Abstract
Fashuai Wu1,*, Deyao Shi1,*, Shidai Mu2, Yu Huang3, Feng Gao1, Xiangcheng Qing1 and Zengwu Shao1
1Department of Orthopaedics, Union Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan 430022, China
2Institute of Hematology, Union Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan 430022, China
3Department of Otorhinolaryngology, Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan 430022, China
*These authors contributed equally to this work
Correspondence to:
Zengwu Shao, email: [email protected]
Keywords: lncRNA; osteosarcoma; prognosis; biomarker; meta-analysis
Received: May 23, 2017 Accepted: December 26, 2017 Published: January 12, 2018
ABSTRACT
Long non-coding RNAs (lncRNAs) contribute to progression of various cancers including osteosarcoma through abnormal regulation of cancer-related cellular processes. Recent researches have shown that various lncRNAs are abnormally expressed in osteosarcoma and associated with the prognosis. With the aim of gaining a better insight into the association between expression level of lncRNAs and prognosis of osteosarcoma, multiple databases including PubMed, Embase, Cochrane Library and Web of Science were carefully searched for available studies up to March 14, 2017. Finally, 19 publications with 1298 patients matched our inclusion criteria and were evaluated in our meta-analysis. The quality of each study was scored using the Newcastle-Ottawa Scale and studies not reaching a minimum threshold were excluded. Results of the meta-analysis demonstrated that abnormal expression level of lncRNAs predicted poor overall survival (pooled HR 3.064, 95% CI: 2.487–3.775) and event-free survival (pooled HR 2.642, 95% CI 1.759–3.970) in osteosarcoma and subgroup analysis showed consistent prognostic value. Furthermore, combining data of Cox multivariable analysis indicated that abnormal expression level of lncRNAs was an independent prognostic marker for overall survival (pooled HR 2.864, 95% CI: 2.246–3.651) in osteosarcoma patients. The clinicopathological parameters analysis further showed that abnormal expression level of lncRNAs was correlated with tumor size, tumor stage, metastasis and differentiation grade of osteosarcoma. Limitations of the meta-analysis included variation of cut-off value definition, Chinese provenance of most studies, publication bias and so on. In conclusion, this meta-analysis suggested that abnormal expression level of lncRNAs has a promising future for predicting the prognosis of patients with osteosarcoma.
INTRODUCTION
Osteosarcoma is an osteoid-producing malignancy of mesenchymal origins. This high-grade tumor is the most common primary bony malignancy and most often originates in the metaphysis of long bones of children and adolescents [1, 2]. The incidence rate of osteosarcoma is about 1.7~4.4/1000,000 per year, and the average male-to-female ratio is 1.22:1 [3]. This devastating tumor is characterized by its high aggressive ability and early systematic metastasis, approximately 20% of osteosarcoma patients present with detectable lung metastatic disease at initial diagnosis, additionally, about 40% of patients free of metastasis will progress to metastasis in later stages during treatment [4]. With the application of chemotherapy, a combination of neoadjuvant chemotherapy, surgery resection and postoperative chemotherapy has clinically become the standard treatment strategy for osteosarcoma and has significantly improved patients’ outcome, the 5-year overall survival of patients with nonmetastatic osteosarcoma of the extremity has approached to 70%~80% over past decades [5, 6]. However, the 5-year overall survival rate of osteosarcoma patients with detectable metastases at diagnosis is as low as 19% [7]. Those patients with recurrent disease also have unsatisfactory 5-year post-relapse-overall-survival (PROS) rate, which is about 30% [8]. The survival of osteosarcoma patients predominantly depends on not only early diagnosis but also timely proper treatment. Therefore, it is highly needed to identify novel diagnostic and prognostic biomarkers for osteosarcoma. Effective prognostic biomarkers predictive of survival or response to therapy will help to predict the outcome more accurately and guide us to select better and more proper treatment options for osteosarcoma patients.
Long non-coding RNAs (lncRNAs) are a class of RNA molecules, which had been dismissed as simply transcriptional ‘junk’ in the past, with a length of more than 200 nucleotides that with little or no protein coding capacity [9, 10]. LncRNAs, which represent a new frontier in molecular biology, play important roles in regulating gene expression at epigenetic, transcriptional and post-transcriptional levels, and contribute to progression of various cancers through abnormal regulation of cancer-related cellular processes, such as proliferation, invasion, metastasis, apoptosis, etc [11–14]. Recent researches have shown that various lncRNAs are abnormally expressed in tissue and blood sample of osteosarcoma, and the expression level of lncRNAs may be associated with prognosis of osteosarcoma, for example, CCAL, MALAT1, TUG1, UCA1, 91H, etc [15–36]. These discoveries indicated that lncRNAs could be promising biomarkers for predicting the prognosis of osteosarcoma. However, most studies reported so far were limited in discrete outcome and sample size. With the aim of gaining a better insight into the prognostic value of lncRNAs in osteosarcoma. We conducted a quantitative meta-analysis to clarify the prognostic value of abnormally expressed lncRNAs in patients with osteosarcoma.
