Introduction

Polycystic ovary syndrome (PCOS) is the most common endocrine and metabolic disorder with principal characteristics of hyperandrogenism, chronic oligoanovulation, and polycystic ovary in reproductive-aged women [1]. These symptoms are usually accompanied by obesity, insulin resistance, and infertility [2–4]. In addition to being an important cause of infertility and exhibiting a higher risk of type 2 diabetes, dyslipidemia, hypertension, and cardiovascular diseases, PCOS is a growing public health concern [5–8]. Although the pathogenesis of PCOS remained unknown, previous studies showed that low adiponectin level was independently associated with insulin resistance or weight in PCOS patients, which suggested that adiponectin might contribute to the development of PCOS [9–13].

Adiponectin, as an important adipocytokine, is the most abundant secreted protein expressed exclusively in adipose tissue and plays a pivotal role in the regulation of insulin sensitivity and metabolism [14]. The major action of adiponectin is to increase insulin sensitivity by stimulating glucose uptake in the liver and muscle, decreasing hepatic gluconeogenesis, and promoting fatty acid β-oxidation in the skeletal muscle [15]. Moreover, adiponectin reduced insulin-induced progesterone and androstenedione production in bovine theca cells in vitro [16]. Indeed, circulating adiponectin levels are lower in obese subjects and are negatively correlated with insulin resistance and testosterone levels [17]. In addition, it has been reported that circulating adiponectin levels were decreased in patients with insulin resistance, type 2 diabetes, cardiovascular disease, and hypertension [17–19]. Therefore, the researchers speculated that circulating adiponectin decreases might play a role in the pathogenesis of PCOS, and devoted to this research [20–23].

Until recently, a number of studies have investigated the associations between adiponectin level and PCOS risk. However, results of these studies were conflicting rather than conclusive. To better understand whether adiponectin level was associative with PCOS, Toulis et al. [24] published a meta-analysis comparing circulating adiponectin levels in women with PCOS and non-PCOS controls matched for similar body mass index (BMI). This analysis revealed that adiponectin levels are lower in women with PCOS than in controls. However, during the last 5 years, many more relevant studies have been published and represented inconsistent results [11–13, 25–46]. As the previous meta-analysis included a limited number of studies, we conducted this updated meta-analysis to investigate the association between circulating adiponectin levels and PCOS from all the relevant studies, including all the newly published studies.

Methods

Search strategy

A comprehensive systematic electronic search was conducted in electronic databases PubMed, EMBASE, and Web of Science up to November 30, 2013. All searches were limited to human subjects without language restriction. The search strategy used the following MeSH terms: “adiponectin,” “adipokines,” “polycystic ovary syndrome,” and “hyperandrogenism.” A manual search of the reference lists of additional relevant articles was performed. Two reviewers (Xiamei Huang and Shan Li) independently completed the literature search, as well as screening of titles, abstracts, full-text articles, and evaluating the relevant articles on the basis of prespecified eligibility criteria. Any discrepancy was resolved by consultation to reach consensus with the third investigator (Xue Qin). Our meta-analysis was conducted according to the Meta-analysis of Observational Studies in Epidemiology guidelines [47].

Eligibility criteria

Eligible studies for this study were those studies that reported serum or plasma adiponectin levels in women with PCOS compared with healthy controls, providing total adiponectin means (M), and standard deviation (SD) or sufficient information to calculate adiponectin means and SD. We excluded literature reviews, letters to the editor, studies on animals or cell lines, studies without healthy control group, as well as studies about genetic variation in adiponectin-related genes. The studies were also excluded if studies enrolled patients with a disease other than PCOS, medication treatment, or pregnant patients. We tried to contact the authors to obtain the relevant information when the data was insufficient or there was doubt about the publications. The study, which included less than 20 subjects, could not be included in this meta-analysis. All case and control subjects in the included studies were matched the BMI and age.

Data extraction

Information was carefully extracted from all eligible studies by two investigators (Xiamei Huang and Shan Li) independently, using a standardized data extraction form. Any disagreements were resolved by discussion during a consensus meeting with a third investigator (Xue Qin). The following information was extracted from each included study: first authors’ name, year of publication, region of study population, total numbers of cases and controls, methods of adiponectin measurement, the type of blood sample, total adiponectin levels (means and SD or standard error, or 95 % confidence interval (95 % CI) of means or median and interquartile range) in cases and controls, the age and the means of BMI estimation. To retrieve the missing data, we also contacted the authors of primary studies.

