Introduction

Due to increasing morbidity and mortality, cancer remains a global and growing, but not uniform, problem1. Despite decades of research, relatively few biomarkers are routinely used in clinics for specific types of cancer (e.g., CA-1252 and PSA3 in ovarian and prostate cancers, respectively). Most patients still have either regional or distant metastatic disease when diagnosed, which always means a complicated therapy and poor prognosis4. There is a demand for reliable and clinically applicable pan-cancer biomarkers to obtain additional prognostic information.

Approximately a quarter of cancer patients show correlations with inflammation, and previous studies have claimed that inflammation is a major hallmark of cancer5,6. Several studies have supported the hypothesis that inflammation is closely related to tumor development, progression, and metastatic dissemination, as well as resistance to treatment7. Compared to one of these traditionally recognized methods, prostate-specific antigen testing for prostate cancer3, which assesses systemic inflammation conditions within the tumor by a peripheral blood test during diagnosis or before treatment, is a relatively cheap and convenient method. Fortunately, recent studies have demonstrated that various inflammatory biomarkers, such as the neutrophil-lymphocyte ratio (NLR) and the platelet-lymphocyte ratio (PLR), play significant roles in various cancers8,9,10. C-reactive protein is a representative and routinely measured inflammatory marker, and elevated levels have been associated with treatment outcomes in different malignancies11,12. Furthermore, the Glasgow Prognostic Score (GPS) and the modified Glasgow Prognostic Score (mGPS), which are determined based on the serum levels of C-reactive protein and albumin, have been linked to outcomes of cancer patients13,14. The C-reactive protein to albumin ratio (CAR), a novel inflammation-based prognostic score, is also based on these two factors15. Recently, several studies have revealed that the CAR may be a pan-cancer prognostic marker in many types of cancers, including hepatocellular carcinoma16 and esophageal squamous cell carcinoma17.

Although there is a relationship between high CAR and human cancer, most studies reported thus far have had restricted sample sizes or discrete outcomes. Here, we performed a meta-analysis of data from published studies to comprehensively and quantitatively evaluate its prognostic value in various cancers.

Materials and Methods

This meta-analysis was conducted following the Preferred Reporting Items for Systematic Reviews and Meta-analysis (PRISMA) criteria18.

Literature search and study selection

Two investigators independently searched for eligible studies in PubMed and Web of Science to evaluate the prognostic value of CAR in patients with cancer. Based on the search strategy, which included the following search terms: (ā€œC-reactive protein Albumin ratioā€ or ā€œC-reactive protein to Albumin ratioā€ or ā€œC-reactive protein/Albumin ratioā€ or ā€œCARā€) and (ā€œcancerā€ or ā€œcarcinomaā€) and (ā€œprognosisā€ or ā€œsurvivalā€), we identified these studies until Oct 6, 2016.

Inclusion and exclusion criteria

Studies with the following criteria were included in the meta-analysis: (1) patients with any type of solid cancers were studied; (2) the prognostic value of the pretreatment CAR was evaluated; (3) hazard ratio (HR) for overall survival (OS) was evaluated with multivariate analysis using the Cox proportional hazard model; (4) a definite cutoff value of CAR was given; (5) publications were full-text studies in English. The exclusion criteria were as follows: (1) hematological malignances; (2) letters, reviews, case report or laboratory studies; (3) insufficient information for data extraction; (4) studies had duplicate data or repeat analysis.

Data extraction and quality assessment

The data from all eligible studies were independently reviewed and extracted by two investigators. Each disagreement was assessed until the investigators reached a consensus to guarantee the accuracy of the information extracted. The extracted data from every study included the first author, year of publication, country of origin, total number of cases, cancer type, study type, cut-off value, cut-off selection methods, range of CAR, treatment strategy, stage, follow-ups, age and HRs for OS and disease-free survival (DFS), as well as their 95% confidence intervals (CIs) and p values for the correlation between CAR and prognosis.

The qualities of the included studies were assessed using the Newcastleā€“Ottawa Quality Assessment Scale (NOS)19. The NOS comprised three parameters of quality: selection (0ā€“4 points), comparability (0ā€“2 points), and outcome assessment (0ā€“3 points). The maximum score is 9 points, and NOS scores of ā‰„7 were defined as high-quality studies20. Any disagreement was resolved by discussion.

