C-reactive protein and risk of breast cancer: A systematic review and meta-analysis

Associations between elevated C-reactive protein (CRP) and breast cancer risk have been reported for many years, but the results remain controversial. To address this issue, a meta-analysis was therefore conducted. Eligible studies were identified by searching the PubMed and EMBASE up to December 2014. Study-specific risk estimates were combined using a random-effects model. Altogether fifteen cohort and case-control studies were included in this meta-analysis, involving a total of 5,286 breast cancer cases. The combined OR per natural log unit change in CRP for breast cancer was 1.16 (95% CI: 1.06-1.27). There was moderate heterogeneity among studies (I2 = 45.9%). The association was stronger in Asian population (OR = 1.57, 95% CI: 1.25-1.96) compared to European (OR = 1.12, 95% CI: 1.02-1.23) and American (OR = 1.08, 95% CI: 1.01-1.16). Prediagnostic high-sensitivity CRP concentrations (OR = 1.22, 95% CI: 1.10-1.35) was superior to common CRP (OR = 1.08, 95% CI: 1.01-1.15) in predicting breast cancer risk. The meta-analysis indicated that elevated CRP levels was associated with increased risk of breast cancer. Further research effort should be performed to identify whether CRP, as a marker of inflammation, plays a direct role in breast carcinogenesis.

sample size or long-term follow-up was performed thereafter. Therefore, a meta-analysis of cohort studies and case-control studies was conducted to further clarify the association between the elevated levels of CRP and breast cancer risk.
Results of the meta-analysis. CRP and breast cancer. The multivariable-adjusted ORs for each study and all studies combined for one unit change in ln(CRP) were shown in Fig. 2. Among the 15 studies included, two showed an insignificant negative association between one unit change in ln(CRP) and breast cancer, and the other thirteen showed positive association, four of which showed statistical significance. The combined OR per natural log unit change in CRP for breast cancer was 1.16 (95% CI: 1.06-1.27). However, there was moderate heterogeneity observed across studies included (Q-test P heterogeneity = 0.027, I 2 = 45.9%).

Influence analysis of individual studies.
To address the potential bias due to the quality of the included studies, we performed the sensitivity analysis by calculating combined OR again when omitting one study at a time. Fig. 3 showed the results of sensitivity analysis. The combined OR per natural log unit change in CRP ranged from 1.13 (95% CI: 1.05-1.22) to 1.19 (95% CI: 1.09-1.30). The meta-analysis result of the combined OR per natural log unit change in CRP for breast cancer was not significantly affected by omission of any of the 15 individual studies, which meaned that each single study didn't influence the stability of combined OR estimate.

Publication bias.
There was no evidence of publication bias as demonstrated by the non-significant P values for Begg's (0.805) and Egger's tests (0.172) and the near-symmetric funnel plot (Fig. 4).

First author
Year

Discussion
This meta-analysis assessed the association between CRP levels and breast cancer risk. Overall, the result supported a significant positive association between the elevated levels of CRP and an increased risk of breast cancer. The overall estimate indicated an 16% increase in risk of breast cancer for a natural log unit increase in CRP levels. Sensitivity analysis further confirmed the robustness of results.
Our summary estimate of CRP and breast cancer risk in cohort studies was similar to that of another meta-analysis, which included 5 prospective studies with only 1,240 cases and reported a unit increase in ln(CRP) was associated with 10% increase in breast cancer risk. However, the result was not statistically significant and considerable heterogeneity was found (I 2 = 51.0%). In contrast to that study, our meta-analysis enlarged breast cancer cases to 5,286 and the summary risk estimate showed smaller heterogeneity (I 2 = 45.9%).
Results from subgroup analyses showed that geographic region, menstrual status, CRP markers and case diagnosis method might be possible sources of heterogeneity. Despite suffering the limitations of observational nature, several findings from subgroup-analysis deserved notable. A higher combined OR per natural log unit change in CRP was found in participants from Asia, which showed that regional differences might exist between the elevated levels of CRP and an increased risk of breast cancer. Results from subgroup analyses stratified by source of menstrual status showed that the elevated levels of CRP could increase the postmenopausal breast cancer, not the premenopausal breast cancer. As we all know, excess weight and obesity convincingly increase the risk of breast cancer in postmenopausal women 24,25 and are established factors that contribute to chronic inflammation 26 . Despite the strong relationship between CRP and body weight 27,28 , the association between CRP levels and breast cancer risk was unlikely to be confounded by BMI, since four of six studies provided risk estimates that were adjusted for BMI. Besides, Hs-CRP, as an inflammatory biomarker, was superior to common CRP in predicting risk of breast cancer.
The present study has several strengths. First, it included a large sample size (5,286 breast cancer cases). Moreover, more comparable dose-response relationship were created for each study, and subgroup analyses stratified by 7 different variants were conducted, thus the effect of potential confounders was minimized. In addition, the combined OR per natural log unit change in CRP for breast cancer was not significantly affected by omission of any of the 15 individual studies, as well as no publication bias was observed in our analyses, indicating that our results were robust.
However, the present meta-analysis has several limitations. First, studies included in this meta-analysis were heterogeneous, which could be explained by differences in populations, CRP markers, and CRP detection method. To address this issue, the random-effects model meta-analysis was reported to combine data whenever significant heterogeneity was noted. We used appropriate well-motivated inclusion criteria to maximize homogeneity, and performed sensitivity and subgroup analyses to investigate potential sources of heterogeneity. Second, information was limited for the results stratified by menstrual status and BMI categories as not all studies involved here provided relevant information. Finally, a meta-analysis is not able to solve problems with confounding factors that may be inherent in the included studies. Although all the included studies presented here were carefully adjusted for potential confounders, including age, BMI, physical activity, smoking, alcohol consumption, HRT use, nonsteroidal anti-inflammatory drug (NSAID) use, it is possible that the associations of circulating CRP with breast cancer risk have been inflated by residual confounding or reverse causality. Insufficient control for confounding factors can skew the results in either direction, to exaggeration or underestimation of risk estimates. Besides, although it has been demonstrated that CRP levels are relatively stable over short   Table 2. Results of subgroup analyses. Abbreviation: OR, odds ratio; CI, confidence intervals; BMI, body mass index; Hs-CRP, High-sensitivity C-reactive protein; ELISA, enzyme-linked immunosorbent assay. * Refers to cohort study and nested case-control study; ** Refers to case-control study; † I 2 is interpreted as the proportion of total variation across studies that are due to heterogeneity rather than chance.
periods of time and have little or no diurnal variation 29 , CRP levels are easily influenced by a variety of physiological and pathological stimulus, such as acute or chronic infection and use of anti-infectious agents. An alternative way to eliminate reverse causality and to minimize residual confounding would be to investigate the associations of breast cancer with genetic variants known to be associated with circulating CRP. As genetic variants are randomly allocated at conception, such investigations would provide unconfounded and unbiased estimates of any associations of inflammatory markers and any cancer outcomes 30,31 .
In conclusion, the findings of this meta-analysis indicated that elevated CRP levels was associated with increased risk of breast cancer, especially among the Asian population. Although causality evidence was insufficient, these results seemed to support a role of chronic inflammation in breast carcinogenesis. Further studies, especially with high-quality and more breast cancer cases involved cohort studies, are needed to identify whether CRP, as a marker of inflammation, does play a direct role in breast carcinogenesis.