Quality evaluation of literatures

Quality evaluation of studies was conducted independently by two reviewers (Xiamei Huang and Shan Li) according to the Newcastle-Ottawa Scale (NOS) [48]. The NOS tool contains nine items and scores ranged 0 to 9 (Table 1). The quality assessment of all the included studies that evaluated the association between circulating adiponectin levels and PCOS is showed in Table 2.

Table 1 Assessment of study quality
Full size table
Table 2 Study characteristics of the included studies of polycystic ovary syndrome and circulating adiponectin levels in meta-analysis
Full size table

Statistical analysis

Adiponectin levels in each study were extracted as mean ± SD. When the studies did not report the means and SD, we tried to communicate the authors for primary studies to retrieve the missing data. In case of no response or unavailability of the requested information, we transformed the reported standard error, or 95 % CI of means, medians, and ranges to means and SD using the formulas as described by Hozo et al. [49]. Homeostasis model assessment of insulin resistance (HOMA-IR) value was used as measures of insulin sensitivity. When not reported, missing HOMA-IR values were estimated according to reported mean glucose and insulin values, using the Oxford Diabetes Trials Unit calculator (http://www.dtu.ox.ac.uk).

Weighted mean differences (WMDs) together with the corresponding 95 % CIs in total adiponectin levels were initially estimated using a fixed-effects model. If there was significant heterogeneity, we turned to use a random-effects model [50]. To investigate the sources of heterogeneity between the results of different studies, we carried out the following tests: heterogeneity tests, stratified analysis, meta-regression analysis, and sensitivity analysis [51, 52]. For heterogeneity tests, we used Cochran Q test and I2 statistic to evaluate statistical heterogeneity among studies [53]. Statistically significant heterogeneity was considered when P value was less than 0.1 and I2 value was more than 50 % [54]. Stratified analysis was conducted on the basis of BMI, population region, sample size, methods of adiponectin measurement, HOMA2-IR ratio, total testosterone ratio (T-ratio), and the type of blood sample. Subsequently, restricted maximum likelihood-based random-effects meta-regression analyses was carried out to evaluate the above potential heterogeneous factors. Univariate meta-regression analysis was conducted first, after which the variables that were significant at the 0.1 level were entered into the multivariable model. To identify potentially influential studies, sensitivity analysis was also performed to examine whether the effect estimate was robust by repeating the random-effects meta-analysis after omitting one study at a time. In addition, cumulative meta-analysis was conducted to evaluate the evolution of the combined estimates over time according to the ascending date of publication. Finally, the possibility of publication bias was assessed by Begg’s funnel plots and Egger’s tests [55].

All statistical analyses were conducted using STATA version 12.0 (StataCorp LP, College Station, Texas, USA). A two-sided P value less than 0.05 was considered statistically significant.

Results

Literature search results

A flow chart showing the study selection is presented in Fig. 1. According to our search criteria, 62 potential eligible studies that estimated the effect of total adiponectin levels on PCOS were identified for full-text assessment. After reading the full texts, 24 studies were excluded as they did not meet the predefined selection criteria. Seven studies were excluded because they were published with insufficient information; we could not extract the data [56–62]. Seven studies were excluded because circulating adiponectin was not measured in the healthy control group [63–69]. Four studies [70–73] with different BMI between PCOS women and controls were also excluded. Three studies [74–76] that published their results in geometric mean (−2SD to +2SD) rather than mean ± SD were subsequently excluded as we were unable to obtain original data from the authors and could not extract original data from the previous meta-analysis. Another two studies [77, 78] that contained less than ten specimens in each group were also excluded. Partial subjects were overlap between two studies [40, 79] by the same authors, so the smaller size one [79] was excluded. Ultimately, we included a total of 38 studies with 3,598 subjects (1,944 PCOS women and 1,654 controls) in this meta-analysis [9–13, 20–23, 25–46, 80–86].