Statistical analysis

This meta-analysis was performed with STATA version 12.0 (Stata Corp LP, TX, USA) and RevMan software (version 5.3; The Cochrane Collaboration). HRs and their 95% CIs were extracted from each study to calculate pooled HRs. When they were not reported directly in the original study, we estimated the HR through the extracted data from the Kaplan-Meier curve using the methods published by Tierney et al.21. The heterogeneity of the pooled results was measured using Cochranā€™s Q test and Higgins I-squared statistic. Significant heterogeneity was defined as pā€‰<ā€‰0.1 or I2ā€‰>ā€‰50%. The random-effects model (DerSimonian-Laird method)22 was used to analyze the pooled HRs when heterogeneity was significant; otherwise, the fixed-effects model (Mantel Haenszel method)23 was applied. Publication bias was formally investigated by three methods, the Beggā€™s24 and Eggerā€™s tests25, and the ā€œtrim and fillā€ method26. The trim-and-fill method estimates the number of missing studies needed. Subgroup analysis was performed on the basis of cutoff value, cutoff value selection, treatment method, country, sample size, stage and cancer type. The differences between the subgroups were assessed using RevMan software. Sensitivity analysis was used to examine the stability of the pooled results using STATA software. Furthermore, linear regression analysis was performed to evaluate the correlation of the CAR cutoff value and log (CAR cutoff value) with the HR for OS using GraphPad Prism Software 5 (GraphPad Software Inc., San Diego, CA, USA).

Results

The selection process is shown in Fig. 1. We identified 420 relevant studies from the first search strategy. After screening the titles and abstracts, 11 potential studies were selected. Among these studies, Masatsune Shibutani et al.27 evaluated the prognostic significance of CAR with relapse-free survival (RFS) and cancer-specific survival (CSF) instead of OS, while one study failed to obtain a definite HR because the tumor patients in this study were classified into three groups based on two different CAR cutoff values28. After reading these studies, we found another publication which evaluated the prognostic value of CAR in patients with small-cell lung cancer (SCLC)29. Finally, 10 eligible studies were selected that met the inclusion criteria in this meta-analysis.

Figure 1
figure 1

The flow chart of literature selection.

The major characteristics of this meta-analysis are shown in Table 1. These 10 retrospective studies compromised 4592 patients with hepatocellular carcinoma (HCC)16, esophageal squamous cell carcinoma (ESCC)17,30, gastric cancer (GC)31, small-cell lung cancer (SCLC)29, colorectal cancer (CRC)32,33, pancreatic cancer34,35 and nasopharyngeal carcinoma (NPC)36. All studies were published between 2015 and 2016 and were from China (nā€‰=ā€‰7) or Japan (nā€‰=ā€‰3). Based on different treatment methods, the studies were divided into three groups, including with-surgery (nā€‰=ā€‰6) and no-surgery (nā€‰=ā€‰4) treatment. Only two study presented the HR for both DFS and OS, and the HRs and their 95% CIs for OS were directly extracted from the rest of the studies. All studies conducted a multivariable analysis of OS. The quality of 10 studies was evaluated using the Newcastleā€“Ottawa Quality Assessment Scale (NOS) (Table 2).

Table 1 Characteristics of the included studies.
Table 2 Assessment of Study Quality.

The pooled results showed that patients with a high pretreatment CAR had significantly poorer OS than those with low CAR (HR: 2.01, 95% CI: 1.58ā€“2.56, I2ā€‰=ā€‰79%, pā€‰<ā€‰0.001, Fig. 2). For further exploration of the heterogeneity, subgroup analyses were conducted.

Figure 2
figure 2

Forest plot of hazard ratio (HR) for the association of CAR with overall survival (OS) and subgroup analysis by cutoff value.

