Prognostic significance of neutrophil to lymphocyte ratio in ovarian cancer: evidence from 4,910 patients

Increasing evidence indicates that elevated neutrophil to lymphocyte ratio (NLR) are related with poor prognosis in various types of tumors. However, the prognostic role of NLR in patients with ovarian cancer (OC) remains controversial. Thus, the current meta-analysis aimed to investigate the prognostic role of NLR in patients with OC. A total of 16 studies with 4,910 patients were included. By pooling hazard ratios (HRs) with 95% confidence intervals (CIs) and odds ratios (ORs) with 95% CIs from each study. The results demonstrated that elevated pretreatment NLR was significantly related to poor OS (HR: 1.50, 95% CI: 1.27-1.77) and PFS (HR: 1.53, 95% CI: 1.28-1.84) in patients with OC. Subgroup analyses was divided by ethnicity, sample size, histologic types, cut-off value of NLR, analysis method and NOS score, but the results did not showed any significant change the main results. This meta-analysis revealed that elevated pretreatment NLR might be a predicative factor of poor prognosis in OC patients.


INTRODUCTION
Ovarian cancer (OC) is the second most common female genital tract cancer and the most lethal form of all gynecological malignancies [1].Based on GLOBOCAN estimates, more than 238,700 new cases and 151,900 women died from ovarian cancer every year in worldwide [2].Due to the lack of specific symptoms and efficiently prognostic biomarkers, over 75% of patients are at advanced stage of the disease (Stage III or IV) when being diagnosed [3,4].About 30%-40% of OC patients can survive for 5 years while 10-year survival rate is only 5%-21% [4,5].Therefore, efficient and reliable biomarkers for individualized prediction of treatment outcomes and prognosis in early stage of OC patients are urgently required.
Accumulating evidence has shown that inflammation plays an essential role in tumor development at different stages including initiation, promotion, malignant conversion, invasion, and metastasis.The systemic inflammation is associated with poor prognosis in most of cancers [6][7][8].The prognostic value of systemic inflammatory response markers has received paramount attention, and a variety of blood-based parameters that reflect the status of systemic inflammation have been extensively explored as prognostic biomarkers in various cancers including OC [9][10][11].These inflammatory markers include C-reactive protein (CRP) [12], platelet count [13], neutrophil to lymphocyte ratio (NLR) [14], lymphocyte to monocyte ratio(LMR) [15], platelet to lymphocyte ratio (PLR) [16] and modified Glasgow prognostic score (mGPS) [9].

Selection and characteristics of included studies
The literature retrieval procedure was illustrated in Figure 1.The initial search collected a total of 260 articles (258 from database search, 2 from other sources), of which 42 duplicate publications were removed.The remaining 218 articles were subjected to next evaluation.After reviewing the titles and abstracts, 192 articles were excluded as they were duplicate reports, abstracts, reviews, case reports or studies irrelevant to the current analysis.Subsequently, the full text of remaining 26 articles was carefully reviewed to assess the eligibility of these articles and 10 articles were further excluded: 1 study had significant overlap, 5 articles were abstracts without detailed data, 2 papers were not prognostic article and 2 papers lacked necessary data for estimating the HR and 95% CI or the cut-off of NLR.Finally, 16 studies were used for the meta-analysis.

NLR and OS in OC
Total 16 studies with 4,910 OC patients evaluated the prognostic significance of NLR for OS.Meta-analysis of these 16 retrospective cohorts showed that the patients with elevated NLR were associated with unfavorable OS (HR: 1.50, 95% CI: 1.27-1.77,p < 0.001) (Figure 2).Due to the extreme heterogeneity between studies (I 2 : 80.2%, P h < 0.001), we conducted subgroup analyses according to the potential confounders, such as ethnicity, sample size, histologic types, cut-off value of NLR and NOS score.Subgroup analyses by ethnicity indicated that elevated NLR predicted poor prognosis for patients in Asian populations (HR: 1.69 95%CI: 1.31-2.21,random effects), but did not have prognostic efficiency for patients in Caucasian

