Effectiveness and safety of dipeptidyl peptidase 4 inhibitors in the management of type 2 diabetes in older adults: a systematic review and development of recommendations to reduce inappropriate prescribing

Database searches were conducted by YVM. We started searching for systematic reviews and meta-analyses (search 1 and 2). During study selection under search 1 and 2, we identified eligible individual studies from excluded systematic reviews and meta-analyses and transferred these to the Search 3A list for potential inclusion. The list of studies in Search 3A was checked for inclusion following the procedures described below under “Selection of studies”. Only one relevant meta-analysis was found from Searches 1 and 2. However, this meta-analysis covered just one type of DPP-4 inhibitor (linagliptin). Therefore, we conducted Search 3B for individual studies published in the last 10 years (2005–2015) [21]. Detailed information about databases and search dates is summarised below:

In addition to database searches, we checked the references of included reviews and studies following the procedures described later under “Selection of studies”. A list of excluded studies after full-text check with reasons for exclusion is provided in Additional file 1.

The PICOS-framework was used to develop the search terms (population: older people with type 2 diabetes, intervention: DPP-4 inhibitors, comparison: any, outcomes: see list above “Types of outcomes” and study design: systematic reviews, meta-analyses, controlled interventional studies and observational studies). We also created search filters specific to different study designs and each filter is described in detail in the protocol [21]. Additional file 2 shows the full search terms for each search (i.e. Searches 1, 2 and 3B).

Literature search results were uploaded to the Endnote X7 reference management software. Endnote was used to import search results and to de-duplicate references.

First, two independent reviewers assessed titles and abstracts from each search and identified studies to include. Second, full manuscripts were obtained for all titles and abstracts that met the inclusion criteria or where there was any uncertainty for inclusion. GS, AV, YVM and REED were involved in this task. Reviewers agreed on which articles should be included and ARG acted as arbitrator when GS, AV, YVM and REED could not reach a full consensus.

GS, YVM and REED independently conducted data extraction from each study using a standardised and piloted data collection form which has been published alongside the protocol [21]. GS, YVM and REED checked each other’s data extraction to look for completeness and accuracy. The data extraction form collected information related to the study design and aim, characteristics of participants (age, sex, setting, comorbidities, use of concomitant medications, functional status, frailty, and cognitive status), the intervention (i.e. DPP-4 inhibitors) and comparison, time to follow-up, and reported outcomes. We also collected information on the involvement of pharmaceutical companies in the included studies.

We used three validated assessment tools to assess the quality of the evidence from each included study: for systematic reviews/meta-analyses the Assessment of Multiple Systematic Reviews tool (AMSTAR) [22, 23], for intervention studies the Cochrane Collaboration’s tool for assessing risk of bias [19], and for observational studies the Critical Appraisal Skills Programme (CASP) [24, 25].

An overall rating for each study was made based on study limitations as suggested by Guyatt et al. (2008) [26], starting with high quality for randomised trials without important limitations (such as lack of allocation concealment; lack of blinding, large loss at follow-up, unmet intention to treat analysis, stopping early for benefit; and failure to report outcomes) and low quality for observational studies without important limitations.

We included all relevant data from publications relating to a single primary study. Due to our staged approach, it was possible that a publication that was part of an included systematic review or meta-analysis, would also be included as a separate individual study, resulting in a risk of “double-counting”. