As is shown in Fig. 1, 1099 literatures were obtained from online databases PubMed, Embase, Web of science (WOS) and Cochrane library. After removing the duplicates, abstract, review, case report, meta-analysis, studies which were not written in English and unrelated researches, 86 articles were subsequently full-text review. Among these, 71 articles were further to removed according to these criteria: studies not on patient (n = 2), studies without survival data (n = 45), survival data from TCGA (n = 5), multiple miRNAs (n = 1), or insufficient data (n = 19). Eventually, a total of 15 studies from 14 articles were included.(Liu et al., 2014; Meng et al., 2017; Gao et al., 2015; Organista-NAVA et al., 2015; Zhao et al., 2015; Wang et al., 2016; Mori et al., 2016; Kerimis et al., 2017; Dong et al., 2018; Su et al., 2018; Pan et al., 2018; Zhou et al., 2018; Liu et al., 2018; Yan et al., 2019) (Table 1). The overall sample size is 1518 patients coming from 25 to 247 which from 4 countries. Among these studies, several types of cancer include lung cancer (n = 3)[33–35], hepatocellular carcinoma (n = 2)[36, 37], colorectal cancer (n = 2)[29, 38], nasopharyngeal carcinoma (n = 2)[39, 40], osteosarcoma (n = 1)[41], ALL (n = 1)[42], AML (n = 1)[42], advanced gastric cancer (n = 1)[28], esophageal cancer (n = 1)[43], head and neck squamous cell carcinoma (n = 1)[44]. As for OS, RFS and DFS, there were seven studies directly provide HRs and its 95% CI [28, 29, 35–39]. In addition, the remaining eight studies only provided Kaplan-Meier curves[28, 33, 34, 40–44]. All studies measured the miR-24-3p expression level by qRT-PCR (quantitative real-time polymerase chain reaction).
3.1. The association between miR-24-3p expression levels and the overall survival (OS)
Ten enrolled articles including eleven studies and 1212 patients were used to investigate the correlation between miR-24-3p expression levels and the OS by using log rank tests and presented the data of univariate. Generally, a significant correlation between miR-24-3p levels and OS (HR = 1.609, CI: 1.291–2.004, Figure. 2a). However, an obvious heterogeneity was also observed (I2 = 85.20%, P = 0.000, Table 2). Hence, the random effects model was followed in succession but the significance was disappeared (HR = 1.507, CI: 0.810–2.803, Table 2), indicating that the heterogeneity significantly influenced the consequence.
In order to explore the source of the heterogeneity, subgroup analyses were applied by factors including population (Asian(Chinese) and Non-Asian), sample size (≥ 100 and < 100), NOS scores (≥ 8 and < 8), specimen (tissue and non-tissue) tumor category 1 (solid tumor and non-solid tumor), tumor category 2 (digestive system and non-digestive system) and tumor (esophageal cancer, osteosarcoma, lung cancer, gastric cancer, colorectal cancer, ALL, AML and hepatocellular carcinoma). As a consequence, the heterogeneity was controlled successfully in six subgroups and all them have significant correlations: 1). The subgroup of non-Asian (HR = 2.615, CI: 1.668–4.099; I2 = 0.000%, P = 0. 693). 2). The specimen derived from non-tissue (HR = 2.399, CI: 1.659–3.470; I2 = 0.000%, P = 0.949). 3) The sample size greater than or equal to 100 (HR = 2.779, CI:2.051–3.766; I2 = 0.000%, P = 0 .873). 4) The patients of hematologic tumor (HR = 2.425, CI: 1.491–3.944; I2 = 0.000%, P = 0.751). 5). The patients of hepatocellular carcinoma (HR = 2.607, CI: 1.756–3.871; I2 = 0.000%, P = 0.637). and 6). The patients of lung cancer (HR = 3.274, CI: 1.422–7.539, I2 = 0.000%, P = 0.698). In addition, significant correlations are also observed in the study of NOS score less than 8 by random effects model, which were consistent with the significance of the results by fixed effects model (Table 2). Moreover, significant correlations were observed between miR-24-3p expression levels and OS in the studies with the population derived from Asian(Chinese) (HR = 1.381, CI: 1.219–2.004), solid tumor (HR = 1.448, CI: 1.131–1.852), digestive system (HR = 1.705, CI: 1.291–2.253) and non-digestive system (HR = 1.461, CI: 1.021–2.090) by fixed effects model, while there were no significances identified in these groups when the random effects model was applied (Table 2). For patients of Colorectal cancer, the prognostic value of miR-24-3p expression levels to the OS was completely different (Kerimis D et al.