Dipeptidyl peptidase-4 inhibitors and risk of heart failure in type 2 diabetes: systematic review and meta-analysis of randomised and observational studies.

OBJECTIVES: To examine the association between dipeptidyl peptidase-4 (DPP-4) inhibitors and the risk of heart failure or hospital admission for heart failure in patients with type 2 diabetes.
DESIGN: Systematic review and meta-analysis of randomised and observational studies. DATA SOURCES: Medline, Embase, Cochrane Central Register of Controlled Trials, and ClinicalTrials.gov searched up to 25 June 2015, and communication with experts. ELIGIBILITY CRITERIA: Randomised controlled trials, non-randomised controlled trials, cohort studies, and case-control studies that compared DPP-4 inhibitors against placebo, lifestyle modification, or active antidiabetic drugs in adults with type 2 diabetes, and explicitly reported the outcome of heart failure or hospital admission for heart failure. DATA COLLECTION AND ANALYSIS: Teams of paired reviewers independently screened for eligible studies, assessed risk of bias, and extracted data using standardised, pilot tested forms. Data from trials and observational studies were pooled separately; quality of evidence was assessed by the GRADE approach.
RESULTS: Eligible studies included 43 trials (n=68,775) and 12 observational studies (nine cohort studies, three nested case-control studies; n=1,777,358). Pooling of 38 trials reporting heart failure provided low quality evidence for a possible similar risk of heart failure between DPP-4 inhibitor use versus control (42/15,701 v 33/12,591; odds ratio 0.97 (95% confidence interval 0.61 to 1.56); risk difference 2 fewer (19 fewer to 28 more) events per 1000 patients with type 2 diabetes over five years). The observational studies provided effect estimates generally consistent with trial findings, but with very low quality evidence. Pooling of the five trials reporting admission for heart failure provided moderate quality evidence for an increased risk in patients treated with DPP-4 inhibitors versus control (622/18,554 v 552/18,474; 1.13 (1.00 to 1.26); 8 more (0 more to 16 more)). The pooling of adjusted estimates from observational studies similarly suggested (with very low quality evidence) a possible increased risk of admission for heart failure (adjusted odds ratio 1.41, 95% confidence interval 0.95 to 2.09) in patients treated with DPP-4 inhibitors (exclusively sitagliptin) versus no use.
CONCLUSIONS: The relative effect of DPP-4 inhibitors on the risk of heart failure in patients with type 2 diabetes is uncertain, given the relatively short follow-up and low quality of evidence. Both randomised controlled trials and observational studies, however, suggest that these drugs may increase the risk of hospital admission for heart failure in those patients with existing cardiovascular diseases or multiple risk factors for vascular diseases, compared with no use. Published by the BMJ Publishing Group Limited. For permission to use (where not already granted under a licence) please go to http://group.bmj.com/group/rights-licensing/permissions.

Introduction

Of over 380 million people with diabetes worldwide, most (85-95%) have type 2 diabetes.1 Dipeptidyl peptidase-4 (DPP-4) inhibitors are a relatively new class of incretin based agents for treating type 2 diabetes. Evidence from randomised controlled trials has established that DPP-4 inhibitors reduce levels of glycated haemoglobin (HbA1c),2 3 do not affect body weight,2 pose a low risk of hypoglycaemia,4 and do not increase the risk of cardiovascular events.5 6 7 The American Diabetes Association and European Association for the Study of Diabetes have recommended this drug class as second line agents for type 2 diabetes management.8

A recent major trial9 (SAVOR-TIMI 53) reported an increased risk of admission to hospital for heart failure (hazard ratio 1.27, 95% confidence interval 1.07 to 1.51) with the DPP-4 inhibitor saxagliptin. Although unexpected, the finding raised concern among professionals and health authorities. In 2014, the US Food and Drug Administration (FDA) requested the clinical trial data from the manufacturer to investigate the potential association between use of saxagliptin and heart failure. The FDA then recommended that “Patients should not stop taking saxagliptin and should speak with their health care professionals about any questions or concerns. Health care professionals should continue to follow the prescribing recommendations in the drug labels.”10

Subsequently, the EXAMINE trial11 testing alogliptin, and the TECOS trial12 testing sitagliptin, reported no significant effect on hospital admission for heart failure. Evidence from observational studies has been inconsistent,13 14 15 16 17 and the effect of DPP-4 inhibitors on heart failure remains controversial.

A systematic review of trials and observational studies offers an opportunity to consider the total body of evidence and potentially resolve the concern. We therefore undertook a systematic review to assess the extent to which DPP-4 inhibitors affect the risk of heart failure or hospital admission for heart failure in patients with type 2 diabetes.

Methods

We followed the standards set by the meta-analysis of observational studies in epidemiology (MOOSE)18 and preferred reporting items for systematic reviews and meta-analyses (PRISMA)19 for the reporting of our study.

Eligibility criteria

We included randomised controlled trials, non-randomised controlled trials, cohort studies, and case-control studies that compared DPP-4 inhibitors against placebo, lifestyle modification, or active antidiabetic drugs in adults with type 2 diabetes. We required follow-up for at least 12 weeks (not applicable to case-control studies), and explicit reporting of the outcome of heart failure or hospital admission for heart failure (either as raw data or adjusted effect estimates with 95% confidence intervals). We classified study designs according to recommendations by the Cochrane Non-Randomised Studies Methods Group. Trials, particularly phase III studies, reported heart failure either as a normal adverse event or a serious adverse event. For serious adverse events, admission for heart failure may have been included. We defined heart failure reported in such trials as an unspecified outcome.

Literature search

We searched Medline, Embase, and the Cochrane Central Register of Controlled Trials (CENTRAL) from inception to 25 June 2015. We combined both MeSH and free text terms for identifying relevant articles. An information expert (DP) developed our search strategies (web appendix 1).

We also searched ClinicalTrials.gov to identify additional eligible studies. Section 801 of the US Food and Drug Administration Amendments Act (FDAAA 801) requires responsible parties to submit summary results of clinical trials, including serious adverse events and adverse events with frequency over 5%, to this trial registry.20 21 In doing so, important information regarding heart failure can be collected. We used generic names of each drug to identify relevant studies, and limited our search to studies labelled as “completed” or “terminated,” in which summary results were available.

We also contacted content experts and industry representatives, and searched for conference abstracts on the American Diabetes Association and European Association for the Study of Diabetes, for additional information.

Study process

Teams of two paired reviewers, trained in health research methods, independently screened titles, abstracts, and full texts for eligibility; assessed risk of bias; and collected data from each eligible study, using standardised, pilot tested forms, together with detailed instructions. Reviewers resolved disagreement through discussion or, if required, by adjudication by a third reviewer (XS).

Risk of bias assessment

We used the Cochrane Collaboration’s tool22 to assess the risk of bias of randomised controlled trials. The items included random sequence generation, allocation concealment, blinding of participants, caregivers, and assessors of outcomes (that is, heart failure or hospital admission for heart failure), and adjudication of the outcomes. By assessing the risk of bias associated with blinding of patients, caregivers, and outcome assessors, we modified the instrument by removing the “unclear” option, an approach that we have previously shown to be reliable and valid.23

We used the Newcastle-Ottawa quality assessment scale24 to assess the risk of bias of cohort studies and case-control studies. We removed the items “representativeness of the exposed cohort” and “was follow-up long enough for outcomes to occur” for cohort studies and the item “representativeness of the cases” for case-control studies because these items relate to applicability of results. We, however, added two items: the ascertainment of type 2 diabetes and the ascertainment of potential confounding factors for these both types of studies, because misclassification could result from suboptimal approaches to these issues. We planned to assess publication bias but were unable to do so owing to very low event rates.