We performed subgroup analysis of OS based on cutoff value because there was a large range of change between each study. First, we evaluated the correlation of cutoff value and HR for OS using linear regression analysis. The results showed that there was no association between cutoff value and HR for OS (r2ā€‰=ā€‰0.0658, pā€‰=ā€‰0.474) (Fig. 3A). Because we did not know whether the CAR in each study was normally distributed, the correlation of log (cutoff value) and HR for OS was analyzed. Moreover, there was no association between log (cutoff value) and HR for OS (r2ā€‰=ā€‰0.0377, pā€‰=ā€‰0.591) (Fig. 3B). We classified the cutoff values into the lower cutoff group, which had a cutoff value lower than 0.1, and the higher cutoff group, where the cutoff values ranged from 0.4 to 0.7. A combined analysis showed that a higher CAR, which was higher than that of cutoff, was associated with poor OS both in the lower cutoff group (HR: 1.81, 95% CI: 1.40ā€“2.34, pā€‰<ā€‰0.001) and the higher cutoff group (HR: 2.26,95% CI: 1.44ā€“3.56, pā€‰=ā€‰0.004). There was no statistically significant difference between these groups (p for subgroup differenceā€‰=ā€‰0.40) (Fig. 2). In all included studies, 7 studies reported that the cutoff value was selected by the receiver operating characteristic (ROC), and 3 studies selected the cutoff value based on Cutoff Finder, which was a web-based system, R software-engineered, designed by Budczies J et al.37. In cutoff selection, subgroup analysis showed that elevated CAR was positively related to poor HR both in the ROC group (HR: 2.05, 95% CI: 1.64ā€“2.57, pā€‰<ā€‰0.001) and the Cutoff Finder group (HR: 1.89, 95% CI: 1.01ā€“3.51, pā€‰=ā€‰0.046) (p for subgroup differenceā€‰=ā€‰0.80). When different treatment methods were considered, elevated CAR was positively related to poor OS both in the with-surgery group (HR: 2.03, 95% CI: 1.56ā€“2.64, pā€‰<ā€‰0.001) and the no-surgery group (HR: 1.97 95% CI: 1.20ā€“3.23, pā€‰=ā€‰0.008) (p for subgroup differenceā€‰=ā€‰0.92). In the subgroup analyses by country, we found increased CAR predicted a worse OS for Chinese (HR: 1.87, 95% CI: 1.40ā€“2.50, pā€‰<ā€‰0.001) and Japanese (HR: 2.47, 95% CI: 1.70ā€“3.59, pā€‰<ā€‰0.001) patients (p for subgroup differenceā€‰=ā€‰0.25). After stratification by sample size, the pooled HRs were 1.69 (95% CI: 1.35ā€“2.12) for studies with more than 300 cases and 2.69 (95% CI: 1.90ā€“3.81) for studies with less than 300 cases (p for subgroup differenceā€‰=ā€‰0.03). When different stages were considered, the hazard ratios for the effect of CAR on OS were 2.00 (95% CIā€‰=ā€‰1.34ā€“2.97) for the no metastasis group, 2.24 (95% CIā€‰=ā€‰1.45ā€“3.47) for the metastasis group, and 2.00 (95% CIā€‰=ā€‰1.42ā€“2.82) for the mixed group consisting of studies that included patients at all stages. A high CAR for subjects with metastasis was associated with a numerically higher value for the hazard ratio than for subjects with no metastasis, but this difference was not statistically significant (p for subgroup differenceā€‰=ā€‰0.91). Cancer type subgroups were generated by the number of studies on same cancer if at least two studies on that cancer were available, while the remaining studies were pooled in a subgroup termed ā€œothers.ā€ The effect of CAR on OS was significant for all cancer types, and there was no difference between these groups (pā€‰=ā€‰0.39). All results of subgroup analyses are illustrated in Table 3.

Figure 3
figure 3

The correlation of cutoff value and log (cutoff value) of the HR for OS using linear regression analysis.

Table 3 Results of subgroup meta-analysis and meta-regression analysis.

We conducted multivariate meta-regression analysis to explore the possible source of heterogeneity. The results suggested that cutoff value (pā€‰=ā€‰0.451), cutoff value selection (pā€‰=ā€‰0.364), treatment method (pā€‰=ā€‰0.338), country (pā€‰=ā€‰0.154), sample size (pā€‰=ā€‰0.888), stage (pā€‰=ā€‰0.194) and cancer type (pā€‰=ā€‰0.682) did not contribute to the heterogeneity (Table 3). A sensitivity analysis was used to determine whether any study could affect the pooled HRs, and the answer was negative (Fig. 4). Beggā€™s test and Eggerā€™s linear regression test were performed to evaluate the publication bias. Evidence for significant publication bias for OS was not found, as the p value for Beggā€™s test was 0.283, and the p value for Eggerā€™s test was 0.325. As estimated by the trim-and-fill method, no missing studies were required to make the filled funnel plots symmetrical (Fig. 5).

Figure 4
figure 4

Sensitivity analysis of the relationship between CAR and OS.

Figure 5
figure 5

Filled funnel plots for publication bias test of OS.

Discussion

To the best of our knowledge, no meta-analyses assessing the correlation of CAR with the prognosis and survival of patients with various tumors have been performed. In this study, we combined the outcomes of 4,592 patients from 10 available studies, indicating that a high pretreatment CAR was significantly associated with poor OS HR (2.01, 95% CI: 1.58ā€“2.56, pā€‰<ā€‰0.001) in different solid cancers, although there was heterogeneity. Subgroup analyses between CAR and OS were performed, and a high CAR was still a negative maker for worse OS when the patients were segregated according to cutoff value, cutoff value selection, treatment method, country, sample size, stage and cancer type.

