Perioperative CRP: A novel inflammation‐based classification in gastric cancer for recurrence and chemotherapy benefit

Abstract Background Perioperative C‐reactive protein (CRP) levels have effects on the prognosis of cancer patients. We intended to determine the prognostic value of combining the two for gastric cancer (GC). Methods Data were extracted from a clinical trial. By calculating the area under the curve (AUC) and the C‐index, the predictive value of CRPs among different time points, including preoperative (pre‐CRP), postoperative days 1, 3, and 5 (post‐CRPs), and postoperative maximum CRP (post‐CRPmax), was derived. Multivariate analysis was performed to further explore the independent variates for recurrence‐free survival (RFS). Results Finally, 401 patients were available in the present study. For RFS, higher AUC (0.692) and concordance index (0.678) of pre‐CRP were observed when compared with those of post‐CRPs. Further, among post‐CRPs, post‐CRPmax had the highest predictive values (AUC: 0.591; concordance index: 0.585) among the other post‐CRPs. The threshold values in predicting RFS for pre‐CRP and post‐CRPmax were 3.1 mg/L and 77.1 mg/L. Multivariate analysis showed both pre‐CRP≥3.1 mg/L (high‐pre‐CRP) and post‐CRPmax≥77.1 mg/L (high‐post‐CRPmax) were risk factors for RFS. Postoperative chemotherapy benefit was further analyzed for patients with stage II/III GC and indicated that patients with pre‐CRP<3.1 mg/L had better prognosis without benefit from postoperative adjuvant chemotherapy (ACT), p = 0.557. In high‐pre‐CRP patients, only patients with post‐CRPmax≥77.1 mg/L but not post‐CRPmax<77.1 mg/L benefited from postoperative ACT (RFS: 33.2% vs 49.9% for non‐chemotherapy group and chemotherapy group, respectively, p = 0.037). Analyses for overall survival obtained the similar outcomes. Conclusions Both high‐pre‐CRP and high‐post‐CRPmax are associated with worse prognosis in GC. ACT seems to only improve the prognosis for stage II/III GC with pre‐CRP≥3.1 mg/L and post‐CRPmax≥77.1 mg/L after radical gastrectomy. Further studies are needed to confirm these findings and explore the potential mechanism.


| INTRODUCTION
Gastric carcinoma (GC) is the fifth most common malignancy worldwide and ranks third in cancer-related mortality 1 and postoperative recurrence remains common for patients after radical gastrectomy, affecting approximately 18% to 45.5% of patients. [2][3][4][5] Postoperative adjuvant chemotherapy (ACT) has been confirmed to improve the recurrence-free survival (RFS) after GC surgery. [6][7][8] However, there is still a lack of simple and practical classifiers that can effectively identify subgroups benefiting from ACT.
Since the correlation between inflammation and cancer was uncovered, 9 increasingly more scholars have carried out studies on inflammatory factors and tumors. 10 A succession of studies confirmed that patients with a high level of preoperative C-reactive protein (CRP) (pre-CRP) were closely related to a poor prognosis for GC, 11,12 indicating that pre-CRP is expected to become an effective forecasting tool alone or in combination with other tumor characteristics or inflammation indices to predict the prognosis of GC. In addition, existing research shows that both pre-CRP and postoperative CRP (post-CRP) levels are associated with the prognoses for cancer patients. 13,14 However, the prognostic value of combining these two variables has not been reported yet, especially for GC.
Therefore, this study used prospective clinical trial data for the first time to explore whether combined pre-CRP and post-CRP levels can be an effective predictor of postoperative recurrence. We also explored their relationship with the efficacy of ACT.

| Patients
From 1 January 2015 to 1 April 2016, 438 patients were recruited to the clinical trial. 15 The details about the inclusion criteria and exclusion criteria, were previously reported 15 and shown in Table S1. The final analysis of the randomized controlled trial (RCT) included 419 patients. The RCT was conducted in accordance with the protocol that was approved by the institutional review boards of Fujian Medical University Union Hospital (FMUUH) (ClinicalTrials.gov number NCT02327481). 15 The present study was also approved by the institutional review boards of FMUUH (IRB number: 2020KY0106). The present study was a subgroup analysis of the previously conducted RCT. 5 After excluding 10 patients with neuroendocrine carcinoma, six patients with palliative surgery and two patients without evidence for GC, the present analysis was restricted to 401 patients with pathologic stage I, II, or III gastric adenocarcinoma (pT1-4aN0-3M0) according to the 7th American Joint Committee on Cancer tumor-node-metastasis (TNM) staging system, as previously described. 16,17 Patients in stage I were excluded from a subset analysis assessing the benefits of ACT.