NLR and PFS in OC
Twelve cohorts with 3,884 subjects reported the data of relationship between NLR and PFS in patients with OC.The pooled result showed that increased NLR was significantly correlated with worse PFS (HR: 1.53, 95% CI: 1.28-1.84,p < 0.001), with extreme heterogeneity (I 2 : 85.2%, P h < 0.001) (Figure 3).We also explored the heterogeneity by conducting subgroup analyses.In the subgroup analysis based on ethnicity, an elevated NLR appeared to be a poor predictor for PFS no matter the patients were Caucasian (HR: 1.25, 95% CI: 1.10-1.42,Fixed) and Asian (HR: 1.59, 95% CI: 1.25-1.84,Random).Stratification by sample size, the obvious relationship between elevated NLR and poor PFS was found in studies with both small sample size (< 200) (HR: 1.83, 95% CI: 1.24-2.69,random effects) and large sample (≥200) (HR: 1.35, 95% CI: 1.13-1.62,random effects).Similarly, this trend was also observed with the stratification of histology type, cut-off value of NLR, analysis method and NOS score (Table 2).

Meta-regression analysis
Heterogeneity among studies was observed in the overall comparisons as well as in the subgroup analyses.We conducted meta-regression analysis by using variables as ethnicity (Asian vs. Caucasian), sample size (<200 vs.≥200), histologic types (Epithelial vs. Mixed), cut-off value of NLR (<3.4 vs.≥3.4) and NOS score (<7 vs.≥7), to investigate the potential source of heterogeneity among studies for OS and PFS.In multivariate analysis, the results showed that ethnicity (P = 0.744), sample size (P = 0.074), histologic types (P = 0.278), cut-off value of NLR (P = 0.526) and NOS score (P = 0.465) did not contribute to the source of heterogeneity for OS.Moreover, the data demonstrated that ethnicity (P = 0.884), sample size (P = 0.585), histologic types (P = 0.346), cut-off value of NLR (P = 0.585), analysis method (P = 0.839) and NOS score (P = 0.785) were not associated for the source of heterogeneity for PFS.

Sensitivity analysis
By omitting the included studies sequentially, a sensitivity analysis was performed to assess the influence of the each included study on the pooled HR on OS and PFS was performed.The pooled HRs for OS and PFS were not substantially change.The results showed that the pooled HRs were stable and robust.The results of the sensitivity analysis were shown in Figure 4.

Publication bias
Begg's funnel plot and Egger's test were performed to evaluate the publication bias of the included studies.The Begg's funnel shape of the funnel plots showed that the dots were not symmetrically distributed(Figure 5), and the results of Egger's test demonstrated that there was significant publication bias in OS and PFS (P>|t| = 0.000 for OS and P>|t| = 0.001 for PFS, respectively).The results revealed publication bias in this metaanalysis.Therefore, we further performed a "trim and fill" analysis to identify the source of the publication bias.It was estimated that there were seven unpublished studies evaluating the role of the NLR in OS and six unpublished studies evaluating the role of NLR in PFS (Figure 5).The recalculated results that combined estimated unpublished studies did not change significantly for OS (HR: 1.08, 95%CI: 1.05-1.12,p < 0.001) and PFS (HR = 1.07, 95%CI: 1.04-1.11,p < 0.001), indicated a positive outcome even though publication bias still exists.