Fig. 1
figure 1

The flow chart of study selection

Full size image

Study characteristics

The main characteristics of the included studies are shown in Table 2. These studies were published between 2003 and 2013. The majority of the trials were conducted in Asia, whereas nine were done in Europe, six in America, and one in Australia. Serum specimens were used to measure total adiponectin levels in almost all studies, and only six studies used plasma specimens. Total adiponectin was measured by enzyme-linked immunosorbent assay in 24 studies and by radioimmunoassay in 14 studies. Furthermore, most studies investigated HOMA-IR and total testosterone levels between PCOS women and controls to account for a difference in adiponectin levels and a percentage of the potential variability of across-study results. There were eight studies without HOMA-IR results, and ten studies without testosterone levels. Among them, four studies [13, 21, 36, 39] reported total adiponectin level only in two subgroups of population stratified by BMI. Then we treated them as two independent studies in the following quantitative synthesis. Therefore, a total of 42 separate comparisons were included in our final meta-analysis.

Quantitative synthesis of data

A meta-analysis of 42 separate comparisons, which reported data on 3,598 subjects (1,944 patients with PCOS and 1,654 controls), was performed. The overall effect of the pooled analyses indicated that adiponectin levels in PCOS patients were significantly lower than healthy controls, with summary WMD of 2.67 μg/ml reduction (95 % CI −3.22 to −2.13, P = 0.000). However, it should be noted that significant heterogeneity across studies was present (I2 = 95.9 %, P = 0.000).

Stratified analysis

Stratified analysis was conducted to explore heterogeneity between studies and assess the robustness of our findings (Table 3). We evaluated potential sources of heterogeneity between studies including BMI, region, HOMA-IR ratio, total testosterone ratio, sample size, assay methods, and the type of blood sample. Among them, eight studies [12, 13, 21, 28, 36, 39, 44, 46] reported adiponectin levels in two subgroups of population stratified by BMI. Therefore, a total of 46 separate comparisons were included in the subgroup analysis stratified by BMI. As seen in Fig. 2, lower adiponectin levels were detected in women with PCOS in both thin (BMI <25 kg/m2, WMD −2.15, 95 % CI −2.87 to −1.43; P = 0.000) and obesity (BMI ≥25 kg/m2, WMD −2.99, 95 % CI −3.71 to −2.28; P = 0.000) women. And the overall pattern in WMD did not vary substantially by the potential sources except region, HOMA-IR ratio, and total testosterone ratio. Only one study in Australia showed that it was not significantly different in adiponectin levels between PCOS patients and healthy controls (WMD −0.10, 95 % CI −2.73 to 2.53; P = 0.941). However, as for study population in Asia, Europe, and America, adiponectin levels in PCOS patients were significantly lower than healthy controls (all P < 0.05). HOMA-IR ratio and total testosterone ratio were used to express the relative different in IR and testosterone levels between groups, respectively. No significant difference in pooled WMD of adiponectin levels was observed in studies with HOMA-IR ratio of less than 1.36 (WMD −2.40, 95 % CI −4.82 to 0.02; P = 0.052), as well as studies with HOMA-IR ratio between 1.71 and 2.12 (WMD −1.42, 95 % CI −3.41 to 0.57; P = 0.162). Similarly, summary WMD of adiponectin levels was not significantly different in studies with total testosterone ratio of more than 2.5, where adiponectin was found lower in healthy controls (WMD 1.88, 95 % CI −1.95 to 5.72; P = 0.336). Unfortunately, all subgroup analysis demonstrated large heterogeneity, and these variables did not seem to be found to contribute significantly to the heterogeneity between studies.

Table 3 Stratified meta-analysis of circulating adiponectin levels and polycystic ovary syndrome
Full size table
Fig. 2
figure 2

Quantitative synthesis of adiponectin levels between polycystic ovary syndrome patients and controls stratified by BMI (random-effects model)

Full size image

Meta-regression

To further investigate the impact of the above study characteristics on the study estimates of WMD in adiponectin, we conducted meta-regression analysis. SMD was used as the dependent variable, and BMI, region, HOMA-IR ratio, total testosterone ratio, sample size, assay methods, and the type of blood sample were entered as explanatory covariates. Univariate meta-regression analysis was performed first. In univariate meta-regression analysis, BMI (42 studies, P = 0.072), region (42 studies, P = 0.563), HOMA-IR ratio (33 studies, P = 0.773), total testosterone ratio (32 studies, P = 0.209), sample size (42 studies, P = 0.395), method (42 studies, P = 0.358), and blood sample (42 studies, P = 0.620) were assessed independently. Results of the univariate analysis are presented in Table 4. If the regression coefficient of the covariate was significant at the level of 0.1, then the covariate was entered into the multivariate meta-regression. Only one covariate (BMI) was found to be a significant factor in univariate analysis. Therefore, subsequent multivariate meta-regression could not be done. The results of meta-regression suggested that BMI might contribute little to the heterogeneity between included studies, and other covariates failed to account for heterogeneity in any of the preplanned comparisons.

