Prognostic value of neutrophil‐to‐lymphocyte ratio and platelet‐to‐lymphocyte ratio for breast cancer patients: An updated meta‐analysis of 17079 individuals

Abstract Aims This study aimed to evaluate the prognostic effect of neutrophil‐to‐lymphocyte ratio (NLR) and platelet‐to‐lymphocyte ratio (PLR) for patients with breast cancer (BC). Methods A literature search was performed by searching medical databases. Basic characteristics and prognostic data were extracted from included studies. Primary outcomes, such as overall survival (OS) and disease‐free survival (DFS), were synthesized and compared. Subgroup analyses were performed according to pathology, geographical region, cut‐off value, and tumor progression. Results A total of 39 studies comprising 17079 BC patients were included in this meta‐analysis. Among them, 28 studies with 142 64 BC patients investigated predicting role of NLR for OS, showing elevated NLR were associated poor prognosis (hazard ratio [HR]: 1.78, 95% confidence interval [CI]: 1.49‐2.13, P < 0.001). Twenty‐seven studies containing 115 04 patients explored the role of NLR in predicting DFS, showing elevated NLR was associated with poor DFS with HR of 1.60 (95% CI: 1.42‐1.96, P < 0.001). Twelve studies explored the role of PLR in predicting OS, showing patients with higher PLR were associated with a significantly worse prognosis with a pooled HR of 1.32 (95% CI: 1.11‐1.57, P = 0.002). Eleven studies with 5013 patients shown patients with elevated PLR were associated shorter DFS (HR: 1.43, 95% CI: 1.09‐1.86, P = 0.009). Subgroup analyses shown a greater magnitude of association between NLR and OS in triple‐negative BC patients than in HER2‐positive ones. Conclusions Our study suggested that elevated NLR and PLR were associated with poor OS as well as high risk of recurrence for BC patients. Subgroup analyses confirmed the prognostic effect of NLR and PLR in HER2‐positive BC patients. As easily accessible parameters, NLR and PLR should be identified as useful biomarkers in the management of BC.


| Search strategy
A comprehensive literature search of relevant studies was performed through the online medical databases PubMed, Embase, Web of Science, the Cochrane Library and Scopus. Studies focused on the correlation of BC and NLR as well as PLR were taken into retrievement. Search terms were confined to the following main words and Medical Subject Headings terms: "neutrophil", "platelet", "lymphocyte", "neutrophil-to-lymphocyte", "platelet-to-lymphocyte", "breast cancer", and "breast carcinoma". Moreover, by using cross-references from the references of primary selected studies and relevant studies, a backward search was also performed to ensure a comprehensive search. Literatures were restricted to those written in English. There was no restriction on geographical region. Two reviewers (Wanying Guo and Xin Lu) completed the electronic search independently.

| Study inclusion and exclusion criteria
Reviewers screened the eligible studies based upon inclusion and exclusion criteria which were prespecified. The final decision on inclusion and exclusion criteria were approved by all the authors. Any disagreements between the two reviewers were solved and made final decision by the senior reviewer (Miao Deng).
The criteria for inclusion were as following: 1. Articles analyzed BC patients. 2. Articles evaluated prognostic value of NLR or PLR. 3. Articles assessed the OS or DFS of BC patients.
The criteria for exclusion were as following: 1. Articles not focused on the prognosis of BC patients. 2. Articles concerned on neither pretreatment NLR nor PLR. 3. Reviews or editorials. 4. Case reports or conference abstracts. 5. Articles without data of interest (OS or DFS).
Articles with duplicate sample set, such as those published by same authors or departments, were picked out for further screening. In this case, only the articles with largest sample size or those published most recently were finally included in the present metaanalysis. However, for those studies analyzed two or more independent sample sets such as training and validation cohorts, the cohorts were enrolled in this study and analyzed independently.

| Data management and statistical analysis
The software of Endnote (version X8) was used for preliminary screening and sorting. Data were extracted from the enrolled literatures by two authors (Wanying Guo and Xin Lu) after reading full text intensively. The baseline information included full list of authors as well as affiliations, year of publication, geographical region, research centers, models of use, sample size, cut-off values for NLR and PLR, mean or median ages, proportion of triple negative patients, indications for surgical treatment, follow-up time, and treatment strategies. The hazard ratios (HRs) with 95% confidential intervals (CIs) were directly extracted from the tables or texts. However, in several studies, the HRs and 95% CIs were not shown directly. In such cases, the software of Engauge Digitizer (version 4.1) was used to extract HRs and 95% CIs by computing the Kaplan-Meier graph. 11,12 The primary outcomes were pooled using the Cochrane Collaboration's Review Manager (version 5.3, Cochrane Collaboration, Oxford, UK). 13 Random effect model was applied routinely only if no obvious heterogeneity was observed among the included literatures (I 2 < 40%). Heterogeneity within studies was explored by using the chi-square test with a P value of 0.10 for significance. Moreover, the heterogeneities were quantified using the I 2 statistics. Sensitivity analyses of main outcomes were conducted by using the software of Stata (version 12.0). 14 The publication bias was investigated using funnel plots. Moreover, the symmetry properties of funnel plot was examined by using Begg and Egger tests. 15

| Risk of bias assessment
All the included studies were critically assessed for methodological quality by 2 researchers independently (Lu X and Guo WY) by using the Quality In Prognosis Studies tool. Each study was graded for the following domains: study participation, study attrition, prognostic factor measurement, outcome measurement, study confounding, and statistical analysis and reporting. The risk of bias for each domain is graded as low (−), moderate (±), or high (+).