Data collection

We collected the following information from each eligible randomised controlled trial:

• General study characteristics: author name, year of publication, total number of patients randomised, number of treatment groups, length of follow-up, study phase, funding source, trial registry number, countries involved, and number of study sites

• Patient characteristics: sex, age, diabetes duration, body mass index, baseline HbA1c level, and fasting plasma glucose values

• Interventions: medications common to all groups (baseline treatment), details of DPP-4 inhibitors treatment and control group (eg, drug generic name, and duration of treatment)

• Outcomes: the definition of heart failure, number of events, and patients included for analyses in each group, as well as adjusted data if available.

For each trial, if the initial treatment assignment was switched (eg, patients in placebo group started receiving DPP-4 inhibitors after 24 weeks), we collected the outcome data before that point. If a trial had multiple reports, we collated all data into one study.25 If a trial had both reports from ClincialTrials.gov and journal publications, we carefully checked data from these two sources for consistency. If outcome data were reported at multiple follow-up points, we used data from the longest follow-up.

For observational studies, we collected data similar to randomised controlled trials (eg, total number of patients, sex, age, diabetes duration, body mass index, baseline HbA1c). We documented, for each observational study, the definition of outcomes and sources of data for the outcomes. In addition, we documented information on:

• Study design (eg, retrospective or prospective cohort study)

• Data source (eg, claims data, electronic medical records)

• Methods used to ascertain type 2 diabetes status (eg, International Classification of Diseases (ICD) code)

• Exposures (eg, DPP-4 inhibitors, and other exposure variables)

• Methods used to control confounding (eg, logistic or cox regression, and control variables).

We collected adjusted estimates and their associated 95% confidence intervals, as well as adjustment factors, in addition to raw event data and exposure time.

Data analysis

We conducted separate analyses for randomised controlled trials and observational studies. We also separately analysed the data on heart failure and hospital admission for heart failure, because those two outcomes, although sharing the same clinical and pathophysiological features, represent differential seriousness of the effect of DPP-4 inhibitors treatment on patients and society. Heart failure could be subclinical and might not be diagnosed; admission for heart failure is, however, always a clinical event and a condition important to patients and clinicians. We considered admission for heart failure as the more important outcome for patients.

For the analysis of trials, we pooled outcome data using Peto’s methods because of very low event rates,26 27 and reported pooled Peto odds ratios and associated 95% confidence intervals. We examined heterogeneity among studies with the Cochrane χ2 test and the I2 statistic. We explored sources of heterogeneity with four prespecified subgroup hypotheses:

• Type of control (placebo v active treatment; larger effect in trials with placebo control)

• Length of follow-up (≤52 v >52 weeks; larger effect in those with longer follow-up)

• Mode of treatment (monotherapy v add-on or combination therapy; larger effect in those with add on or combination therapy)

• Individual DPP-4 inhibitors (different DPP-4 inhibitors v control).

We carried out sensitivity analyses by using alternative effect measures (odds ratios v risk ratios), pooling methods (Peto v Mantel-Hanszel method), and statistical models regarding heterogeneity (random v fixed effects).

In the analysis of observational studies, we qualitatively summarised the data for heart failure, because of the substantial variations in the comparison (that is, type of control) and patient populations in those studies. We pooled adjusted estimates of hospital admission for heart failure from cohort and nested case-control studies using a random effects model. Although the effect measures differ for those two designs (hazard ratios for cohort studies and odds ratios for nested case-control studies), they are relative measures and the effect estimates are close when the event rate is low (<5%).

Quality of evidence

We used the grading of recommendations assessment, development, and evaluation (GRADE) methodology to rate quality of the evidence for heart failure and hospital admission for heart failure as high, moderate, low, or very low.28 Randomised controlled trials begin as high quality evidence, but can be rated down because of risk of bias,29 imprecision,30 inconsistency,31 indirectness,32 and publication bias.33 Observational studies begin as low quality evidence, but can be rated up for a large magnitude of effect, a dose-response gradient, or presence of plausible confounders or other biases that increase confidence in the estimated effect.34

Patient involvement

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.

Results

Of 11 440 potentially relevant reports identified, after title and abstract screening, 820 reports proved potentially eligible. On full text screening, 55 studies proved eligible, including 43 randomised controlled trials, representing 68 775 patients, reported in 77 reports9 11 12 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 (45 from journal reports, 31 from the ClinicalTrials.gov website, and one conference abstract) and 12 observational studies,13 14 15 16 17 109 110 111 112 113 114 115 involving 1 777 358 patients, reported in nine cohort studies and three nested case-control studies (nine from journal reports, one from a trial registry, and two conference abstracts; fig 1). Two cohort studies15 116 analysed patient data from the same claims database, one presenting a subpopulation of the other; we included only the larger cohort study.15

Fig 1 Flowchart of article selection

Evidence from randomised controlled trials

Trials reporting heart failure

Of the 43 randomised controlled trials, 38 reported heart failure, of which 33 (87%) were international studies, and 35 (92%) were clearly labelled as phase III trials. These 38 trials enrolled 109-2789 patients (total n=31 680; mean age range 49.7-72.6 years, mean body mass index 24.0-32.8, mean baseline HbA1c 7.1-9.9%, mean fasting plasma glucose 7.7-11.1 mmol/L, and mean duration of diabetes 1.7-17.5 years; table 1). Nine trials used DPP-4 inhibitors as monotherapy, 27 as add-on or combination therapy, and two as both monotherapy and combination therapy. Length of follow-up ranged from 12 to 206 weeks (median 52; table 2).

