C-Reactive Protein Is an Important Biomarker for Prognosis Tumor Recurrence and Treatment Response in Adult Solid Tumors: A Systematic Review

1Approximately 1,638,910 new cancer diagnoses and about 577,190 deaths occurred in the US in 2012, mostly from solid tumors. Prognostication in cancer can be either subjective or objective. In the former, dependent on clinician skill and experience, it is often inaccurate and usually overly optimistic. Prognostication is an important clinical skill for oncologists. Despite advances in medical technology and biology, it is still an inexact science, even with extensive and expensive investigations. Objective determination of prognosis can be based on a combination of tumor, patient, and environmental factors. The use of biological tumor markers to help prognostication (alone or combined with other parameters) has appeal. An ideal potential tumor marker should have a long half-life, be measured accurately and precisely by a simple and inexpensive blood test. It is also important that it be sensitive to change so that it can be followed over time through serial measurements. A few biologic markers meet these criteria. C-reactive protein (CRP) is one.

2CRP is an acute phase reactant, which reflects tissue injury. The half-life is 19 hours in both health and disease. CRP secretion by hepatocytes appears controlled by interleukin 6 (IL-6). Interleukin-1 (IL-1) and tumor necrosis factor (TNF) also stimulate CRP synthesis. CRP is a stable downstream marker of inflammation, unlike the pro-inflammatory cytokines, which have short half-lives (minutes). In chronic inflammatory diseases, serial CRP levels have been correlated with disease severity, and response to therapy. Many large prospective studies now support the role of CRP in prediction of coronary artery disease, though controversies exist.

3Chronic inflammation has been linked to cancer at tumor initiation, but may also be associated with invasive potential and disease progression. A relationship has been proposed between systemic inflammation and various cancer symptoms. A strong positive correlation between high CRP and high IL-6 levels was shown in advanced pancreatic cancer. Elevated CRP levels have been linked to shorter survival in several common cancers.

4In this paper, we describe the results of a systematic review of the relationship between elevated serum CRP and life expectancy in people with solid tumors. We also examined its role in the prediction of treatment response and risk of tumor recurrence.

5Only articles in English were included. Original reports of any studies of solid malignancies in adults were scrutinized. All study designs were included. The following articles were excluded: all non-English literature, basic research, animal research, all pediatric and hematological malignancies, and studies where prognostic parameters were not assessed, or serum CRP levels not measured. Editorial letters and comments were also excluded. Review papers were consulted, but for discussion purposes only.

6Electronic databases included: PubMed (1966 to December 2012); EMBASE (1988 to 2012); Web of Science (1980 to 2012); SCOPUS (1965 to 2012); and the EBM-Cochrane Central Register of Controlled trials and EBM-Cochrane Database of Systematic Reviews (Up to 2012). The search was repeated at the end of data analysis.

7PubMed search of CRP or related MeSH terms (c—reactive protein/c-reactive) with (AND) neoplasm/neoplasms/cancer in all fields with (AND) prognosis/mortality/survival OR survival rate/treatment outcome/treatment/outcome was done including other databases. Search words/terms were as follows:

8A qualified medical librarian (see Acknowledgements) reviewed the search strategy. The first literature screen (Fig 1) was based on article title. If that was irrelevant, the abstract was also reviewed (by NBB, SS and DW) before an exclusion decision. Abstracts (and when necessary the full text) of the remaining articles were then assessed. The reviewers (NBB, SS, ST and DW) met periodically to discuss reasons of exclusion or inclusion of selected papers. Retained articles were then subjected to quality assessment (S1 Appendix).

9A quality assessment (QA) system was developed. This was based on existing guidelines for observational cohort prognostic studies. The tool combined five criteria:

10Study design

11Patient selection

12Prognostic variables

13Follow-up

14Data analysis

15A score of 0–2 was assigned to each—if a study met the conditions in full (score of 2), partially (score of 1), or not at all (score of 0) (S1 Appendix). The total was expressed as a percentage of the maximum possible score. A score of 80–100% was ranked as an adequate study; ≥50 but <80% as intermediate; articles that scored <50% were considered inadequate and excluded from the review (S2 Appendix).

