Incretin based treatments and mortality in patients with type 2 diabetes: systematic review and meta-analysisBMJ 2017; 357 doi: https://doi.org/10.1136/bmj.j2499 (Published 08 June 2017) Cite this as: BMJ 2017;357:j2499
- Jiali Liu, postgraduate1,
- Ling Li, research associate1,
- Ke Deng, research associate1,
- Chang Xu, research associate1,
- Jason W Busse, associate professor2 3 4,
- Per Olav Vandvik, associate professor5 6,
- Sheyu Li, associate professor7,
- Gordon H Guyatt, distinguished professor2 8,
- Xin Sun, professor1
- 1Chinese Evidence-based Medicine Centre and CREAT Group, State Key Laboratory of Biotherapy, West China Hospital, Sichuan University and Collaborative Innovation Centre, Chengdu, 610041, Sichuan, China
- 2Department of Health Research Methods, Evidence and Impact, McMaster University, Hamilton, ON L8S 4K1, Canada
- 3Department of Anesthesia, McMaster University, Hamilton, ON L8S 4K1, Canada
- 4Michael G DeGroote Institute for Pain Research and Care, McMaster University, Hamilton, ON L8S 4K1, Canada
- 5Norwegian Knowledge Centre for the Health Services, N-0130 Oslo, Norway
- 6Department of Medicine, Innlandet Hospital Trust, 2819 Gjøvik, Norway
- 7Department of Endocrinology and Metabolism, West China Hospital, Sichuan University, Chengdu, 610041, Sichuan, China
- 8Department of Medicine, McMaster University, Hamilton, ON L8S 4K1, Canada
- Correspondence to: X Sun
- Accepted 11 May 2017
Objective To assess the impact of incretin based treatment on all cause mortality in patients with type 2 diabetes.
Design Systematic review and meta-analysis of randomised trials.
Data sources Medline, Embase, the Cochrane Central Register of Controlled Trials (CENTRAL), and ClinicalTrials.gov.
Eligibility criteria Randomised controlled trials that compared glucagon-like peptide-1 (GLP-1) receptor agonists or dipeptidyl peptidase-4 (DPP-4) inhibitors with placebo or active anti-diabetic drugs in patients with type 2 diabetes.
Data collection and analysis Paired reviewers independently screened citations, assessed risk of bias of included studies, and extracted data. Peto’s method was used as the primary approach to pool effect estimates from trials, sensitivity analyses were carried out with other statistical approaches, and meta-regression was applied for six prespecified hypotheses to explore heterogeneity. The GRADE approach was used to rate the quality of evidence.
Results 189 randomised controlled trials (n=155 145) were included, all of which were at low to moderate risk of bias; 77 reported no events of death and 112 reported 3888 deaths among 151 614 patients. Meta-analysis of 189 trials showed no difference in all cause mortality between incretin drugs versus control (1925/84 136 v 1963/67 478; odds ratio 0.96, 95% confidence interval 0.90 to 1.02, I2=0%; risk difference 3 fewer events (95% confidence interval 7 fewer to 1 more) per 1000 patients over five years; moderate quality evidence). Results suggested the possibility of a mortality benefit with GLP-1 agonists but not DPP-4 inhibitors, but the subgroup hypothesis had low credibility. Sensitivity analyses showed no important differences in the estimates of effects.
Conclusions Current evidence does not support the suggestion that incretin based treatment increases all cause mortality in patients with type 2 diabetes. Further studies are warranted to examine if the effect differs between GLP-1 agonists versus DPP-4 inhibitors.
