Prognostic significance of cyclin D3 expression in malignancy patients: a meta-analysis

All procedures mentioned below were conducted based on a standard guideline for meta-analysis involving cancer biomarker prognosis trials [36, 37]. Two researchers performed each step individually, while any disagreement was solved by group conference. Databases of EMBASE, PubMed and Web of Science were independently searched by two researchers to acquire the related literatures involving the prognosis significance of cyclin D3 expression among malignancy sick personnel. For the sake of enhancing sensitivity of our search, not only free-text words but also MeSH terminology were utilized in current meta-analysis. The search strategy was: “CCND3 or cyclin D3” AND “neoplasms or neoplasias or tumors or cancers or carcinoma or malignancies or malignant neoplasms” AND “prognoses or prognostic factors or prognostic or prognosis or survival or outcome”. And related references of searching concerned literatures were also screened to identify potentially eligible literatures.

Researches that complied with the undermentioned criteria were eventually enrolled: (1) Patients were pathologically diagnosed with any type of malignancy. (2) The expression levels of cyclin D3 were identified in tissues samples. (3) Patients were classified into negative and positive expression or low and high expression group in line with the cyclin D3 of expression levels, the connection between expressing level of cyclin D3 and survival results was examined. (4) Hazard ratios (HR) and their 95% confidence intervals (CI) for survival times were computed by included articles which can provide enough data or survival curves. (5) Officially published and English-written literatures until July 2018. The eliminated criteria as follows: 1. Duplicated articles; 2. Reviews, laboratory articles, case-reports and conference abstracts; 3. Insufficient data about survival analysis.

Two investigators extracted related data respectively and came to an agreement on the following items. Original data of elementary demographic characteristics (year of publication; authors of article; region; the category of carcinoma; detection method; cyclin D3 level; sample size; age; follow-up duration; Newcastle–Ottawa Scale score (NOS); cut-off value and endpoints) were exhaustively extracted from included literatures involving Kaplan–Meier curves, test words and tables. In term of endpoints, overall survival (OS), progression-free survival (PFS), recurrence-free survival (RFS), disease free survival (DFS) and disease specific survival (DSS) were considered as terminal events. For the purpose of evaluating the effect of the expression level of cyclin D3 for cancer patients in prognosis, HR was adopted via abiding by a methodology recommended previously [38]. Furthermore, original data was also obtained by contacting the authors of the included literatures.

Two investigators individually assessed qualities of all enrolled researches by utilizing the Newcastle–Ottawa Scale. The critical scale is divided into three categories consisting of outcomes, selection of subjects and comparability of trial groups to assess a study. We regarded the included studies with at least six score as high-quality in methodology. And if the scores were less than 6, those articles were considered as low-quality studies.

Our quantitative calculation was conducted based on Stata Software 14.0. We applied pooled HRs (high/low) along with its related 95% CIs to evaluate the association between the prognostic value and the expression levels of cyclin D3 in different malignancies. By utilizing Cochran’s Q and I² statistics, the heterogeneity of enrolled literatures can be evaluated precisely. Additionally, we regarded an I² value larger than 50% or a p value no more than 0.10 as statistically significance. An insignificant heterogeneity (p > 0.10, I² < 50%) was changed via the fixed-effects model for analysis. On the contrary, we would select the random-effects model. In order to explore the source of heterogeneity, we also performed subgroup analysis and meta-regression. Furthermore, sensitivity analysis was implemented to confirm the steadiness of collected results. Finally, we assessed publication bias by means of utilizing both Begg’s and Egger’s test. What’s more, if the p value is no more than 0.05, the results above all can be regarded as statistical significance.

Eventually, we selected 13 studies [39‐51] from the initial retrieved 276 literatures consisting of 16 cohorts. The specific selection flow chart of each steps was illustrated in Fig. 1. In addition, the entire sample size of included studies ranged from 66 to 391, which added up to 2395 participants. What’s more, the range of follow-up duration in all enrolled cohorts was from 30 to 168 months. In term of various types carcinomas, there were three different kinds of tumors that were the most including breast neoplasms (n = 2), urinary bladder neoplasms (n = 2) and lymphoma (n = 3). Among these studies, OS (n = 11), DFS (n = 2), RFS (n = 2), FRS (n = 2) and DSS (n = 1) were regarded as survival outcome. When it comes to the detection methods for analyzing the expression of cyclin D3, immunohistochemistry (IHC) was the most frequent method, following by polymerase chain reaction (PCR) and the rarest was western blot (WB). Finally, the values of cut-off were varied from each study owing to the various definitions for cut-off. Further details about baseline features were recorded in Table 1.

