NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

Norris SL, Carson S, Thakurta S, et al. Drug Class Review: Thiazolidinediones: Final Report Update 1 [Internet]. Portland (OR): Oregon Health & Science University; 2008 Aug.

Cover of Drug Class Review: Thiazolidinediones

Drug Class Review: Thiazolidinediones: Final Report Update 1 [Internet].

Show details


Literature Search

To identify relevant citations for the original report, 2 independent reviewers identified potentially relevant titles and abstracts from the Cochrane Central Register of Controlled Trials (3rd quarter 2005), Cochrane Database of Systematic Reviews, DARE, MEDLINE (1966 to July, week 4, 2005), and EMBASE (3rd quarter 2005). Search terms included drug names and indications. (See Appendix A for complete search strategies.) To identify additional studies, we also searched reference lists of included studies and reviews and we reviewed dossiers submitted by pharmaceutical companies. All citations were imported into an electronic database (EndNote 9.0.0, Thomson Reuters).

For the update the original search terms were used, but titles and abstracts and then full-text articles were screened to include additional active-control studies that address the updated key questions and new head-to-head and placebo-controlled studies. Updated searches were conducted in November 2007 (Appendix A). Electronic searches were supplemented by hand searches of dossiers received from the makers of pioglitazone and rosiglitazone, and medical and statistical reviews available on the Food and Drug Administration website.

Articles deemed potentially relevant after review of titles and abstracts were retrieved in full-text form. Two independent reviewers achieved consensus on all included and excluded articles. Excluded articles were coded in the EndNote database with the reason for exclusion.

Study Selection

The pharmacotherapeutic agents reviewed were the 2 thiazolidinediones currently available in the United States: pioglitazone hydrochloride (Actos®) and rosiglitazone maleate (Avandia). Muraglitazar (Pargluva) was not reviewed as it was not available in the United States as of January 1, 2008.

Participants in included studies were adults with type 2 diabetes, prediabetes, or the metabolic syndrome. As noted above, various definitions exist for the metabolic syndrome. Any study examining persons with the metabolic syndrome was included if the authors used 1 of the widely accepted definitions mentioned above (see Table 1).

Included studies examining type 2 diabetes had to present 1 or more of the primary outcomes of interest to this review: glycemic control (A1c), time to initiation of insulin for glycemic control, progression or occurrence of microvascular disease (nephropathy, retinopathy, and neuropathy), progression or occurrence of macrovascular disease (cardiovascular disease, cerebral vascular disease, amputation), other complications of diabetes, mortality, and quality of life.

Included studies examined either effectiveness or efficacy of the 2 included drugs. The purpose of this report was primarily to examine effectiveness; however, since there were very few data available on effectiveness, efficacy studies were included and reviewed in detail.

For efficacy, effectiveness, and safety, published and unpublished English-language reports in any geographic setting were included if they had a total sample size of ten or more participants. We included letters if primary data were presented and there was sufficient detail to evaluate quality. We excluded abstracts and conference proceedings, as these publications generally do not have sufficient detail to assess internal or external validity. Theses were not included as the full text is frequently difficult to retrieve.

Selection criteria for the original report

For the assessment of efficacy and effectiveness in the original report, we included reports of randomized controlled trials and controlled clinical trials. We included trials comparing rosiglitazone and pioglitazone (head-to-head trials), as well as trials comparing either one of these drugs to placebo. We also included trials comparing these drugs to another pharmacotherapeutic agent (active-control trials) only if they examined effectiveness outcomes or population subgroups.

For examination of efficacy and effectiveness among subgroups, we expanded our inclusion criteria to encompass all study designs (that is, observational, before-after, case-control studies, and time series) where data were available. We used this approach because few controlled trials were available that examined subgroups; therefore, we expanded our inclusion criteria in order to examine the best available evidence, recognizing that study designs that do not involve randomization are weaker designs and are more likely to be biased or confounded by known or unknown factors affecting the outcomes of interest.

For the assessment of tolerability and adverse effects, we included observational studies, including case series with a sample size greater than ten, before-after studies, randomized controlled trials, and controlled clinical trials. Clinical trials are often not designed to assess adverse events, may select low-risk patients (in order to minimize drop-out rates), or may have too short a follow-up period in which to adequately assess safety. Observational studies designed to assess adverse event rates may include broader populations, carry out observations over a longer time period, use higher quality methodological techniques for assessing adverse events, or examine larger sample sizes.

Safety and tolerability were examined using data provided on overall and serious adverse events, withdrawals due to adverse effects, and other relevant specific adverse events including hypoglycemia, liver toxicity, heart failure, pulmonary edema, weight gain, and edema.

Selection criteria for the updated report

For the updated report we expanded our inclusion criteria with respect to study designs for effectiveness outcomes in order to be consistent with criteria used in the Agency for Healthcare Research and Quality report. Most notably, we expanded our examination of active-control comparisons, which was previously restricted by sample size, follow-up interval, or outcomes. These criteria are listed in Table 3, where they are contrasted with those of the prior report and of the Agency for Healthcare Research and Quality report.