Table 1

Characteristics of included randomised controlled trials

Author (year)	International study	No of countries involved	No of study sites	Study phase	Total No of patients randomised	Length of follow-up (weeks)	Male patients (No, %)	Mean age (years)	Mean body mass index	Mean HbA1c (%)	Mean FPG (mmol/L)	Mean diabetes duration (years)
Trials reporting heart failure
Arjona Ferreira (2013)a35, 36	Yes	NR	NR	III	426	54	158 (57)	64.5	26.8	7.8	8.1	10.4
Arjona Ferreira (2013)b37, 38	Yes	12	31	III	129	54	77 (59.7)	59.5	26.8	7.8	9.0	17.5*
Bosi (2011)39	Yes	NR	NR	III	803	52	414 (51.6)	55.1	31.5	8.2	9.0	7.2
Ferrannini (2009)40	Yes	24	402	III	2789	52	1490 (53.4)	57.5	31.8	7.3	9.2	5.7
Fonseca (2013)41	Yes	12	58	III	313	26	195 (62.3)	56.0	29.9	9.8	9.8	NR
Garber (2007)42	Yes	2	123	III	463	24	199 (50)	54.0	32.4	8.7	10.1	4.7
Henry (2014)43, 44	NR	NR	NR	III	1615	54	912 (56.5)	NR	30.9	8.8	10.0	7.9
Iwamoto (2010)45, 46	Yes	1	97	II	363	12	224 (61.7)	59.8	24.5	7.6	8.2	5.4
NCT00094770 (2009)47, 48, 49	Yes	NR	173	III	1172	104	694 (59.2)	56.7	31.2	7.7	9.2	6.4
NCT00103857 (2009)50, 51	Yes	NR	140	III	1091	104	539 (49.4)	53.5	NR	8.8	11.1	NR
NCT00121641 (2011)52, 53	NR	6	135	III	403	206	204 (50.9)	53.5	31.7	7.9	9.7	2.6
NCT00121667 (2011)54, 55	Yes	9	154	III	745	206	377 (50.7)	54.6	31.4	NR	NR	NR
NCT00286442 (2011)56, 57	Yes	15	115	III	527	26	265 (50.3)	54.8	32.0	7.9	9.5	6.0
NCT00286468 (2011)58, 59	Yes	15	125	III	585	26	261 (52.2)	56.6	30.1	NR	NR	7.7
NCT00295633 (2009)60, 61, 62	Yes	8	133	III	565	76	643 (49.2)	52.0	30.2	9.5	11.1	1.7
NCT00327015 (2009)63, 64, 65	Yes	13	211	III	1309	24	643 (49.2)	52.0	30.2	9.5	11.1	1.7
NCT00395343 (2009)66, 67	Yes	24	100	III	641	24	326 (50.9)	57.8	31.0	8.7	9.8	12.5
NCT00482729 (2009)68, 69, 70	Yes	2	229	III	1250	44	708 (56.8)	49.7	NR	9.9	NR	NR
NCT00575588 (2010)71, 72, 73	Yes	11	130	III	858	104	444 (51.7)	57.5	31.4	7.7	9.0	5.4
NCT00614939 (2010)74, 75, 76	Yes	13	75	III	170	52	73 (42.9)	66.5	30.7	8.3	9.9	16.7
NCT00622284 (2011)77, 78	Yes	16	209	III	1560	104	933 (60.2)	56.6	30.2	7.7	9.1	715 (47.1)†
NCT00642278 (2013)79, 80	Yes	13	85	II	451	12	236 (52.3)	52.9	31.5	7.8	9.0	NR
NCT00707993 (2013)81, 82	Yes	15	110	III	441	54	198 (44.9)	69.9	29.8	7.5	8.1	6.1
NCT00757588 (2011)83, 84	Yes	10	72	III	457	24	188 (41.3)	57.2	32.3	8.7	9.6	11.9
NCT00798161 (2011)85, 86	Yes	14	133	III	791	24	426 (53.9)	55.3	29.1	8.7	10.9	562 (74.3)†
NCT00838903 (2014)87, 88	Yes	10	289	III	1049	164	482 (47.6)	54.5	32.6	8.1	9.2	6.0
NCT00856284 (2013)89, 90	Yes	32	310	III	2639	104	1312 (49.7)	55.4	31.2	7.6	NR	5.5
NCT00954447 (2012)91	Yes	19	167	III	1263	52	658 (52.2)	60.0	31.0	8.3	8.3	NR
NCT01006603 (2013)92	Yes	13	152	IV	720	52	445 (61.8)	72.6	NR	NR	NR	NR
NCT01189890 (2013)93	Yes	NR	NR	III	480	30	202 (42.1)	70.7	NR	7.8	9.4	NR
NCT01263483 (2011)94, 95	No	1	31	II and III	230	12	142 (61.7)	62.1	24.0	8.0	NR	7.8
NCT01289990 (2014)96	Yes	19	243	III	2700	76	1492 (55.3)	55.6	NR	NR	NR	NR
Pratley (2009)97, 98	Yes	14	125	III	493	26	287 (58.2)	55.4	32.8	8.0	NR	7.6
Pratley (2014)99	Yes	13	198	III	784	26	374 (47.7)	53.5	30.7	NR	NR	4.0
Rosenstock (2006)100	Yes	17	NR	III	353	24	196 (55.5)	56.3	31.5	8.0	9.2	NR
Rosenstock (2010)101	Yes	23	268	III	655	26	320 (48.9)	52.6	31.1	8.8	10.6	3.2
Seino (2012)102	No	1	30	III	288	12	198 (68.8)	52.6	25.9	8.0	NR	6.3
Yang (2015)103	No	1	25	III	109	24	57 (54.3)	56.2	25.0	7.1	7.7	3.6
Trials reporting hospital admission for heart failure
Green (2015) (TECOS)12	Yes	38	673	III	14  735	156‡	10 374 (70.7)	65.5	30.2	7.2	NR	11.6
Krum (2014) (VIVIDD)104	NR	NR	NR	NR	253	52	NR	63	NR	7.8	NR	NR
Laakso (2015)105	Yes	9	52	III	235	52	149 (63.4)	66.6	NR	8.1	NR	NR
Scirica (2013) (SAVOR-TIMI 53)9, 106	Yes	26	788	IV	16  492	109‡	11 037 (66.9)	65.0	31.1	NR	8.7	10.3*
Zannad (2015) (EXAMINE)11, 107, 108	Yes	49	898	III	5380	76‡	3651 (67.9)	60.9	29.5	NR	NR	9.2

FPG=fasting plasma glucose; NR=not reported.

*Median diabetes duration (years).

†No (%) of patients with no more than five years’ duration.

‡Median follow-up time (weeks).