RESULTS
Included studies and characteristics
According to our search strategy, a total of 263 studies were retrieved from Pubmed, Embase, Cochrane Library and Web of Science and from the reference list of relevant articles. Of these, 129 duplicated articles were excluded; through title and abstract review, studies describing unrelated topics (n = 80) which refers to those studies that were not research articles or those studies unrelated to osteosarcoma, lncRNA , or patient prognosis were removed; after reading the remaining 54 full-text articles, studies lacking relevant data (n = 32), studies presented in language other than English (n = 1), full-text unavailable studies (n = 1), and the studies of which included patients were reported in previous studies (n = 1) were excluded. Eventually, 19 studies (containing 1298 patients) met the inclusion criteria and were evaluated in this analysis [18–36]. The screening process and results was shown in Figure 1.
Figure 1: The flow diagram of the studies identified, included and excluded.
Table 1 represented the main characteristics of all the included 19 studies ranging from 2015 to 2016. Among these studies, 2 studies had a sample size larger or equal to 100 and 17 studies had a sample size less than 100 patients. 18 studies were from China and 1 study was from Brazil. Quantitative real-time polymerase chain reaction assays (qRT-PCR) were performed to quantify the level of lncRNAs in all of the studies. Specimens were composed of tissue in 18 studies and plasma in 2 studies. Of the total 18 lncRNAs, only lncRNA HULC were investigated by at least two studies, and the remaining 17 lncRNAs were studied in a single report. The cut-off definitions were various in different studies. In 14 studies, the median of lncRNA level were regarded as cut-off value; others based on ROC analysis or fold change. There are 2 articles that reported the negative relationship between expression level of lncRNAs and survival of osteosarcoma patients. 58% (11/19) of the NOS scores for the included studies on prognosis were ≥ 7. Cox multivariate analysis was used in 12 studies including 904 patients, and 8 studies used both univariate and multivariate analysis. In 14 studies, none of the patients received chemotherapy or radiotherapy before surgery. 1 study showed patients received multidrug chemotherapy before surgery and the left 4 studies did not mention about it. HR values and 95% CIs could be obtained directly from 10 publications. In the other publications, HR values or 95% CIs should be calculated or retracted from Kaplan–Meier curves. All of 19 studies assessing the prognostic value of 18 abnormally expressed lncRNAs in patients with osteosarcoma. The prognostic value of lncRNAs expression was assessed by analyzing overall survival (OS) in 1298 patients and event free survival (EFS) in 301 patients. The association between lncRNAs expression and clinicpathological characteristics of osteosarcoma was estimated in 79% (15/19) of the studies.
Table 1: Characteristics of studies included in the meta-analysis
Abbreviations: OS: overall survival; EFS: event free survival; ROC: receiver operating characteristic; NOS: Newcastle-Ottawa Scale; qRT-PCR: quantitative reverse transcription PCR; NA: not available;
Tumor staging: a TNM staging, bEnneking staging, c staging system not stated clearly in paper; Follow-up: #median;
Method: 1 denoted as obtaining HRs directly from publications; 2 denoted as HRs calculated from the total number of events and its p value; 3 denoted as HRs extracted from Kaplan–Meier curves.
lncRNA expression and prognosis of osteosarcoma
A total of 19 studies reported OS of osteosarcoma based on abnormal expression of lncRNAs in 1298 patients. The pooled hazard ratio (HR) for OS was 3.064 (95% CI: 2.487−3.775, p < 0.00001) (Figure 2), which indicated abnormal expression level of lncRNAs was significantly associated with OS in patients with osteosarcoma. The test for heterogeneity gave no significant results (χ2= 5.90, p = 0.997; I2 = 0.0%). In order to further explore the association between abnormal expression level of lncRNAs and OS in osteosarcoma patients, subgroup analysis was performed based on the following factors: region (China or Brazil), sample size (more than 100 or fewer than 100), expression level relevant to poor prognosis (high or low), pre-operative treatment (none, yes or unclear) and paper quality (NOS scores ≥ 7 or < 7) (Supplementary Figure 1). The results of subgroup analysis showed the significant association between abnormal expression level of lncRNAs and OS of osteosarcoma patients were not altered with all the factors above. No significant heterogeneity was found across studies in all the subgroup analysis (Table 2). In order to assess publication bias, a Begg’s funnel plot was presented for the visual inspection. The funnel plot showed asymmetry and Begg’s test (p < 0.001) indicated significant publication bias for OS. An Egger’s test (p < 0.001) was also performed which indicated significance among the included studies (Supplementary Figure 3).
Figure 2: Meta-analysis of the pooled HRs of overall survival (OS) with abnormally expressed lncRNAs.
