Testosterone Therapy in Men With T2D and Metabolic Syndrome
Testosterone Therapy in Men With T2D and Metabolic Syndrome
In this study, we followed the reporting recommendations made in the PRISMA (Preferred reporting items for systematic reviews and meta-analyses) statement. The PRISMA statement was designed to improve the quality of systematic reviews or meta-analyses. The statement lists 27 items to include when conducting and reporting such a study. In this study, all 27 items were included.
Eligible studies were defined in a protocol as fully published (English language) double-blind randomized controlled clinical trials that assessed the effects of testosterone therapy in men with diagnosed metabolic syndrome and/or T2D on measures of glucose metabolism.
We conducted a comprehensive search of the literature using the electronic databases Medline, Embase and the Cochrane register of controlled trials from inception to July 2014. The search strategy was developed in consultation with an experienced medical research librarian using a broad range of relevant search terms (full research strategies are available in the supplementary information). In addition, reference lists of potentially eligible articles and relevant reviews were searched by hand. Study selection was conducted by two independent reviewers (M.G. and B.B.Y). Studies included by both reviewers were compared and disagreement resolved by consensus and third party adjudication. Only placebo-controlled double-blind RCTs were eligible.
Two investigators (M.G and B.B.Y) independently extracted the relevant data using a standardized form. Data extracted from each eligible RCT included demographic information, diagnosis of T2D and/or the metabolic syndrome, number of participants, baseline and on treatment testosterone levels, type and duration of treatment. Disagreements were resolved by consensus and third party adjudication.
Two reviewers (M.G and B.B.Y), working in duplicate, assessed the methodological quality of each eligible RCT using the full 25-item CONSORT checklist of information to be included when reporting a RCT. The CONSORT checklist was designed to improve the quality of RCT reporting. Therefore, the quality of a published RCT can be assessed by quantifying how many of the 25 recommended criteria are reported. A high quality RCT will report all 25 items, and the quality of a RCT correlates inversely with the number of reported items.
The primary outcomes of interest were the mean differences (MD) in insulin resistance (assessed by homeostatic model assessment of insulin resistance, HOMA-IR) and in glycaemic control (assessed by HbA1c) between treatment and control groups. In addition, standardized mean differences (SMD) were derived. For each eligible RCT, MD ± Standard Deviation (SD) from baseline to end of trial in each group, treated and controls were retrieved for HOMA-IR and HbA1c. Where SD was not given, it was estimated from SEM or from the 95% confidence interval of the MD. For obtaining SMD, mean differences of individual trials were standardized prior to meta-analysing. In RCTs reporting an open-label extension phase, only the initial double-blind placebo-controlled phase was considered.
The secondary outcomes were effects of testosterone on symptoms, cardiovascular risk markers and adverse effects. We meta-analysed constitutional symptoms reported by Aging Male Symptom score. Due to between-trial heterogeneity and inconsistent reporting, effects of testosterone on sexual symptoms, cardiovascular risk markers and adverse effects could not be meta-analysed, but were instead reported in a descriptive fashion.
Three investigators (M.G., B.B.Y and G.W.) independently verified and collated the extracted data to provide a descriptive synthesis of key characteristics and a quantitative synthesis of effect size estimates for each RCT.
The consistency or heterogeneity of the results among various trials in a meta-analysis is an important statistical measure. Hence, the variation in effect beyond chance was tested by using both the Cochran's Q chi-squared test and the I test, which describes the percentage of the variability in effect estimates that is due to heterogeneity rather than sampling error, therefore providing a quantitative measure of nonrandom differences observed across studies. An I > 30% indicates a moderate intertrial heterogeneity. Given the fact that considerable inconsistency existed among the trials, we used a random effects model to estimate the average true difference (MD) in HOMA-IR or HBA1c in the treatment group, compared to placebo. The choice of differing HOMA-IR modelling employed in the trials, HOMA1 or HOMA2, was assessed in the meta-analysis by adding this information as a moderator variable. SMD are based on Hedges' g with appropriate correction for a negative bias. The Hedges g' with appropriate correction for negative bias was used to calculate the SMD, as recommended by the Cochrane Library, which in contrast to the MD relates the size of the intervention effect to the variability in each study. In a moderator analysis, we examined the impact of a covariate (moderator variable) on the effect. In addition, we used a weighted restricted maximum-likelihood estimator for fitting the models, and the model was implemented by the metafor package (version 1.9.4) in the R statistical program (R for Mac version 3.1.1).
Graphical data presentation included Forest plots and Funnel plots. A Funnel plot is a scatterplot of treatment effect against a measure of study size-related imprecision (such as the standard error used here) on the vertical axis and serves as a visual aid for detecting bias. In addition to this graphical method, a regression test (regtest) was employed to formally test for the presence of asymmetry in the Funnel plot.
The study is an analysis of published data that does not require specific approval by an ethics committee.