Incretin based treatments, including dipeptidyl peptidase-4 (DPP-4) inhibitors and glucagon-like peptide-1 (GLP-1) receptor agonists, represent important options for treating people with type 2 diabetes.1 The American Diabetes Association and the European Association for the Study of Diabetes (EASD) have recommended these drug classes as second line agents for treatment of type 2 diabetes.2 Their effects on glucose control are well established,345 with additional benefits of weight loss, antihypertensive effects, and minimal risk of hypoglycemia.4567891011
A recent large randomised trial (SAVOR-TIMI 53 study12) including patients with type 2 diabetes with established, or at risk for, cardiovascular disease, however, suggested possible increased mortality with saxagliptin versus placebo (5.1% v 4.6%). In response, the US Food and Drug Administration released the following statement in 2015: “A potential increase in all cause mortality with saxagliptin was observed—The ITT on-study analysis suggested an increase in all-cause mortality (HR=1.11, 95.1% CI [0.96 to 1.27]) based on about 800 observed deaths—Sensitivity analyses that censored subjects after treatment exposure showed more unfavorable trends in the risk of all cause mortality—Such trends were observed for both CV and non‐CV related causes of death.”13
This observation raised concern as to whether incretin based treatments could be associated with increased mortality; however, findings from other large trials were inconsistent. The TECOS14 and the EXAMINE trial15—testing effects of sitagliptin and alogliptin—found no significant increase in mortality. Evidence from observational studies is also inconsistent.161718192021 We therefore carried out a systematic review and meta-analysis of randomised controlled trials to determine the effect of incretin based treatments on mortality in patients with type 2 diabetes.
We followed the reporting standards for systematic reviews and meta-analyses of randomised controlled trials according to the PRISMA statements.22
We included randomised controlled trials that compared GLP-1 agonists or DPP-4 inhibitors against placebo, lifestyle modification, or active anti-hyperglycaemic drugs in patients with type 2 diabetes. Eligible studies reported ≥12 weeks’ follow-up and explicitly reported data on all cause mortality.
We searched Medline, Embase, and the Cochrane Central Register of Controlled Trials (CENTRAL) to identify relevant studies from inception to 18 February 2017, without language restrictions. We used database specific subject headings (such as MeSH terms) and free texts terms to search for potentially eligible studies (appendix 1). We searched ClinicalTrials.gov to identify additional relevant clinical trials and confirmed mortality data from all eligible published trials. This trial registry documents all phase II-IV drug trials as required by section 801 of the US Food and Drug Administration Amendments Act (FDAAA 801)23 and typically includes extensive lists of adverse events.24
Paired reviewers, trained in research methods, independently and in duplicate screened titles/abstracts and full texts for eligible articles, assessed risk of bias, and extracted data from each eligible study using standardised pilot tested forms with detailed instructions. Reviewers dealt with discrepancies through discussion or, if necessary, arbitration by a third reviewer.
Risk of bias assessment
Two reviewers independently assessed risk of bias of randomised controlled trials using a modified Cochrane risk of bias instrument that includes response options of “definitely or probably yes” (assigned a low risk of bias) or “definitely or probably no” (assigned a high risk of bias), an approach we have previously validated.252627 The items included random sequence generation; allocation concealment; blinding of participants, caregivers, outcome assessors and outcome adjudicators; infrequent missing outcome data; selective outcome reporting; and other sources of bias (details available at www.evidencepartners.com/resources/methodological-resources/).
For randomised controlled trials, we collected information regarding study characteristics (such as author name, year of publication, study design, sample size, length of follow-up), intervention characteristics (such as baseline treatment, type, dose, and duration of study treatment); patients’ characteristics (such as age, sex, duration of type 2 diabetes, cardiovascular disease, body mass index (BMI), baseline HbA1c, fasting plasma glucose); and mortality outcomes (that is, number of deaths and patients included for analyses in each treatment group). If a published trial did not report the number of deaths, while the corresponding registry report from ClinicalTrials.gov reported mortality, we used outcome data from the registry report. In an extension phase of a trial, if the initial treatment assignment was switched, we collected the outcome data before that point. For trials with multiple follow-up points, data, or reports, we collected outcome data at the longest follow-up.28
Data analysis and rating quality of evidence
We conducted a meta-analysis of all included trials using the reported deaths from treatment and control groups. Given the low event rates in many trials, we used Peto’s method as the primary analysis to pool effect estimates across studies.29 We excluded studies in which no deaths occurred in either study arm.