| Pre-and postoperative crp levels
We obtained preoperative serum CRP values within the 7 days prior to surgery and postoperative days (POD) 1, 3, and 5 from the patients' records. All the patients had CRP levels tested at least once preoperatively and postoperatively. Post-CRP max was defined as the postoperative maximum CRP value from surgery until hospital discharge in the present study.

| Follow-up
The median follow-up time was 42 months (range 3-51 months) in this cohort as of the data cut-off date (April 2019, at least 3 years after enrollment of the last patient). All surviving patients were followed-up at least 3 years. Postoperative follow-ups were routinely performed every 3 months for 2 years and then every 6 months from years 3 to 5. Recurrence was diagnosed based on the radiologic findings or the biopsies with suspicious lesions when possible, as previously described 5 .

K E Y W O R D S
adjuvant chemotherapy, C-reactive protein, gastric cancer, prognosis, recurrence

| Statistical analyses
Categorical variables were compared using the χ 2 or Fisher's exact test and continuous variables by t test. Receiver operating characteristic (ROC) curves were established to estimate the optimal cut-off values for preoperative and postoperative CRP levels as risk factors for recurrence. By calculating the area under the curve (AUC) and the C-index, the discriminative ability of CRPs during different periods was compared. RFS was assessed using the Kaplan-Meier method. The AIC was performed to compare the prognostic value of the different models. 21 . Factors related to RFS were identified by the Cox proportional hazards regression model. To explore that whether the postoperative complication affected the relationship between postoperative CRP max and RFS, we included in our primary Cox model an additional interaction term 22 between postoperative complication and postoperative CRP max ( Table 2). Decision curve analysis was also performed, which could evaluate the clinical usefulness of a prediction model by calculating its net benefit using the rate of true and false positives in varied risk thresholds for screening. 23 All the statistical analyses were performed using SPSS v.18.0 for Windows (SPSS Inc.) and R (https://www.r-proje ct.org/). Table 1 showed the baseline characteristics of the cohorts. A total of 109 (27.2%) patients experienced recurrence after radical gastrectomy. The CRP levels for the recurrence patients were higher than those without recurrence at all time points, but significant differences were only observed for preoperative, POD5, and post-CRP max (all p < 0.05). Further, patients with recurrence were associated with poor clinical features, and all the p-values were less than 0.05. The median OS and RFS for the whole cohorts have not been reached. The 3-year RFS and OS rate were 72.9% (95%CI: 68.6%-77.2%) and 77.3% (95%CI: 73.2%-81.4%), respectively ( Figure S1A,B). Figure S2 shows the CRP values at different time points before and after surgery. Figure 1A-E shows the ROC curves for five different CRPs, indicating that pre-CRP had the highest AUC (0.692) compared with CRPs at different time periods (all p < 0.05). Similarly, pre-CRP had the highest Concordance index (0.678). In addition, among the post-CRPs, post-CRP max had the highest AUC (0.591) and Concordance index (0.585) (Table S2). Therefore, post-CRP max was selected to represent the post-CRPs for subsequent analyses. The optimal cut-off values for pre-CRP and post-CRP max for RFS were 3.1 and 77.1 mg/L, respectively. According to these cut-off values, patients were defined by the following categories: low-pre-CRP patients with pre-CRP<3.1 mg/L, high-pre-CRP patients with pre-CRP≥3.1 mg/L, low-post-CRP max patients with post-CRP max <77.1 mg/L, and highpost-CRP max patients with post-CRP max ≥77.1 mg/L. Table  S3 showed the association among complication status, preoperative CRP level, and postoperative CRP max level. It was found that patients who experienced postoperative complication were more likely to be with high-post-CRP max status.