Materials and Methods
In this study, we followed the reporting recommendations made in the PRISMA (Preferred reporting items for systematic reviews and meta-analyses) statement. The PRISMA statement was designed to improve the quality of systematic reviews or meta-analyses. The statement lists 27 items to include when conducting and reporting such a study. In this study, all 27 items were included.
Eligibility Criteria
Eligible studies were defined in a protocol as fully published (English language) double-blind randomized controlled clinical trials that assessed the effects of testosterone therapy in men with diagnosed metabolic syndrome and/or T2D on measures of glucose metabolism.
Search Strategy
We conducted a comprehensive search of the literature using the electronic databases Medline, Embase and the Cochrane register of controlled trials from inception to July 2014. The search strategy was developed in consultation with an experienced medical research librarian using a broad range of relevant search terms (full research strategies are available in the supplementary information). In addition, reference lists of potentially eligible articles and relevant reviews were searched by hand. Study selection was conducted by two independent reviewers (M.G. and B.B.Y). Studies included by both reviewers were compared and disagreement resolved by consensus and third party adjudication. Only placebo-controlled double-blind RCTs were eligible.
Data Extraction
Two investigators (M.G and B.B.Y) independently extracted the relevant data using a standardized form. Data extracted from each eligible RCT included demographic information, diagnosis of T2D and/or the metabolic syndrome, number of participants, baseline and on treatment testosterone levels, type and duration of treatment. Disagreements were resolved by consensus and third party adjudication.
Quality Assessment
Two reviewers (M.G and B.B.Y), working in duplicate, assessed the methodological quality of each eligible RCT using the full 25-item CONSORT checklist of information to be included when reporting a RCT. The CONSORT checklist was designed to improve the quality of RCT reporting. Therefore, the quality of a published RCT can be assessed by quantifying how many of the 25 recommended criteria are reported. A high quality RCT will report all 25 items, and the quality of a RCT correlates inversely with the number of reported items.
Primary Outcomes
The primary outcomes of interest were the mean differences (MD) in insulin resistance (assessed by homeostatic model assessment of insulin resistance, HOMA-IR) and in glycaemic control (assessed by HbA1c) between treatment and control groups. In addition, standardized mean differences (SMD) were derived. For each eligible RCT, MD ± Standard Deviation (SD) from baseline to end of trial in each group, treated and controls were retrieved for HOMA-IR and HbA1c. Where SD was not given, it was estimated from SEM or from the 95% confidence interval of the MD. For obtaining SMD, mean differences of individual trials were standardized prior to meta-analysing. In RCTs reporting an open-label extension phase, only the initial double-blind placebo-controlled phase was considered.
Secondary Outcomes
The secondary outcomes were effects of testosterone on symptoms, cardiovascular risk markers and adverse effects. We meta-analysed constitutional symptoms reported by Aging Male Symptom score. Due to between-trial heterogeneity and inconsistent reporting, effects of testosterone on sexual symptoms, cardiovascular risk markers and adverse effects could not be meta-analysed, but were instead reported in a descriptive fashion.
Data Synthesis and Statistical Analysis
Three investigators (M.G., B.B.Y and G.W.) independently verified and collated the extracted data to provide a descriptive synthesis of key characteristics and a quantitative synthesis of effect size estimates for each RCT.
The consistency or heterogeneity of the results among various trials in a meta-analysis is an important statistical measure. Hence, the variation in effect beyond chance was tested by using both the Cochran's Q chi-squared test and the I test, which describes the percentage of the variability in effect estimates that is due to heterogeneity rather than sampling error, therefore providing a quantitative measure of nonrandom differences observed across studies. An I > 30% indicates a moderate intertrial heterogeneity. Given the fact that considerable inconsistency existed among the trials, we used a random effects model to estimate the average true difference (MD) in HOMA-IR or HBA1c in the treatment group, compared to placebo. The choice of differing HOMA-IR modelling employed in the trials, HOMA1 or HOMA2, was assessed in the meta-analysis by adding this information as a moderator variable. SMD are based on Hedges' g with appropriate correction for a negative bias. The Hedges g' with appropriate correction for negative bias was used to calculate the SMD, as recommended by the Cochrane Library, which in contrast to the MD relates the size of the intervention effect to the variability in each study. In a moderator analysis, we examined the impact of a covariate (moderator variable) on the effect. In addition, we used a weighted restricted maximum-likelihood estimator for fitting the models, and the model was implemented by the metafor package (version 1.9.4) in the R statistical program (R for Mac version 3.1.1).
Graphical data presentation included Forest plots and Funnel plots. A Funnel plot is a scatterplot of treatment effect against a measure of study size-related imprecision (such as the standard error used here) on the vertical axis and serves as a visual aid for detecting bias. In addition to this graphical method, a regression test (regtest) was employed to formally test for the presence of asymmetry in the Funnel plot.
The study is an analysis of published data that does not require specific approval by an ethics committee.