We conducted sensitivity analyses to examine the impact of using alternative effect measures (odds ratio versus relative risk), pooling methods (Peto versus Mantel-Hanszel (M-H)), statistical models (fixed versus random effects), continuity correction of 0.5 for trials with no events when pooling with the M-H method (versus excluding such trials),3031 and analysis of trials with two or more events (that is, excluding trials with no events or only one event).
We generated a funnel plot and applied Egger’s test to examine publication bias. We also conducted an additional test for publication bias by excluding trials with only one event.
We used the Cochran’s χ2 test and the I2 statistic to examine statistical heterogeneity. Following recent guidance for a credible subgroup effect,32 we planned six hypotheses to explain variability in effect estimates of all cause mortality between studies: risk of cardiovascular disease at baseline (with low versus high versus unclear risk; larger effect in patients with low risk of cardiovascular comorbidities at baseline); type of incretin drug (DPP-4 inhibitors versus GLP-1 agonists; larger effect in trials testing GLP-1 agonists); length of follow-up (52 weeks or shorter versus over 52 weeks; larger effect in trials with longer follow-up); type of control (placebo versus active treatment; larger effect in trials with placebo control); mode of treatment (monotherapy versus add on/combination; larger effect in trials with add on/combination treatment); and individual incretin drugs (different incretins). We conducted univariable random effects meta-regression for each of the six hypotheses, when there were at least 10 trials available for analysis. We also conducted a post hoc multiple regression analysis adjusting for risk of cardiovascular disease at baseline, type of incretin drug, length of follow-up, type of control, and mode of treatment. To explore consistency of an apparent subgroup effect, we additionally conducted two univariable exploratory subgroup analyses by type of incretin drug, one on all cause mortality and another on cardiovascular events, using data from the six large cardiovascular outcome trials that compared either GLP-1 agonists or DPP-4 inhibitors versus placebo.
We used the GRADE approach to rate the quality of evidence and generate absolute estimates of effect for the outcomes.33 To calculate the absolute increase in risk for mortality, we estimated the baseline risk for death from a large cohort study16 that enrolled patients without previous myocardial infarction or stroke.
No patients were involved in setting the research question or the outcome measures, nor were they involved in developing plans for design or implementation of the study. No patients were asked to advise on interpretation or writing up of results. There are no plans to disseminate the results of the research to study participants or the relevant patient community.
Of 19 250 unique reports, reviewers judged 1187 as potentially eligible at title and abstract screening; of these, 189 randomised controlled trials involving 155 145 participants proved eligible (fig 1⇓).
All 189 randomised controlled trials were funded by industry. Among those, 126 enrolled patients with type 2 diabetes at low risk of cardiovascular disease at baseline (specifically excluding patients with certain cardiovascular diseases), and 55 enrolled patients with unclear risk of cardiovascular disease at baseline (without mention of excluding patients with cardiovascular disease). The eight others enrolled patients at high risk of, or with established, cardiovascular disease at baseline, including six large cardiovascular outcomes trials (tables A and B in appendix 2).
The six large cardiovascular outcome trials (TECOS,14 LEADER,34 SUSTAIN-6,35 ELIXA,36 SAVOR-TIMI 53,12 and EXAMINE15 trial) enrolled 3297 to 16 492 patients at high risk of, or with established, cardiovascular disease at baseline, followed patients for a median of 1.5 to 3.8 years, and evaluated one of six incretin agents (sitagliptin, liraglutide, semaglutide, lixisenatide, saxagliptin, or alogliptin) versus placebo. The trials enrolled similar patients: mean age range 60.3-65.0, mean BMI 29.5-32.8, mean baseline HbA1c 7.2-8.7%, mean fasting plasma glucose 8.2-8.7 mmol/L, and a mean or median duration of diabetes of 7.2-13.9 years (table A in appendix 2).