Table 4 Univariate meta-regression analysis for the potential variables between studies
Full size table

Sensitivity analysis

We also conducted a sensitivity analysis using random-effects estimates by omitting one study at a time and calculating the summary SMD for the remaining studies. We found that there was no change in the direction of effect when any one study was excluded, which indicated that the results of our meta-analysis was reliable and stable (data not shown).

Cumulative meta-analysis

In overall included studies, the random-effects pooled WMD was always insignificantly larger or smaller than zero from the year 2003 to the first trial in 2008 by Aroda et al. [84], representing no statistically significant difference in adiponectin level between PCOS patients and healthy controls. But a statistically significant effect was consistently observed from the study by Glucelik et al. in 2008 [10], and it changed little after that study, indicating the stability of the association (data not shown).

Publication bias

Publication bias of literatures was evaluated by Begg’s funnel plots and Egger’s tests. As shown in Fig. 3, the Begg’s funnel plots are slightly asymmetrical in distribution, which could raise the possibility of publication bias, although the Egger’s regression test suggests no statistically significant asymmetry of the funnel plot (t = −1.82, P = 0.076). We further undertook analysis using the trim-and-fill method [87]. The results from the trim-and-fill analysis did not change the summary estimate of effect, suggesting that publication bias is unlikely to affect the results of the meta-analysis.

Fig. 3
figure 3

Funnel plots for publication bias in the studies of the meta-analysis of adiponectin level and polycystic ovary syndrome

Full size image

Discussion

PCOS is considered to be the most common cause of anovulatory infertility, which affects 5–10 % in reproductive-aged women [88]. Moreover, PCOS is frequently associated with the metabolic syndrome such as insulin resistance, dyslipidemia, obesity, and an increased risk of type 2 diabetes, dyslipidemia, hypertension, and cardiovascular diseases [17–19]. Although increasing number of studies are in an effort to the research of PCOS, the pathogenesis of PCOS is still unknown. As adiponectin presents the effects of insulin sensitivity, antidiabetic and anti-inflammatory, and decreasing circulating adiponectin levels are present in women with obesity and type 2 diabetes, it has been suggested that adiponectin might play a role in the pathogenesis of PCOS. This study aimed to systematically estimate the association between adiponectin levels and the incidence of PCOS by combining the primary data from all relevant studies. The overall results of our meta-analysis with 38 including studies demonstrated that total adiponectin levels were significantly lower in women with PCOS than in healthy controls (random-effects WMD, −2.67, 95 % CI −3.22 to −2.13; I2 = 95.9 %), which were consistent with the results of the previous meta-analysis (16 studies, random-effects WMD, −1.71, 95 % CI −2.82 to −0.6; I2 = 80.7 %) [24].

Mechanisms of adiponectin and PCOS

Adiponectin decreases the triglyceride content of muscles, regulates insulin signalization, activates peroxisome proliferator-activated receptor, increases the use of fatty acids and energy, increases muscle fat oxidation and transport via activation of adenosine monophosphate-activated protein kinase, inhibits gluconeogenic enzyme expression, and decreases hepatic glucose production [89]. Because of these properties, adiponectin has an increasing effect on insulin sensitivity as well as anti-atherogenic, anti-inflammatory, and anti-diabetic effects.

Evidence in previous animal studies has shown that adiponectin may exert advantageous effects on increasing insulin sensitivity and antidiabetic effects. Yamauchi et al. [90] found that decreased expression of adiponectin correlated with insulin resistance in mouse models of altered insulin sensitivity, and adiponectin decreased insulin resistance by decreasing triglyceride content in muscle and liver. The phenotypes of adiponectin-deficient and transgenic adiponectin-overproducing animal models underscore the role of adiponectin in the maintenance of glucose and lipid homeostasis [91]. Moreover, pharmacologic adiponectin treatments in rodents increased insulin sensitivity, which indicates the replenishment of adiponectin that might provide a noble treatment modality for insulin resistance and type 2 diabetes [91].

