Insulin Sensitizers for Improving the Endocrine and Metabolic Profile in Overweight Women With PCOS

Chuan Xing; Chunzhu Li; Bing He


J Clin Endocrinol Metab. 2020;105(9) 

In This Article

Materials and Methods

Literature Search

This network meta-analysis (NMA) was registered with the PROSPERO international prospective register of systematic reviews (registration number CRD 156677). We searched medical literature for relevant studies using PubMed, EMBASE, the Cochrane Library, the WanFang Database, the WeiPu Database, and China National Knowledge Infrastructure (CNKI) from their dates of establishment to September 2019. A total of 5050 records were identified using electronic search strategies; in addition, 2 relevant records were obtained from the reference lists of the included studies. We used the search terms polycystic ovary syndrome with metformin, glucagon-like peptide-1 receptor agonists, and thiazolidinediones and limited the publication type to randomized controlled trials (RCTs). There were no language or location restrictions.

Inclusion and Exclusion Criteria

To be included in this NMA, the studies needed to meet the following inclusion criteria: (1) women were diagnosed with PCOS, aged 18 to 49 years, with the lower inclusion limit of body mass index (BMI) ≥25 kg/m2; (2) the diagnosis of PCOS was based on the Rotterdam European Society for Human Reproduction and Embryology/American Society for Reproductive Medicine standard or the National Institute of Child Health and Human Development standard with no limits in terms of disease duration or ethnicity; (3) RCT study design; (4) the study included at least one of the outcomes of menstrual frequency, sex hormone parameters (including total testosterone [TT], FT, SHBG and androstenedione [AND]), glucose metabolism parameters (including fasting blood glucose [FG] and fasting insulin [FINS]) and obesity-related parameters (including BMI and waist circumference [WC]); (5) comparisons between the relevant interventions, including metformin, GLP-1 receptor agonists, TZDs, metformin + GLP-1 receptor agonists, and metformin + TZDs; and (6) the intervention period was at least 12 weeks.

Exclusion criteria included the following: (1) combinations with other related drugs, such as ovulation-inducing agents or other contraceptives; or (2) patients with other diseases such as congenital adrenal hyperplasia, Cushing syndrome, androgen-secreting neoplasm, hyperprolactinemia, diabetes, and kidney or liver disease.

Outcomes, Data Extraction, and Quality Assessment

Outcomes included changes in menstrual frequency, sex hormone, glucose metabolism, and obesity-related parameters. Data were extracted according to study, author, year, region, size, interventions, follow-up, and efficacy (Table 1).

The included trials were assessed for quality by 2 authors (X.C. and L.C.Z.) using Cochrane Risk of Bias Tool 2.0 for RCTs.[29] We assessed the methodological quality of the included RCTs according to standard criteria of The Cochrane Collaboration. Seven domains related to risk of bias were assessed for each study, including: (1) random sequence generation; (2) allocation concealment; (3) blinding of participants and personnel; (4) blinding of outcome assessment; (5) incomplete outcome data; (6) selective reporting; and (7) other bias. Every question was answered "yes," "no," or "unclear," and 2 reviewers assessed each trial. In case of a disagreement, judgment was made through open discussion. According to criteria of The Cochrane Collaboration, we divided the studies into 3 categories: (1) low risk of bias: low risk of bias for all key domains; (2) unclear risk of bias: unclear risk of bias for one or more key domains; and (3) high risk of bias: high risk of bias for one or more key domains. Discrepancies in risk assessments were discussed with the third author (H.B.).

Statistical Analysis

Data Synthesis and Analysis. Both traditional meta-analysis (TMA) and NMA were performed to simultaneously compare the efficacies of different treatment options for managing changes in menstrual frequency, sex hormone parameters (including TT, FT, SHBG, and AND), metabolic parameters (including FG and FINS), and obesity-related parameters (including BMI and WC).[30] For dichotomous outcomes, we calculated pooled odds ratios (ORs) with 95% confidence intervals (CIs). For continuous outcomes, we calculated weighted mean differences (WMDs) with 95% CIs. All tests were two-tailed, and a P value < 0.05 was deemed statistically significant.

Traditional Meta-analysis. We performed TMA using RevMan version 5.3 (Nordic Cochrane Center) to directly compare the efficacies of metformin, GLP-1 receptor agonists, TZDs, metformin + GLP-1 receptor agonists, and metformin + TZDs. The WMD and 95% CI between 2 groups were synthesized using the mean difference (MD) and standard deviation (SD), and the sample size or the data were converted from the MD and sample SD at the endpoint between 2 groups.[31] Interstudy heterogeneity between the trials was assessed using the chi-squared test for analysis, combining I2 and P value to determine the level of heterogeneity where I2 > 50% or P < 0.10 suggested a high level of heterogeneity. For all the parameters included in this study, the random effect models were used. Although we planned to assess publication bias by using the Egger regression asymmetry test, we were not able to conduct formal testing because of the small number of studies available from head-to-head comparisons. Funnel plots created by Stata software (version 15.0, Stata Corp, College Station, TX) were used to examine publication bias.

Network Meta-analysis. The NMA was performed to combine results from direct and indirect comparisons of treatment effect in a single analysis with Stata software (version 15.0, Stata Corp, College Station, TX). The analysis model was based on the definition of likelihood and prior probability, and we confirmed the convergence and calculated the pooled estimates of MDs and 95% credibility intervals (CrIs) to compare the 5 different interventions to each other. Additionally, the surface under the cumulative ranking (SUCRA) curve was calculated to rank the different interventions. Compared with other interventions, 1 intervention showed a higher SUCRA value, so it might have a greater effect on the endpoint.[32] The consistency assumption of direct evidence and indirect evidence was assessed by the node-splitting method. When the results showed that the direct evidence of the outcomes was consistent with indirect evidence (P values > 0.05), the consistency model was adopted.