Of the 183 other trials, 153 (83.6%) were clearly labelled as phase III studies. The length of follow-up was 12-234 weeks (median 24 weeks; interquartile range 24-52 weeks); the mean age of participants range was 49.7-74.9; mean or median BMI 21.7-37.1; mean baseline HbA1c 6.6%-10.2%; mean fasting plasma glucose 6.2-12.2 mmol/L; and mean or median duration of diabetes 1.0-15.9 years (table A in appendix 2); 119 tested DPP-4 inhibitors, 68 GLP-1 agonists, and four tested both agents; 130 tested incretin drug versus placebo, 69 versus active comparator, 16 versus both placebo and active comparator; 71 used incretin drugs as monotherapy, 121 as add on/combination treatment, and nine administered both treatment options (table C in appendix 2).
Risk of bias assessment
Among the six large cardiovascular outcome trials, all adequately generated their randomisation sequence, concealed allocation, blinded patients and caregivers, and were free from reporting bias. Five (83.3%) trials had infrequent missing outcome data. The baseline characteristics were generally similar between treatment groups across all the trials (table D in appendix 2).
Of the 183 other randomised controlled trials, 177 (96.7%) adequately generated their randomisation sequence, 170 (92.9%) concealed allocation, and 155 (84.7%) blinded patients and caregivers. The baseline characteristics were generally similar between treatment groups in each trial. Only 47 (25.7%), however, were free from frequent missing outcome data (table D in appendix 2).
Effects on all cause mortality
Of the 189 trials, 77 (40.7%) reported that no deaths occurred during the course of study. The 112 other trials (59.3%) reported 3888 deaths among 151 614 patients, of which 3592 (92.4%) were reported from the six large cardiovascular outcome trials.
Our meta-analysis including the 112 trials that reported at least one death showed no difference in mortality between incretin drug versus control (1925/84 136 v 1963/67 478; odds ratio 0.96, 95% confidence interval 0.90 to 1.02, I2=0%; risk difference 3 fewer events (95% confidence interval 7 fewer to 1 more) per 1000 patients over five years; moderate quality evidence, rated down because of inconsistency between GLP-1 agonists and DPP-4 inhibitors) (fig 2⇓ and table 1⇓). Trials that compared incretin treatment against placebo or no treatment contributed 97% of the weight to the analysis, and trials with an active comparator 4%.
Our funnel plot and statistical test showed no evidence of publication bias (fig 3⇓, Egger’s test P=0.21). Our analysis excluding trials with no events or one event across arms found a similar result (fig A in appendix 3, Egger’s test P=0.33). Sensitivity analyses using an alternative effect measure (M-H relative risk 0.95, 95% confidence interval 0.90 to 1.01), pooling method (M-H odds ratio 0.95, 0.89 to 1.01), statistical model (random effects M-H odds ratio 0.95, 0.89 to 1.02), continuity correction of 0.5 for trials with no events (M-H odds ratio 0.95, 0.89 to 1.01), and by excluding trials with no events or one event across arms (Peto odds ratio 0.96, 0.90 to 1.02) showed similar results (table 2⇓).
Univariable meta-regression showed no association between all cause mortality and any of our six subgroup factors (P=0.38 for length of follow-up; P=0.20 for risk of cardiovascular disease at baseline; P=0.52 for type of incretin drug; P=0.05 for type of control; P=0.80 for mode of treatment; P=0.86 for individual incretin agents; table 3⇓). Subgroup analyses by type of incretin drug (GLP-1 agonists or DPP-4 inhibitors), with data from six large randomised trials that compared incretin agents versus placebo, showed similar finding for all cause mortality (interaction test P=0.09; GLP-1 agonists versus placebo: 654/9350 v 730/9355; M-H odds ratio 0.89, 95% confidence interval 0.80 to 0.99; DPP-4 inhibitors versus placebo: 1120/18 313 v 1088/18 230; 1.02, 0.91 to 1.14; fig 4⇓) and composite cardiovascular events (interaction test P=0.19; GLP-1 agonists versus placebo: 1113/9350 v 1229/9355; 0.88, 0.74 to 1.04; DPP-4 inhibitors versus placebo: 1527/18 313 v 1527/18 230; 0.99, 0.92 to 1.07; fig 5⇓).