It is clear that most women with PCOS have marked insulin resistance, as well as dyslipidemia and obesity. In humans, many studies reported that adiponectin levels were statistically significantly lower in women in PCOS, which were negatively correlated with insulin resistance and positively correlated with obesity in PCOS [9, 13, 28, 37, 84]. Some studies, however, have not found an association between adiponectin and PCOS [21, 22, 25, 26, 31]. Several studies reported that adiponectin levels reduced in women with PCOS independent of BMI and severity of insulin resistance [11, 30]. In addition, adiponectin circulates in different multimer complexes, and the high-molecular weight (HMW) multimer is the most biologically potent form, which is decreased in women with PCOS compared with normal controls [11, 13, 38, 84]. Tao et al. [13] revealed that HMW adiponectin was a stronger predictor of insulin resistance than total adiponectin in both women with PCOS and normal women.

Sources of heterogeneity

Heterogeneity is a potential problem that may affect the interpretation of the results. The present meta-analysis showed that there was large heterogeneity between studies. Subsequent subgroup analysis stratified by seven potential sources was performed. Unfortunately, as seen in Table 3, large heterogeneity was identified as well. We then conducted a meta-regression and found BIM might contribute little to the overall heterogeneity (P = 0.072). In addition, sensitivity analysis suggested that no single study influenced the pooled SMD qualitatively. In subgroup analysis, we found no significant differences in circulating adiponectin levels between PCOS women and healthy controls only in Australia population. Inconsistent results were also presented in the subgroup analysis stratified by HOMA-IR ratio and T-ratio (Table 3). These conflicting results and the heterogeneity between studies were likely due to geographical differences and variability in PCOS diagnostic criteria. Other factors such as study designs and limited sample size might also contribute to the heterogeneity. A potential confounding factor in the present study was that some of the cases have higher HOMA-IR compared with the controls. This might be one of the reasons for the between-group differences in adiponectin levels.

Study strengths

To our knowledge, this is the most comprehensive meta-analysis to date to evaluate the association between adiponectin levels and PCOS. Substantial number of cases and controls were pooled from all publications concerned with circulating adiponectin levels and PCOS, which greatly increased statistical power of the analysis and provided enough evidence for us to draw a safe conclusion. Although obvious heterogeneity was observed across the studies, sensitivity analysis suggested that no single study influenced the pooled SMD qualitatively. And cumulative meta-analysis showed that no substantive change had occurred in pooled WMD after the study was published in 2008, indicating the stability of the association between low adiponectin levels and PCOS women. In addition, no publication bias was detected in this meta-analysis, which indicated that the pooled results of our study should be reliable. Taken together, these data further confirm the reliability and stability of the meta-analysis results.

Limitations of this meta-analysis

Some limitations of the present meta-analysis should be considered. Firstly, substantial heterogeneity across the studies for the pooled estimates was recorded, which was unsatisfactorily explained although we conducted subgroup analysis, meta-regression analysis, and sensitivity analysis to find the sources. The heterogeneity might reflect clinical heterogeneity related to variability in PCOS diagnostic criteria, ethnicity, or diet and physical activity. Secondly, our results were based on unadjusted estimates, whereas a more precise evaluation should consider the confounding factors such as smoking status, alcoholic consumption, environmental factors, and other diet lifestyle. Thirdly, dyslipidemia is also relative with adiponectin and PCOS, but we were not able to conduct further analysis because of data limitation. Last but not the least, some of the included studies contained small numbers of cases and the backgrounds of patients varied, which would result in low statistical power and the inconsistent results between studies. With these limitations in mind, caution should be applied in extrapolating the results mentioned above for wider application.

Conclusions

In conclusion, the current evidence suggested that circulating adiponectin levels in women with PCOS were significantly lower than those in healthy controls, which indicated that circulating adiponectin might play a role in the development of PCOS. However, for the subgroup of HOMA-IR ratio and total testosterone ratio, inconsistent results were presented. The well-designed studies with larger sample size, adjusted with confounding factors are needed to answer the question of whether low circulating adiponectin levels are relative with IR and testosterone.