When stratified by stage, there was a trend for the association of increased CAR with a worse OS to be greater for patients with metastasis than the patients without metastasis. The reason may be that greater tumor burden resulted in more prolonged chronic inflammation. In a NPC study, patients with Stage III-IV disease had a significantly higher pretreatment CRP/Alb ratio than patients with Stage I-II disease. There was no single study affecting the results in our meta-analysis as determined by the sensitivity analysis. Although three methods of testing the publication bias were performed, and the results showed that there was no significant publication bias in our study, the tests may have false negatives due to the small number of studied included in our study. Taken together, the results suggested that some publication bias was likely to be still present and the actual effect sizes could be smaller than we reported.

A critical problem of our meta-analysis was the large range of cutoff values. To address this problem, we analyzed the correlation between the cutoff value and HR for OS. The results showed there was no relationship between the cutoff value and HR for OS. Next, we performed a subgroup analysis based on a lower cutoff value group and a higher cutoff value group. The results showed that there was no difference between these two groups. Another key point was to analyze the selection of the cutoff value. Of the ten included studies, seven studies reported that the cutoff was selected by the receiver operating characteristic (ROC), and the rest 3 studies were based on Cutoff Finder. Furthermore, cutoff selection subgroup analysis showed that elevated CAR was positively related to poor HR both in the ROC group and Cutoff Finder group.

During the initiation of carcinogenesis, the reactive oxygen species (ROS) and reactive intermediates (RNI) released by inflammatory cells induce DNA damage and genomic instability. During this process, cancer cells often over-express proinflammatory mediators, including proteases, cytokines, and chemokines, which activate many inflammatory signaling pathways. The essential role of the inflammatory microenvironment in tumors has been emphasized over the past decades38,39,40. CRP is synthesized by the liver, secreted into the circulation and extensively influenced by proinflammatory cytokines, such as interleukin-6, interleukin-1, tumor necrotic factor-Ī± and transforming growth factor-Ī²41. It has been shown to be a sensitive prognosis predictor of colorectal cancer42, prostate cancer43 and others. However, inflammation and malnutrition may inhibit the production of albumin44. The serum albumin level at later stages of tumorigenesis could be significantly decreased by tumor necrotic factor increased permeability of the microvasculature and interleukin -Ib and interleukin-6 induced suppressed albumin synthesis, whereas there was no or slight hypoalbuminemia at the beginning of the disease. Therefore, serum albumin is also good indicator of cancer prognosis45,46. Thus, we hypothesized that CAR may be a potential biomarker of outcomes for different cancers.

The CAR was first proposed by Fairclough et al.15 to identify acutely sick patients, and it was later shown to independently predict the mortality of patients with severe sepsis or septic shock47, until Kinoshita et al.16 combined its prognostic value with hepatocellular carcinoma. Previous researchers have demonstrated that GPS, mGPS, NLR and PLR predicted the prognosis of patients with cancer48,49,50. The CAR reflects the ratio of the CRP and albumin levels continuously compared with GPS or mGPS, which may be underestimated or overestimated in some patients. Additionally, CAR showed the highest area under the receiver operating characteristic (AUROC) among these clinical characteristics, such as NLR, CEA, and pathological differentiation in terms of OS32.

There were many limitations in this study that need to be carefully considered. First, the number of included studies was limited, and the pooled results may be less powerful. Second, the heterogeneity among these studies was relatively high and could not be eliminated completely. However, a meta-regression analysis was performed, and we did not find that heterogeneity was caused by cutoff value or other factors included in the analysis. The heterogeneity of the study was probably due to other factors, such as different start time to follow-up (four studies reported that the start date was diagnosis, in one study, the start date was the surgery date, and the others study did not report the start date). We could not analyze all the different factors because only summarized data rather than individual patient data could be used. Third, all of the eligible studies were retrospective due to a lack of a relevant prospective studies. Moreover, Zhen Chenā€™s and Masatsune Shibutaniā€™s studies might contain important parameters for our study. However, we could not include them because of insufficient data. Fourth, publication bias may be present in our meta-analysis because there were no studies with negative results included in this study. Although Eggerā€™s test and the ā€œtrim and fillā€ method were performed to evaluate the publication bias, there may be false negative results due to the limited number of studies. Therefore, whether CAR is a potential prognostic predictor in patients with cancer requires further investigation.

In conclusion, our meta-analysis showed that high CAR was significantly associated with poor OS in patients with cancer. However, further large prospective studies should be conducted to explore the critical role of CAR for survival in cancer patients. If replicated in further large-scale and well-designed studies, our findings will support the clinical use of CAR as a pan-cancer prognostic marker.

Additional Information

How to cite this article: Li, N. et al. Prognostic Role of the Pretreatment C-Reactive Protein/Albumin Ratio in Solid Cancers: A Meta-Analysis. Sci. Rep. 7, 41298; doi: 10.1038/srep41298 (2017).

Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.