Multiple meta-regression, adjusted for length of follow-up, risk of cardiovascular disease at baseline, type of control, and mode of treatment, suggested that GLP-1 agonists, but not DPP-4 inhibitors, are associated with lower all cause mortality (P=0.01 for interaction).
Findings and interpretations
Our systematic review and meta-analysis provides no support for the hypothesis that incretin based treatment is associated with increased mortality in patients with type 2 diabetes. This finding should reassure patients and clinicians and refutes the concern raised by the SAVOR-TIMI trial suggesting increased mortality.
In our univariable meta-regression, we found no differential effects among the six prespecified hypotheses; however, multiple meta-regression suggested that GLP-1 agonists, but not DPP-4 inhibitors, could be associated with lower mortality. When checked against 11 criteria32 for assessing the credibility of an apparent subgroup effect, the following supports the hypothesis. The subgroup analysis was one of the small number of prespecified hypotheses tested; the characteristics were measured at baseline; the test for interaction, based on multiple meta-regression, was significant and the additional exploratory subgroup analysis of a subset of the six large cardiovascular outcome trials showed a similar trend; the subgroup findings seemed to be consistent across related outcomes (such as cardiovascular events, fig 5⇑); and, in terms of biological rationale, animal studies and randomised controlled trials in humans have shown that GLP-1 agonists are associated with a larger sustainable reduction of HbA1c,37 weight loss, and systolic blood pressure control than DPP-4 inhibitors.3839
On the other hand, the subgroup analysis is based on a comparison between trials; it is not robust across statistical analyses (the effect did not appear in the univariable analysis but only in the meta-regression, and the meta-regression is at high risk of over-fitting—six trials account for 92.5% of the weight), and was not prespecified; the apparent reduction in mortality with GLP-1 is both small and of borderline significance. We thus conclude that the subgroup hypothesis has low credibility.
Strengths and limitations
Strengths of our review include a systematic and rigorous approach to the identification of randomised controlled trials investigating the impact of incretin based treatment on mortality. We conducted a limited number of preplanned subgroup analyses to explore for differences in risk of mortality. We used the GRADE approach to assess the quality of evidence that showed convincing evidence that incretin drugs do not increase mortality (table 1⇑).
Our study also has limitations. Firstly, studies might selectively report data regarding death in their full publications, which could lead to risk of selection bias; we attempted to mitigate this risk by reviewing records on ClinicalTrials.gov for unreported deaths and included outcome data from 23 trial registry reports. Secondly, we were unable to assess the long term effects of these drugs: among the 189 trials, only 40 had long term follow-up over 52 weeks, and the longest follow-up in the large trials was 3.8 years. Our subgroup analysis suggesting a possible reduction in mortality with GLP-1 agonists but not with DPP4 inhibitors has limited credibility. We identified 11 trials reporting head-to-head comparison,4041424344454647484950 but with limited information—15 events from 6736 participants—and thus estimates were uninformative.
Comparison with other studies
Three previous meta-analyses515253 have reported the effect of GLP-1 agonists, versus placebo or active treatments, on all cause mortality among patients with type 2 diabetes. One, including 33 trials with more than 12 weeks’ follow-up, nine of which reported at least one death, showed that GLP-1 receptor agonists did not seem to be associated with increased mortality (odds ratio 0.67, 95% confidence interval 0.26 to 1.78)51; another one, including 33 trials with more than 24 weeks’ follow-up, 14 of which reported at least one death, found similar results (0.89, 0.46 to 1.70).52 The other recently published meta-analysis, including five trials with the outcome of mortality, found no effect of GLP-1 agonists on mortality (relative risk 0.90, 95% confidence interval 0.70 to 1.15).53 In comparison, our results, primarily based on large cardiovascular outcome trials, suggest that GLP-1 agonists could be associated with lower mortality in patients at high risk of, or with, established cardiovascular disease.
