Diabetes Ther https://doi.org/10.1007/s13300-018-0443-z REVIEW Meta-Analysis of Sulfonylurea Therapy on Long-Term Risk of Mortality and Cardiovascular Events Compared to Other Oral Glucose-Lowering Treatments . . W. Ryan Powell Cindy L. Christiansen Donald R. Miller Received: April 11, 2018 The Author(s) 2018 and experimental studies comparing the safety ABSTRACT of SUs to that of other diabetes medications in people with type 2 diabetes mellitus through Introduction: Among the most pressing clini- December 15, 2015. Studies with at least 1 year cal decisions in type 2 diabetes treatments are of follow-up, which explicitly examined major which drugs should be used after metformin is cardiovascular events or death in patients who no longer sufﬁcient, and whether sulfonylureas showed no evidence of serious conditions at (SUs) should remain as a suitable second-line baseline, were selected for inclusion in meta- treatment. In this article we summarize current analyses. evidence on the long-term safety risks associ- Results: SU treatment was associated with an ated with SU therapy relative to other oral glu- elevated risk relative to treatment with met- cose-lowering therapies. formin (METF), thiazolidinedione (TZD), Methods: The MEDLINE database and Clini- dipeptidyl peptidase-4 inhibitor (DPP-4), and caltrials.gov were searched for observational glucagon-like peptide-1 (GLP-1) agonist classes, either when compared alone (as a monother- Enhanced Digital Features To view enhanced digital apy) or when used in combination with METF. features for this article go to https://doi.org/10.6084/ Signiﬁcant ﬁndings were almost entirely m9.ﬁgshare.6244904. derived from nontrial data and not conﬁrmed Electronic supplementary material The online by smaller, efﬁcacy designed randomized con- version of this article (https://doi.org/10.1007/s13300- trolled trials whose effects were in the same 018-0443-z) contains supplementary material, which is available to authorized users. direction but much more imprecise. Conclusion: Although much of the evidence is W. R. Powell (&) C. L. Christiansen D. R. Miller derived and will continue to come from obser- Center for Healthcare Organization and vational studies, the methodological rigor of Implementation Research, Bedford VA Medical such studies is questionable. A key challenge for Center, Bedford, MA, USA evaluators is the extent to which they should e-mail: WilliamRyan.Powell@va.gov incorporate evidence from study designs that W. R. Powell C. L. Christiansen D. R. Miller are quasi-experimental. Department of Health Law, Policy and Management, Boston University School of Public Health, Boston, MA, USA Keywords: All-cause mortality; Cardiovascular disease; Meta-analysis; Sulfonylurea; Type 2 D. R. Miller diabetes mellitus Department of Dermatology, Boston University School of Medicine, Boston, MA, USA Diabetes Ther Abbreviations METHODS CVD Cardiovascular disease DPP-4 Dipeptidyl peptidase-4 Data Sources ES Effect size GLP-1 Glucagon-like peptide-1 The MEDLINE database (via PubMed) was sear- MEGL Meglitinide ched for studies comparing the safety of SUs METF Metformin (monotherapy or in combination) relative to RCT Randomized controlled trials other oral diabetes medications in patients with RR Relative risk type 2 diabetes from 1965 to December 15, SGLT-2 Sodium-glucose co-transporter 2 2015, using the search terms reported in Elec- SU Sulfonylurea tronic Supplementary Material (ESM) S1. Clini- TZD Thiazolidinediones caltrials.gov, a public database that registers clinical trials, was also searched for unpublished data. In addition, the reference lists of the rel- INTRODUCTION evant articles identiﬁed by the search of these databases were examined for studies not The current clinical consensus is to treat people retrieved from the other search strategies. with type 2 diabetes with metformin (METF) Finally, references from previous meta-analyses when diet and exercise have failed to control and Cochrane reviews were examined. This glucose levels. However, to date, the question of article is based on previously conducted studies whether sulfonylureas (SUs), a class of oral anti- and does not contain any studies with human hyperglycemic agents, are a suitable option for participants or animals performed by any of the second-line therapy remains a focus of con- authors. tention. While SUs