16Data were extracted using a custom designed extraction sheet (till 2010) and later utilizing the Research Electronic Data Capture (REDCap) forms with same sheet. REDCap is a secure web application to create and manage data. The following information was collected: author name, publication date, study title, study design, quality assessment score and grade, number of patients and controls (when available), cancer type, extent of disease, main outcome, main results related to prognosis or treatment outcomes, CRP cut-off points, assay method, mean CRP value, survival definition, median survival duration, CRP sensitivity and specificity as a predictor of prognosis, treatment outcomes or recurrence, other parameters assessed for prognosis, strongest predictors of prognosis, statistical analysis used, and possible co-morbid contributors to increased CRP levels other than cancer (like infection, chemo- or radiotherapy, surgery). The descriptor term “cutoff” refers to the value an investigator/author used to determine an “elevated” CRP even if that level was within the biochemical reference range.

17There was substantial variation in study design and cancer primary sites. A meta-analysis was therefore inappropriate. For each study article we estimated the minimum sample size necessary to detect a difference at P ≤ 0.05. We used the general rule of n = 10 per variable. The estimated minimum sample size was compared to the actual size of the study. Studies with insufficient sample sizes were considered underpowered. Predictors by multivariate analysis were stratified by relative risk (RR) and statistical significance (p-value):

18When RR<2 or >0.5

19When RR>2 or <0.5

20When RR>5 or <0.2

21When RR>10 or <0.1

22When p<0.05

23When p<0.01

24When p<0.001

25We followed the PRISMA statement (S1 PRISMA Checklist) to design and report our systematic review.

26The search identified three thousand and eighty nine (3089) citations: fourteen hundred sixty-six (1466) in PubMed, eight hundred two (802) in Web of Science, three hundred twenty (320) in SCOPUS, three hundred eleven (311) in EMBASE and one hundred ninety (190) in the Cochrane database. After removal of duplicates, fifteen hundred twenty six (1526) remained. Irrelevant studies were then removed. These included those where survival or prognosis was not an outcome, studies where CRP was not studied as a prognostic marker, animal/cell-line based studies, letters and editorials, and those that did not fit our inclusion criteria. Seven hundred thirty one (731) papers were left. Next, three hundred (300) studies in hematologic malignancies, ten (10) non-English articles and forty two (42) pediatric reports were removed. Subsequent to the quality assessment (QA) of the three hundred seventy nine (379) studies retained, thirty five (ten prospective, twenty five retrospective) were inadequate by QA score and excluded. Then two hundred seventy one (271) research studies and seventy-three (73) review papers remained (Fig 1). Survival and outcome measures differed between studies. As a result, no direct study comparisons were possible.

27Two hundred seventy one original articles constituted the final analysis (Fig 1). Only seven of these scored ≥ 80% in the QA (all were prospective and longitudinal, and with a control group in three). Two hundred sixty four (264) had an intermediate QA score. One hundred twenty nine (129) of the 271 did not describe their patient selection procedures. Examples included whether patients were screened for infections, the timing of CRP measurement in relation to factors that could raise CRP level (like chemo- or radiation therapy), and invasive procedures. The sensitivity and specificity of the predictive prognostic value of CRP were reported in only four studies, two in melanoma, one each in cancer of the esophagus and lung. A power analysis was described in two; CRP was an independent prognostic marker of survival in one but not the other. The reference level of CRP for evaluation of responses varied both for RCC and GI studies.

28Forty five percent (n = 122) of the studies were prospective and 52% (N = 142) of the 271 studies were retrospective; the remaining 3% (N = 7) combined retrospective and prospective design. In the prospective studies, median sample size was 121 (range 15–9605) versus 146 (range 32–9608) in the retrospective. High CRP predicted prognosis in 82% (100/122) of the prospective studies. In 13% (16/122) of prospective studies, this was by univariate analysis only. In 18% (22/122), CRP was not prognostic of survival (Table 1). Only 16% (20 of 122) of the prospective studies had a control group (CG) (Table 1). Overall CRP predicted prognosis in 90% (245/271) of studies; 80% by multivariate analysis (MVA) and 20% by univariate analysis (UVA) (Table 2).

29Fifty one (19%) studies looked at renal cell carcinoma. Of these, CRP was prognostic in 90% (46 of 51). In 12 of 51 (24%), CRP predicted prognosis on univariate analysis only. CRP was not predictive of prognosis in five studies (Table 2, Fig 2).

301.1 Prognosis: Forty-six of the 51 studies in renal cell carcinoma (90%) had prognosis as a primary outcome. In thirty three of the 46, CRP was a strong predictor of survival by multivariate analysis (Table 3). In the other 12 of the 46, CRP predicted prognosis by univariate analysis only. One of these was underpowered; none of the other eleven studied the prognostic value of CRP as a primary outcome measure (Table 3).

