metastasis.28 In the
80 present association study, we cannot distinguish between elevated
Overall Survival (%)

CRP being a marker of occult cancer or inflammation and elevated

CRP being causative in carcinogenesis. However, mean time from
60
CRP, mg/L blood sampling to diagnosis of cancer was 5.8 years in the CRP cate-
<1 gory more than 3 mg/L, suggesting that elevated levels of CRP might
40 be more than just a marker of occult cancer. Conversely, when we
1 to 3 excluded incident cancer cases diagnosed within 2 years of blood
>3 sampling, the association between CRP levels and incident any cancer
20 was attenuated. Also in support of CRP being a marker of occult
cancer, we found that elevated CRP levels were associated with any,
lung, and colorectal cancer with metastases at diagnosis, but not with
0 2 4 6 8 10 localized cancer only.
Thus far, seven prospective nested case-control studies and five
Time Since Diagnosis (years) prospective cohort studies have been published, but the results are
No. at risk
hsCRP, mg/L
<1 106 71 49 28 17 9 inconsistent.4-13,15,16 A recent meta-analysis found an odds ratio of
1 to 3 1,050 565 417 288 193 119 1.10 (95% CI, 1.02 to 1.18) for any cancer and of 1.32 (95% CI, 1.08 to
>3 468 218 162 107 69 41
1.61) for lung cancer for a log unit increase in CRP level.6 Despite the
Fig 4. Survival after a diagnosis of cancer by levels of C-reactive protein (CRP). considerable heterogeneity in the meta-analysis (I2 ⬎ 70%), these
P value for log-rank trend test examines whether increasing levels of CRP are findings are in support of our findings of baseline CRP levels being
associated with decreasing survival. Results are unadjusted. Multifactorially
adjusted hazard ratios for early death in patients with cancer by levels of CRP are associated with risk of any cancer and lung cancer. In support of our
presented in Table 2. hs, high-sensitivity assay. finding of an increasing risk of lung cancer with increasing levels of

2222 © 2009 by American Society of Clinical Oncology JOURNAL OF CLINICAL ONCOLOGY

Downloaded from ascopubs.org by 157.46.111.191 on March 25, 2023 from 157.046.111.191

Copyright © 2023 American Society of Clinical Oncology. All rights reserved.
 C-Reactive Protein, Cancer Risk, and Prognosis

Table 2. Risk of Early Death in Patients With Cancer From the Copenhagen City Heart Study 1991 to 1994 and/or 2001 to 2003 Examination According
to Plasma Levels of CRP at Study Entry
Plasma Levels of CRP, mg/L

⬍1 1-3 ⬎3

Stratification No. of Patients No. of Events P, Interaction HR HR 95% CI HR 95% CI P, Trend

None, all cancer 1,624 998 1.0 1.5 1.0 to 2.2 1.8 1.2 to 2.7 .002
Cancer type
Lung cancer 230 209 .83 1.0 1.9 0.7 to 4.9 2.1 0.8 to 5.4 .26
Colorectal cancer 173 114 1.0 1.3 0.3 to 5.3 1.2 0.3 to 5.1 .92
Breast cancer 202 65 1.0 0.6 0.1 to 3.5 1.1 0.2 to 6.7 .17
Other cancer 1,019 610 1.0 1.4 0.9 to 2.3 1.8 1.1 to 2.9 .006
Cancer stage
Localized cancer 647 317 .03 1.0 2.2 1.1 to 4.5 2.5 1.2 to 5.3 .03
Metastases 455 400 1.0 0.7 0.4 to 1.3 0.8 0.4 to 1.4 .95
Cancer histology
Adenocarcinoma 513 352 .77 1.0 1.2 0.6 to 2.3 1.4 0.7 to 2.8 .15
Squamous carcinomaⴱ 327 150 1.0 2.3 0.8 to 6.8 3.5 1.1 to 10.8 .009
Other carcinoma 192 151 1.0 1.3 0.3 to 5.6 1.6 0.4 to 6.8 .33
Other histology 129 92 1.0 1.0 0.4 to 2.9 1.3 0.4 to 3.8 .49
Time from blood sampling to
diagnosis, years
⬍2 262 202 .61 1.0 2.1 0.9 to 4.9 2.6 1.1 to 6.2 .04
2-8 785 499 1.0 1.2 0.7 to 2.1 1.5 0.8 to 2.7 .06
⬎8 577 297 1.0 1.9 0.9 to 4.1 2.4 1.1 to 5.3 .02