represent a common, inex- A ﬂow chart of the selection process resulting pensive, and effective treatment to manage from the MEDLINE and other search strategies glucose levels [1, 2], they have become increas- is shown in Fig. 1. A total of 1982 articles were ingly controversial because of long-term safety extracted from MEDLINE and a further 264 concerns. Emerging evidence links the use of SUs articles were culled from the other search with elevated risks for cardiovascular events and strategies, resulting in a total of 2246 articles mortality compared to other glucose-lowering extracted for assessment. The abstracts of each drug therapies, but expert opinion remains divi- of these articles were then reviewed for eligi- ded on whether SUs should remain a suit- bility, following which 172 of these articles able therapy in the clinical setting [3, 4]. This were reviewed in their entirety. Finally, a total difference in opinions may be attributed, in part, of 50 articles met the eligibility requirements to to the fact that a number of studies reporting be included in the series of meta-analyses (see elevated risks are observational in nature and ESM S2 for a list of all studies included). thus open to challenge in terms of their Information on the effect size (e.g., hazard methodological rigor. This factor and the lack of ratio, odds ratio, relative risk [RR]) or the raw safety and efﬁcacy measures in randomized information required to calculate it (e.g., num- controlled trials (RCTs) designed to evaluate ber of major cardiovascular events, number of long-term outcomes while also reﬂecting actual people who died), the standard deviation (or clinical populations have likely contributed to 95% conﬁdence interval [CI]), sample size the adoption of different clinical guidelines. (number of people in treatment group), and The aim of the study reported here is to pool study characteristics relevant to the population, the existing evidence to summarize the risk of outcome, and exposure were extracted from (1) cardiovascular events and (2) mortality (all- each study if provided. Adjusted estimates of cause and cardiovascular) associated with SU the effect size were used if provided; otherwise use relative to other therapies within a broad unadjusted estimates were extracted. Authors of range of indicated populations by conducting a individual articles were not contacted to obtain series of meta-analyses. Diabetes Ther Fig. 1 Flow chart of the study selection process. RCTs Randomized controlled trials information if missing. For the purposes of the date of the ﬁrst prescription were included for meta-analyses reported here, hazard ratios, odds assessment. ratios, and relative risk were treated as equiva- Studies were excluded from the meta-analy- lent measures when pooling estimates. Article ses if they met any of the following criteria: extraction and the culling of information were included only patients with serious conditions conducted by one of the authors who is a health at baseline, such as a history of major cardio- services researcher (WRP), with consultation or vascular events or renal failure; had a treatment further reviews by the other two authors (CLC, population of only children (younger than DRM) who are senior health services 18 years of age) or only type 1 diabetes patients; researchers. did not include an active comparator (e.g., diet/ exercise, placebo); had a case–control design; involved research only on animals; written in a Study Selection language other than English. For studies for which there were more than one publication, Randomized controlled trials and observational the article with the most complete data or cohort studies were included in this meta-anal- which involved the most recent follow-up was ysis. All studies explicitly examining all-cause selected. For observational studies, an attempt mortality, cardiovascular-related mortality, or to address confounding factors must have been major cardiovascular events were examined. implemented (matched in the design or model Some heterogeneity in the deﬁnition of major adjustment) by including basic demographic cardiovascular composite endpoints used across information (i.e., age, sex, and race) and rele- studies existed; for clarity, each study deﬁnition vant comorbidities at baseline (those adjusting is given in ESM S3. Since the aim was to