311.2. Treatment response: Thirteen of the 51 studies in renal cell carcinoma had treatment response and prognosis as a primary outcome. In 12 of the thirteen, CRP independently predicted both treatment response and prognosis. Six studies investigated CRP and treatment response as a primary outcome. High CRP predicted treatment response in all except one. This study was also underpowered, and the primary objective was not treatment response (Table 3). Treatment responses (ill-defined) were evaluated after resection of localized tumors and after cytokine based therapies (IL-2 infusions, IF-α) in metastatic RCC. Low CRP level was associated with better treatment responses overall in 11 of thirteen studies.

321.3. Tumor recurrence: Six of the 51 renal cell carcinoma studies investigated recurrence and survival as primary outcomes. In three studies, elevated CRP independently predicted both tumor recurrence and prognosis (Table 3). One of these examined CRP kinetics (change in CRP over time) and identified non-normalization of postoperative CRP as a predictor of recurrence.

33Of 90 studies, 48 were in colorectal and 42 in esophageal, gastric or gastroesophageal cancers. In colorectal cancer, high CRP strongly predicted survival in 36 (75%) studies. High CRP was an independent prognostic indicator in most reports, 31 of 36 (65%). Only two were negative. In another ten studies CRP predicted prognosis by univariate analysis only; one of these was underpowered (Table 2, Fig 2). Elevated CRP independently predicted prognosis in thirty five of the 42 (71%) studies in gastroesophageal cancer.

342.1. Prognosis: Most studies (81 of 90) in gastrointestinal malignancies had survival as a study outcome. Eighty percent (65 of 81) investigated CRP and prognosis as the primary outcome. High CRP was an independent predictor of survival in 56% (45 of 81) and a strong predictor in 25% (20 of 81). It was a predictor on univariate analysis only in 20% (16 0f 81). In two studies, CRP predicted prognosis (but not independently of disease stage). Once this was considered, in those two, CRP was not a statistically significant prognostic predictor (Table 4).

352.2. Treatment response or tumor stage: CRP predicted treatment response in six studies (Table 4). It did not predict stage in one study but this was underpowered (Table 4). Treatment responses were evaluated after curative resection followed by adjuvant 5-Flurouracil (5-FU) in localized GI tumors. Responses after neo-adjuvant chemotherapy, chemo-radiotherapy, and IL-2 infusions (with either 5FU or surgery) were observed in advanced tumors. In 4 of the five studies, high CRP level was associated with poorer responses.

362.3. Tumor recurrence: Ten of the 90 GI studies investigated recurrence as a primary outcome. In six of the ten, high CRP independently predicted recurrence. One study did not; it included both retrospective and prospective cohorts. Furthermore, CRP prediction of recurrence was not the main outcome (Table 4). High CRP was a strong predictor of recurrence in the rest of the other studies.

3724 studies (each) investigated CRP and prognosis in pancreatic and lung cancer. CRP predicted prognosis in 23 of 24 (96%) studies in pancreatic cancer, 22 of 24 (92%) in lung cancer, all 10 in hepatocellular carcinoma (HCC), all 5 in melanoma, 4 of 7 (57%) in breast cancer, 12 of 12 (100%) in bladder cancer, 7 of 9 (78%) in prostate cancer and 21 of 24 (88%) others (cervical cancer, ovarian cancer, bone and soft tissue etc.). 14 of 15 (93%) studies of heterogeneous cancers found high CRP to be a predictor of prognosis (Table 2, Fig 2).

38CRP as a prognostic indicator was investigated as the primary outcome in most of these studies. Eighteen percent of all studies (48 of 271) found CRP prognostic only by univariate analysis. The forty eight consisted of 12 in renal cell carcinoma; 10 in colorectal cancer; 6 in gastroesophageal; 7 in pancreas; 2 each in lung and bladder; 2 in heterogeneous groups; 1 in hepatocellular cancer and 6 in others (ovarian, primary bone and soft tissue cancers, oral squamous cell carcinoma, hepatocellular carcinoma and malignant histiocytoma) (S3 Appendix). The median sample size was one hundred fifty five (range 38–9608). Thirty included various disease stages, and another 18 advanced, or metastatic/recurrent disease. One had an adequate quality score, forty three intermediate. 4 were underpowered.