There were eleven studies exploring the association between aberrantly expressing cyclin D3 with OS in this meta-analysis. At the same time, we applied random-effect model to reckon the pooled HR. And result demonstrated that higher expression level of cyclin D3 was significant correlated to reduction of OS among malignancy patients (HR 1.88; 95% CI 1.31–2.69, p = 0.001) (Fig. 2). Owing to obvious heterogeneity from all included studies (I² = 76.2%, p < 0.001), subgroup analysis was further performed by factors of type of malignancy (lymphoma or non-lymphoma), sample size (more than 100 or less than 100), access of HR (directly or indirectly), follow-up duration (over 100 or less than 100 months) and the quality of enrolled literatures (NOS scores < 7 or ≥ 7) to investigate sources of heterogeneity (Fig. 3a–f). Consequently, our outcomes of subgroup analysis demonstrated that relationship between cyclin D3 redundancy and worse OS of malignancy patients remained notable except for the subgroup of less than 100 participants (p = 0.338) (Table 2). To further explore the source of heterogeneity, we performed meta-regression by the covariates including above factors. But, the results of meta-regression did not reveal p values no more than 0.05 in above covariates, which indicated that all above factors were not the sources of heterogeneity (Table 2).

Additionally, the prognosis role of the expression levels of cyclin D3 in two kinds of cancers was assessed systemically. Our outcomes suggested that elevated cyclin D3 level implicated an unfavorable OS in lymphoma (HR 3.72; 95% CI 2.18–6.36, p < 0.000) (Fig. 4a). Nevertheless, regarding breast neoplasms, our results revealed that no significant relationship between the cyclin D3 redundancy with OS was obtained (HR 2.12; 95% CI 0.76–5.91, p = 0.149) (Fig. 4b).

Among the included studies, two researches estimated the relevance between cyclin D3 expression level with DFS, RFS and PFS, respectively. And only one study including three cohorts evaluated the connection between the expression level of cyclin D3 with DSS. Among this meta-analysis, we showed that cyclin D3 increasingly expressing had significantly worse outcome in worse DSS (HR 2.68; 95% CI 1.35–5.31, p = 0.005) (Fig. 5d). However, no matter how high or low expression level of cyclin D3, there existed no differentiation in DFS (HR 2.65; 95% CI 0.83–8.46, p = 0.099) (Fig. 5a), RFS (HR 2.86; 95% CI 0.82–9.96, p = 0.099) (Fig. 5b) and PFS (HR 5.24; 95% CI 0.46–60.25, p = 0.184) (Fig. 5c). Moreover, we did not carry out the subgroup analysis thanks to the finite numbers of included trials.

In order to assess the impacts of single study on the total outcomes, sensitivity analysis was conducted. With respect to OS, our result of sensitivity analysis revealed that the outcomes originating from Moller et al. and Levidou et al. influenced consequences remarkably, demonstrating that the critical source of heterogeneity was likely to come from the above studies. The list of pooled HRs and 95% CIs after excluding single study one by one indicated robustness of our results (Fig. 6a). Furthermore, with regard to DFS (Fig. 6b), RFS (Fig. 6c), PFS (Fig. 6d) and DSS (Fig. 6e) our sensitivity analysis identified that all selected studies influenced outcomes greatly, which suggested that the outcomes of DFS, RFS, PFS and DSS were not stable because of the limited number of studies included in each analysis. Thus, more and more related studies were needed to explore the effects of cyclin D3 on DFS, RFS, PFS and DSS in human malignancy.

By using Begg’s test and Egger’s test, we systemically assessed publication bias of all above included studies. The result of Begg’s test (p = 0.273) (Fig. 7a) and Egger’s test (p = 0.547) (Fig. 7b) about OS revealed that there existed no significant publication bias among enrolled documents. In terms of DFS, RFS, PFS and DSS, we didn’t perform the publication bias just because of the small amount of selected literatures, no more than 10 in each analysis.