evaluate for cardiovascular disease [CVD] risk at a mini- long-term cardiovascular and mortality risks, mum). This resulted in 24 RCTs and 26 only studies with C 1 year of follow-up from the Diabetes Ther observational cohort studies being included in should not impact the magnitude of the effect this study. size. Therefore, the ﬁxed effect model combines all study information together without taking into account that studies can vary between each Data Extraction and Quality Assessment other as well as between different study designs. Weights given to each study are determined Details on potential biases in each RCT included only by its within-study variance (study in the meta-analyses were assessed using items weight = 1/within study variance). Since vari- from the Jadad scale, which assesses the ance is a function of sample size, smaller studies methodological quality of RCTs in terms of will contribute less information to the weighted study design and its appropriateness (random- estimate than larger studies. ization, double blind) as well as whether a In the random effects model, the weights description of the dropouts from the study is given to each study are determined not only by included . The quality of observational the within-group variability (as for ﬁxed effects) cohort studies was rated using the eight items but also by the between-group variability. The from the Newcastle–Ottawa Scale , which implication is that relatively greater weight assesses quality in three domains: sample tends to be given to smaller studies than it selection, comparability of groups, and out- would be in a ﬁxed effect model approach since come assessment. An additional item for both the weights for each study now account for study designs examined whether industry between-study design variability. In general, funding explicitly sponsored the study. since random effects models also include Details of the quality assessments are pre- between-study variation, they will tend to have sented in ESM S4 and S5. Results from the relatively wider conﬁdence intervals compared Newcastle–Ottawa Scale suggest that all studies to ﬁxed effects models . The inverse variance met most of the quality assessments in each and the DerSimonian–Laird methods were used domain. Regarding the RCTs, all studies were to estimate the ﬁxed and random effects, randomized, 20 of the 24 were double blind, respectively, using the METAN command in the and in 23 a description of the participant Stata version 14.1 data analysis and statistical dropouts was provided. However, industry software . funding was judged to be high in 64% of all A particular challenge for researchers is how studies (23/24 RCTs; 9/26 observational stud- to synthesize results that are produced from two ies). With the exception of industry-funded inherently different study designs, namely, studies, most studies were assessed as being at a RCTs and quasi-experimental observational low risk of bias on the domains assessed, sug- designs. Therefore, to address this method- gesting that the overall quality was fair to good ological challenge, we used a two-level hierar- in the selected studies. Total scores from the chical Bayesian design to synthesize result quality assessments were not used to exclude estimates across RCTs and observational studies from the meta-analyses. designs. This is a random effects model approach and assumes that the effects derived Data Synthesis and Analysis from different study designs will be similar and also different to some extent. The combined Each outcome and comparison required two or effect is the weighted average of these two more studies. For RCTs and studies with obser- common effect sizes. vational designs, both ﬁxed effects and random Overall pooled estimates were estimated effect models were conducted and reported. In a using the ‘bayesmh’ command with random ﬁxed effects model, the assumption is that each effect of study design in Stata 14.1 . Thus, the study provides evidence towards one common model accounts for heterogeneity from the dif- effect size; that is, the model assumes the effect ferent study design. This is similar to the size should be the same and that the features of approach used by Peters et al.  and involved the study (e.g., study design, population) Markov chain Monte Carlo estimation using a Diabetes Ther Metropolis–Hastings