NOTE. Participants with a diagnosis of any cancer before study entry were excluded from the analyses. During follow-up, 1,624 participants developed cancer, and
of these, 998 died. All hazard ratios are multifactorially adjusted for age at diagnosis, sex, cancer type, cancer stage, cancer histology, and time from blood sampling
to diagnosis. P values for trend tests examine whether increasing levels of CRP are associated with decreasing survival. P values for interaction tests examine
whether an interaction exists between the covariate stratified for and levels of CRP in categories. Information about stage and histology was available for 1,102 and
1,161, respectively, of the 1,624 patients with cancer. In the multifactorial Cox regression model, a category including unknown stage and a category including
unknown histology were included.
Abbreviations: CRP, C-reactive protein; HR, hazard ratio.
ⴱ
Exclusion of 29 patients with lung cancer of 327 patients with squamous carcinoma resulted in multifactorially adjusted hazard ratios of 2.1 (95% CI, 0.7 to 6.3)
for CRP 1 to 3 versus ⬍ 1 mg/L and 3.5 (95% CI, 1.1 to 11.2) for CRP ⬎ 3 versus ⬍ 1 mg/L (P for trend ⫽ .007).

CRP, studies have suggested an increased risk of lung cancer in indi- Several studies have reported that CRP can be used as a prognos-
viduals with asthma as well as in individuals with tuberculosis or other tic marker in patients with cancer, but so far only two studies with few
infectious inflammation of the lung.30 participants have measured CRP with a high-sensitivity assay.7,19
An increased risk of colorectal cancer has been observed in pa- Il’yasova et al7 found a hazard ratio of 1.4 (95% CI, 1.1 to 1.8) for early
tients with inflammatory bowel diseases,15,28 suggesting that chronic death of patients with cancer for a log unit increase in CRP level, and
inflammation and elevated CRP levels could be associated with risk of Heikkilä et al19 reported similar estimates. The results of our study,
colorectal cancer. Although we did not find a statistically significant which included far more participants, support these findings and
association between elevated levels of CRP and incident colorectal consequently the value of CRP as a prognostic marker in patients with
cancer, the hazard ratios for colorectal cancer were of a magnitude cancer. Interestingly, our results do not seem to suggest any differen-
comparable to that for any cancer and lung cancer; the number of tial effect on the association between elevated levels of CRP and early
cases was just smaller for colorectal cancer. Furthermore, we found an death according to different cancer types or histology, or according to
association between elevated levels of CRP and risk of colorectal can- time from blood sampling to diagnosis. However, our results suggest
cer with metastases. A recent meta-analysis found an odds ratio of 1.09 that the prognostic value of CRP as a marker of early death is present
(95% CI, 0.98 to 1.21) for a log unit increase in CRP level, whereas primarily in patients with localized cancer without known metastases
another recent meta-analysis found that levels of CRP were weakly at time of diagnosis. Consequently, CRP does not merely seem to be a
associated with an increased risk of colorectal cancer (odds ratio, 1.12; surrogate for known metastases at diagnosis, and CRP might provide
95% CI, 1.01 to 1.25) for a log unit increase in CRP level.6,14 In support prognostic information beyond that provided by cancer type, stage,
of our finding of no association of CRP levels with risk of breast cancer, and histology.
a recent meta-analysis found an odds ratio for breast cancer of 1.10 We studied the value of elevated CRP levels as a risk factor for
(95% CI, 0.97 to 1.26) for a log unit increase in CRP level.6 One could cancer and as a prognostic marker in patients with cancer in a large
suspect that many of the breast cancer cases in the present study were prospective cohort study with up to 16 years of follow-up and had no
screening detected (early stage) and that CRP could be associated with losses to follow-up. In addition, we were able to adjust for regression
late stage at breast cancer diagnosis. However, we did not find an dilution bias, because CRP was measured twice in approximately 41%
association between elevated levels of CRP and incident breast cancer of the participants. Potential limitations of our study include con-
with metastases. founding and selection bias. However, we included several important

www.jco.org © 2009 by American Society of Clinical Oncology 2223

Downloaded from ascopubs.org by 157.46.111.191 on March 25, 2023 from 157.046.111.191

