Augmented expression of Ki-67 is correlated with clinicopathological characteristics and prognosis for lung cancer patients: an up-dated systematic review and meta-analysis with 108 studies and 14,732 patients

Background Lung cancer ranks as the leading cause of cancer-related deaths worldwide and we performed this meta-analysis to investigate eligible studies and determine the prognostic effect of Ki-67. Methods In total, 108 studies in 95 articles with 14,732 patients were found to be eligible, of which 96 studies reported on overall survival (OS) and 19 studies reported on disease-free survival (DFS) with relation to Ki-67 expression in lung cancer patients. Results The pooled hazard ratio (HR) indicated that a high Ki-67 level could be a valuable prognostic factor for lung cancer (HR = 1.122 for OS, P < 0.001 and HR = 1.894 for DFS, P < 0.001). Subsequently, the results revealed that a high Ki-67 level was significantly associated with clinical parameters of lung cancer including age (odd ratio, OR = 1.246 for older patients, P = 0.018), gender (OR = 1.874 for males, P < 0.001) and smoking status (OR = 3.087 for smokers, P < 0.001). Additionally, significant positive correlations were found between Ki-67 overexpression and poorer differentiation (OR = 1.993, P = 0.003), larger tumor size (OR = 1.436, P = 0.003), and higher pathologic stages (OR = 1.867 for III-IV, P < 0.001). Furthermore, high expression of Ki-67 was found to be a valuable predictive factor for lymph node metastasis positive (OR = 1.653, P < 0.001) and advanced TNM stages (OR = 1.497 for stage III-IV, P = 0.024). Finally, no publication bias was detected in any of the analyses. Conclusions This study highlights that the high expression of Ki-67 is clinically relevant in terms of the prognostic and clinicopathological characteristics for lung cancer. Nevertheless, more prospective well-designed studies are warranted to validate these findings. Electronic supplementary material The online version of this article (10.1186/s12931-018-0843-7) contains supplementary material, which is available to authorized users.


Background
Lung cancer is the most frequent diagnosed malignant neoplasms, and it was the first cause of cancer death in 2016 globally [1]. In the United States, lung cancer accounted for 27% of all cancer deaths in 2016. Non-small cell lung cancer (NSCLC), accounting for over 80% of all lung cancers, is the major cause of death worldwide [2]. Although the treatment of NSCLC patients includes surgery, radiotherapy and chemotherapy, the progress in lung cancer treatment is still slow, for which the 5-year relative survival is currently 18%. Diagnosis at an advanced stage is the major reason for this low survival rate [3]. Several prognostic factors were well characterized in lung cancer including sex, age, loss of weight, TNM stage, LDH, neutrophilia, haemoglobin as well as serum calcium [4]. Importantly, The IASLC Lung Cancer Staging Project in 2015 also carried out the 8th edition of the anatomic classification of lung cancer, which redefined the tumor-size cut-points in TNM stage for the lung cancer patients. The prognostic value of reclassification of tumor size were confirmed in 70,967 nonsmall-cell lung cancer patients from 1999 to 2010 [5]. To improve the survival of lung cancer patients, the choice of targeted treatments is increasingly being based on oncogenic drivers including ALK rearrangements, KRAS and epidermal growth factor receptor (EGFR) mutations [6,7]. Additionally, BRAF mutations represent promising new therapeutic targets for lung cancer [8]. Likewise, several driver biomarkers also shed new light on the target treatment for lung cancer patients such as Her2 [9], NUT [10], DDR2 [11], FGFR1 [12], and PTEN [13]. Furthermore, several new molecular targets been highlighted in lung cancer, including ROS1 fusions [14], NTRK1 fusions [15] and exon 14 skipping mutations [16]. Recently, the checkpoint inhibitors targeting programmed death protein 1 (PD-1) have shown for durable clinical responses in NSCLC patients with advanced stage [17]. The immunomodulatory monoclonal antibodies against cytotoxic T-lymphocyte-associated antigen 4 (CTLA-4) also present the promising results for the treatment of advanced -stages lung cancer patients. However, due to the complex molecular mechanism of lung cancer, the identification of biomarkers in a large proportion of lung cancer patients is still required for a deeper understanding of the underlying epigenetic heterogeneity, and this will benefit the discovery of targeted therapies against lung cancer.
Ki-67, encoded by the MKI67 gene, is expressed throughout the cell cycle in proliferating but absent in in quiescent (G0) cells [18]. Ki-67 appears in the middle or late G1 stage, and then its expression increases through the S and G2 stage until it reaches a peak during the M stage [19]. The high expression of Ki-67 may contribute to aggressive and infiltrative growth of lung squamous carcinoma (SQC), cervical SQC and laryngeal SQC [20][21][22]. In addition, overexpression of Ki-67 has been positively associated with lymph node metastasis in gastric carcinoma and breast cancer [23,24]. Ki-67 expression has been reported to be associated with a poor outcome in many malignancies including prostate, bladder and breast cancer [25][26][27][28][29].
Although two meta-analyses have previously reported that Ki-67 could be a possible indicator of short-term survival in lung cancer patients [30,31], studies on a larger number of lung cancer patients and more reliable evidence are still needed to confirm the prognostic and clinicopathological role of Ki-67 for patients with lung cancer. Thus, our investigation aimed to evaluate the prognostic value and clinicopathological significance of Ki-67 in lung cancer patients via the review of previously published articles.