algorithm and Gibbs sam- pling with vague conjugate prior distributions speciﬁed on unknown parameters. Convergence diagnostics suggested fairly rapid convergence with no trend in trace plots, low autocorrela- tion, and acceptance rates for the Metropolis–Hastings algorithm of around 75% (well above the 10% rule of thumb) and efﬁ- ciencies of [ 1% for all analyses. Heterogeneity across the studies was assessed using the I statistic, with values of [ 50% benchmarked as indicating substantial hetero- geneity . This statistic represents the per- centage of variance in the effect size attributable to heterogeneity, with larger values indicating less overlap in conﬁdence intervals across studies. A beneﬁt of the statistic is that the number of studies involved in each meta- analysis has little inﬂuence on the I statistic, unlike other estimates. In drug comparisons that included C 10 studies, publication bias was assessed by testing for asymmetry in funnel plots (scatterplot for Fig. 2 Pooled relative risks (RR) for all-cause mortality. the log effect size by the log standard error) Inverse variance ﬁxed effect estimates are shown for pooled using Egger’s tests  via the METABIAS Stata estimates by study design, and two-level hierarchical command . Tests for funnel plot asymmetry Bayesian estimates are shown for overall pooled estimates. are not recommended in comparisons with\10 RR and the 95% conﬁdence interval (CI) are presented for studies since power may be too low to detect results by study design, and RR and 95% credible intervals moderate asymmetry . are presented for overall pooled estimates. DPP-4 Dipep- tidyl peptidase-4, ES effect size, GLP-1 glucagon-like peptide-1, MEGL meglitinide, METF metformin, Obs RESULTS observational, SGLT-2 sodium–glucose co-transporter 2, SU sulfonylurea, TZD thiazolidinedione A total of 24 randomized clinical trials and 26 observational cohort studies were included in METF (all-cause mortality: RR 1.38, 95% CI the series of meta-analyses. Meta-analysis sum- 1.35, 1.41; cardiovascular mortality: 1.21 95% maries of the effect size (and 95% CIs or credible CI 1.16, 1.27; cardiovascular composite RR 1.18, intervals) for each comparison and outcome are 95% CI 1.15, 1.22), thiazolidinedione (TZD) presented in Figs. 2, 3, and 4. Further informa- (all-cause mortality: RR 1.28, 95% CI 1.13, tion, including both ﬁxed and random effect 1.45), and combination METF ? TZD (all-cause models for each analysis, is presented in ESM S6. mortality: RR 1.76, 95% CI 1.41, 2.20; cardio- vascular composite: RR 1.99, 95% CI 1.47, Pooled Effects by Design 2.69). There were also differential risks when SU Observational Cohort Design combination therapy was evaluated against SU Sixteen meta-analyses (from eight drug-to-drug and METF monotherapy, respectively. A lower comparisons) of only observational cohort risk was associated with METF ? SU combina- studies suggest that treatment with SUs poses a tion therapy when compared to SU monother- greater risk than other therapies. Three of these apy (all-cause mortality: RR 0.75, 95% CI 0.71, comparisons involved SU monotherapy against Diabetes Ther Fig. 3 Pooled RR for cardiovascular mortality. Inverse Fig. 4 Pooled RR for cardiovascular composite events. variance ﬁxed effect estimates are shown for pooled Inverse variance ﬁxed effect estimates are shown for pooled estimates by study design, and two-level hierarchical estimates by study design, and two-level hierarchical Bayesian estimates are shown for overall pooled estimates. Bayesian estimates are shown for overall pooled estimates. RR and the 95% CI are presented for results by study RR and the 95% CI are presented for results by study design, and RR and 95% credible intervals are presented design, and RR and 95% credible intervals are presented for overall pooled estimates for overall pooled estimates 0.80; cardiovascular mortality: RR 0.80, 95% CI inverse variance method and the DerSimo- 0.66, 0.97; cardiovascular composite: RR 0.84, nian–Laird random effect method, such that the 95% CI 0.77, 0.93), and a higher risk was asso- added between-study variance included in the ciated with SU ? METF combination therapy random effects estimates produced wider con- compared against METF monotherapy (all-cause ﬁdence intervals for the pooled effect in all mortality: RR 1.15, 95% CI 1.08, 1.22; cardio- cases, giving