The impact of metformin use on survival in prostate cancer: a systematic review and meta-analysis

Background Metformin has been implicated to reduce the risk of prostate cancer (PCa) beyond its glucose-lowering effect. However, the influence of metformin on prognosis of PCa is often controversial. Results A total of 13 cohort studies encompassing 177,490 individuals were included in the meta-analysis. Data on overall survival (OS) and cancer-specific survival (CSS) was extracted from 8 and six studies, respectively. Comparing metformin users with non-metformin users, the pooled hazard ratios (HRs) for OS and CSS were 0.79 (95% confidence interval [CI] 0.63–0.98) and 0.76 (95% CI 0.57–1.02), respectively. Subgroup analyses stratified by baseline charcteristics indicated significant CSS benefits were noted in studies conducted in USA/Canada with prospective, large sample size, multiple-centered study design. Five studies reported the PCa prognosis for recurrence-free survival (RFS) and metformin use was significantly associated with patient RFS (HR 0.74, 95% CI, 0.58–0.95). Methods Relevant studies were searched and identified using PubMed, Embase and Cochrane databases from inception through January 2017, which investigated associations between the use of metformin and PCa prognosis. Combined HRs with 95% CI were pooled using a random-effects model. The primary outcomes of interest were OS and CSS. Conclusions Our findings provide indication that metformin therapy has a trend to improve survival for patients with PCa. Further prospective, multi-centered, large sample size cohort studies are warranted to determine the true relationship.


INTRODUCTION
Biguanides, commonly known as metformin, are one type of the most widely prescribed drugs mainly to lower blood glucose for patients with type 2 diabetes. Experimental studies have shown that metformin has antineoplastic effects in several malignant tumors, including breast cancer, pancreatic cancer, and prostate cancer (PCa) [1][2][3].
Metformin has been implicated to restrain mitochondrial complex [1], reducing mitochondrial ATP production, leading to cellular energetic stress [3], which can activate AMPK, resulting in the inhibition of tumor growth through an anti-proliferative phenotype [3,4]. Metformin can also act as a chemosensitizer. In breast cancer xenograft models, metformin has been shown to enhance the effect of chemotherpy and prolong remission in breast cance cell line. In colon cancer cell lines, metformin

Meta-Analysis
can enhance the chemosensitivity of 5-fluorouracil and oxaliplatin [5,6]. Moreover, metformin has also been shown to improve survival in diabetic patients with advanced endometrial cancer and non-small cell lung cancer [7,8].
However, these results were controversial. Therefore, we updated the systematic review and meta-analysis to reappraise the prognostic value of metformin in PCa.

Description of the search and selection of studies
A total of 561 citations were identified for eligibility through the systematic literature search. After exclusion of duplicate publications and full text review of the relevant studies, A total of 13 cohort studies encompassing 177,490 individuals, with a mean sample size of 13,653 (range 250 to 105,245) were included in the quantitative synthesis. Data on overall survival (OS) and cancer-specific survival (CSS) were available from 8 and 6 studies, respectively [10][11][12][13][14][15][16][17][18][19][20][21][22] (Figure 1 and  Supplementary Tables 1-4). Table 1 provides the baseline characteristics of each study that met our inclusion criteria. All studies were published between 2010 and 2016 in English peerreviewed journals. Five of the included studies were population-based cohort studies and eight were hospitalbased cohort studies. Nine studies has retrospective designs, and four studies has prospective designs. Ten studies were performed in USA or Canada, two in Europe and one in Asia. Five studies involved single-center data, whereas eight were multi-center studies. Assessment of methodological quality by NOS yielded a mean score of 7 (range, 6 to 9), and 8 of 10 studies had a score of 7 or above ( Table 2). As shown in Figure 2A, the pooled hazard ratio (HR) for the OS comparing metformin use versus non-use was 0.79 (95% CI 0.63-0.98), and there was significant interstudy heterogeneity (I 2 = 79.5%, P < 0.001). The subgroup analysis limited study region to USA/Canada showed similar result (n = 6, HR 0.72, 95% CI 0.57-0.90). We also found that studies with retrospective design, sample size less than 10,000, hospital-based study, single center study, with patients including only diabetics and metformin use calculated as ever versus never use have similar results with the main analysis. However, due to the limited studies included in some subgroups, though the trend of the survival benefits were noted, significant differences were not reached (Table 3A). Figure 2B showed that the pooled HR for the CSS comparing metformin use versus non-use was 0.76 (95% CI 0.57-1.02), and there was significant inter-study We also find that studies with prospective design, larger sample size more than 10,000, population-based study and multiple center study have similar trends of survival benefits for metformin use with the main analysis. Due to the limited studies included in the main analysis and some subgroups, through the trend of the survival benefits were found, further large prospective studies need to be conducted to test this association (Table 3B). Five studies investigated the association between metformin use and recurrence-free survival (RFS), we found that metformin use was significant associated with improved RFS for PCa Patients (n = 5, HR 0.74, 95% CI 0.58-0.95).