Study selection and characteristics
As shown in Fig. 1, two investigators read the full text and considered 108 studies in 95 articles [18,25, consisting of 14,732 patients as applicable. The baseline characteristics of the included articles are indicated in Table 1. Ninety-six studies included data regarding OS, and nineteen studies included data regarding DFS. The number of cohorts of each study ranged from 32 to 778. In terms of the study region, 62 studies were from Asia, 30 were from Europe, and 15 were from America. In total, 104 studies consisted of 14,596 NSCLC patients: of these, 28 studies reported on ADC patients, and 10 studies reported on SQC patients. All the studies detected Ki-67 reactivity using IHC.  Fig. 5) was observed among patients with high expression of Ki-67, while the heterogeneity using the random effects model was obvious (I 2 = 78.30%, P < 0.001).
To investigate the source of heterogeneity, subgroup analyses of publication year, region, histological type, sample size, cut-off value of Ki-67 and estimated method for HR determination were performed. From the subgroup analysis of OS, no heterogeneity was found in the small cell lung cancer (SCLC) group (I 2 = 22.30%, P = 0.277). Next, a reduction in heterogeneity was observed after performing subgroup analysis of DFS according to the study region, especially in the studies from America (I 2 = 8.40%, P = 0.351) and Asia (I 2 = 25.60%, P = 0.179). As indicated in the subgroup of cutoff value, there was a low degree of heterogeneity in both Ki-67 low expression (I 2 = 25.30%, P = 0.196) and high expression groups (I 2 = 19.40%, P = 0.287). Furthermore, we also performed meta-regression analysis to explore the original of the heterogeneity in the studies. Consistent with the subgroup analysis, the results revealed that the regions and cut-off values might be the potential bias for the heterogeneity (P = 0.017 and P = 0.022, respectively). Altogether, we concluded that the different regions and inconsistent cut-off values might have contributed to the heterogeneity in the results of analyses for DFS. The regression also revealed that the heterogeneity originated from regions and inconsistent cut-off values ( Table 4).

Publication bias
To identify potential publication bias, Begg's test and funnel plots were used. No publication bias was found in the analysis for OS (p = 0.444, Fig. 6a) and DFS (P = 0.246, Fig. 6b). Moreover, there was no publication bias   Figure S4, Additional file 5: Figure S5, Additional file 6: Figure S6 and Additional file 7: Figure S7).