statistically non-signiﬁcant esti- vascular mortality: RR 1.47, 95% CI 1.18, 1.82). mates. Thus, substantial heterogeneity existed The remaining analyses found elevated within each of these analyses, with the I effects for SU ? METF combination therapy statistic ranging from 74 to 93%. All of these relative to other METF combinations, such as analyses involved METF ? SU combination METF ? TZD (all-cause mortality: RR 1.20, 95% therapy compared to monotherapies, and they CI 1.08, 1.34; cardiovascular composite: found a lower risk when compared to SU alone RR 1.12, 95% CI 1.03, 1.23), METF ? dipeptidyl (all-cause, cardiovascular composite) and a peptidase-4 (DPP-4) (all-cause mortality: higher risk when compared to METF RR 1.45, 95% CI 1.32, 1.59; cardiovascular monotherapy (on all-cause mortality. cardio- composite: RR 1.46, 95% CI 1.28, 1.68), and vascular death). With the exception of this last METF ? glucagon-like peptide-1 (GLP-1) (all- drug comparison, all of the inconsistent com- cause mortality: RR 1.42, 95% CI 1.00, 2.01). parisons had a similar magnitude and directions In addition, pooled results were statistically of the estimated pooled effects between random inconsistent in four analyses between the ﬁxed effects and ﬁxed effects estimates (see ESM S6). Diabetes Ther Randomized Controlled Trials all of which are known risk factors for CVD. One signiﬁcant elevated effect was found in the Certain SUs affect vascular ATP-sensitive potas- series of analyses using only RCTs. People ran- sium channels (KATP channels); this results in domized to receive the combination METF ? interference with ischemic preconditioning and SU had an 86% increased risk of a cardiovascular the KATP channels possibly not being selective composite event than those assigned combined for pancreatic b-cells and rather binding to therapy with METF ? DPP-4 (pooled RR 1.86, receptors in other tissues, such as cardiomy- 95% CI 1.18, 2.93). All other pooled estimates of ocytes and vascular smooth muscle cells . RCT design studies failed to detect a difference These ﬁndings, together with mounting evi- in risk between SU therapy and other regimens dence from epidemiologic studies, have further for all outcomes. While most comparisons had raised concerns over the use of SUs. the same direction in the effect as pooled The pooled results of the series of meta- observational cohort estimates, precision was analyses reported here suggest that SU therapy often worse in the RCT than in its pooled is associated with an elevated health risk rela- observational cohort counterpart. tive to METF, TZD, GLP-1 agonists, and DPP-4 inhibitors when either compared as a monotherapy or when used in combination Overall Combined Across Study Design with METF. These ﬁndings are almost entirely None of the analyses suggested an elevated derived from observational data (with one effect for SUs when results were combined exception). across RCT and observational cohort study While most RCT-derived estimates were in designs according to all two-level hierarchical the same direction as and had a similar magni- Bayesian models. While the overall direction tude to those for their observational cohort and magnitude of the effect estimates are simi- counterpart, the uncertainty surrounding each lar to that of the pooled estimates from obser- effect was much larger for the former. There- vational cohort designed studies, overall pooled fore, when evidence was pooled using both estimates have considerably wider credible types of study design, there was high variability intervals. This is most likely a result of the around the effect estimates (wide credible added variation existing between study designs. intervals) as a result of the imprecise estimates reported from prior RCT studies. Across all RCTs Publication Bias in this study, the majority that evaluated long- term safety outcomes had small sample sizes Assessing publication bias was limited since with relatively few or no events in a given drug most analyses were excluded if there were \ 10 group occurring during the follow-up period. As studies included. There was no signiﬁcant test a result, existing RCTs were not sufﬁciently result suggesting publication bias according to powered to evaluate long-term safety outcomes. Egger’s test. Pooled estimates from the observational studies suggest worse outcomes for