DISCUSSION
Nowadays, many prognostic factors have been investigated in an attempt to improve treatment outcomes in OC patients, such as age, FIGO stage, residual tumor, lymph node metastasis, vascular invasion and resection margin involvement.However, these factors have some limitations including difficulty in obtaining tumor samples, time consuming, high cost to identify, and insufficient specificity or sensitivity for specific OC, which limit their extensive application in clinical settings [4,5,34,35].Therefore, new clinical predictive and prognostic markers are required.The strong linkage between inflammation and cancer progression has been well established in the past decade [6,8,10].Interestingly, recent studies demonstrated that inflammation-related neutrophils and immunocytes mediate the communication between tumor cells and tumor microenvironment [13,49].The pretreatments have become the potential attractive indicator of prognosis for estimating the inflammatory response and outcomes of cancer patients.However, these two cell types play different roles in the inflammation response and cancer progression [50].Neutrophils promote the proliferation and survival and of cancer cells by collecting more inflammation mediators such as interleukin-1, interleukin-6, vascular endothelial growth factor and tumour necrosis factor, and large number of neutrophils suppress cytolytic activity of lymphocytes, natural killer cells and T-cell proliferation function [51,52].However, lymphocytes have a vital role in the immune defense against tumour cells, which inhibit tumor cell proliferation and metastases [53,54].As an indicator of concisely reflecting the balance between neutrophils and immunocytes, NLR is a promising prognostic factor [22,23,26].Additionally, it is a convenient and cost-effective and reproducible index to measure the NLR in clinical practice, which makes NLR an attractive biomarker for OC prognostication.
Although one recent meta-analysis have reported that high NLR is associated with an adverse OS and EDS in patients with gynecologic malignancies [55], OC has not been discussed emphatically and additional original studies with larger sample sizes have been published since then [40,42,44,47], which led to its limitations.To provide a more precise assessment of the prognostic role of NLR for patients with OC, we performed a comprehensive meta-analysis by including the most recent and relevant articles.In the present study, 16 studies with 4,910 patients were included to calculate pooled HR.The results demonstrated that elevated pretreatment NLR was significantly related to poor OS and PFS in patients with OC.These findings are similar to the previous metaanalyzes, which also evaluated the prognostic value of NLR in patients with other malignancies [11, 14-17, 19, 21-24, 27, 28, 30, 31, 56] or gynecologic cancers.Due to the extreme heterogeneity between studies, subgroup analyses by ethnicity, sample size, histologic types, primary treatment, cut-off value of NLR, analysis method and NOS score did not show obviously significant change in the main results.Moreover, our meta-regression analysis showed that none of the variables listed above contributed to the heterogeneity.Furthermore, in terms of the correlations between NLR and clinicopathological characteristics, we did not conduct quantitative synthesis because of a relatively small amount of studies and sample size.Because of these deficiencies, although we are able to conclude that the pooled estimate of available studies indicates that a higher NLR is associated with a poor OC prognosis, whether the NLR can serve prospectively as a clinical marker of prognosis will need further investigation.In addition, the publication bias was not significant, indicating that the results of our meta-analysis are reliable.To the best of our knowledge, this is the first PROSPERO meta-analysis exploring the prognostic effects of increased NLR in OS and PFS in patients with OC.
This meta-analysis has several limitations that should be acknowledged.First, a large part of included studies were retrospective analysis and thus more susceptible to the bias in data selection.Second, the patients in most of included publications were from Asian ethnicity, which may cause selection bias for general population.Third, some studies included all pathological types of OC patients, which may cause heterogeneity and interfere with the execution of subgroup analysis and meta-regression.Fourth, the cut-off values for determining elevated NLR in each individual study were inconsistent, which may cause the heterogeneity to some extent.Finally, some studies provided the corresponding HRs but their 95% CIs could only be retrieved from univariate analysis or calculated from Kaplan-Meier curves rather than from multivariate analysis, which may impair the accuracy of the pooled estimates.
In conclusion, our meta-analysis demonstrated that elevated NLR by pretreatment was closely associated with poor survival outcome in OC patients.However, attention should be paid cautiously due to the limitations listed above.To better understand the role of NLR in the prognosis of OC patients, more large-scale and welldesigned investigations should be conducted.

Literature search
The protocol of this review was registered with PROSPERO (No. CRD 42016052250) (Supplementary Table 1).This analysis was carried out in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Supplementary Table 2).We conducted thorough literature search using the PubMed, Embase and Cochrane Library for related articles that evaluated the prognostic value of NLR for ovarian cancer.The literature search was last updated on January 31, 2017.The following medical subject headings (MeSH) and keywords employed in the literature search included: ('ovarian neoplasms' OR 'ovarian carcinoma' OR 'ovarian tumor' OR 'ovarian cancer') AND ('NLR' OR 'neutrophil to lymphocyte ratio' OR 'neutrophil-tolymphocyte ratio' OR 'neutrophil lymphocyte ratio') AND ('prognosis' OR 'prognostic' OR 'survival' OR 'outcome' OR 'mortality').There were no restrictions in language or publication date.In addition to the literatures searched from the above sources, the articles listed in the reference lists of the reviewed articles were also manually reviewed to obtain additional eligible studies.