| Prognosis according to pre-CRP and post-CRP max status
Overall, the prognoses of the high-pre-CRP group and the highpost-CRP max group were significantly worse than the low-pre-CRP group or low-post-CRP max group (3-year RFS: 55.7% vs 87.8%, p < 0.001; 69.6% vs 79.8%, p = 0.041, respectively) ( Figure S3A and S3B). In order to eliminate the potential effect of postoperative complication on prognosis, multivariate analysis which included postoperative complication showed that both pre-CRP≥3.1 (HR: 1.728, 95% CI: 1.076-2.733, p = 0.024) and post-CRP max ≥77.1 (HR: 1.631, 95% CI: 1.019-2.611, p = 0.041) were independent risk factors of postoperative recurrence, and postoperative complication was not associated with postoperative recurrence. In addition, the interaction between postoperative complication and post-CRP max was not significant (p = 0.369), suggesting that postoperative complication did not influence the relationship between post-CRP max and RFS (Table 2). Furthermore, in a separate analysis of each clinicopathological factor, the prognostic value of the pre-CRP and post-CRP max results was consistent (Figure 2A,B).
Since both pre-CRP and post-CRP max were closely associated with the prognosis, further survival analyses were evaluated according to combined pre-CRP and post-CRP max status. As shown in Figure 3A, favorable prognosis was observed in the low-pre-CRP patients regardless of post-CRPmax status (3-year RFS: 90.0% [low-post-CRP max ] vs 86.9% [high-post-CRP max ], p = 0.541). In the high-pre-CRP group, however, low-post-CRP max was related to a better prognosis than high-post-CRP max (RFS: 68.5% vs 50.5%, p = 0.033).

| Incorporation of pre-CRP and post-CRP max levels into PTNM stage
By combining the pre-CRP, post-CRP max , and pTNM stage, a novel predictive model was established for RFS (model A).

| Clinical utility of the predictive model consisting of PTNM stage and pre-CRP and post-CRP max levels
When we compared the net benefit of model A (comprising pre-CRP, post-CRP max , and pTNM stage) with model B, it was observed that, in a wide range of threshold probabilities (33% ~83%), the clinical net benefit of the former was greater than the latter. ( Figure S4).  Figure 3B, Figure S5A). Because the prognosis of high-pre-CRP GC differed by post-CRP max status, we evaluated whether the benefit of ACT also differed according to post-CRP max status in high-pre-CRP GC. As Figure 3C and Figure S5B show, a significant improved prognosis from ACT was observed when compared with surgery only in the high-post-CRP max subgroup of highpre-CRP GC (3-year RFS: 50.1% vs 30.8% in chemotherapy and surgery only groups, respectively; p = 0.014; 3-year OS: 47.0% vs 23.1% in chemotherapy and surgery only groups, respectively; p = 0.003). However, there was no significant 3-year OS: 63.8% vs 50.8% in chemotherapy and surgery only groups, respectively, p = 0.636). Therefore, we established a suggested treatment algorithm for stage II/III GC after R0 resection according to pre-CRP and post-CRP max statuses ( Figure S6).