Table 2: Results of subgroup analysis of pooled hazard ratios of overall survival of patients with osteosarcoma based on abnormal expression level of lncRNAs
Subgroup analysis | No. of studies | No. of patients | Pooled HR (95% CI) | Heterogeneity | |
|---|---|---|---|---|---|
I2 | p-value | ||||
Region | |||||
China | 18 | 1265 | 3.019 [2.447–3.725] | 0.0% | 0.999 |
Brazil | 1 | 33 | 8.724 [1.497–50.832] | - | - |
Sample size | |||||
< 100 | 17 | 1001 | 3.051 [2.440–3.816] | 0.0% | 0.989 |
≥ 100 | 2 | 297 | 3.151 [1.766–5.621] | 0.0% | 0.993 |
Expression level relevant to poor prognosis | |||||
High | 17 | 1152 | 3.112 [2.498–3.877] | 0.0% | 0.992 |
Low | 2 | 146 | 2.663 [1.373–5.164] | 0.0% | 0.693 |
Pre-operative treatment | |||||
None | 14 | 1017 | 2.973 [2.358–3.748] | 0.0% | 0.967 |
Yes | 1 | 60 | 3.457 [1.216–9.829] | - | - |
Unclear | 4 | 221 | 3.495 [2.036–6.000] | 0.0% | 0.978 |
NOS score | |||||
≥ 7 | 11 | 834 | 3.043 [2.332–3.971] | 0.0% | 0.948 |
< 7 | 8 | 464 | 3.098 [2.214–4.334] | 0.0% | 0.964 |
Additionally, in order to ensure the stability of the results, sensitivity analyses were performed including influence analysis, file-drawer analysis and trim and fill analysis. As for the influence analysis, HRs and their 95% CIs did not change significantly after the exclusion of any of the studies (Supplementary Figure 2). Since there was significant publication bias, we performed file-drawer analysis and the calculation showed the fail-safe number (Nfs) should be 776, which was obviously larger than 19, the number of included studies. And the “trim and fill” method was used for further analysis. The result showed that 6 studies evaluating the prognostic value of abnormally expressed lncRNAs in OS of osteosarcoma patients remained unpublished. The filled meta-analysis for OS (HR 2.772, 95% CI:2.291−3.355, p < 0.001) supported our original result (Supplementary Figure 4). All above sensitivity analyses showed our results were stable.
The prognostic value of abnormal expression of lncRNAs in EFS of patients with osteosarcoma was evaluated in 5 studies including 301 patients. We found that abnormal expression level of lncRNAs was significantly associated with EFS (HR 2.642, 95% CI: 1.759–3.970, p < 0.00001) (Figure 3). Moderate heterogeneity existed among studies (χ2 = 6.92, p = 0.140; I2 = 42.2%). Subgroup analysis, sensitivity analysis and appraisal of publication bias were not performed due to the limited number and relative homogeneity of the studies.
Figure 3: Meta-analysis of the pooled HRs of event-free-survival (EFS) for osteosarcoma patients with abnormally expressed lncRNAs.
Using Cox multivariate analysis in 11 studies including 881 patients, we found that abnormal expression level of lncRNAs was an independent prognostic factor for OS of patients with osteosarcoma (pooled HR 2.864, 95% CI: 2.246–3.651, p < 0.00001), and no significant heterogeneity was detected among studies (χ2 = 3.14, p = 0.978; I2 = 0.0%) (Figure 4). Subgroup analysis was not performed because of the relative homogeneity and similar property of the studies. As for the publication bias, asymmetry was shown on the funnel plot, Begg’s test (p < 0.001) and Egger’s test (p < 0.001) showed significant publication bias (Supplementary Figure 3).
Figure 4: Meta-analysis of the independent role of abnormal expression level of lncRNAs in overall survival (OS) of osteosarcoma.
Sensitivity analyses were performed likewise, the influence analysis showed no obvious change of the HRs and their 95% CIs after the exclusion of any of the studies (Supplementary Figure 3). The file-drawer analysis showed the fail-safe number (Nfs) should be 285, which was obviously larger than 11, the number of included studies. The “Trim and fill” analysis was performed and the result estimated that none study evaluating the independent prognostic value of abnormal expression of lncRNAs in OS of osteosarcoma remained unpublished. Thus the result of filled meta-analysis maintained unchanged. Based on above results of sensitivity analyses, our results were stable to some extent.
lncRNA expression and clinicopathological characteristics of osteosarcoma
As shown in Table 3, several studies examined the association between lncRNA expression and the clinicopathological characteristics of osteosarcoma. In aspect of tumor staging, TNM staging system and Enneking staging system were used in five and nine studies respectively (Staging system not stated clearly in paper was regard as using Enneking staging). There was a significant association between lncRNA expression and tumor staging of osteosarcoma (Advanced (III/IV) vs Early (I/II) staging: pooled OR 4.592, 95% CI: 2.984−7.068, p < 0.00001; pooled OR 4.720, 95% CI: 3.187−6.992, p < 0.00001 respectively). Fifteen studies examined the association between lncRNA expression and tumor metastasis, and significant association was observed as pooled OR 3.025 (95% CI: 2.274−4.024, p < 0.00001). There were fifteen studies examining the association between lncRNA expression and tumor location, while no significance was observed (pooled OR 0.876, 95% CI: 0.670−1.145, p = 0.332). Eleven studies were included to examine tumor size. LncRNA expression was significantly associated with enlarged tumor size (pooled OR 1.806, 95% CI: 1.157−2.820). There were only two studies examining tumor differentiation grade. We observed significant association between lncRNA expression and tumor differentiation grade (pooled OR 2.738, 95% CI: 1.453–5.158, p < 0.00001). Corresponding heterogeneity tests were all shown in Table 3. Subgroup analysis, sensitivity analysis and appraisal of publication bias were not performed due to the limited number and relative homogeneity of the studies.