[38] HR = 4.070, CI: 1.25–13.19; Gao Y et al.[29] HR = 0.285, CI: 0.139–0.584). Due to insufficient data, the consequence was lack of efficiency and the heterogeneity was also significant (I2 = 93.00%, P = 0.000). Therefore, more relevant studies are required to perform the analysis. Built on the above consequences, meta regression was further used, but there was no meaningful contribution identified to impact on the heterogeneity (Table 2). Subsequently, the sensitivity analysis was performed, but there was also no positive consequence (Figure. 2c). Next, funnel plots, Begg’s test and Egger’s test were implemented to assess the potential publication bias and two studies as the outliers were identified eventually (Figure. 2d) (Liu et al.[41] and Gao et al.[29] ). After removing them, dramatically decline of the heterogeneity was observed (I2 = 34.60%, P = 0.141) in the overall analysis, and the significance of the prognostic effects of miR-24-3p expression was still obvious (Figure. 2b).
3.2. The independent role of miR-24-3p expression levels as a prognostic indicator
Five studies containing 775 patients implemented the cox multivariate regression to assess the prognostic value of miR-24-3p expression levels in carcinoma patients by adjusting other factors. The significant correlation of miR-24-3p expression levels to the OS (HR = 2.384, CI: 1.813–3.134) was observed by fixed effects model. However, the heterogeneity was relatively obvious (I2 = 82.30%, P = 0.000, Table 3) and the significance was vanished by random effects model (HR = 1.994, CI: 0.991–4.015). Homoplastically, Subgroup analyses were applied to reduce the heterogeneity. As a result, the homogeneity was reached within the studies of sample size greater than or equal to 100 (I2 = 0.000%, P = 0.861), NOS less than 8 (I2 = 45.50%, P = 0.176) and the patients of hepatocellular carcinoma (I2 = 45.50%, P = 0.176). And the significant association was identified between miR-24-3p expression levels and OS with the sample size greater than 100 (HR = 3.369, CI: 2.414–4.701), NOS less than 8 (HR = 3.041, CI: 2.150–4.300) and the patients of hepatocellular carcinoma (HR = 3.041, CI: 2.150–4.300). In addition, the significant correlations were identified between miR-24-3p expression levels to the OS in the population from Asian(Chinese) (HR = 2.373, CI: 1.813–3.134), the specimen derived from tissue (HR = 2.448, CI: 1.804–3.323) and NOS larger than or equal 8 by fixed effects model, which become to no significance within those subgroups by random effects model (Table 3). For patients of Colorectal cancer (HR = 0.752, CI: 0.212–0.978), the prognostic value of miR-24-3p expression levels to the OS was also opposite (Kerimis D et al.[38] HR = 2.60, CI: 0.780–8.660; Gao Y et al.[29] HR = 0.456, CI: 0.394–1.434). Thus, more pertinent studies are required to perform the analysis. Similarly, there was no noteworthy contribution identified to greatly influence the variation of HR by meta regression (Table 3). But the sensitivity analysis suggested that Gao et al.[29] has significant impact on the result (Figure. 4c). The heterogeneity was vanishing (I2 = 0.000%, P = 0.591, Figure. 4b)by removing this outlier and the correlation of miR-24-3p expression levels to the OS was also significant (HR = 3.039, CI: 2.268–4.074,Figure. 4b). Finally, funnel plots, Begg’s test (P = 0.734) and Egger’s test (P = 0.460) indicated that there was no bias. But, the number of enrolled studies was few, more data are needed to reinforce this result.