| DISCUSSION
We present the post hoc analyses of a clinical trial and showed that both pre-CRP and post-CRP max were closely associated with the recurrence for GC after R0 resection, and the predictive value of pre-CRP was significantly greater than post-CRPs. Compared with pTNM stage alone, incorporation of pre-CRP and post-CRP max levels with pTNM stage significantly improved the predictive ability and clinical utility of the predictive model. Furthermore, our analyses for stage II/III GC showed that patients with pre-CRP<3.1 had good prognosis and received unclear benefits from ACT. However, ACT should be focused on patients with pre-CRP≥3.1.
Since the correlation between inflammation and cancer was uncovered, 9 increasingly more evidence has shown that tumor progression is not only associated with the intrinsic properties of the cancer cells but is also related to the inflammatory immune response. 24 Several studies confirmed that inflammatory indices, such as neutrophil-lymphocyte ratio, CRP-albumin ratio, and CRP-prealbumin ratio, were closely associated with the prognosis of GC. [25][26][27] However, study focusing on the prognostic value of combined pre-CRP and post-CRP for GC recurrence has not been reported.
CRP is mainly produced by liver cells and, to a lesser degree, by kidney, monocytes, and neutrophils. 28 The rapid increase in serum concentration is associated with IL-6, TNF-α, and other proinflammatory cytokines. 29 These proinflammatory cytokines accelerate angiogenesis, which in turn enhances the progression and metastasis of malignant tumors. 11,30 Cancer cells could also produce several cytokines and chemokines, thus leading to inflammatory cell infiltration into the tumor microenvironment (TME) and increasing the serum CRP concentration. 31 Therefore, pre-CRP level can reflect the TME status and its relationship with host immunity. In addition, CRP binds to the surface of apoptotic cells and activates the classic complement pathway, enhancing opsonization and phagocytosis of CRP-tagged targets. 32,33 CRP can also recruit C4b-binding protein, the main inhibitor of the classic complement pathway, and regulates the activity of immune cells, such as macrophages, neutrophils, and monocytes. 28 Therefore, CRP may also be considered a regulator of innate immunity, rather than merely an indicator of inflammation.
Recently, serum CRP level has been shown to be closely related to the prognosis of a variety of malignant tumors, including breast cancer, colorectal cancer, and thymic epithelial tumors. [34][35][36][37] However, most study only focused on CRP levels before treatment. Few reports explored the relationship between posttreatment CRP levels and long-term outcomes in patients with cancer. Pastorino et al. found for the first time T A B L E 2 Univariate and multivariate analyses of factors associated with recurrence-free survival that baseline and postoperative CRP levels were closely related to prognosis in patients with resectable lung cancer. 13 At present, no study reported the influence of perioperative CRP levels on the prognosis of patients with GC. For the first time, the present study demonstrated that both pre-CRP level and post-CRP max were closely related to postoperative recurrence of GC by multivariate analysis. By constructing ROC curves and calculating the AUC and C-index, we found an interesting phenomenon that pre-CRP had a significantly higher predictive value than post-CRPmax for GC recurrence. The reason for this finding may be that post-CRP levels could be affected by surgical stress and complications. 38,39 As a result, although post-CRP max has independent prognostic value for prognosis, there could be additional confounding factors beyond pre-CRP. In addition, previous study showed that postoperative complication had a negative effect on recurrence and prognosis for GC patients. 40,41 However, in the present study, although the postoperative complication is related to the high-post-CRP max , it failed to affect the postoperative recurrence. Further interaction analysis suggested that postoperative complication did not influence the relationship between post-CRP max and RFS. Research conducted by Saito showed the similar results. 42 These findings suggested that the post-CRP max level in the early postoperative phase may be helpful for the postoperative complication and for the prediction of long-term prognosis. Further study is needed to validate the results.