Copyright © 2023 American Society of Clinical Oncology. All rights reserved.
 Allin, Bojesen, and Nordestgaard

potential confounders associated with CRP levels in the Cox regres-

AUTHORS’ DISCLOSURES OF POTENTIAL CONFLICTS
sion model. Despite this, we naturally cannot exclude all possible OF INTEREST
confounders and the possibility that CRP may act as a surrogate for
imperfectly measured risk factors for cancer as a result of not measur-
The author(s) indicated no potential conflicts of interest.
ing exposure during the etiologically relevant time or misspecification
of exposure, like body mass index as a surrogate for body fat. Selection
bias of healthy individuals would draw the results in a direction toward
AUTHOR CONTRIBUTIONS
the null hypothesis and therefore cannot explain our positive results.
Despite adjustment for age at diagnosis, sex, cancer type, cancer stage,
Conception and design: Kristine H. Allin, Stig E. Bojesen,
cancer histology, and time from blood sampling to diagnosis, the
Børge G. Nordestgaard
interpretation of the positive association found between elevated levels Financial support: Børge G. Nordestgaard
of CRP and early death after a cancer diagnosis is diminished by the Provision of study materials or patients: Stig E. Bojesen,
heterogeneity of the cancer diagnoses and lack of information Børge G. Nordestgaard
about treatment. Collection and assembly of data: Kristine H. Allin, Stig E. Bojesen,
In conclusion, we have demonstrated that elevated levels of CRP Børge G. Nordestgaard
in apparently cancer-free individuals are associated with increased risk Data analysis and interpretation: Kristine H. Allin, Stig E. Bojesen,
Børge G. Nordestgaard
of incident cancer of any type, lung cancer, and possibly colorectal Manuscript writing: Kristine H. Allin, Stig E. Bojesen,
cancer. Moreover, we have demonstrated that elevated levels of CRP at Børge G. Nordestgaard
baseline are associated with early death after a diagnosis of cancer, Final approval of manuscript: Kristine H. Allin, Stig E. Bojesen,
particularly in patients without metastases. Børge G. Nordestgaard

