2.1. Materials and methods
2.1. 1. Search strategies
A systematic literature search of the PubMed, Embase and Web of Science databases for relevant articles was performed with the deadline of December 29, 2019. Search terms used in the search strategy included the keywords “bone sarcoma”, “bone cancer”, “bone neoplasms”, “chondrosarcomas”, “sarcoma, Ewing's”, “sarcoma, Ewings”, “Ewing's sarcoma”, “Ewings sarcoma”, “Ewing's tumor”, “Ewings tumor”, “tumor, Ewing's”, “Ewing sarcoma”, “Ewing tumor”, “tumor, Ewing”, “osteosarcomas”, “osteosarcoma tumor, “osteosarcoma tumors”, “tumor, osteosarcoma”, “tumors, osteosarcoma”, “sarcoma, osteogenic”, “osteogenic sarcomas”, “sarcomas, osteogenic”or“osteogenic sarcoma” combined with “neutrophil to lymphocyte ratio”, “neutrophil-lymphocyte ratio” or “NLR” and“prognosis” or “outcome” or“survival”. In addition, the reference lists of the relative articles were carefully scanned for potentially eligible studies.
2.1.2. Selection criteria
The following eligibility criteria were as follows: (1) the diagnosis of all patients with bone sarcomas was confirmed depended on histological evidence; (2) studies investigated the association of pretreatment NLR with overall survival (OS); (3) reported a cut-off value for NLR; (4) the study provided sufficient information to calculate the HR and 95% CI. The exclusion criteria were as follows: (1) articles that were letters, conference abstracts, case reports, editorials, laboratory studies, expert opinions and reviews; (2) lack of sufficient data for further analysis; (3) repeated analyses and duplicate publications; (4) non-English articles.
2.1.3. Data extraction and quality assessment
Studies were assessed for eligibility and quality and the data extracted by three independent reviewers (HZH, XH and WBY), and any conflicts between them was resolved by discussion. The following information was were collected from the 6 included studies: first author’s name, the year of publication, country, ethnicity, number of patients, age, gender, cut-off values, stage, time of follow-up, the survival data and the and the relevant information regarding the bone sarcomas.
The quality of the eligible studies was assessed according to the Newcastle-Ottawa quality assessment scale (NOS) [14]. The NOS scores ≥6 were defined as high-quality studies.
2.1.4. Statistical analysis
All statistical analyses were carried out by STATA software version 12.0 (STATA Corporation, College Station, TX, USA). The combined HR and 95% CIs were used to evaluate the association between NLR and OS based on the imformation extracted from the eligible studies. The between-study heterogeneity was assessed by using the chi-square test and the I2 statistic. An I2 value of >50% indicated a significant heterogeneity. We further conducted sensitivity analyses and publication bias to assess the stability of results.
2.2. Results
2.2.1. Study characteristics
The study selection process is shown in the flow diagram (Fig. 1). A total of 256 potential articles were acquired from the three databases (PubMed, Embase and Web of Science) through expanding the search strategy. 168 studies were left after duplicates removed. Of these studies, 92 were excluded by reviewing the titles and abstracts, leaving 76 articles for the full-text review. In the review, 70 studies were excluded for the reasons as follows: 3 were not relevant to NLR or bone tumor, 28 were eliminated for no relevant outcomes reported, 13 studies were letter, reviews or meta-analysis, 6 were animal experiments, and 8 were of insufficient data for analysis. Finally, 6 eligible studies involving 1131 patients that met the inclusion criteria were enrolled into our meta-analysis [15-20].
All the included studies were published between 2016 and 2017. The sample sizes ranged from 100 to 359. Of the 6 studies, four studies came from China, one in Peru, and one in Denmark. All the included studies regarded the OS as the endpoint, and two studies presented progression-free survival (PFS) and disease-specific mortality (DSM) respectively. Quality assessment results of the eligible studies varied from 7 to 8, with average 7.5 (Supplementary Table 1). The detailed information of the eligible studies is shown in Table 1.
2.2.2. Meta-analysis
2.2.3. Overall survival
The present results revealed that high NLR was significantly related with a poor prognosis for patients with bone sarcoma (OS: HR=2.26, 95%Cl: 1.83-2.69, p<0.001) (Fig. 2), and the fixed-effect model was utilized for no significant heterogeneity among the studies (I²=0.0%, p=0.691). We made subgroup-analysis to further explore the relationship between high NLR and OS based on the following parameters: type of cancer, ethnicity, sample size (≥200 or <200), cut-off values for NLR (≥3 or <3), follow-up time (≥30 months or <30) and paper quality (NOS scores ≥8 or <8). The subgroup-analysis illustrated the same outcomes that the significant relationship between high NLR and poor OS was not altered with all the factors above (Table 2). Moreover, no significant heterogeneity was detected across studies.
2.2.4. Diagnosis analysis
Forest plots of the sensitivity and specificity of NLR for predicting overall survival of patients with bone sarcoma are shown in Fig. 3. A random effect model was utilized with an obvious heterogeneity (I²=68.31% for sensitivity and I²=37.81% for specificity). The summary outcomes are as follows: sensitivity (SEN), 0.63 (95%CI 0.55-0.70); specificity (SPE), 0.80 (95%CI 0.75-0.84); positive likelihood ratio (PLR), 3.10 (95%CI 2.30-4.20); negative likelihood ratio, 0.46 (95%CI 0.36-0.58); and overall diagnostic odds ratio (DOR), 7.0 (95%CI 4.0-11.0). Furthermore, we made a summary receiver operator characteristic (SROC) curve (Fig. 4) and calculated the area under the curve (AUC) (0.80, 95%CI 0.76-0.83). To sum up, the study suggested that NLR had a relatively high diagnostic accuracy for the prognosis of malignant bone tumor patients. Whereas, more studies were warranted to verify our findings.
2.2.5. Sensitivity analysis and publication bias
In the present study, we quantitatively performed Begg’s and Egger’s tests to assess the publication bias. No evidence of publication bias was observed from Begg’s funnel plot (p=0.260) (Supplementary Figure 1) and Egger’s test (p=0.223) (Supplementary Figure 2). Accordingly, the possibility of publication bias could be excluded. Furthermore, the sensitivity analysis revealed that the outcomes did not change greatly when omitting studies one by one (Supplementary Figure 3). We made a Deeks’ funnel plot asymmetry test and no evidence of publication bias (p=0.27) existed (Supplementary Figure 4).