Table 2

Interventions tested and event rates in randomised controlled trials

Author (year)	Drug treatments used across groups	DPP-4 inhibitors			Control		Duration of treatment (weeks)
		Type	Events/analysed patients (No)		Type	Events/analysed patients (No)
Trials reporting heart failure
Arjona Ferreira (2013)a35, 36	None	Sitagliptin	0/210		Glipizide	4/212	54
Arjona Ferreira (2013)b37, 38	None	Sitagliptin	2/64		Glipizide	2/65	54
Bosi (2011)39	Metformin, and pioglitazone 30 mg	Alogliptin	2/404		Add-on pioglitazone 15 mg	1/399	52
Ferrannini (2009)40	Metformin	Vildagliptin	2/1389		Glimepiride	2/1383	52
Fonseca (2013)41	Metformin and pioglitazone	Sitagliptin	0/157		Placebo	0/156	26
Garber (2007)42	Pioglitazone	Vildagliptin	1/304		Placebo	1/158	24
Henry (2014)43, 44	Pioglitazone	Sitagliptin	2/691		No additional drugs	0/693	54
Iwamoto (2010)45, 46	None	Sitagliptin	1/290		Placebo	0/73	12
NCT00094770 (2009)47, 48, 49	Metformin	Sitagliptin	2/588		Glipizide	1/584	104
NCT00103857 (2009)50, 51	Metformin	Sitagliptin	1/372		No additional drugs	0/364	104
NCT00121641 (2011)52, 53	None	Saxagliptin	1/306		Placebo	0/95	206
NCT00121667 (2011)54, 55	Metformin	Saxagliptin	3/564		Placebo	2/179	206
NCT00286442 (2011)56, 57	Metformin	Alogliptin	1/423		Placebo	0/104	26
NCT00286468 (2011)58, 59	Glyburide	Alogliptin	1/401		Placebo	0/99	26
NCT00295633 (2009)60, 61, 62	TZD	Saxagliptin	0/381		Placebo	1/184	76
NCT00327015 (2009)63, 64, 65	Metformin	Saxagliptin	0/643		No additional drugs	2/328	24
NCT00395343 (2009)66, 67	Insulin with or without metformin	Sitagliptin	0/322		Placebo	2/319	24
NCT00482729 (2009)68, 69, 70	Metformin	Sitagliptin	1/625		No additional drugs	0/621	44
NCT00575588 (2010)71, 72, 73	Metformin	Saxagliptin	1/428		Glipizide	1/430	104
NCT00614939 (2010)74, 75, 76	OADs and/or insulin	Saxagliptin	1/85		Placebo	0/85	52
NCT00622284 (2011)77, 78	Metformin	Linagliptin	3/776		Glimepiride	2/775	104
NCT00642278 (2013)79, 80	Metformin	Sitagliptin	0/65		Placebo	0/65	12
		Sitagliptin	0/65		Canagliflozin	1/321
NCT00707993 (2013)81, 82	None	Alogliptin	1/222		Glipizide	1/219	52
NCT00757588 (2011)83, 84	Insulin with or without metformin	Saxagliptin	2/304		Placebo	0/151	24
NCT00798161 (2011)85, 86	None	Linagliptin	0/142		Placebo	0/72	24
	Metformin	Linagliptin	1/286		No additional drugs	0/291
NCT00838903 (2014)87, 88	Metformin	Sitagliptin	1/302		Glimepiride	1/307	156
		Sitagliptin	1/302		Placebo	0/101
NCT00856284 (2013)89, 90	Metformin	Alogliptin	3/1751		Glipizide	1/878	104
NCT00954447 (2012)91	Basal insulin and/or OADs	Linagliptin	3/631		Placebo	2/630	52
NCT01006603 (2013)92	None	Saxagliptin	1/359		Glimepiride	3/359	52
NCT01189890 (2013)93	None	Sitagliptin	0/241		Glimepiride	1/236	30
NCT01263483 (2011)94, 95	Voglibose	Alogliptin	0/155		Placebo	1/75	12
NCT01289990 (2014)96	None	Sitagliptin	1/223		Placebo	0/223	76
		Sitagliptin	1/223		Empagliflozin	0/453
Pratley (2009)97, 98	Pioglitazone or pioglitazone, plus metformin or SU	Alogliptin	3/397		Placebo	0/97	26
Pratley (2014)99	None	Alogliptin	0/222		Placebo	0/106	26
	Metformin	Alogliptin	0/220		No additional drugs	0/220
Rosenstock (2006)100	Pioglitazone	Sitagliptin	0/175		Placebo	0/178	24
Rosenstock (2010)101	Pioglitazone	Alogliptin	0/327		No additional drugs	0/163	26
Seino (2012)102	Metformin	Alogliptin	1/188		Placebo	0/100	12
Yang (2015)103	None	Anagliptin	0/68		Placebo	1/40	24
Trials reporting hospital admission for heart failure
Green (2015) (TECOS)12	One or two OADs (metformin, pioglitazone, or SU) or insulin with or without metformin	Sitagliptin	228/7332		Placebo	229/7339	156*
Krum (2014) (VIVIDD)104	Standard diabetes treatment	Vildagliptin	13/128		Placebo	10/124	52
Laakso (2015)105	None	Linagliptin	7/113		Placebo or glimepiride	6/120	52
Scirica (2013) (SAVOR-TIMI 53)9, 106	Antihyperglycaemic drugs	Saxagliptin	289/8280		Placebo	228/8212	109*
Zannad (2015) (EXAMINE)11, 107, 108	Standard of care treatment for type 2 diabetes mellitus	Alogliptin	85/2701		Placebo	79/2679	78*

BG=biguanide; TZD=thiazolidinedione; OADs=oral antidiabetic drugs; SU=sulfonylurea.

*Median treatment time (weeks).

All 38 trials were industry funded. Most (n=24) were identified from ClinicalTrials.gov, of which four91 92 93 96 have not been published in a peer reviewed journal. Because of the limited information in the trial registry, we were unable to adequately assess the risk of bias for these four trials. On the basis of the information we collected, 16 (42%) trials adequately generated their randomisation sequence; 11 (29%) adequately concealed allocation; all trials blinded patients, caregivers, and outcome assessors; eight (21%) adjudicated heart failure events; and four (11%) used blinded assessors to adjudicate heart failure (web appendix 2). The treatment groups of each included trial were generally balanced with respect to demographic and clinical characteristics.

Effects on heart failure

The 38 trials reported 75 heart failure events occurring in 28 292 patients who were treated with at least one drug (raw event rate 0.27%). The definition of heart failure was available in only one trial37; 33 (87%) trials reported heart failure as serious adverse events. The pooling of data from these trials showed no significant difference in the risk of heart failure between DPP-4 inhibitors treatment and control. Event rates were 0.27% for DPP-4 inhibitors versus 0.26% for controls (odds ratio 0.97 (95% confidence interval 0.61 to 1.56), I2=0%; risk difference 2 fewer (19 fewer to 28 more) events per 1000 patients with type 2 diabetes over five years; fig 2 and table 3). We rated the quality of evidence as low because of risk of bias and imprecision (table 3).

Fig 2 Risk of heart failure in patients with type 2 diabetes who received DPP-4 inhibitors versus control from randomised controlled trials

Table 3

GRADE evidence profile of DPP-4 inhibitors and risk of heart failure in type 2 diabetes

Quality assessment							Summary of findings					Quality of evidence
No of participants (studies), follow-up period	Risk of bias	Inconsistency	Indirectness	Imprecision	Publication bias		Study event rates		Relative risk (95% CI)	Anticipated absolute effects (5- year time frame)§
							With control	With DPP-4 inhibitors		Risk with control	Risk difference with DPP-4 inhibitors (95% CI)
Heart failure
28 292 (38), 12-206 weeks	Serious limitation, due to risk of bias*	No serious limitations	No serious limitations	Serious limitation, confidence interval includes important benefit and harm	Undetected		33/12 591 (0.26%)	42/15 701 (0.27%)	Odds ratio 0.97 (0.61 to 1.56)	50 per 1000†	2 fewer (19 fewer to 28 more)	⊕⊕ΟΟLowdue to risk of bias and imprecision
Hospital admission for heart failure
37 028 (5), 1-3 years	No serious limitations	No serious limitations	No serious limitations	Serious limitation, confidence interval includes no harm and important harm	Undetected		552/18 474 (3%)	622/18 554 (3.4%)	Odds ratio 1.13 (1.00 to 1.26)	60 per 1000‡	8 more (0 more to 16 more)	⊕⊕⊕ΟModerate due to imprecision

*Most trials had unclear risk of bias on random sequence generation and allocation concealment (web appendix 2), and the follow-up (median of 52 weeks) was not long enough for heart failure to occur in patients at low risk of cardiovascular disease.

†Baseline risk estimate for heart failure in a five year time frame comes from the control arm of the cohort study we identified to best represent our target population (Kannan 2015111), with 528 events of heart failure in 13 185 participants (4.0%) at four year follow-up across the control and intervention arms.