Inclusion and exclusion criteria
Selection criteria studies included in the metaanalysis need to meet the following criteria: (1) evaluated the prognostic value of NLR for OC; (2) patients with a pathologically confirmed diagnosis of OC; (2) the value of NLR was acquired from a peripheral blood before treatment; (3) HRs and 95% CIs for NLR in OS and (or) PFS were reported, or the data sufficient to estimate the HR and 95% CIs in studies; (4) the cut-off value of NLR was clearly reported.A study must meet all four inclusion criteria for inclusion.The exclusion criteria were: (1) letters, conference abstracts or review articles; (2) animal studies, non-clinical studies or case reports; (3) studies had overlapping or duplicate data; (4) no access to the full text for quality assessment and insufficient data to estimate HRs and 95% CIs; (4) did not present the cut-off value for elevated NLR.

Data extraction and quality assessment
Eligible publications were reviewed by two investigators independently (Q Zhou and Z He) and the relevant data were extracted in a standardized table.The following data was extracted: surname of the first author's surname, publication year, study country, ethnicity, duration, sample size, subtype of OC, tumor stage, tumor grade, treatment method, cut-off value defining elevated NLR and HRs with corresponding 95% CIs for PFS/ RFS and(or) OS.The disagreements were resolved by consulting the third investigator (Li Hong).
The quality of each included study was assessed independently by two investigators assessed according to the Newcastle-Ottawa Scale (NOS) (http://www.ohri.ca/programs/clinical_epidemiology/oxford.asp)[57].The NOS consists of three parameters of quality: selection (0-4 points), comparability (0-2 points), and outcome assessment (0-3 points).The maximum possible score is 9 points and NOS scores ≥7 are considered as high-quality studies.In the case of conflicting evaluations, a third investigator (Li Hong) was consulted, and disagreement was resolved by multilateral discussion.

Statistical analysis
All data were analyzed using STATE 12.0 software (State Corporation, College Station, TX, USA), and all P values were two-sided.HRs and 95% CIs for OS and PFS were directly obtained from individual articles or calculated from indirect data according to the methods illustrated by Tierney et al [58].When analyzing the relationship between NLR and clinicopathological factors, odds ratios (OR) and 95% CI were synthesized as the effective value.The heterogeneity among studies was estimated with the χ2-based Q test and Higgins'Ι 2 statistic.A p-value < 0.1 for the Q-test or Ι 2 >50% indicated significant heterogeneity, and the random-effects model was used, otherwise, the fixed-effects model was used.Sources of inter-study heterogeneity were explored using subgroup analysis.Sensitivity analyses were conducted to evaluate the stability of the results.Publication bias of literatures was evaluated using Begg's funnel plot and Egger's test.

Figure 1 :
Figure 1: Flow chart showing the process for selecting eligible studies in the meta-analysis.

Figure 2 :
Figure 2: The forest plot between elevated NLR and OS in patients with OC.

Figure 3 :
Figure 3: The forest plot between elevated NLR and PFS in patients with OC.

Figure 4 :
Figure 4: Sensitivity analysis of NLR on OS (A) and PFS (B) in OC patients.

Figure 5 :
Figure 5: Funnel plot adjusted using the trim and fill method for OS (A) and PFS (B) in patients with OC.Circles: included studies.Diamonds: presumed missing studies.

Table 2 : Summary of meta-analysis results. Factors No. of studies No. of patients Effects model HR(95% CL) P Heterogeneity
P denotes P value for statistical significance based on Z test; Ph denotes P value for heterogeneity based on Q test.HR hazard ratio; CI confidence interval.