Discussion
As previously mentioned, there are two meta-analyses showing that high expression of Ki-67 predicts worse prognosis in lung cancer patients [30] and early-stage NSCLCs [31].  Compared to the previous meta-analysis, our meta-analysis included results on all subtypes of lung cancer and represented broader ethnicity; in addition, subgroups were classified according to region, cut-off value, number of patients and histological type. With the inclusion of high-quality studies and a larger number of patients, the results derived from our study are more convincing. Ki-67 is present in the active phases of the cell cycle (G1, S, and G2), as well as during mitosis, but it is not expressed in the G0 phase. Thus, it has become an excellent operational marker for the estimation of the proportion of proliferative cells in a given cell population [125]. Our study demonstrated that Ki-67 expression was lower in ADC compared with that in SQC, suggesting it is a useful biomarker for distinguishing ADC from SQC. Consisted with our result, Ki-67 was also revealed to be higher in ADC than in SQC in the same stage, duo to the different tumor biology of histological subtypes in NSCLC [49,64]. The high-grade Ki-67 was proved to be significantly correlated with a more aggressive tumor infiltration patterns in lung SQC, Indicating the strong association between tumor invasiveness and cell proliferation [20]. It has been reported in the literature that inverted papillomas with high levels of Ki-67 also include squamous cell carcinomas. Regarding the SCLC, Vasudha Murlidhar et al. recently carried out a result that Ki-67 could contribute to the early detection of metastasis in circulating lung cancer cells.
Ki-67 was also demonstrated as a potential diagnostic factor for histopathological definition of SCLC [126]. Contrary to our study, previous study revealed that maternal cigarette smoking could dramatically decrease the expression of Ki-67 in cytotrophoblasts [127]. Interestingly, further study also found that Ki-67 was lower expressed in smokers and smokers with COPD compared to the non-smokers. The authors hypothesized that the permanent cellular damage might play a crucial role in the destruction of bronchiolar tissue [128]. Nevertheless, the mechanisms that govern how Ki-67 expression contribute to the tumorigenesis and progression of lung cancer remain to be unveiled.
The authors suspected that Ki-67 might affect cyclin-dependent kinase1 (CDK1), leading to the entry of inverted papilloma cells into the active phase of cell cycle(G1)and resulting in malignant transformation [129]. Interestingly, our study found that Ki-67 expression in male patients was significantly higher comparing with that in female patients. Previous reports have suggested that testosterone can promote the growth of cancer cells that express androgen receptors, which negatively regulates the Ki-67 level in lung cancer patients [130,131]. Many previous studies have also revealed that Ki-67 is significantly associated with histopathologic parameters in other tumor because of the correlation between proliferation and those parameters [132][133][134]. It was found that p53 regulated the p53and Sp1-dependent pathways, leading to the inhibition of Ki-67 promoter [135,136]. A recent study confirmed that there was a correlation between the specific Ki-67  [137]. Based on these results, we concluded that Ki-67 serves as a valuable indicator for the aggressiveness and prognosis of lung cancer.
Heterogeneity was significant in this meta-analysis. To eliminate the heterogeneity, subgroup analyses according to region, cut-off value, number of patients and histological type were carried out using random effects models. As a result, we revealed that the source of heterogeneity originated from the publication region and the cutoff value by the meta-regression analysis. Our meta-analysis was limited to publications in English or Chinese; nevertheless, the researchers typically tend to publish studies with negative results in local journals and in the native language of the study region. In addition, although detailed exclusion criteria were established to avoid duplication, our meta-analysis was not able to avoid the same patient cohorts in different publications. Other methodological factors might also affect heterogeneity, such as the antibody and cut-off value used in the study. Although anti-MIB-1 antibody is the most frequently used antibody in studies, most of the included studies stratified high and low levels of Ki-67 using a median value varying from 3 to 75%, which might have influenced the results. Several studies used the Ki-67 cut-off less than 10% to assess the prognostic impact of Ki-67 after surgical resection with curative intent in early-stages lung cancer patients. Most of the included studies used the median value of Ki-67 index as cutoff value, which could divide the patient into equally the group but did not reflect the clinical relevant use. A cut-off value of Ki-67 maximizing the hazard ratio across the groups could be used for the clinical management for the diagnosis and prognosis of lung cancers. More importantly, Multiple clinical laboratories have reported the Ki-67 cutoff values between 10 and 14% could be recommended as the gold standard identify the high risk of the survival outcome in cancers [138][139][140][141]. Additionally, microarrays were used in several studies, for which the sensitivity of assessment of Ki-67 expression is generally poor. Another possibility of bias may be related to the method of extrapolating the HR; the HR extracted from survival curves was less reliable than direct analysis of variance. Additionally, the subgroup analysis for different stages because most of the included studies recruited the lung cancer patients within three or more tumor stages. Thus, we could not classify the patients into the early stages and advanced stages for the subgroup analysis.

Conclusion
In conclusion, our meta-analysis demonstrated that high expression of Ki-67 is associated with worse prognosis and disease progression in lung cancer patients. Ki-67 can be an independent biological marker for predicting the prognosis of lung cancer patients. Subsequent studies are required to investigate the prognoses and clinical characteristics of lung cancer patients to confirm our findings.