Four previous meta-analyses have reported the effect of DPP-4 inhibitors, versus placebo or active treatments, on all cause mortality among patients with type 2 diabetes.54555657 Monami and colleagues, who included 41 trials with more than 12 weeks’ follow-up, 18 of which reported at least one death, suggested that DPP-4 inhibitors did not increase mortality (odds ratio 0.78, 95% confidence interval 0.40 to 1.51).54 Savarese and colleagues found no effect of DPP-4 inhibitors on mortality regardless of length of follow-up (relative risk 1.06, 95% confidence interval 0.56 to 2.01, for follow-up <29 weeks; 1.01, 0.91 to 1.13, for follow-up ≥29 weeks). 55 Wu and colleagues, who included 43 randomised controlled trials with 50 982 patients and 1228 deaths, found similar results (1.01, 0.91 to 1.13).56 The final study, also by Savarese and colleagues, included 66 trials with sample size more than 200 patients and report of at least one death, and again found no effect on mortality (odds ratio 1.01, 95% confidence interval 0.93 to 1.09).57 Our study, including 122 smaller trials and three larger trials, showed, with much narrower confidence intervals, that DPP-4 inhibitors had minimal or no impact on mortality.
Compared with these studies, we included a large number of additional studies (38 additional randomised controlled trials of GLP-1 agonists and 56 of DPP-4 inhibitors) and conducted more thorough analyses. The addition of large cardiovascular outcome trials provided more reliable estimates of effects on mortality, and subgroup analysis raised the possibility of differential effects of the two classes of incretin drugs—though the subgroup hypothesis has low credibility.32
Our results provide no support for the hypothesis that incretin based treatment are associated with increased all cause mortality in patients with type 2 diabetes. Additional large well designed randomised trials with adequate follow-up will be necessary to definitively establish or refute possible differences in the effect of GLP-1 agonists and DPP-4 inhibitors on all cause mortality.
What is already known on this topic
Concern has arisen as to whether incretin based treatments are associated with increased mortality, given a recent large randomised trial (SAVOR-TIMI 53 study) that suggested possible increased mortality with saxagliptin versus placebo
Although previously meta-analyses have explored this question, they had important methodological limitations and did not consider much of the currently available evidence
What this study adds
The current evidence provides no support for the hypothesis that incretin based treatment increases all cause mortality in patients with type 2 diabetes
Further studies are warranted to examine if the effect differs between GLP-1 agonists and DPP-4 inhibitors
We thank Daphne Plaut for developing the search strategy and conducting the initial literature search.
Contributors: JL, LL, and KD contributed equally to this work. XS, LL, and POV conceived the study. XS acquired the funding. XS, JL, LL, and KD had full access to all of the data in the study, and take responsibility for the integrity of the data and the accuracy of the data analysis. JL, LL, KD and CX conducted the literature searches and extracted the data. All authors conducted the analysis and interpreted the data and critically revised the manuscript. JL, LL and KD drafted the manuscript. XS is guarantor.
Funding: This study was supported by National Natural Science Foundation of China (grant No 71573183), “Thousand Youth Talents Plan” of China (grant No D1024002), and Sichuan Province, and Young Investigator Award of Sichuan University (grant No 2013SCU04A37). These funders had no role in the study design, writing of the manuscript, or decision to submit this or future manuscripts for publication. JWB is funded by a New Investigator Award from the Canadian Institutes of Health Research and Canadian Chiropractic Research Foundation. SL is funded by the National Natural Science Foundation of China (grant No 81400811 and 21534008).
Competing interests: All authors have completed the ICMJE uniform disclosure form at www.icmje.org/coi_disclosure.pdf and declare: no support from any organisation for the submitted work; no financial relationships with any organisations that might have an interest in the submitted work in the previous three years, and no other relationships or activities that could appear to have influenced the submitted work.
Ethical approval: Not required.
Data sharing: No additional data available.
Transparency: The lead author (the manuscript’s guarantor) affirms that the manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned have been explained.
This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.