‡Baseline risk estimate for hospital admission for heart failure in a five year time frame comes from control arms of the five trials we identified to best represent our target population (fig 3) with 552 events in 18 474 participants (30 per 1000) over a 2.5 year follow-up period, in the absence of observational studies providing more credible baseline risk estimates.

§Units are no of events per 1000 patients with type 2 diabetes mellitus over a five year time frame.

Fig 3 Risk of hospital admission for heart failure in patients with type 2 diabetes who received DPP-4 inhibitors versus control from randomised controlled trials

The subgroup analysis by type of control (placebo v active drugs) showed no difference in treatment effects (interaction P=0.57; comparison with placebo, odds ratio 1.17 (95% confidence interval 0.58 to 2.33); comparison with active drugs, 0.89 (0.47 to 1.66); fig A in web appendix 3). The subgroup analyses of the other three prespecified hypotheses showed no difference in treatment effects (figs B-D in appendix 3). Sensitivity analysis using alternative effect measures, statistical methods, and analysis models did not show important changes in pooled effects (figs E-G in web appendix 3).

Trials reporting hospital admission for heart failure

We included three large trials9 11 12 (SAVOR-TIMI 53, EXAMINE, and TECOS) and two small trials104 105 reporting hospital admission for heart failure; all were designed to assess the cardiovascular safety of DPP-4 inhibitors compared with placebo (table 1). The SAVOR-TIMI 53 trial investigated saxagliptin in patients with diabetes who had a renal impairment and cardiovascular disease or multiple risk factors for vascular disease. The EXAMINE trial recruited patients receiving alogliptin with type 2 diabetes and a recent acute coronary syndrome. The TECOS trial examined sitagliptin in patients with type 2 diabetes and cardiovascular disease. In addition, one small trial104 assessed vildagliptin in patients with type 2 diabetes as well as heart failure and a left ventricular ejection fraction less than 40%; the other small trial105 assessed linagliptin in patients with type 2 diabetes with moderate to severe renal impairment.

All three large trials were international studies. The median length of follow-up ranged from 76 to 156 weeks (table 1). Those trials enrolled 5380-16 492 patients (total n=36 607; mean age range 60.9-65.5 years, mean body mass index 29.5-31.1, and duration of diabetes 9.2-11.6 years). The two small trials followed up patients for 52 weeks; mean age ranged from 63 to 66.6 years and mean HbA1c levels ranged from 7.8% to 8.1%.

All trials, but one104 (which had unclear details because it was presented as an abstract), adequately generated their randomisation sequence and adequately concealed allocation; all trials blinded patients, caregivers, outcome assessors, and centrally adjudicated hospital admission for heart failure outcome through a clinical events classification committee who were blinded to treatment allocation. All trials were funded by industry (web appendix 2).

Effects on hospital admission for heart failure

All five trials9 11 12 104 105 reported unadjusted rates of hospital admission for heart failure. Overall, 1174 events of admission for heart failure occurred in 37 028 patients (raw event rate 3.4% for DPP-4 inhibitors v 3.0% for controls; table 3). Pooling across trials showed a borderline increase in the risk of hospital admission for heart failure in patients with type 2 diabetes using DPP-4 inhibitors versus control (odds ratio 1.13 (95% confidence interval 1.00 to 1.26), I2=0%; risk difference 8 more (0 more to 16 more) per 1000 patients with type 2 diabetes over five years; fig 3 and table 3). We rated the quality of evidence as moderate due to imprecision (table 3). Sensitivity analysis by use of alternative effect measures, statistical methods, and analysis models did not show important changes in the pooled effects (figs H-J in web appendix 3).

Evidence from observational studies

Of 12 observational studies, four109 110 111 112 reported heart failure, and eight13 14 15 16 17 113 114 115 reported hospital admission for heart failure; nine13 14 15 109 110 111 113 114 115 were cohort studies and the other three16 17 112 were nested case-control studies (fig 1).

Observational studies reporting heart failure

Of the four studies reporting heart failure, two prospective cohort studies109 110 compared DPP-4 inhibitors versus sulfonylureas and sitagliptin versus sulfonylureas. One retrospective cohort study111 assessed DPP-4 inhibitors versus sulfonylureas and reported the findings from the subgroup of DPP-4 inhibitors. Finally, one nested case-control study112 using claims data investigated use of sitagliptin versus no use in patients admitted to hospital for acute coronary syndrome (table 4 and table 5). Sample sizes ranged from 616 to 13 185, and the mean or median length of follow-up ranged from one to four years. Enrolled patients had a mean or median age ranging from 55 to 65.8 years. None of the studies explicitly defined provided diagnostic criteria for heart failure.

Table 4

Characteristics of included observational studies

Author (year)	Study design	Data source	Countries	Funding	Total No of patients	Follow-up (years)	Male patients (No (%))	Mean age (years)	Mean body mass index	Mean HbA1c (%)	Mean FPG (mmol/L)	Mean diabetes duration (years)	CVD at baseline
Studies reporting heart failure
Gitt (2013)109	Prospective cohort study	Registry data	Germany	Private for-profit funding	616	1	3097 (49.8)	65.8†	NR	7.3†	7.7†	4.9†	Included patients had CVD or had no CVD at baseline
NCT01357135 (2014)110	Prospective cohort study	Electronic medical records	France	Private for-profit funding	3453	3	2004 (58.0)	63.5	NR	NR	NR	NR	NR
Kannan (2015)*, 111	Retrospective cohort study	Electronic health records	USA	No funding	13 185	4†	7827 (54.6)	60.6	32.6†	NR	NR	NR	Included patients had no history of CVD or congestive heart failure at baseline
Eurich (2014)112	Nested case-control study	Claims data	USA	NR	5027	NA	3268 (65)	55	NR	NR	NR	NR	Included patients had no history of heart failure in the 3 years before admission to hospital for an acute coronary syndrome event
Studies reporting hospital admission for heart failure
Fadini (2015)13	Retrospective cohort study	Registry data	Italy	Public funding	127 555	2.6	66 201 (51.9)	67.0	NR	NR	NR	NR	Included patients had CVD or no CVD at baseline
Fu (2015)14	Retrospective cohort study	Claims data	USA	NR	218 556	0.5	NR	NR	NR	NR	NR	NR	Included patients had CVD or no CVD at baseline
Seong (2015)113	Retrospective cohort study	Claims data	South Korea	No funding	349 476	0.6	191 167 (54.7)	58.3	NR	NR	NR	NR	Included patients had no history of CVD within 2.5 years before cohort entry
Suh (2015)114	Retrospective cohort study	Claims data	South Korea	NR	935 519	0.9	518 614 (55.4)	59.4	NR	NR	NR	NR	NR
Velez(2015)*, 115	Retrospective cohort study	Electronic medical records	USA	Public funding	4224	2.0†	2265 (53.6)	60.8	NR	8.0	NR	2.5	Included patients had CVD or no CVD at baseline
Wang (2014)15	Retrospective cohort study	Claims data	Taiwan	Public funding	16 576	1.5†	8615 (52.0)	64.3	NR	NR	NR	8.6	Included patients had CVD or no CVD at baseline
Weir (2014)16	Nested case-control study	Claims data	USA	NR	45 434	NA	27 013 (59.5)	54.6	NR	7.5	NR	NR	Included patients were recently diagnosed with heart failure
Yu (2015)*, 17	Nested case-control study	Electronic medical records	UK	Public funding	57 737	NA	32 795 (56.8)	61.6	NR	NR	NR	2.3	Included patients had CVD or no CVD at baseline

FPG=fasting plasma glucose; CVD=cardiovascular disease; NR=not reported; NA=not applicable.