39Overall, CRP was not prognostic in 26 of 271 studies (17 prospective, 9 retrospective) (S4 Appendix). These included 9 in digestive tumors; 5 in renal cell carcinoma; 3 in breast; 2 each in lung and prostate; 1 in pancreas; 1 in heterogeneous and 3 in other cancers patients. Median sample size was one hundred thirty eight (range 31–329). 15 of the 26 included various disease stages. The others were resectable/unresectable or advanced/locally advanced and/or metastatic disease. Although all had intermediate quality scores, three were also underpowered. In most negative studies, CRP as a prognostic indicator was not the primary outcome measure (S4 Appendix).

40CRP was used alone in 6% (15 of 271). Many studies considered more than one parameter for prognostic purpose. Demographic characteristics (age, gender, sex) were included as prognostic parameters in 66% of studies (170/256). Common clinicopathologic parameters included with CRP were: stage (TNM, Dukes, others) 23% (59/256); metastasis (lymph node, liver, others) 17%; performance status (ECOG, KPS, others) 16%; tumor characteristics (histology, site, diameter, size) 16%; WBC 13%. Biochemical parameters used with CRP (specifically in renal cell carcinoma) were: albumin (alone or as hypoalbuminemia), LDH, and interleukins (IL-6, IL-8, IL-2). In digestive tumors common biochemical parameters used were: albumin (alone or hypoalbuminemia), carcinoembryonic antigen (CEA), cancer antigen 19–9 (CA19-9) and interleukins (IL-6, IL-8, IL-2).

41Efforts to improve prognostication in cancer had limited success. The number of cancer prognostic biomarkers validated as clinically useful is small, despite extensive research. Many studies have been underpowered. These studies are also difficult to interpret and compare because of heterogeneous study designs. This has prevented meta-analyses of prognostic biologic markers. We encountered this same difficulty during this systematic review.

42Although thirty four percent of the studies (92 of 271) used an elevated CRP cut-off point of >10mg/L, the rest varied. The cut-off value was not reported at all in twenty one studies, and simply as present/absent, or positive/negative in others. Reported cut-off values extended over a wide range: 0.5ng/ml, 1ng/ml, >94nmol/L, >2mg/L, >5mg/L, >8mg/L, > 11 or 12 mg/L, > 35 or 50 mg/L. This made meaningful study comparisons difficult. We tried to standardize if not, cut-off values, then at least the units used. All the studies (except one of the high sensitivity CRP) used CRP. Most (>90%) of CRP levels were reported either in milligram per liter or milligram per deciliter (mg/L or mg/dL).

43High sensitivity CRP (hs-CRP), tumoral CRP and CRP kinetics have also been utilized for disease progression and prognosis. Increased hs-CRP has been associated with late recurrence in renal cell carcinoma and with increased mortality in breast cancer and in men with lung cancer. Tumoral CRP (increased locally within the tumor) may be superior to serum CRP for prognosis and recurrence. Determined by CRP gene expression, tumoral CRP values are more personalized and rather a target for individualized therapy. CRP kinetics may predict survival, recurrence and clinical course in cancer. Human CRP gene is located on the chromosome 1q21-23, spans 1.9 kb and has two exons. CRP gene polymorphism has been associated with increased cancer risk and worse prognosis, mainly in colorectal cancer.

44Various prognostic scoring systems and instruments have been developed utilizing CRP along with other clinical parameters. Prognostic Inflammatory Nutritional Index, PINI (CRP, Alpha-1 Acid Glycoprotein, albumin and prealbumin); Glasgow Prognostic Scale or Modified Glasgow Prognostic Scale (CRP, albumin); Fuzzy Logic Based Prognostic Score (CRP and albumin); Biomarker Based Score (CRP, albumin, Gamma- Glutamyl Transferase (GGT) and HDL). GPS/mGPS and Fuzzy score only differ by CRP cutoffs. We included studies which utilized the Glasgow Prognostic Score or modified Glasgow Prognostic Score, as identified by the search criteria. We have not included studies which utilized Fuzzy score except for discussion purpose. The dominant biochemical component in both GPS and mGPS is CRP. One study defined mGPS as an Inflammation Based Index (IBI) and utilized it as a validated prognostic index for HCC.

45CRP is a non-specific marker of inflammation. It can be elevated for many reasons: infection, invasive procedures, or medications. Inadequate screening for known non-cancer CRP-modifying factors may have significantly influenced values. In addition, it is accepted that sensitivity, specificity, positive and negative predictive values should be used to validate and compare any test against a gold standard. Only two studies reported this data.