SUs versus older type 2 diabetes drug classes. For the DISCUSSION monotherapy regimens, a higher pooled rela- tive risk was reported for SU monotherapy in Cardiovascular disease is the main cause of comparison to METF on all three safety out- death in people with diabetes, yet evidence on comes, and for TZD on all-cause mortality. The whether particular drug therapies contribute to results also suggest a higher risk for both SU an increase in cardiovascular events and mor- monotherapy and METF ? SU combination tality has been unclear and insufﬁcient. Early therapy than with METF ? TZD combination evidence for concerns over SU use came from therapy for all-cause mortality and cardiovas- the UK Prospective Diabetes Study  and from cular composite events. studies showing that their use is associated with Beginning in 2008, all novel type 2 diabetes weight gain, ﬂuid retention, and hypoglycemia, medications have to undergo a trial focused on Diabetes Ther cardiovascular outcomes. These studies have than previously reported [18, 19], there is typically involved the enrollment of patients increasing pressure to include evidence derived with high cardiovascular risks (those with from non-randomized designs . With this numerous CVD risk factors or with existing increasing demand, a methodological challenge CVD). Evidence from most studies indicate that for researchers is how/whether evidence from novel agents do not pose an increased cardio- observational cohort and RCTs can be com- vascular risk compared to placebo (exception bined to inform key treatment decisions. In our being saxagliptin, for which an increased risk of study, we used a two-level Bayesian model to hospitalization for heart failure has been shown explore how results can be synthesized across ). However, there are several shortcomings study designs. Since there were fewer RCTs than to these studies, with criticism focused on their observational studies in our meta-analyses, this lack of a clear interpretation of the cardiovas- strategy tended to give more weight to RCTs cular risk among the broader indicated popula- than otherwise would occur if the results were tion, an insufﬁcient study period duration simply combined without any consideration of which does not allow understanding of the the study design. However, this strategy also cardiovascular safety proﬁle (there is no included additional variance in the form of mandatory minimum duration set for these between-study design variance in the Bayesian studies), and the fact that placebo-controlled models. trial designs do not provide insight into clini- Future studies should explore whether there cally relevant questions . are other suitable methods to account for Meta-analysis of the results of observational uncertainty and pooling estimates across study cohort studies suggests that SUs have higher designs for the purpose of advancing empirical long-term risks than do the newer potential knowledge and informing evidence-based second-line drug classes on one or more out- medicine practice. In particular, Bayesian mul- comes. Compared to the combination METF ? tilevel models that use informed prior distribu- DDP-4, our results suggest that the combination tions that are formally speciﬁed to reﬂect the METF ? SU poses an increased risk for cardio- relative strength of RCT designs compared to vascular composite events (which is in agree- observational designs would be most beneﬁcial. ment with the RCT pooled results) as well as for Such an approach would assign less weight to all-cause mortality, and compared to the com- study design types that are more susceptible to bination METF ? GLP-1, there was an elevated bias (e.g., observational designs) relative to RCT risk for all-cause mortality. designs. Empirically, these weights might be While new evidence from trials on second- developed via meta-regression examining how line medications after METF with a long follow- effect estimates vary by study design, as has up would ideally be a welcome addition, such been suggested previously . Additionally, studies are typically neither feasible nor timely. expert judgments may be elicited via survey or This is particularly true for any study investi- using a Delphi or group consensus approach, gating comparative safety among the older drug where this information may be quantiﬁed in classes, such as SUs and