*Three studies accessed incretin agents (both glucagon-like peptide 1 receptor agonists and DPP-4 inhibitors) and the risk of heart failure, so the data above were the characteristics of total patients included.

†Median value.

Table 5

Exposures, outcomes, and results of included observational studies

Author (year)	Exposure of interest	Control group	No of events or cases	Total no of analysed patients	Adjusted estimates (95% CI)	Adjusted covariate
Studies reporting heart failure
Eurich (2014)112	Sitagliptin use	No use	457	5027	OR 0.75 (0.38 to 1.46)	Demographics, clinical and laboratory data, pharmacy claims, healthcare use and propensity scores (conditional probability of being treated with metformin or sulfonylurea or insulin or sitagliptin)
Kannan (2015)111	DPP-4 inhibitors (combined with metformin)	Sulfonylureas (combined with metformin)	528*	13 185 (55 110 person years)*	HR 1.10 (1.04 to 1.17)	Age, sex, race, body mass index, number of encounters, median household income, smoking status, systolic and diastolic blood pressure, hypertension, dyslipidaemia, cerebral vascular event, presence of neuropathy, retinopathy, dementia, chronic obstructive pulmonary disease, cancer, atrial fibrillation, antihypertensive drugs, lipid lowering agents, antiplatelet agents, and propensity for being on metformin and sulfonylureas at baseline, lipid profile, estimated glomerular filtration rate
Gitt (2013)109	DPP-4 inhibitors	Sulfonylureas	11	616	NR	NR
NCT01357135 (2014)110	Sitagliptin (combined with metformin)	Sulfonylureas (combined with metformin)	2	2607	NR	NR
Studies reporting hospital admission for heart failure
Fadini (2015)13	DPP-4 inhibitors	Sulfonylureas	1181	110 757	HR 0.78 (0.62 to 0.97)	Age, sex, use of certain medications (drugs for hypertension, dyslipidaemia, chronic obstructive pulmonary disease, non-steroidal anti-inflammatory drugs, and antiplatelet drugs), presence of previous hospital admissions, Charlson index level grouped into three categories, previous use of oral glucose lowering drugs, cotreatment with metformin, and adherence level categorised on the basis of the medication possession ratio (MPR (%); <80% v ≥80%)
Fu (2015)14	DPP-4 inhibitors	Sulfonylureas	495	218 556	No CVD at baseline:HR 0.59 (0.38 to 0.89); CVD at baseline:0.95 (0.78 to 1.15)	Adjusted covariates of Cox proportional hazard models were not stated explicitly; each comparison consisted of patients matched 1:1 on a propensity score based on demographics, general clinical characteristics, and hospital admission for heart failure risk factors from one year before baseline; analyses were stratified by presence of CVD
Seong (2015)113	DPP-4 inhibitors	Sulfonylureas and pioglitazone	212	349 476 (211 959 person years)	DPP-4 inhibitors v sulfonylureas: adjusted HR 0.93 (0.62 to 1.41); DPP-4 inhibitors v pioglitazone: 0.21 (0.15 to 0.28)	Adjusted factors included age, sex, duration of diabetes at baseline; comorbidities in year before the index date (microvascular complications of diabetes (retinopathy, neuropathy, or nephropathy), peripheral vascular disease, hypertension, and dyslipidemia), and associated Charlson score; diabetes related hospital admission and total number of hypoglycaemic drug classes used in year before the index date; and use of the following drug classes in year before the index date: hypoglycaemic, lipid lowering, antihypertensive, antiplatelet (drug names not listed here)
Suh (2015)114	DPP-4 inhibitors	Pioglitazone	998	935 519	Sitagliptin v pioglitazone: adjusted HR 0.97 (0.80 to 1.16); vildagliptin v pioglitazone: 1.22 (0.99 to 1.50)	Age and sex
Velez (2015)115	DPP-4 inhibitors	Control (no details)	127	3987	HR 0.58 (0.38, 0.88)	Propensity score, number of antidiabetic drugs, duration of diabetes, baseline beta blocker use, and use of angiotensin converting enzyme inhibitor or angiotensin receptor blocker
Wang (2014)15	Sitagliptin use	No use	614	16 576	HR 1.21 (1.04 to 1.42)	Adjusted covariates of Cox proportional hazard models were not stated explicitly; potential confounding were mitigated by the propensity score matching approach, and covariates included age, sex, duration of diabetes, antidiabetic drugs used, comorbidities, and outpatient visit
Weir (2014)16	Sitagliptin use	No use	824	9062	OR 1.84 (1.16 to 2.92)	Demographics (age, sex, and socioeconomic status), most recent clinical laboratory data (HbA1c, low and high density lipoprotein cholesterol, triglycerides, estimated glomerular filtration rate, albuminuria, and haemoglobin concentrations), history of CVD (ischaemic heart disease, myocardial infarction, dyslipidaemia, hypertension, arrhythmia, and valve disease), and prescription drug use (antiplatelet drugs, anticoagulants, statins, calcium channel blockers, β blockers, angiotensin converting enzyme inhibitors, renin inhibitors, diuretics, and nitrates)
Yu (2015)17	DPP-4 inhibitors (sitagliptin, vildagliptin, and saxagliptin, alone or in combination with other antidiabetic drugs)	Other oral antidiabetic drugs	1118*	18 744*	OR 0.88 (0.63 to 1.22)	Sex, body mass index, excessive alcohol use, smoking status, HbA1c level, comorbidities (neuropathy, renal disease, retinopathy, atrial fibrillation, cancer (other than non-melanoma skin cancer), chronic obstructive pulmonary disease, coronary artery disease, dyslipidaemia, hypertension, previous myocardial infarction, peripheral arteriopathy, previous coronary revascularisation, peripheral vascular disease, and previous stroke), number of prescriptions, number of physician visits, and use of the following drugs in the year before cohort entry: angiotensin converting enzyme inhibitors, angiotensin receptor blockers, beta blockers, calcium channel blockers, diuretics, fibrates, statins, aspirin, and other non-steroidal anti-inflammatory drugs

NR=not reported; HR=hazard ratio; OR=odds ratio; CVD=cardiovascular disease.

*These two studies accessed incretin drugs and the risk of heart failure, and data of events/cases and total number of analysed patients regarding glucagon-like peptide 1 receptor agonists and DPP-4 inhibitors were not reported separately, so the data above were of total study patients.

Four studies used registry data, electronic health or medical records, or claims data for their analyses. Patients with type 2 diabetes were ascertained by physicians in one prospective cohort study109 or by ICD-9 Clinical Modification (CM) codes in one nested case-control study112; the other two cohort studies110 111 did not explicitly report the ascertainment of type 2 diabetes. None of these studies mentioned the ascertainment of exposure to DPP-4 inhibitors agents and other confounding variables; the accuracy of ascertaining exposure and confounding factors was unclear. Of these three cohort studies, only one111 demonstrated that the outcome of interest was not present at start of study, and mentioned the method used to assess the outcome of interest. Of these four studies, two111 112 controlled for the effect of confounding factors (web appendices 4 and 5).