Sensitivity analyses and publication bias
The tests for funnel plot asymmetry in OS and CSS subset indicated the absence of publication bias, which were further confirmed by Egger's test (P = 0.69 for OS, P = 0.32 for CSS), and Begg's test (P = 1.00 for OS, P = 0.26 for CSS). The adjusted estimates calculated using the trim-and-fill method were similar with the original analyses for both OS and CSS (Supplementary Table 5). We did not explore the publication bias for RFS due to the limited number of studies involved.

Principal findings of this study
This present systematic review and meta-analysis represents the most comprehensive review to date on the association between metformin use and PCa prognosis by including 13 cohort studies enrolling 177,490 individuals. Overall, we find that metformin intake has a trend to improve survival for patients with PCa in terms of OS, CSS and RFS. Significant CSS benefits were noted in studies conducted in USA/Canada with prospective, large sample size, multiple-centered study design.

Comparisons with previous studies
The result of this study is similar with that of two previous meta-analyses. The first meta-analysis by Stopsack et al [24] found metformin use was associated with improved OS and RFS for patients with PCa by meta-analysing 9 studies. By pooling 8 studies, Hwang et al [25] found that PCa patients who used metformin had RFS benefits compared with those who did not use metformin. However, due to small number of included studies and limited sample size, no statistical significance was found for other outcomes such as CSS. For the present meta-analysis, we have tried to explore the potential between-study heterogeneity by conducting subgroup analyses in terms of OS subset. Though no significant decrease in heterogeneity of the subgroups, we still could not exclude the potential heterogeneity from these origins. Moreover, the trim-and-fill method further confirmed the robustness of results for OS and CSS. However, it do add the implications that metformin could influence survival in specific individuals with PCa, not in others. We found that metformin use might have overall survival effects in selected patients and well-designed studies, such as in patients involving only diabetics and metformin use calculated as ever versus never use, etc.

Potential mechanisms
Several potential mechanisms for the antineoplastic action of metformin have been noted. Metformin, as an activator of AMP-activated protein kinase (AMPK), may play an important role in cancer metabolism. AMPK pathway is reported to inhibit mTOR signaling and result in fatty acid synthesis, inhibition of protein synthesis, and cell proliferation [26]. It has been reported that fatty acid synthase is overexpressed in PCa, breast cancer and pancreatic cancer, which is necessary for de novo fatty acid biosynthesis and malignant phenotype. AMPK activation can reduce the expression of fatty acid synthase and acetyl-CoA carboxylase, which diminishes the metabolization and growth of PCa cells [27]. Zadra et al [28] also suggested that suppression of de novo lipogenesis affected AMPK-mediated inhibition of PCa growth. In addition, metformin plays a role in cyclin-dependent kinase (CDK) induction of autophagy, cell cycle arrest, and apoptosis. Metformin can reduce the activity of cyclin D1, leading to the inhibition of PCa cell lines [29]. It has been vertified that the cyclin D1 pathway can serve as a regulator of androgendependent transcription and cell cycle progression in PCa cells [30].