TZDs. For example, the the form of a prior probability distribution. one trial of second-line TZD use that had a fol- Finally, it is important to note that there are low-up of [ 1 year (TOSCA.IT) was underpow- several shortcomings in existing comparative ered, with only about one-third of the actual safety analyses that need to be explored in events necessary to detect a 20% reduction in future research. While SU therapy is commonly the cardiovascular composite outcome with compared to metformin and TZD, there is lim- 80% power . In addition, the forthcoming ited comparative safety research on how the GRADE study does not have a TZD treatment newer classes of medications compare against arm . SU therapy (e.g., sodium–glucose co-transporter While ongoing trials such as the GRADE and 2 inhibitors). There are even fewer comparative CAROLINA trials may provide evidence on safety analyses that parse out the different newer classes of drugs with a longer follow-up sequencing possibilities involving SU Diabetes Ther combination therapy, such as whether existing ACKNOWLEDGEMENTS therapy (e.g., often METF monotherapy) is dis- continued or augmented when a second-line The authors would like to thank Lewis Kazis, therapy is introduced. Manuel Cifuentes, and Varsha Vimalananda for Also, there were few comparisons that their helpful feedback. included C 10 studies to examine publication bias, and so this factor cannot be ruled out. In Funding. No funding or sponsorship was addition, other biases beyond the types assessed received for this study or publication of this in this study could inﬂuence study effect sizes. article. The article processing charges were In future work, meta-regression is one way to funded by the authors. explore the inﬂuence that various study char- acteristics as well as other effect-modifying fac- Authorship. All named authors meet the tors have on these estimates. International Committee of Medical Journal Editors (ICMJE) criteria for authorship for this article, take responsibility for the integrity of CONCLUSIONS the work as a whole, and have given their approval for this version to be published. While the results from previous studies suggest that type 2 diabetes medications other than SUs Authorship Contributions. W. Ryan Powell appear to have equal glucose-lowering efﬁcacy designed the study, performed the data analysis, both alone and when combined with METF , interpreted the results, wrote the manuscript, further research is needed to determine whether and approved the ﬁnal version. Donald R. Miller they also provide greater long-term safety. In and Cindy L. Christiansen contributed to the this study, meta-analyses using only observa- design, reviewed/edited the manuscript, and tional cohort evidence suggest that SUs pose an gave ﬁnal approval. W. Ryan Powell is the elevated risk when compared to other drug guarantor of this work and, as such, had full classes. RCTs to date have been poorly designed access to all the data in the study and takes to evaluate long-term outcomes with type 2 responsibility for the integrity of the data and diabetes medications, resulting in few events the accuracy of the data analysis. and providing little evidence. The focus of many of these trials has been to make direct Disclosures. The authors (W. Ryan Powell, head-to-head comparisons to assess which Cindy L. Christiansen, and Donald R. Miller) medications work best at managing glucose have nothing to disclose. levels, and they were not designed to examine long-term risks. These trials have typically been Compliance with Ethics Guidelines. This small in size with relatively short follow-up article is based on previously conducted studies periods, thereby limiting the ability to obtain and does not contain any studies with human precise estimates of risk. participants or animals performed by any of the While much of the evidence is derived and authors. will continue to come from observational database studies, the methodological rigor of Data Availability. All study-speciﬁc effect such studies is questionable (e.g., internal sizes analyzed in this series of meta-analyses are threats to validity such as selection bias and included in ESM S6. unmeasured confounding are possible). Since evidence from RCTs on the long-term risks is Open Access. This article is distributed typically not feasible or underpowered, a