All three cohort studies109 110 111 reported unadjusted rates of heart failure, involving 541 events among 16 408 patients (raw event rate 3.3%). Because of the heterogeneous and indirect nature of the identified evidence—with substantial variations in comparisons and types of patients—we did not pool data across studies. The outcome information is presented in table 5.

One retrospective cohort study111 and one nested case-control study112 reported adjusted data. The retrospective cohort study, including 13 185 patients and with a median follow-up of four years, suggested that, compared with sulfonylureas, DPP-4 inhibitors was statistically associated with an increased risk of congestive heart failure (adjusted hazard ratio 1.10, 95% confidence interval 1.04 to 1.17). The nested case-control study, selecting 457 heart failure cases and 4570 controls, showed no statistical difference in the risk of heart failure between use and no use of sitagliptinin the 90 days before acute coronary syndrome (adjusted odds ratio 0.75, 95% confidence interval 0.38 to 1.46, table 5). Using GRADE, we rated the quality of evidence in the identified studies as very low, owing to risk of bias, indirectness, and imprecision in addition to the inherent risk for confounding given the observational design.

Observational studies reporting hospital admission for heart failure

Of the eight studies reporting hospital admission for heart failure, six retrospective cohort studies,13 14 15 113 114 115 using registry data, claims data, or electronic medical records, assessed DPP-4 inhibitors versus active drugs (eg, sulfonylureas, pioglitazone), and the use of sitagliptin versus no use of sitagliptin. The other two nested case-control studies16 17 assessed use of sitagliptin versus no use of sitagliptin, and incretin based drugs (including the DPP-4 inhibitors subgroup) versus other oral antidiabetic drugs (tables 4 and 5). The sample sizes of these eight studies ranged from 4224 to 935 519, and the mean or median length of follow-up ranged from 0.5 to 2.6 years. Enrolled patients had a mean age ranging from 54.6 to 67 years, mean baseline HbA1c level ranging from 7.5% to 8.0%, and a mean duration of diabetes ranging from 2.3 to 8.6 years.

The eight studies used registry data, claims data, or electronic medical records for analyses. Only two studies15 113 explicitly reported use of ICD codes to ascertain patients with diabetes; one study13 ascertained exposure to DPP-4 inhibitors by using anatomical therapeutic chemical classes; three studies13 15 113 explicitly stated use of ICD codes to ascertain other confounding variables. Four studies13 15 113 114 used ICD-9 or ICD-10 codes to assess outcomes. Three cohort studies13 113 114 clarified that the outcome of interest was not present at enrolment. All eight studies controlled for potential confounding factors, but failed to specify the extent to which the data were complete in the database (web appendices 4 and 5).

All but one retrospective cohort study115 reported unadjusted rates of hospital admission for heart failure. The five cohort studies13 14 15 113 114 included 3500 events among 1 630 884 patients (raw event rate 0.2%; 1466 events (0.2%) in 912 309 patients from the DPP-4 inhibitors group, and 2034 events (0.3%) in 718 575 patients from the control group). The two nested case-control studies16 17 involved 1942 cases among 27 806 patients. Because of the variety of confounding factors investigated in the studies, we did not pool the unadjusted data.

All eight studies reported adjusted estimates of hospital admission for heart failure. Of these, six studies—five cohort studies and one nested case-control study—compared DPP-4 inhibitors with active drugs (sulfonylureas, pioglitazone, other oral antidiabetic drugs). Pooling of adjusted estimates from these six studies showed that DPP-4 inhibitors were associated with reduced risk of hospital admission for heart failure (adjusted odds ratio 0.85, 95% confidence interval 0.74 to 0.97; I2=31%). However, pooling of the cohort study15 (16 576 patients and 614 events), and the nested case-control study16 (824 cases and 8238 controls) suggested a non-significant trend for increased risk of admission for heart failure compared with no use of sitagliptin (adjusted odds ratio 1.41, 0.95 to 2.09; I2=65%). There was significant subgroup effect by type of control (interaction P=0.02, fig 4). Using GRADE, we rated the quality of evidence as very low, due to risk of bias, heterogeneity, and imprecision in addition to the inherent risk for confounding given the observational design.

Fig 4 Risk of hospital admission for heart failure in patients with type 2 diabetes who received DPP-4 inhibitors versus control based on adjusted data from observational studies. SE=standard error; IV=inverse variance

Table 6 summarises the evidence regarding the effects of DPP-4 inhibitors on heart failure or hospital admission for heart failure.

Table 6

Risk of heart failure or hospital admission for heart failure among patients with type 2 diabetes receiving DPP-4 inhibitor treatment

Comparison	No of studies (events or cases, patients)	DPP-4 inhibitors (events/patients)	Control (events/patients)	Effect estimate (95%CI)	Cardiovascular morbidities at baseline
Heart failure
Randomised controlled trials
DPP-4 inhibitors v control	38 (75, 28 292)	42/15 701	33/12 591	Pooled OR 0.97 (0.61 to 1.56)	Typically without CVD
Observational studies
DPP-4 inhibitors v SU	1 (11, 616)	8/436	3/153	Unadjusted OR 0.88 (0.22 to 3.48)	With or without CVD
DPP-4 inhibitors v SU	1 (528, 13 185)	NR	NR	Adjusted HR 1.10 (1.04 to 1.17)	No history of CVD or congestive heart failure
Sitagliptin v SU	1 (2, 2607)	1/1874	1/733	Unadjusted OR 0.39 (0.02 to 6.26)	NR
Sitagliptin use v no use	1 (457, 5027)	—	—	Adjusted OR 0.75 (0.38 to 1.46)	Admission to hospital for an acute coronary syndrome event
Hospital admission for heart failure
Randomised controlled trials
DPP-4 inhibitors v control	5 (1174, 37 028)	622/18 554	522/18 474	Pooled OR 1.13 (1.00 to 1.26)	CVD or multiple risk factors for vascular disease
Observational studies
DPP-4 inhibitors v active control (pooled estimates)	6 (4341, 1 618 295)	—	—	Pooled adjusted OR 0.85 (0.74 to 0.97)	With or without CVD
DPP-4 inhibitors v SU	3 (1875, 657 596)	380/202 292	1495/455 304	Adjusted HR 0.84 (0.74 to 0.96)	With or without CVD
DPP-4 inhibitors v pioglitazone	2 (1060, 1 031 432)	796/776 449	264/254 983	Adjusted HR 0.67 (0.57 to 0.78)	With or without CVD
DPP-4 inhibitors v other OADs	1 (1118, 18 744)*			Adjusted OR 0.88 (0.63 to 1.22)	With or without CVD
DPP-4 inhibitors v control	1 (127, 3987)	NR	NR	Adjusted HR 0.58 (0.38, 0.88)	With or without CVD
Sitagliptin use v no use (pooled estimates)	2 (1438, 25 638)	—	—	Pooled adjusted OR 1.41 (0.95 to 2.09)	—
Sitagliptin use v no use	1 (614, 16 576)	339/8288	275/8288	Adjusted HR 1.21 (1.04 to 1.42)	With or without CVD
Sitagliptin use v no use	1 (824, 9062)*			Adjusted OR 1.84 (1.16 to 2.92)	Heart failure at baseline

CVD=cardiovascular disease; SU=sulfonylurea; OR=odds ratio; HR=hazard ratio; NR=not reported; OADs=oral antidiabetic drugs.