Strengths and limitations of the study
There were several limitations in our study. First, the statistical analysis of publication bias was insufficiently powered due to the small number of included studies for OS (n = 8) and CSS (n = 6) subsets, although the results were adjusted by the trim-and-fill model. Secondly, the sensitivity analyses could not be carried out related to the tumor site, disease stage and follow-up period because of unavailability of these data from the included studies, and these factors can also affect the prognosis of PCa patients. Thirdly, the accuracy and precision of the summary estimates could be influenced by the different survival analysis approaches. Although most of the studies used multivariate Cox proportional hazards model, other studies did not report the statistical models [17,20], while another study did not utilize multivariate analysis [11]. In addition, adjustment variables between the included studies are not completely consistent. Fourthly, we were not able to contact the authors or sponsors of some studies to retrieve the data which were excluded from our analyses [12,20]. This might lead to publication bias for pooled estimates.
Several important strengths of our study are presented as follows. Firstly, we performed a comprehensive search of the relevant studies in several main databases without language, publication date or publication type (both full text and abstract) limits, enabling us to select the maximal number of suitable studies for analysis. Secondly, the large sample size including over 100,000 individuals enabled us to quantitatively assess the association between metformin use and PCa prognosis, making it the most powerful and comprehensive synthesis of the evidence on this issue to date. Thirdly, we performed appropriate subgroup analyses for some key study characteristics, such as the study design, study setting, and Newcastle-Ottawa scale (NOS) scale for study quality. Fourthly, we selected and cross-checked the identified studies, developed the data abstraction forms, abstracted the data and assessed the study quality at least by two independent authors to avoid subjectivity to the greatest extent, making the process of the systematic review more objectively.
In summary, our current systematic review and metaanalysis found that metformin was beneficial for survival in patients with PCa, although the true association still need further confirmation based on the existing evidence. Nevertheless, this report indeed provides a direction for clinicians in the treatment of PCa. In future, larger prospective cohort studies, or even randomized controlled trials with longer follow-up period are needed to confirm the associations between metformin intake and PCa survival.

Literature search
A search strategy in line with the preferred reporting items for systematic reviews and meta-analysis (PRISMA) statement was developed [31]. We performed systematic  Tables 1-3 present the above  three database search strategies by using the combinations  of following terms: 'metformin', 'biguanides', 'prostate',  'prostatic', 'cancer', 'carcinoma', 'mortality', 'prognosis', 'prognostic' and 'survival'. We also performed manual reference search of the reference lists from the initial identified relevant studies, reviews and meta-analysis. We restricted the publication language only to English language studies, given the fact that studies published in other languages were often not available for both authors and readers.

Study selection
Two authors (Liu and Chu) independently assessed the searched all the citations through the primary literature search, then identified the final relevant studies for eligibility. Agreement was reached for the discrepancies through discussion or by a senior author (Hao or Xu) if necessary. Studies were considered eligible for inclusion if the following criteria were met: prospective or retrospective cohort studies reported prognostic effects in PCa patients comparing metormin users with non-users, and survival estimates HRs/ risk ratios (RRs) with 95% CIs could be abstracted or calculated using the method reported by Parmar [32]. We used the most detailed or recent information for publications with overlapped data.

Data extraction
The characteristics of each study included were extracted including the first author, publication year, study region and design, study setting, hospital number involved, sample size, follow-up duration, survival endpoints, and HRs or RRs with corresponding 95% CIs and adjusted variables.

Quality assessment
Methodological quality assessment for each study included was performed by two authors (Liu and Chu) and was scored them using the NOS [33]. The two authors scored the study quality of reviewed studies independently, and reach a consensus value for each item.