Table 3

Table 3

Inclusion criteria for the original and updated reports

Data Abstraction

The following data were abstracted from included trials into a relational database developed for this review: study design; setting; population characteristics (including sex, age, race/ethnicity, diagnosis, duration of type 2 diabetes, A1c, weight, and body mass index); eligibility and exclusion criteria; drug dosage and frequency; treatment duration; comparison group care; numbers screened, eligible, enrolled, and lost to follow-up; and results for each prespecified outcome. Similar data were abstracted for studies that were not controlled trials and which examined adverse events.

We recorded results achieved with an intention-to-treat analytic approach, when reported. If only per protocol results were reported, we specified the nature of these results and reported them. In trials with crossover, outcomes for the first intervention were recorded if available. This was because of the potential for bias due to differential withdrawal prior to crossover, the possibility of a “carryover effect” (from the first treatment) in studies without a washout period, and a “rebound” effect from withdrawal of the first intervention.

Quality Assessment

We assessed the internal validity (quality) of controlled clinical trials using the predefined criteria listed in the quality assessment tool found in Appendix C. These criteria are based on those used by the US Preventive Services Task Force29 and the National Health Service Centre for Reviews and Dissemination.30 For each included trial we assessed methods for the following charateristics: randomization; allocation concealment; blinding of participants, investigators, and assessors of outcomes; the similarity of comparison groups at baseline; adequate reporting of attrition, crossover, adherence, and contamination; post-allocation exclusions; and use of intention-to-treat analysis.

We based assessment of observational and other study designs with adverse event data on unbiased selection of patients, loss to follow-up, unbiased and accurate ascertainment of events, and control for potential confounders (Appendix C).

These criteria were then used to categorize studies as good-, fair-, and poor-quality studies. Studies that had a significant flaw in design or implementation such that the results were potentially not valid were categorized as “poor”. Studies that met all quality criteria were rated good quality. The remainder were rated fair. As the “fair quality” category is broad, studies with this rating vary in their strengths and weaknesses.

Studies were not excluded on the basis of poor quality as there is a lack of empirical evidence for a relationship between criteria thought to measure validity and actual study outcomes.31 Studies rated as poor-quality were carefully examined and the potential sources of bias and its potential impact are presented in the evidence tables. If data were sufficient, a sensitivity analysis was performed to compare results between studies with high and low risk of bias.

External validity of studies was assessed by examining the following: adequacy of population description; inclusion and exclusion criteria; and whether the treatment received by the comparison group was reasonably representative of standard practice.

Systematic reviews that fulfilled inclusion criteria were rated for quality using predefined criteria (see Appendix C) to ensure the following: clear statement of the questions and inclusion criteria; adequate search strategy; adequate assessment of individual trials; adequate provision of information; and appropriate methods of synthesis.

Data Analysis and Synthesis

Important descriptive information about the population, setting, intervention, and quality assessment of studies are presented in tables, and synthesis is presented in narrative. When there were sufficient data on the primary outcome of A1c and studies were considered to be homogeneous with respect to important variables (population characteristics, drug dosage, follow-up interval, and the application of any co-intervention), we performed a meta-analysis. We also performed a meta-analysis of two key outcomes related to adverse events: the total number of withdrawals and the withdrawals related to adverse events.

We recorded the mean difference between baseline and follow-up measures for control and intervention groups and the standard error of each difference. If the standard error of the difference for each group was not given, it was estimated from the standard error of the groups at baseline, assuming a correlation between baseline and follow-up of 0.75. If data were presented only in graphs, point estimates were determined from published graphs. Pooled effects of the randomized controlled trials were determined with each study weighted by the inverse of the study variance, using a random effects model with the DerSimonian and Laird formula for calculating between-study variance.32 The R statistical environment and Review Manager (RevMan) was used for the meta-analysis.

An adjusted indirect comparison was performed for the outcome of A1c by combining the results of the meta-analysis comparing pioglitazone to placebo with the results of the meta-analysis comparing rosiglitazone with placebo. The variance of the estimate of effect was estimated as the sum of the variances of the 2 meta-analyses being pooled.33

Heterogeneity between trial results was tested for using a standard chi-squared test using a significance level of alpha=0.1, in view of the low power of such tests.31 We also examined inconsistency among studies with I2, which describes the percentage of the variability in effect estimates that is due to heterogeneity rather than sampling error (that is, chance).34 A value >50% may be considered substantial heterogeneity. If heterogeneity was found, we attempted to determine potential sources by examining individual study characteristics. If heterogeneity was too great to meaningfully pool the results in a quantitative manner, the results are presented in narrative.

In the original report (and not in the update), meta-regression was performed to determine whether the study-level characteristics duration of intervention and study sponsorship (industry or private) affected between-group change in A1c in placebo-controlled trials. For studies using a combination of a thiazolidinedione and another hypoglycemic agent, we examined the effects of insulin, metformin, and sulfonylurea on A1c. For the meta-regression we used STATA (version 9, StataCorp LP, College Station, Texas).

Copyright © 2008, Oregon Health & Science University, Portland, Oregon.
Bookshelf ID: NBK10741
PubReader format: click here to try


Other titles in this collection

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...