Table 3: Results of meta-analysis of abnormal lncRNA expression and clinicopathological features in osteosarcoma
Clinicopathological features | No. of studies | No. of patients | Pooled OR | Heterogeneity | ||
|---|---|---|---|---|---|---|
Fixed | Random | I2 | p-value | |||
TNM staging (Advanced vs Early) | 5 | 397 | 4.592 [2.984–7.068] | 4.587 [2.978–7.067] | 0.0% | 0.939 |
Enneking staging* (Advanced vs Early) | 9 | 637 | 4.720 [3.187–6.992] | 4.680 [3.155–6.941] | 0.0% | 0.979 |
Metastasis (Present vs Absent) | 15 | 1013 | 3.025 [2.274–4.024] | 3.014 [2.183–4.162] | 13.8% | 0.299 |
Tumor location (Femur/Tibia vs Elsewhere) | 15 | 1013 | 0.876 [0.670–1.145] | 0.885 [0.666–1.176] | 6.0% | 0.385 |
Tumor Size ( ≥ 8 cm vs < 8 cm) | 11 | 805 | 1.691 [1.276–2.241] | 1.806 [1.157–2.820] | 54.6% | 0.015 |
Tumor differentiation grade (Poor vs Moderate/Well) | 2 | 181 | 2.738 [1.453–5.158] | 2.737 [1.452–5.157] | 0.0% | 0.822 |
*Staging system not stated clearly in paper was regard as using Enneking staging. Tumor staging: I/II and III/IV were defined as early staging and advanced staging, respectively.
DISCUSSION
Osteosarcoma, which is often fatal in both children and adolescents and accounts for 5 % of child malignancies and 9 % of cancer-related deaths in children, is the most common primary malignancy of bone [2, 3]. Although advances in tumor treatment strategy have significantly raised the survival rate of osteosarcoma patients, the survival of osteosarcoma patients with metastatic disease or recurrences and those in advanced stage is still quite poor [3, 6–8]. Thus, novel promising biomarkers of osteosarcoma that can help in diagnosis and prognosis evaluation are still urgently needed. They will help to detect osteosarcoma patients at early stage and contribute to recognize those patients whose tumor will resist chemotherapy or whose tumor will metastasize or recur. These potential biomarkers predictive of survival or response to therapy will be helpful to guiding the individualized treatment of osteosarcoma patients and finally improve their outcome. In recent years, studies have demonstrated that lncRNAs are involved in various biological processes of osteosarcoma, including progression and metastasis, and aberrant expression of multiple lncRNAs was found to have the potential value for predicting outcome of osteosarcoma patients [37, 38].
In this meta-analysis, we examined the prognostic value of abnormally expressed lncRNAs in osteosarcoma and the relation between lncRNAs and clinicopathological characteristics. We examined 19 independent studies comprising data from a total of 1298 patients. We found that abnormal expression level of lncRNAs was associated with OS in patients with osteosarcoma (pooled HR 3.064, 95% CI: 2.487−3.775, p < 0.00001). In detail, of all studies included, subgroup analysis indicated that the significant association did not alter with factors of the region, sample size, lncRNA expression level relevant to poor prognosis (high or low), pre-operative treatment or paper quality. And poorer EFS was found related to abnormal expression level of lncRNAs (HR 2.642, 95% CI: 1.759−3.970, p < 0.00001). Moreover, by combining HRs from studies using Cox multivariate analysis we found abnormal expression level of lncRNAs was an independent prognostic factor for OS of patients with osteosarcoma (pooled HR 2.864, 95% CI: 2.246−3.651, p < 0.00001).
In the aspect of clinicopathological characteristics of osteosarcoma, our results revealed that abnormal expression level of lncRNAs was significantly associated with advanced tumor staging, enlarged tumor size, tumor metastasis and poor tumor differentiation grade, but not related with tumor location.
In our study, a few limitations should be underlined. First, the cut-off value dividing lncRNA expression into high and low groups were different among studies, although most of them were set to median. Second, the staging system for osteosarcoma was regarded as Enneking staging system, when it was not stated clearly in publication, because Enneking staging system are widely used as clinical staging system for osteosarcoma in China now. Third, we only searched English papers. And most studies were from China, the results might mainly represent Chinese patients. Chinese provenance of most studies essentially affected generalizability. Fourth, differences of paper quality among the studies might cause bias in the meta-analysis although subgroup analysis or sensitive analysis did not show obvious change. Fifth, HRs of six studies could not be directly obtained from the publications. Calculating them ourselves or to reconstruct the survival curves to extract the HR estimates might not be precise enough. Sixth, nearly all of the studies included in this analysis reported positive results, so our results might overestimate the prognostic value of abnormally expressed lncRNAs in osteosarcoma to some degree. Therefore, large-scale, multicenter, and high-quality studies are highly needed to confirm our findings. In our opinion, with the updating of gene chip and microarray platform technology, such as TCGA and GEO databases, and an explosion of lncRNAs research in osteosarcoma, a significant extension of our finding and re-analysis, which will include more patients from different regions of the world, would be accomplished in near future.
In summary, this meta-analysis evaluated the prognostic value of abnormally expressed lncRNAs in patients with osteosarcoma. We demonstrated abnormal expression level of lncRNAs was closely associated with poor outcomes of osteosarcoma and it was an independent prognostic factor for overall survival. Abnormal expression level of lncRNAs also reflected some malignant clinicopathological characteristics of osteosarcoma. Despite the limitations mentioned above, although there is a long way to clinical application, we believe that abnormally expressed lncRNAs will be potential prognostic biomarkers for osteosarcoma patients. With future development of lncRNAs research in osteosarcoma, it will play a role in guiding the individualized treatment of osteosarcoma patients in near future.