| Subgroup analysis
To identify the sources of heterogeneity, subgroup analyses were conducted according to etiologies, geographical region,

| Overall and subgroup analyses for PLR
As shown in Table 3 (Figure 3). Subgroup analyses shown there were no statistically significant correlation between PLR and OS in patients with HER2-positive expression or patients with early stage tumor. Subgroup analysis based on cut-off value suggested a value that more than 185 was better for PLR in predicting prognosis of BC patients (HR: 1.81; 95% CI: 1.33-2.46; P < 0.001). In subgroup analyses for PLR in predicting DFS, no significant association was found (Table 3).

| Sensitivity analyses and publication bias
The sensitivity analyses was conducted to investigate the stabilities of the pooled HRs of OS and DFS by omitting enrolled studies in turn. The results showed that the pooled HRs did not alter significantly after eliminating the included studies in sequence which suggested the steady of our findings ( Figure 4).
The funnel plots showed no obvious publication bias among the enrolled studies (studies distributed around the center line symmetrically). Moreover, 2 kinds of statistical tests further validated the dissymmetry of the funnel plot by using Egger's test and Begg's test (Table 4).

| DISCUSSION
Recently, more and more studies focused on correlation between inflammation and solid malignancies revealed that tumor initiation, progression, and metastasis were affected by host systemic inflammatory response as well as tumor microenvironment. 4 Although the underlying molecular mechanisms have not been adequately illuminated. Neutrophil was considered to be related to cancer-associated inflammation with the potential mechanism of responding to the ectopic interleukin-8 released in tumor proliferation, progression and metastasis. 53 Moreover, cancer-associated cytokines like tumor necrosis factor-α and interleukin-6 contribute to neutrophilia in solid cancers. 54,55 Neutrophilia inhibits the cytotoxic activity of immune cells like lymphocytes, natural killer cells and T cells which would counteract the anti-tumor immune response. 56,57 A high platelet counts was considered to be related to metastasis of BC cells with the mechanism of contribution to lysophosphatidic aciddependent metastasis. 58 Platelets could also promote tumor angiogenesis and stroma formation by secreting vascular endothelial growth factor and facilitating migration of inflammatory cells. 59,60 Loi et al found elevated lymphocytic infiltration in BC was associated with favorable prognosis, especially in those node-positive and HER2-negative BC. 61 The lymphocytes played an important role in cell-mediated anti-tumor immune responses and tumor immunological surveillance. [62][63][64] In view of the heterogeneity of BC, subgroup analyses according to different subtypes like triple-negative and HER2-positive. The results shown a greater magnitude of association between NLR and OS in triple-negative BC patients than in HER2-positive ones. And negative prognostic effect was found for NLR and PLR in HER2-positive BC  65 Their process of extracting data were not rigorous enough that several HRs with 95% were not consistent with original researches. 10,47 The pooled results of our study were more credible and stable because of more rigorous in data extraction and subgroup analyses. Nonetheless, future prospective studies with large sample size were in need to confirm our outcomes especially the prognostic effect of NLR and PLR on HER2-positive BC patients. Our subgroup analyses also suggested that a cut-off value no less than 185 for PLR in predicting OS was more preferable. Individualized therapy based on tumor-associated biological characteristics and host circumstance was advocated in treatment strategies for BC. Several treatments such as surgical resection combined with adjuvant chemotherapy or neoadjuvant chemotherapy, endocrine therapy, and targeted therapy were optional for BC patients with different tumor stages. This study performed subgroup analysis based on tumor stage suggested comparative effect of NLR and PLR in predicting OS. The data were insufficient to conduct subgroup analysis based on treatment strategies.
To our knowledge, this was the most comprehensive metaanalysis with largest sample size to estimate the prognostic role of PLR as well as NLR for BC. However, there were still several limitations should be taken into consideration when interpreting our findings. First, although there was no obvious publication bias, all the included studies were retrospectively designed. High proportion of retrospective individual studies would give rise to inherent bias inevitably. It would be preferable for future studies to design and collect data prospectively. Second, in the subgroup analysis of pathology, only triple-negative and HER2-positive BC patients were enrolled. Furthermore, all the included studies were written in English which would result in potential publication bias. Finally, future international multi-center studies with larger sample size were in need to confirm our results. Nevertheless, the present meta-analysis was performed at an appropriate time as an adequate number of studies with sufficient data in a large patient cohort investigating the prognostic effect of NLR and PLR for BC patients have been accumulated, allowing evaluation through meta-analysis. A meta-analysis is considered to be a statistical inspection of scientific studies which is associated with higher evidence level than the individual studies themselves. 66 Subgroup analyses were performed to minimize heterogeneity stemming from different BC-specific subtypes, optional cut-off values, geographical regions, and tumor stages.
In conclusion, this study suggested that elevated NLR and PLR were associated with poor OS as well as high risk of recurrence for BC patients. Subgroup analyses confirmed negative prognostic value of NLR and PLR for HER2-positive BC patients. As easily accessible parameters, NLR and PLR should be identified as useful biomarkers in the management of BC.