F I G U R E 2 The relationship between (A) pre-CRP, (B) post-CRP max status and 3-year RFS in various subgroups
To further evaluate the predictive value of pre-CRP and post-CRP for the prognosis of gastric cancer (GC), we constructed a predictive model, model A, including pre-CRP, post-CRP max , and pTNM stage. The C-index was significantly higher, and the AIC value was lower than model B (pTNM stage only). Since the decision curve analysis was proposed in 2006, 23 it has been widely accepted by scholars to calculate net benefits and evaluate the clinical utility of predictive models, including the utility of models for cancer patients. [43][44][45][46] The DCA further confirmed the clinical utility of model A.
Numerous studies have demonstrated that ACT can significantly improve the RFS 6-8 for GC. However, it is also essential for doctors to identify patients who could benefit from ACT. At present, several scholars have carried out relevant research. Sohn found that GC with chromosomal instability molecular subtypes could benefit from ACT. 47 Cheong et al. constructed a single patient classifier to screen out GC patients who benefited from ACT 48 Ji and Young et al. found that microsatellite instability was related to the efficacy of ACT. 49,50 Compared with the above results, CRP, an inflammation-based marker, has been routinely tested in clinic, and it is cheaper and easier to obtain. Based on the present findings, we recommend using pre-CRP status and post-CRP max status for risk stratification and deciding on the appropriateness of ACT for patients with stage II/III GC ( Figure S6). For GC patients with low-pre-CRP, regardless of their post-CRP max level, we suspected that ACT after radical gastrectomy might not be an optimal strategy due to this form of GC having a favorable prognosis and an unclear benefit from ACT. However, due to the small sample size, this hypothesis needs to be validated in more larger cohorts in future. For high-pre-CRP patients with high-post-CRP max status, the prognosis was poor, but the benefit of ACT was obvious. Therefore, ACT should be strongly recommended for this subgroup. Although the survival of high-pre-CRP GC with low-post-CRP max was better than that of high-pre-CRP GC with high-post-CRP max , it was still worse than low-pre-CRP GC, while no benefit from ACT was observed.
Therefore, in order to further improve the prognosis of this subgroup, a large sample of clinical trials is warranted for this particular subgroup to evaluate the potential use of different ACT regimens. What is more, according to previous (B) in patients with stage II/III gastric cancer according to pre-CRP status and treatment arm; and (C) in patients with stage II/III gastric cancer according to post-CRP max status and treatment arm for high-pre-CRP gastric cancer patients. ACT indicates adjuvant chemotherapy reports, dynamic changes in inflammatory markers, such as NLR, and tumor markers, such as CA19-9, before and after neoadjuvant chemotherapy are also associated with longterm prognosis of patients with tumors. [51][52][53][54] Similarly, future clinical trials can be conducted to further assess whether high-pre-CRP GC can benefit from neoadjuvant chemotherapy and to assess the predictive value of the dynamic change in CRP levels before and after neoadjuvant chemotherapy, which would assist clinicians in formulating the best treatment options for such patients.
Considering previous evidence that statins or metformin can reduce serum CRP levels, 55,56 our findings open new prospects for ACT in GC after radical gastrectomy. CRP levels could be used to determine the efficacy of anti-inflammatory drugs, such as COX-2 inhibitors, metformin, or statins, 57 in randomized clinical trials to improve the long-term prognosis for GC. Prospective trials could further explore the CRP as a reliable intermediate biomarker to predict the longterm effects of such drug interventions. In addition, we hope to confirm our findings through further prospective clinical trials. Further basic studies will be conducted to determine how CRP participates in the progress of GC and how CRP affects the efficacy of ACT.
Recently, several studies showed that serum amyloid A (SAA) before treatment is associated with the prognosis of patients with solid tumors, including GC. 58,59 Due to the inherent defects of retrospective study, SAA levels were not detected in this cohorts. Therefore, we were unable to explore the relationship between SAA and CRP levels and long-term outcomes in patients with GC. Further large studies are warranted in the future to explore the correlation among SAA and CRP levels and prognosis of patients with GC.
This study had several potential drawbacks. First, it was a monocentric exploratory study without external validation, and the sample size was small. Second, due to the small sample size, we did not analyze that whether the detailed cycles and regimens of ACT altered the relationship between CRP levels and prognosis. Third, the median follow-up of 41 months was relatively short. However, when it occurs, recurrence of GC usually develops within the first 2 years after surgery, 5,60 so it may be sufficient considering the recurrence patterns in the present study. As far as we know, this study is the first to combine pre-CRP and post-CRP levels and investigate their relationship with postoperative recurrence and the efficacy of ACT, which provides a basis and direction for further clinical trials in the future.
In conclusion, through post hoc analysis of a prospective clinical trial, we demonstrated that pre-CRP and post-CRP max statuses were clinically actionable as potential prognostic indicators and could be used as supplements to traditional pTNM staging. In addition, pre-CRP and post-CRP max statuses were helpful to optimize the treatment decision for GC. For stage II/III GC patients with pre-CRP ≥3.1 mg/L and post-CRP max ≥77.1 mg/L, ACT should be strongly recommended. Further multicenter validation studies are warranted.

DISCLOSURE
There are no conflict of interest or financial ties to disclose from any authors.

AUTHOR CONTRIBUTIONS
Jun Lu, Bin-bin Xu, and Ping Li conceived the study, analyzed the data, and drafted the manuscript; Zhen Xue, Jianwei Xie, Chao-hui Zheng, and Chang-ming Huang helped critically revise the manuscript for important intellectual content; Jun Lu, Bin-bin Xu, Chao-hui Zheng, Chang-ming Huang, and Ping Li helped collect data and design the study.

DATA AVAILABILITY STATEMENT
The dataset analyzed for this study is available from the corresponding author on reasonable request.