MATERIALS AND METHODS
Literature retrieval strategy
The present review was performed in accordance with the PRISMA statement Supplementary Table 1 (2009) and the standard guidelines for meta-analyses and systematic reviews of tumor marker prognostic studies [39–41]. The research databases PubMed, Embase, Cochrane Library and Web of Science were independently searched by two researchers (Shidai Mu and Yu Huang) to obtain all relevant articles about the prognostic value of abnormally expressed lncRNAs in patients with osteosarcoma. The literature search ended in March 14, 2017. The search strategy used both MeSH terms and free-text words to increase the sensitivity of the search. The search strategy was: (“long non-coding RNA or lncRNA or RNA long non-coding or lincRNA or long intergenic non-coding RNA”) AND (“osteosarcoma or osteogenic sarcoma”). All included studies were retrieved in English database. We also retrieved articles from other sources, for example, retrieving from the reference lists of relevant articles. Conflicts were solved through group discussion.
Inclusion and exclusion criteria
Studies included in this analysis had to meet the following inclusion criteria: 1) Patients were pathologically diagnosed with osteosarcoma. 2) The expression of lncRNAs was determined in tissues or plasma samples from patients using quantitative reverse transcription polymerase chain reaction (qRT-PCR). 3) Patients were divided into high and low lncRNA expression groups, the prognostic value of one lncRNA was investigated and the relationship between lncRNA expression and survival was examined. 4) Sufficient data or the survival curve was provided to estimate hazard ratios (HRs) for survival rates and their 95% confidence intervals (95% CIs).
Studies were excluded from the analysis if they met any of the following exclusion criteria: 1) They were retracted articles, letters, case reports, reviews, meta-analysis, commentary, conference reports or expert opinions. 2) Articles presented in languages other than English. 3) The article was not found in full text. 4) The article was a repeated study or included patients was reported in a previous study. 5) Did not provide sufficient descriptions of the required data or the data could not be received from the original article or from the authors. 6) Sample population consisted of less than 30 cases. 7)Used nonhuman samples. All the eligible studies were carefully identified by the same two researchers (Fashuai Wu and Deyao Shi), and discrepancies were resolved by discussing with a third researcher (Feng Gao).
Quality of all included studies was assessed independently by four researchers (Fashuai Wu, Deyao Shi, Feng Gao and Xiangcheng Qing) according to the Newcastle-Ottawa Scale (NOS), which contains 9 items, and disagreements were resolved through discussion. Studies with NOS scores more than 7 were considered to be of high quality. And studies not reaching a minimum threshold (NOS scores <=5) would be excluded from the analysis to guarantee the quality of included papers.
Data extraction
Four researchers (Fashuai Wu, Deyao Shi, Shidai Mu and Yu Huang) extracted data independently, and conflicts were resolved through discussion. For each eligible article, extracted information included: author, journal name, year of publication, lncRNA signature, expression associates with poor prognosis, NOS score, method of obtaining HRs and characteristics of the study population (including country of the population enrolled, number of patients (high/low), tumor size, tumor location, tumor differentiation grade, tumor stage, lymph node metastasis, distant metastasis, follow up (month), endpoint, assay method, sample type, cut off value, pre-operation treatment and survival analysis). The primary outcome was overall survival (OS), which was measured from the initiation of therapy until death from any cause. The secondary endpoint was event-free survival (EFS). Data regarding disease-free survival (DFS), progress-free survival (PFS), recurrence-free survival (RFS) were obtained from articles, and were redefined as EFS, which was measured from the date of initiation of therapy to the date of recurrence or metastasis [42, 43].
The effect of lncRNA expression on survival was assessed using HR. We extracted HRs following a methodology suggested previously [44]. HRs and their 95% CIs were extracted directly from the publication, if provided by the authors. Otherwise, we calculated the HRs and their 95% CIs from the published data including p values of the log-rank tests and number of events. Besides, we used the Engauge Digitizer version 9.8 to read survival rates in Kaplan–Meier Curves, then we inputted the extracted survival rates into the spreadsheet set up by Tierney JF et al. to obtain HRs and their 95% CIs, assuming that patients were censored at a constant rate during follow-up [45]. If possible, we asked for original data directly from the authors of the relevant studies.
Statistical methods
1) Pooled HRs and their associated 95% CIs were merged using a fixed-effect model (Mantel-Haenszel). While the random-effect model was applied if significant heterogeneity was present. An HR > 1 indicates that the patients have a poor prognosis. 2) The clinicopathological characteristics were assessed by the pooled odds ratios (ORs). 3) The test for heterogeneity of combined HRs and ORs was performed using a χ2 based Cochran Q test and Higgins I2 statistic. A p value of less than 0.05 or an I2 value of larger than 50% was considered statistically significant. 4) The presence of publication bias was evaluated by using both Begg’s test and Egger’s test. A p value of less than 0.05 was considered statistically significant [46]. 5) Sensitivity analyses including influence analysis, file-drawer analysis (fail-safe number) and trim and fill analysis, were used to examine the stability of results. By excluding any of the whole studies one at a time, influence analysis was used to trace the potential heterogeneity sources. To evaluate the influence of publication bias, file-drawer analysis (fail-safe number) and trim and fill analysis were carried out. Fail-safe number (Nfs) was firstly prompted by Rosenthal in 1979 [47]. Nfs refers to the number of unpublished studies needed to reverse the result of a meta-analysis and Nfs was calculated by the specific formula according to Rosenthal’s article. When Nfs is much larger than actual included studies, we consider the result is stable. The trim and fill method is a nonparametric data augmentation technique firstly proposed by Duval and Tweedie [48]. This rank-based data augmentation technique can be used to estimate the number of missing studies that might exist in a meta-analysis and the effect that these studies might have had on its outcome. The method formalizes the use of funnel plots and yields an effect adjusted for funnel plot asymmetry. Once the number of ‘missing’ studies is estimated, a recalculation was performed to assess the weighted mean effect size and its variance when the ‘missing’ studies are included [49, 50]. 6) Statistical analysis and graphical representation were performed using Stata software statistical software version 14.0 (Stata Corporation, College Station, TX, USA).