3.3. The correlation of miR-24-3p expression levels to the RFS /DFS
Except OS as a prognostic indicator, RFS and DFS are also be accepted as an evaluation criterion. Here, four studies reported RFS including 393 patients applied log rank tests, while only one also utilized cox multivariate regression. After pooling the HR, we observed a significant association between miR-24-3p expression levels to the RFS of log rank tests (HR = 2.315, CI: 1.491–3.594, figure. 5a) by fixed effects model. However, the heterogeneities were quite obvious (I2 = 66.70%, P = 0.290, Table 4). The random effects model was further implemented but the significance was disappeared (HR = 1.814, CI: 0.741–4.440), indicating that the heterogeneity influenced the consequences significantly. Furthermore, owing to limited number of statistics from cox multivariate regression, the sensitivity analysis and publication bias were only applied to analysis with data extracted from log rank tests. The sensitivity analysis result indicated that no studies had significant influence on the consequent (Figure. 5c). However, the investigation of potential publication bias identified an outlier (Figure. 5d, Wang S et al.[40]). After deleting this study, the heterogeneity was obvious declined (I2 = 45. 30%, P = 0.161) and the significance of correlation between miR-24-3p expression levels and the RFS was not altered (HR = 2.575, CI: 1.642–4.029, Figure. 5b). Due to the limit included studies, more data are needed in order to enhance the result. In addition, there were only two studies containing 226 patients revealed the DFS statistics and almost no heterogeneity in both log rank tests and cox multivariate regression (I2 = 3.600%, P = 0.309, I2 = 0.000%, P = 0.330, respectively, Table 4) by used a fixed effects model. We also observed significant strong correlation between miR-24-3p expression levels to the DFS of both log rank tests (HR = 2.361, CI: 1.390–4.012) and cox regression tests (HR = 2.313, CI: 1.315–4.067) by fixed effects model.
3.4. Correlations between miR-24-3p levels and clinicopathological features among various carcinomas
Six studies containing 536 patients investigated the correlation of miR-24-3p expression levels to different clinical characteristics. As showed in Table 5, miR-24-3p expression levels were significant correlation with tumor size (OR = 1.655, CI: 1.124–2.437) by the fixed effects model with lesser heterogeneity (I2 = 37.50%, P = 0.184). In addition, there were no significance identified in the correlation between age (OR = 0.684, CI: 0.357–1.310), gender (OR = 1.286, CI: 0.758–2.107), lymph node metastasis (OR = 1.591, CI: 0.758–3.339) or TNM stage (OR = 1.437,CI: 0.959–2.154) with the expression levels of miR-24-3p. There were no heterogeneity in the analysis of age (I2 = 0.000%, P = 0.525) and gender (I2 = 0.000%, P = 0.842), but the heterogeneity of lymph node metastasis and TNM stage were obviously (I2 = 70.90%, P = 0.064; I2 = 85.50%, P = 0.000, respectively). In order to decrease the heterogeneity, sensitivity analysis and publication bias were further investigated to each of them. As a result, an outlier was found (Liu et al.[36]) in the TNM stage. After removing the outlier, the heterogeneity was dramatically decreased from 85.50–0.000% and the associations between high miR-24-3p expression levels to advanced TNM stage were significant (OR = 2.328, CI: 1.490–3.637). (Figure. 6). Moreover, there was no potential publication bias about TNM stage by funnel plot, Begg’s test (P = 0.086) and Egger’s test (P = 0.734). For the analyze of lymph node metastasis, there were only two studies and have obvious opposite result (Pan et al.[33], OR = 2.974, CI: 1.101–8.037; Zhou et al.[35], OR = 0.725, CI: 0.238–2.208). Due to insufficient data, the consequence would be lack of efficiency and the reasons of heterogeneity were unacceptable. Thus, more pertinent studies are required to perform the analysis.