11. Siemes C, Visser LE, Coebergh JW, et al: predicts cancer-specific and non-cancer survival in
REFERENCES C-reactive protein levels, variation in the C-reactive patients with cancer. Nutr Cancer 41:64-69, 2001
protein gene, and cancer risk: The Rotterdam Study. 21. Nielsen HJ, Christensen IJ, Sorensen S, et al:
1. Burtis CA, Ashwood ER, Bruns DE: Tietz Text- J Clin Oncol 24:5216-5222, 2006 Preoperative plasma plasminogen activator inhibitor
book of Clinical Chemistry and Molecular Diagnostics 12. Suzuki K, Ito Y, Wakai K, et al: Serum heat type-1 and serum C-reactive protein levels in pa-
(ed 4). St Louis, MO, Elsevier Saunders, 2006 shock protein 70 levels and lung cancer risk: A tients with colorectal cancer. The RANX05 Colorec-
2. Pepys MB, Hirschfield GM: C-reactive protein: case-control study nested in a large cohort study. tal Cancer Study Group. Ann Surg Oncol 7:617-623,
A critical update. J Clin Invest 111:1805-1812, 2003 Cancer Epidemiol Biomarkers Prev 15:1733-1737, 2000
3. Heikkilä K, Ebrahim S, Lawlor DA: A system- 2006 22. Scott HR, McMillan DC, Forrest LM, et al: The
atic review of the association between circulating 13. Trichopoulos D, Psaltopoulou T, Orfanos P, et systemic inflammatory response, weight loss, per-
concentrations of C reactive protein and cancer. al: Plasma C-reactive protein and risk of cancer: A formance status and survival in patients with inop-
J Epidemiol Community Health 61:824-833, 2007 prospective study from Greece. Cancer Epidemiol erable non-small cell lung cancer. Br J Cancer 87:
4. Erlinger TP, Platz EA, Rifai N, et al: C-reactive Biomarkers Prev 15:381-384, 2006 264-267, 2002
protein and the risk of incident colorectal cancer. 14. Tsilidis KK, Branchini C, Guallar E, et al: 23. Nordestgaard BG, Benn M, Schnohr P, et al:
JAMA 291:585-590, 2004 C-reactive protein and colorectal cancer risk: A sys- Nonfasting triglycerides and risk of myocardial in-
5. Gunter MJ, Stolzenberg-Solomon R, Cross
tematic review of prospective studies. Int J Cancer farction, ischemic heart disease, and death in men
AJ, et al: A prospective study of serum C-reactive
123:1133-1140, 2008 and women. JAMA 298:299-308, 2007
protein and colorectal cancer risk in men. Cancer
15. Zhang SM, Buring JE, Lee IM, et al: C-reactive 24. Schnohr P, Jensen JS, Scharling H, et al:
Res 66:2483-2487, 2006
protein levels are not associated with increased risk Coronary heart disease risk factors ranked by impor-
6. Heikkilä K, Harris R, Lowe G, et al: Associations
for colorectal cancer in women. Ann Intern Med tance for the individual and community: A 21 year
of circulating C-reactive protein and interleukin-6 with
142:425-432, 2005 follow-up of 12000 men and women from The
cancer risk: Findings from two prospective cohorts
16. Zhang SM, Lin J, Cook NR, et al: C-reactive Copenhagen City Heart Study. Eur Heart J 23:620-
and a meta-analysis. Cancer Causes Control 20:15-26,
protein and risk of breast cancer. J Natl Cancer Inst 626, 2002
2009
99:890-894, 2007 25. Storm HH: The Danish Cancer Registry, a
7. Il’yasova D, Colbert LH, Harris TB, et al:
Circulating levels of inflammatory markers and can- 17. Clarke R, Shipley M, Lewington S, et al: self-reporting national cancer registration system
cer risk in the health aging and body composition Underestimation of risk associations due to regres- with elements of active data collection. IARC Sci
cohort. Cancer Epidemiol Biomarkers Prev 14:2413- sion dilution in long-term follow-up of prospective Publ 220-236, 1991
2418, 2005 studies. Am J Epidemiol 150:341-353, 1999 26. Storm HH, Michelsen EV, Clemmensen IH, et
8. Ito Y, Suzuki K, Tamakoshi K, et al: Colorectal 18. Al Murri AM, Bartlett JM, Canney PA, et al: al: The Danish Cancer Registry: History, content,
cancer and serum C-reactive protein levels: A case- Evaluation of an inflammation-based prognostic quality and use. Dan Med Bull 44:535-539, 1997
control study nested in the JACC Study. J Epidemiol score (GPS) in patients with metastatic breast can- 27. Storm HH: Completeness of cancer registra-
15:S185-S189, 2005 (suppl 2) cer. Br J Cancer 94:227-230, 2006 tion in Denmark 1943-1966 and efficacy of record
9. Otani T, Iwasaki M, Sasazuki S, et al: Plasma 19. Heikkilä K, Ebrahim S, Rumley A, et al: Asso- linkage procedures. Int J Epidemiol 17:44-49, 1988
C-reactive protein and risk of colorectal cancer in a ciations of circulating C-reactive protein and 28. Coussens LM, Werb Z: Inflammation and can-
nested case-control study: Japan Public Health interleukin-6 with survival in women with and with- cer. Nature 420:860-867, 2002
Center-based prospective study. Cancer Epidemiol out cancer: Findings from the British Women’s 29. Balkwill F, Mantovani A: Inflammation and
Biomarkers Prev 15:690-695, 2006 Heart and Health Study. Cancer Epidemiol Biomar- cancer: Back to Virchow? Lancet 357:539-545, 2001
10. Rifai N, Buring JE, Lee IM, et al: Is C-reactive kers Prev 16:1155-1159, 2007 30. Engels EA: Inflammation in the development
protein specific for vascular disease in women? Ann 20. McMillan DC, Elahi MM, Sattar N, et al: Mea- of lung cancer: Epidemiological evidence. Expert
Intern Med 136:529-533, 2002 surement of the systemic inflammatory response Rev Anticancer Ther 8:605-615, 2008

■ ■ ■

Acknowledgment
We thank Anja Jochumsen for technical assistance.

2224 © 2009 by American Society of Clinical Oncology JOURNAL OF CLINICAL ONCOLOGY

Downloaded from ascopubs.org by 157.46.111.191 on March 25, 2023 from 157.046.111.191

Copyright © 2023 American Society of Clinical Oncology. All rights reserved.