Author contributions
F.W. and D.S. contributed equally to this study and share the first authorship. S. M. and Y.H independently searched the databases. F.W., D.S. and F.G. identified the eligible studies. F.W., D.S., F.G. and X.Q assessed the quality of included studies. F.W., D.S., S.M. and Y.H. extracted the data. F.W., and D.S. analysed the data and made the figures and tables. F.W., D.S. and Z.S. wrote the manuscript, and all authors commented on it at every stage. Z.S. oversaw the study.
ACKNOWLEDGMENTS
We would like to thank the researchers and study participants for their contributions.
CONFLICTS OF INTEREST
The authors declare no Conflicts financial interests.
REFERENCES
1. Bielack SS, Kempf-Bielack B, Delling G, Exner GU, Flege S, Helmke K, Kotz R, Salzer-Kuntschik M, Werner M, Winkelmann W, Zoubek A, Jürgens H, Winkler K. Prognostic factors in high-grade osteosarcoma of the extremities or trunk: an analysis of 1,702 patients treated on neoadjuvant cooperative osteosarcoma study group protocols. J Clin Oncol. 2002; 20:776–90. https://doi.org/10.1200/JCO.2002.20.3.776.
2. Lindsey BA, Markel JE, Kleinerman ES. Osteosarcoma Overview. Rheumatol Ther; 2016. https://doi.org/10.1007/s40744-016-0050-2.
3. Mirabello L, Troisi RJ, Savage SA. Osteosarcoma incidence and survival rates from 1973 to 2004: data from the Surveillance, Epidemiology, and End Results Program. Cancer. 2009; 115:1531–43. https://doi.org/10.1002/cncr.24121.
4. PosthumaDeBoer J, Witlox MA, Kaspers GJ, van Royen BJ. Molecular alterations as target for therapy in metastatic osteosarcoma: a review of literature. Clin Exp Metastasis. 2011; 28:493–503. https://doi.org/10.1007/s10585-011-9384-x.
5. Picci P, Ferrari S, Bacci G, Gherlinzoni F. Treatment recommendations for osteosarcoma and adult soft tissue sarcomas. Drugs. 1994; 47:82–92. https://doi.org/10.2165/00003495-199447010-00006.
6. Ferrari S, Meazza C, Palmerini E, Tamburini A, Fagioli F, Cozza R, Ferraresi V, Bisogno G, Mascarin M, Cefalo G, Manfrini M, Capanna R, Biagini R, et al. Nonmetastatic osteosarcoma of the extremity. Neoadjuvant chemotherapy with methotrexate, cisplatin, doxorubicin and ifosfamide. An Italian Sarcoma Group study (ISG/OS-Oss). Tumori. 2014; 100:612–19. https://doi.org/10.1700/1778.19262.
7. Mialou V, Philip T, Kalifa C, Perol D, Gentet JC, Marec-Berard P, Pacquement H, Chastagner P, Defaschelles AS, Hartmann O. Metastatic osteosarcoma at diagnosis: prognostic factors and long-term outcome—the French pediatric experience. Cancer. 2005; 104:1100–09. https://doi.org/10.1002/cncr.21263.
8. Bacci G, Briccoli A, Longhi A, Ferrari S, Mercuri M, Faggioli F, Versari M, Picci P. Treatment and outcome of recurrent osteosarcoma: experience at Rizzoli in 235 patients initially treated with neoadjuvant chemotherapy. Acta Oncol. 2005; 44:748–55. https://doi.org/10.1080/02841860500327503.
9. Spizzo R, Almeida MI, Colombatti A, Calin GA. Long non-coding RNAs and cancer: a new frontier of translational research? Oncogene. 2012; 31:4577–87. https://doi.org/10.1038/onc.2011.621.
10. Wright MW, Bruford EA. Naming ‘junk’: human non-protein coding RNA (ncRNA) gene nomenclature. Hum Genomics. 2011; 5:90–98. https://doi.org/10.1186/1479-7364-5-2-90.
11. Zhu J, Fu H, Wu Y, Zheng X. Function of lncRNAs and approaches to lncRNA-protein interactions. Sci China Life Sci. 2013; 56:876–85. https://doi.org/10.1007/s11427-013-4553-6.
12. Zhu H, Zhou X, Chang H, Li H, Liu F, Ma C, Lu J. CCAT1 promotes hepatocellular carcinoma cell proliferation and invasion. Int J Clin Exp Pathol. 2015; 8:5427–34.
13. Li H, Yu B, Li J, Su L, Yan M, Zhu Z, Liu B. Overexpression of lncRNA H19 enhances carcinogenesis and metastasis of gastric cancer. Oncotarget. 2014; 5:2318–29. https://doi.org/10.18632/oncotarget.1913.