46Inflammatory cells are tumor promoters. They produce an attractive environment for tumor growth, induce DNA damage, promote angiogenesis, and favor neoplastic spread and metastasis, and so may affect prognosis. Several explanations exist for the proposed relationship between inflammation and the natural history of cancer. First, tumor growth itself can cause inflammation of surrounding tissue and increase plasma CRP. Second, tumor cells produce various cytokines and chemokines that attract leukocytes. Some cancer cells express CRP and secrete interleukin-6 and interleukin-8, which stimulate liver CRP production. Studies have also shown that IL-6 blocks p-53 induced apoptosis. CRP-positivity develops a favorable microenvironment for the tumor cells through acute inflammatory cytokine network system maintenance. Finally, CRP may be part of the host tumor immune response. Evidence also suggests a causal role for chronic inflammation in several malignancies.

47Cytokines and their surrogate markers (like CRP and IL-6 receptor) can be elevated both locally and systemically in different solid tumors. In renal cell carcinoma, the imbalance between pro-inflammatory cytokines and their anti-inflammatory counterpart is the therapeutic rationale behind immunotherapy. Colorectal cancer seems linked to chronic inflammation (both local and systemic) from genesis to progression. Similar observations have been made in pancreatic and lung cancers. Those tumors are also highly associated with the cancer anorexia-cachexia syndrome, which itself may in part be due to inflammation.

48The role of CRP as a prognostic marker for cardiovascular risk is widely known. Although studies have included large sample sizes, some skepticism still exists. It is noteworthy that a recently published study of 270,000 hospital patients, showed that high CRP levels not only predicted all-cause mortality (compared to the low/or normal CRP group), but also higher cancer mortality. This study was retrospective and may have suffered from selection bias; those who had CRP measured were sicker and so had a higher risk of death.

49In our review, most studies (over half) which met inclusion and quality criteria were in gastrointestinal and renal cell carcinoma. We were surprised not to see more investigations in lung and pancreatic cancer, since they are often considered clinically to have an inflammatory component. This was perhaps influenced by publication bias and selective reporting, i.e. positive studies published while some negative studies may not even be submitted for publication.

50In our review, CRP appeared to be a valuable prognostic predictor particularly in digestive tumors and renal cell carcinoma. It may also help predict tumor recurrence and treatment response in those diseases. CRP was compared to other clinical and biochemical factors in these tumors. In renal cell carcinoma, grade, TNM staging, albumin and lactate dehydrogenase (LDH), were among the strongest prognostic predictors by multivariate analysis. Age, Dukes’ stage, albumin, carcinoembryonic antigen (CEA) and the Glasgow Prognostic Score were amongst those in gastrointestinal tumors.

51Does CRP add any extra information to these other predictors? CRP can be easily and reliably measured. However, it is a non-specific marker. Levels can rise for numerous reasons independent of the cancer; this also reduces the value of single versus serial CRP measurements. Longitudinal studies of CRP values were largely absent. Since CRP cut-off points differed among studies, and the sensitivity and specificity comparisons with different prognostic variables were unreported, it was impossible to conclude with certainty whether CRP was a better predictor than others. In the negative studies, the role of CRP as a prognostic predictor was not the primary outcome and most were underpowered to detect a difference.

52This review had several limitations. Survival and treatment outcomes in the literature were defined and reported inconsistently. Identification of studies depended on CRP being indexed, so we may have been more likely to identify positive studies. Quality assessment was conducted with no cross-validation. The QA system had been piloted on 10 studies picked randomly before the review. This showed it could distinguish between studies in the three QA categories (see Appendices). No meta-analysis or direct study comparisons were done because of the methodological issues described. For similar reasons side by side study comparisons were not possible.

53Increased CRP level predicted prognosis in most (90%) of the studies in solid tumors which met inclusion and quality criteria identified in this systematic review. More than half of all studies (52%) were in gastrointestinal malignancies or renal cell carcinoma. High CRP predicted prognosis in most reports (90%) in these two tumor groups. In addition CRP predicted prognosis in most reports in other solid tumors, so it may also be a clinically useful predictor in lung, pancreas, hepatocellular, and bladder cancers. CRP appeared to be a valuable (and probably under-recognized) prognostic predictor in these tumors. It may also have a role in determining treatment response, and tumor recurrence. The balance of evidence supports wider (and perhaps routine) use of CRP by oncologists for staging, assessment of tumor response and prognostication in at least these two tumor types. These conclusions and recommendations must be tempered by the intermediate quality of most studies.

54Despite some methodological issues, CRP appears valuable to help predict prognosis and other important clinical outcomes in many solid tumors. Better quality prospective longitudinal studies on the role of CRP as a prognostic indicator are needed to confirm these observations.