Methods
Literature search strategy. A systematic search up to December of 2014 was conducted in MEDLINE (via PubMed) and Excerpta Medica database (EMBASE) to identify relevant articles. Search terms included "C-reactive protein" or "C reactive protein" or "CRP" combined with "breast cancer". Additional relevant references cited in retrieved articles were also evaluated.
Inclusion and exclusion criteria. All papers were reviewed by two authors independently.
Uncertainties and discrepancies were resolved by consensus after discussing with a senior researcher. All studies included in the final meta-analysis satisfied the following criteria: (a) cohort or case-control study design; (b) report results on blood CRP levels; (c) breast cancer incidence as the outcome of interest; (d)  report RR (or odds ratio [OR] estimates in case-control studies) or hazard ratios (HR) estimates with their corresponding 95% CI (or sufficient data to calculate of these effect measure). If the study was reported in duplication, the one published earlier or provided more detailed information was included. Review articles and editorials were included if they contained original data. Abstracts were excluded. Data extraction. Two of the authors performed the data extraction from each article and discrepancies were resolved by consensus. For studies meeting inclusion criteria, a standardized data extraction form was used to extract the following data: the first author's name, year of publication, country of origin, study design, cohort study name, participants enrolled criteria, period of enrollment, the length of follow-up for cohort study, the number of participants (or person-years) and cancer cases, participants characteristics (gender composition, mean age, mean body mass index [BMI], menstrual status when blood was collected), CRP measurement methods, and RR or OR estimates with corresponding 95% CIs for CRP as a continuous variable or at least 3 categories of CRP levels. For each study, we extracted the risk estimates that were adjusted for the greatest number of potential confounders.
Statistical analysis. The RR or OR per natural log unit change in CRP with 95% CI was used to compute the combined OR of elevated CRP levels and the risk of breast cancer. A fix-effect or random-effect model was used to combine the data, based on the Mantel-Haenszel method 32 and the DerSimonian and Laird method 33 , respectively. These two models provide similar results when between-studies heterogeneity is absent; otherwise, random-effect model is more appropriate. For studies reporting no risk estimate for one unit change in ln(CRP), we used the method proposed by Orsini 34 and Greenland 35 to estimate the ln(RR) or (OR) for one unit increase in ln(CRP).
Cochrane Q test (P < 0.10 indicated a high level of statistical heterogeneity) and I 2 ( values of 25%, 50% and 75% corresponding to low, moderate and high degrees of heterogeneity, respectively) was used to assess the heterogeneity between eligible studies, which test total variation across studies that was attributable to heterogeneity rather than to chance 36 . Subgroup analyses for one unit increase in ln(CRP) and the risk of breast cancer were subsequently carried out by study type, geographical region, menstrual status, BMI categories, CRP markers, CRP assay methodology and case diagnosis method. Sensitivity analysis was also conducted to assess the influence of each individual study on the strength and stability of the meta-analytic results. To show each study's independent impact on the combined effect, only one study in the meta-analysis was excluded each time. Funnel plots and statistical tests (Begg adjusted rank correlation test and Egger regression asymmetry test) for funnel plot asymmetry were performed to test any existing publication bias.
All statistical analyses were performed using STATA version 12 for Windows (StataCorp LP, College Station, TX, USA). A two-tailed P < 0.05 was considered statistically significant.