14. Wu L, Jin L, Zhang W, Zhang L. Roles of Long Non-Coding RNA CCAT2 in Cervical Cancer Cell Growth and Apoptosis. Med Sci Monit. 2016; 22:875–79. https://doi.org/10.12659/MSM.897754.
15. Wang Y, Yang T, Zhang Z, Lu M, Zhao W, Zeng X, Zhang W. Long non-coding RNA TUG1 promotes migration and invasion by acting as a ceRNA of miR-335-5p in osteosarcoma cells. Cancer Sci. 2017; 108:859–67. https://doi.org/10.1111/cas.13201.
16. Zhang JH, Kang XH, Lu P, Miao ZH, Sun GS, Cao XJ, Cao F. Expression of long non-coding RNA MALAT1 in osteosarcoma and its effect on invasiveness and metastatic potential of osteosarcoma cells. [Article in Chinese] Zhonghua Bing Li Xue Za Zhi. 2016; 45:561–65.
17. Pasic I, Shlien A, Durbin AD, Stavropoulos DJ, Baskin B, Ray PN, Novokmet A, Malkin D. Recurrent focal copy-number changes and loss of heterozygosity implicate two noncoding RNAs and one tumor suppressor gene at chromosome 3q13.31 in osteosarcoma. Cancer Res. 2010; 70:160–71. https://doi.org/10.1158/0008-5472.CAN-09-1902.
18. Zhou DK, Yang XW, Li H, Yang Y, Zhu ZJ, Wu N. Up-regulation of long noncoding RNA CCAL predicts poor patient prognosis and promotes tumor metastasis in osteosarcoma. Int J Biol Markers. 2017; 32:e108–12. https://doi.org/10.5301/jbm.5000240.
19. Chen F, Mo J, Zhang L. Long noncoding RNA BCAR4 promotes osteosarcoma progression through activating GLI2-dependent gene transcription. Tumour Biol. 2016; 37:13403–12. https://doi.org/10.1007/s13277-016-5256-y.
20. Sun J, Wang X, Fu C, Wang X, Zou J, Hua H, Bi Z. Long noncoding RNA FGFR3-AS1 promotes osteosarcoma growth through regulating its natural antisense transcript FGFR3. Mol Biol Rep. 2016; 43:427–36. https://doi.org/10.1007/s11033-016-3975-1.
21. Zhao H, Hou W, Tao J, Zhao Y, Wan G, Ma C, Xu H. Upregulation of lncRNA HNF1A-AS1 promotes cell proliferation and metastasis in osteosarcoma through activation of the Wnt/β-catenin signaling pathway. Am J Transl Res. 2016; 8:3503–12.
22. Li F, Cao L, Hang D, Wang F, Wang Q. Long non-coding RNA HOTTIP is up-regulated and associated with poor prognosis in patients with osteosarcoma. Int J Clin Exp Pathol. 2015; 8:11414–20.
23. Sun XH, Yang LB, Geng XL, Wang R, Zhang ZC. Increased expression of lncRNA HULC indicates a poor prognosis and promotes cell metastasis in osteosarcoma. Int J Clin Exp Pathol. 2015; 8:2994–3000.
24. Uzan VR, Lengert A, Boldrini É, Penna V, Scapulatempo-Neto C, Scrideli CA, Filho AP, Cavalcante CE, de Oliveira CZ, Lopes LF, Vidal DO. High Expression of HULC Is Associated with Poor Prognosis in Osteosarcoma Patients. PLoS One. 2016; 11:e0156774. https://doi.org/10.1371/journal.pone.0156774.
25. Wang B, Su Y, Yang Q, Lv D, Zhang W, Tang K, Wang H, Zhang R, Liu Y. Overexpression of Long Non-Coding RNA HOTAIR Promotes Tumor Growth and Metastasis in Human Osteosarcoma. Mol Cells. 2015; 38:432–40. https://doi.org/10.14348/molcells.2015.2327.
26. Li W, He X, Xue R, Zhang Y, Zhang X, Lu J, Zhang Z, Xue L. Combined over-expression of the hypoxia-inducible factor 2α gene and its long non-coding RNA predicts unfavorable prognosis of patients with osteosarcoma. Pathol Res Pract. 2016; 212:861–66. https://doi.org/10.1016/j.prp.2016.06.013.
27. Gao KT, Lian D. Long non-coding RNA MALAT1 is an independent prognostic factor of osteosarcoma. Eur Rev Med Pharmacol Sci. 2016; 20:3561–65.
28. Tian ZZ, Guo XJ, Zhao YM, Fang Y. Decreased expression of long non-coding RNA MEG3 acts as a potential predictor biomarker in progression and poor prognosis of osteosarcoma. Int J Clin Exp Pathol. 2015; 8:15138–42.
29. Zhu KP, Zhang CL, Shen GQ, Zhu ZS. Long noncoding RNA expression profiles of the doxorubicin-resistant human osteosarcoma cell line MG63/DXR and its parental cell line MG63 as ascertained by microarray analysis. Int J Clin Exp Pathol. 2015; 8:8754–73.
30. Zhao H, Zhao Y, Tao J, Ma C, Zhang J, Xu H, Dong Y. Up-regulated expression of lncRNA NEAT1 promotes progression of osteosarcoma by regulating the activity of Wnt/beta-catenin pathway. Int J Clin Exp Pathol. 2016; 9:11466–72.