*Nested case-control study.

Discussion

Main findings

The only evidence of moderate quality from our results is from randomised controlled trials that examined the effect of DPP-4 inhibitors on hospital admission for heart failure. These studies suggested a small increase, in both relative and absolute terms, in heart failure admissions in patients using DPP-4 inhibitors than those not. The results, however, are of borderline significance. Evidence from observational studies is of very low quality, and thus has little bearing on any inferences about DPP-4 inhibitor effects on heart failure admission.

With respect to the incidence of heart failure, trial evidence leaves uncertainty regarding the relative effect of DPP-4 inhibitors. Because the follow-up was relatively short and the baseline risk of patients was very low in those trials, the incidence of heart failure was very low (well under 1% per year), and with the small number of events, the confidence intervals around relative effects are wide. In addition, heart failure was unspecified in all but one of the phase III trials. Many (87%) reported heart failure as serious adverse events, in which admission for heart failure might have been included according to the definition of serious adverse events. The pooled estimate could thus represent a composite of heart failure with or without admission for heart failure. The observational studies again provide very low quality evidence and have little effect on inferences, although results are consistent. Overall, the current evidence provides no support for the hypothesis that DPP-4 inhibitors increase the incidence of heart failure.

Strengths and limitations

Our study has several strengths. Firstly, we used rigorous methods to systematically identify and include data from both randomised and non-randomised studies to examine the effect of DPP-4 inhibitors on risk of heart failure and hospital admission for heart failure. Secondly, in addition to published reports, we have identified additional data from ClinicalTrials.gov. Our study included four randomised controlled trials and three observational studies that were not published in journals. Thirdly, we instituted a rigorous approach to ensure the data were accurate. In particular, we carefully checked the data reported in ClinicalTrials.gov and journal publications for consistency. Fourthly, we addressed several prespecified subgroup analyses to explore sources of heterogeneity. Finally, we used GRADE to assess the quality of the body of evidence.

Our study also had some limitations. Firstly, for various reasons, some trials are likely not to report outcome data in their full publications. However, we have obtained additional data through the search of the ClinicalTrials.gov and conference abstracts, which minimised the risk of outcome reporting bias. Secondly, given the limitations of reported data, we were unable to confirm whether the increased risk of hospital admission for heart failure was a class effect or a specific effect of saxagliptin. Other limitations included those of the primary studies, such as the risk of bias of observational studies, the potentially variable specification of outcomes (heart failure and hospital admission for heart failure), and the likelihood of variable and incomplete ascertainment of heart failure in the clinical trials.

Comparison with other studies

Four previous meta-analyses7 117 118 119 have explored the effect of DPP-4 inhibitors on the risk of heart failure. Of those studies, one7 found that treatment with DPP-4 inhibitors for 29 weeks or longer was associated with an increased risk of new onset of heart failure (risk ratio 1.16, 95% confidence interval 1.01 to 1.33), but not with treatment for less than 29 weeks (0.67, 0.32 to 1.40). The second117 included 24 randomised controlled trials that enrolled no less than 100 patients and followed up patients for 24 weeks; the third118 exclusively included 37 trials for analysis; the fourth119 included trials and observational studies. All the last three studies found that DPP-4 inhibitors were statistically associated with an increased risk of heart failure (risk ratio 1.16 (1.01 to 1.33), odds ratio 1.19 (1.03 to 1.37), odds ratio 1.15 (1.02 to 1.29), respectively).

Compared with these studies, our review has added substantial information. Firstly, we separately addressed heart failure and hospital admission as a result of heart failure. Secondly, we included both observational studies and randomised controlled trials. With respect to the trials, two important large trials11 12 were published subsequent to the previous reviews and allowed us to analyse the effect of DPP-4 inhibitors on hospital admission for heart failure. We also included additional large observational studies that carry important information regarding the risk of heart failure or admission for heart failure.

Our findings regarding the effect of DPP-4 inhibitors on heart failure were not consistent with previous meta-analyses. This difference is probably due to the fact that the previous studies were dominated by large trials reporting positive association with hospital admission for heart failure (eg, SAVOR TIMI-53), and more recent trials that have failed to find an effect were not considered.

We also found all four meta-analyses in our study7 117 118 119 to have several methodological issues. Firstly, these reviews have pooled data for heart failure and hospital admissions for heart failure. We believe that a more appropriate analysis should consider the two outcomes separately. We identified varying results when analysing the two outcomes separately. More importantly, the pooling of the two outcomes together would probably result in misleading effect estimates, when the authors aimed to assess the effect of DPP-4 inhibitors on the risk of heart failure. Another meta-analysis7 investigated DPP-4 inhibitors on the risk of new onset of heart failure, but this study included trials, such as SAVOR TIMI-53 and EXAMINE that already included patients with heart failure at baseline. The third meta-analysis117 failed to include outcome data published in ClinicalTrials.gov. The final meta-analysis119 combined randomised controlled trials and observational studies to generate grand effect estimates. Because of the substantial differences in the design and analysis of the type of studies, and the considerable variation in observational studies, the grand pooling will introduce misleading findings.

Implications for practice

The current evidence suggests a possible increased risk of hospital admission for heart failure in those patients with type 2 diabetes treated with DPP-4 inhibitors and with cardiovascular diseases or multiple risk factors for vascular diseases at baseline. Although the effect is small if it exists, and the associated confidence interval includes no effect, our results suggest the advisability of caution in the use of DPP-4 inhibitors for patients with type 2 diabetes who are at high risk for heart failure.

Conclusions

The relative effect of DPP-4 inhibitors on heart failure remains uncertain in patients with type 2 diabetes, given the relatively short follow-up and low quality of evidence. The current evidence suggests a small increase in the risk of hospital admission for heart failure in patients with existing cardiovascular diseases or multiple risk factors for vascular diseases. Additional randomised controlled trials enrolling patients with existing cardiovascular diseases or multiple risk factors for vascular diseases will be required to definitively assess the effect of DPP-4 inhibitors on such patients. Such trials, if enrolling patients at high risk of exacerbation and admission, may be feasible. In the meantime, the possible increase in hospital admission for heart failure could be one issue that patients and clinicians consider in choosing antidiabetic drug treatment for patients with existing cardiovascular diseases.

Several occurrences of heart failure or hospital admission for heart failure have been reported in patients with type 2 diabetes taking DPP-4 inhibitors

Systematic reviews of randomised controlled trials and observational studies have suggested an increased risk of heart failure or admission for heart failure associated with the agents

The relative effect of DPP-4 inhibitors on the risk of heart failure is uncertain

Current evidence from trials and observational studies suggests a small increase in risk of admission for heart failure in patients with type 2 diabetes who have existing cardiovascular diseases or multiple risk factors for vascular diseases, relative to no use