greater under the terms of the Creative Commons emphasis on designing frameworks for com- Attribution-NonCommercial 4.0 International parative safety research that incorporate evi- License (http://creativecommons.org/licenses/ dence from well-designed, rigorous by-nc/4.0/), which permits any noncommercial observational studies is needed. use, distribution, and reproduction in any Diabetes Ther 10. Higgins JPT, Thompson SG, Deeks JJ, Altman DG. medium, provided you give appropriate credit Measuring inconsistency in meta-analyses. BMJ. to the original author(s) and the source, provide 2003;327(7414):557–60. a link to the Creative Commons license, and indicate if changes were made. 11. Egger M, Davey Smith G, Schneider M, Minder C. Bias in meta-analysis detected by a simple, graphi- cal test. BMJ. 1997;315(7109):629–34. 12. Harbord RM, Harris RJ, Sterne JAC. Updated tests REFERENCES for small-study effects in meta-analyses. Stata J. 2009;9(2):197–210. 1. Bolen S, Tseng E, Hutﬂess S, et al. Diabetes medi- 13. Higgins, JPT, Green S. Cochrane handbook for sys- cations for adults with Type 2 diabetes: an update. tematic reviews of interventions (5.1.0 [updated Rockville: Agency for Healthcare Research and March 2011]). Hoboken: John Wiley & Sons. http:// Quality (US); 2016. www.handbook.cochrane.org. 2. United Kingdom Prospective Diabetes Study 14. Gribble FM, Tucker SJ, Seino S, Ashcroft FM. Tissue (UKPDS) Group. Intensive blood-glucose control speciﬁcity of sulfonylureas: studies on cloned car- with sulphonylureas or insulin compared with diac and b-cell K(ATP) channels. Diabetes. conventional treatment and risk of complications 1998;47:1412–8. in patients with type 2 diabetes (UKPDS 33). Lancet 1998;352(9131):837–853. 15. Scirica BM, Braunwald E, Raz I, et al. Heart failure, saxagliptin and diabetes mellitus: observations 3. Abrahamson MJ. Should sulfonylureas remain an from the SAVOR-TIMI 53 randomized trial. Circu- acceptable ﬁrst-line add-on to metformin therapy lation 2014;130(18):1579–88. in patients with type 2 diabetes? Yes, they continue to serve us well! Diabetes Care. 2015;38(1):166–9. 16. Regier EE, Venkat MV, Close KL. More than 7 years of hindsight: revisiting the FDA’s 2008 guidance on 4. Genuth S. Should sulfonylureas remain an accept- cardiovascular outcomes trials for Type 2 diabetes able ﬁrst-line add-on to metformin therapy in medications. Clin Diabetes. 2016;34(4):173–80. patients with type 2 diabetes? No, it’s time to move on! Diabetes Care. 2015;38(1):170–5. 17. Vaccaro O, Masulli M, Nicolucci A, et al. Effects on the incidence of cardiovascular events of the addi- 5. Jadad AR, Moore RA, Carroll D, et al. Assessing the tion of pioglitazone versus sulfonylureas in patients quality of reports of randomized clinical trials: is with type 2 diabetes inadequately controlled with blinding necessary? Control Clin Trials. metformin (TOSCA.IT): a randomised, multicentre 1996;17:1–12. trial. Lancet Diabetes Endocrinol. 2017;5(11):887–97. 6. Wells G, Shea B, O’Connell D, et al. The New- castle–Ottawa Scale (NOS) for assessing the 18. Nathan DM, Buse JB, Kahn SE, et al. Rationale and quality of nonrandomised studies in metaanalyses design of the glycemia reduction approaches in [Internet]; 2001. http://www.medicine.mcgill.ca/ diabetes: a comparative effectiveness study rtamblyn/Readings%5CThe%20Newcastle%20-% (GRADE). Diabetes Care. 2013;36(8):2254–61. 20Scale%20for%20assessing%20the%20quality% 20of%20nonrandomised%20studies%20in%20meta- 19. Marx N, Rosenstock J, Kahn SE, et al. Design and analyses.pdf. baseline characteristics of the CARdiovascular Outcome Trial of LINAgliptin Versus Glimepiride in 7. Borenstein M, Hedges LV, Higgins JPT, Rothstein Type 2 Diabetes (CAROLINA). Diab Vasc Dis Res. HR. Introduction to meta-analysis. Chichester: 2015;12(3):164–174. Wiley; 2009. 20. 114th U.S. Congress. 21st Century Cures Act; H.R. 8. StataCorp LP. Stata statistical software: release 14. 34 (114-255 Public Law 114–255); 2015. College Station: StataCorp LP; 2015. 21. Goodman SN. Bayesian methods for evidence 9. Peters JL, Rushton L, Sutton AJ, Jones DR, Abrams evaluation. Circulation. 2013;127(24):2367–9. KR, Mugglestone MA. Bayesian methods for the cross-design synthesis of epidemiological and toxi- cological evidence. J R Stat Soc Ser C Appl Stat. 2005;54(1):159–72.
– Springer Journals
Published: May 28, 2018