31. Zhou Q, Chen F, Zhao J, Li B, Liang Y, Pan W, Zhang S, Wang X, Zheng D. Long non-coding RNA PVT1 promotes osteosarcoma development by acting as a molecular sponge to regulate miR-195. Oncotarget. 2016; 7:82620–33. https://doi.org/10.18632/oncotarget.13012.
32. Li W, Xie P, Ruan WH. Overexpression of lncRNA UCA1 promotes osteosarcoma progression and correlates with poor prognosis. J Bone Oncol. 2016; 5:80–85. https://doi.org/10.1016/j.jbo.2016.05.003.
33. Cong M, Li J, Jing R, Li Z. Long non-coding RNA tumor suppressor candidate 7 functions as a tumor suppressor and inhibits proliferation in osteosarcoma. Tumour Biol. 2016; 37:9441–50. https://doi.org/10.1007/s13277-015-4414-y.
34. Ma B, Li M, Zhang L, Huang M, Lei JB, Fu GH, Liu CX, Lai QW, Chen QQ, Wang YL. Upregulation of long non-coding RNA TUG1 correlates with poor prognosis and disease status in osteosarcoma. Tumour Biol. 2016; 37:4445–55. https://doi.org/10.1007/s13277-015-4301-6.
35. Liu C, Lin J. Long noncoding RNA ZEB1-AS1 acts as an oncogene in osteosarcoma by epigenetically activating ZEB1. Am J Transl Res. 2016; 8:4095–105.
36. Xia WK, Lin QF, Shen D, Liu ZL, Su J, Mao WD. Clinical implication of long noncoding RNA 91H expression profile in osteosarcoma patients. Onco Targets Ther. 2016; 9:4645–52. https://doi.org/10.2147/OTT.S103376.
37. Yang Z, Li X, Yang Y, He Z, Qu X, Zhang Y. Long noncoding RNAs in the progression, metastasis, and prognosis of osteosarcoma. Cell Death Dis. 2016; 7:e2389. https://doi.org/10.1038/cddis.2016.272.
38. Chen R, Wang G, Zheng Y, Hua Y, Cai Z. Long non-coding RNAs in osteosarcoma. Oncotarget. 2017; 8:20462–75. https://doi.org/10.18632/oncotarget.14726.
39. Altman DG, McShane LM, Sauerbrei W, Taube SE. Reporting Recommendations for Tumor Marker Prognostic Studies (REMARK): explanation and elaboration. PLoS Med. 2012; 9:e1001216. https://doi.org/10.1371/journal.pmed.1001216.
40. McShane LM, Altman DG, Sauerbrei W, Taube SE, Gion M, Clark GM, and Statistics Subcommittee of the NCI-EORTC Working Group on Cancer Diagnostics. REporting recommendations for tumor MARKer prognostic studies (REMARK). Nat Clin Pract Urol. 2005; 2:416–22. https://doi.org/10.1038/ncponc0252.
41. Moher D, Liberati A, Tetzlaff J, Altman DG, and PRISMA Group. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. J Clin Epidemiol. 2009; 62:1006–12. https://doi.org/10.1016/j.jclinepi.2009.06.005.
42. Pak K, Suh S, Kim SJ, Kim IJ. Prognostic value of genetic mutations in thyroid cancer: a meta-analysis. Thyroid. 2015; 25:63–70. https://doi.org/10.1089/thy.2014.0241.
43. Kim YH, Goh TS, Lee CS, Oh SO, Kim JI, Jeung SH, Pak K. Prognostic value of microRNAs in osteosarcoma: A meta-analysis. Oncotarget. 2017; 8:8726–37. https://doi.org/10.18632/oncotarget.14429.
44. Parmar MK, Torri V, Stewart L. Extracting summary statistics to perform meta-analyses of the published literature for survival endpoints. Stat Med. 1998; 17:2815–34. https://doi.org/10.1002/(SICI)1097-0258(19981230)17:24<2815::AID-SIM110>3.0.CO;2-8.
45. Tierney JF, Stewart LA, Ghersi D, Burdett S, Sydes MR. Practical methods for incorporating summary time-to-event data into meta-analysis. Trials. 2007; 8:16. https://doi.org/10.1186/1745-6215-8-16.
46. Egger M, Davey Smith G, Schneider M, Minder C. Bias in meta-analysis detected by a simple, graphical test. BMJ. 1997; 315:629–34. https://doi.org/10.1136/bmj.315.7109.629.
47. Rosenthal R. The “File Drawer Problem” and Tolerance for Null Results. Psychol Bull. 1979; 86:638–41. https://doi.org/10.1037/0033-2909.86.3.638.
48. Duval S, Tweedie R. Trim and fill: A simple funnel-plot-based method of testing and adjusting for publication bias in meta-analysis. Biometrics. 2000; 56:455–63. https://doi.org/10.1111/j.0006-341X.2000.00455.x.
49. Jennions MD, Møller AP. Publication bias in ecology and evolution: an empirical assessment using the ‘trim and fill’ method. Biol Rev Camb Philos Soc. 2002; 77:211–22. https://doi.org/10.1017/S1464793101005875.
50. Mayo-Wilson E, Imdad A, Herzer K, Yakoob MY, Bhutta ZA. Vitamin A supplements for preventing mortality, illness, and blindness in children aged under 5: systematic review and meta-analysis. BMJ. 2011; 343d5094. https://doi.org/10.1136/bmj.d5094.