Literature search and selection
The databases that we searched included PubMed, Web of Science, EMBASE, and Chinese datasets (WanFang, China National Knowledge Infrastructure and Chinese VIP) until June 1, 2017. The key words identifying the articles were as follows: (Ki-67 OR Ki67 OR MIB-1 OR "proliferative index" OR "proliferative activity" OR "mitotic index" OR "labeling index" OR "mitotic count" OR "proliferative marker" OR "mitotic figure" OR "mitotic activity") AND (Cancer OR carcinoma OR adenocarcinoma OR tumour OR tumor OR malignanc* OR neoplas*) AND (Lung OR pulmonary OR respiratory OR respiration OR aspiration OR bronchi OR bronchioles OR alveoli OR pneumocytes OR "air way").

Selection criteria
Publications were included if they met the following inclusion criteria: (1) the patients enrolled had been diagnosed with lung cancer; (2) the results for the study included the correlation between Ki-67 and overall survival (OS) or disease-free survival (DFS); (3) the samples used in the studies were human lung tissue, serum or sputum but not animals or cell lines; (4) the techniques used to measure the expression level of Ki-67 in cancer tissue or tumors of the patients were immunohistochemistry (IHC), PCR/ RT-PCR, ELISA or western blotting; (5) the study provided hazard ratios (HRs) and their 95% confidence intervals (CIs) or sufficient information for estimating these parameters; (6) the article was fully written in English or Chinese; and (7) the sample size was larger than 30. Studies were excluded if they met the following exclusion  (2) articles in languages other than English or Chinese; and (3) studies did not include the key information for survival analyses such as HRs and 95%CIs. To avoid data duplication, when the same patient cohort was reported in different publications or the same article was found in different journals, only the most recent and complete publication was included.

Data extraction and quality assessment
All the articles were independently reviewed and selected by two investigators. Discrepancies were resolved by discussion and arbitrated by a third investigator. The following information was extracted from each publication: first author's name, year of publication year, pathology type, tumor stage, number of patients, sample type, cut-off value of Ki-67, determination assay, method to extract HR and survival type. Additionally, we also obtained the clinicopathological characteristics of the lung cancer patients in the included studies including age (old/young), gender (male/female), histological type (adenocarcinoma/squamous carcinoma, ADC/SQC), smoking status (smoker/ non-smoker), differentiation (poor/well or moderate), pathologic stage (III-IV/I-II), tumor size (large/small), tumor stage (T3-4/T1-T2), metastasis (yes/no), lymph node (N1-Nx/N0), TNM stage (T3-4/T1-T2) and invasion (yes/no). Especially, the tumor stage was used to describe the size and extent of tumor. And the TNM stage were used to define the progression of cancer based on the size and extension tumor, lymphatic involvement and metastasis status. Based on the Newcastle Ottawa Scale (NOS) criteria [142], studies with NOS scores higher than 6 are considered high-quality studies, whereas those with NOS scores less than 5 are defined as low-quality studies. This study was strictly performed according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) [143] and the PRISMA checklist were also provided in the Additional file 8.

Statistical methods
HR and 95%CIs were used to measure the relationship between Ki-67 expression and prognosis of lung cancer patients. The most accurate determination was made when the study directly provided the HRs and 95%CIs. The multivariate HRs were calculated by using the Cox proportional hazards model, which could independently predict the survival outcome for the lung cancer patients. We preferentially chose the multivariate values when the study provided both univariate and multivariate HRs. If the above data were not available, we used the Engauge Digitizer version 4.1 to extract the survival rates from KM curves and estimated the HR according the method as Tierneyet al. described [144,145]. Moreover, we calculated the HR from the original survival data that the study provided using SPSS. An observed HR > 1 indicated worse prognosis for the lung cancer patients with high expression of Ki-67, if the 95%CIs for the overall HR was not 1, we considered the prognostic effect of Ki-67 on to be statistically significant. Odds ratio (OR) with 95%CIs were used to analyze the degree of association between Ki-67 level and clinicopathological characteristics. Heterogeneity was determined using the χ 2 and inconsistency (I 2 ) tests [146]. I 2 > 50% or P < 0.1 indicated substantial heterogeneity among the studies, in which case a random effects model was applied; otherwise, we utilized the fixed effects model. Then, subgroup analysis and meta-regression analysis were used to investigate any source of heterogeneity. Moreover, publication bias was assessed using Begg's test and funnel plots, and P-values < 0.05 indicated statistically significant publication bias [147].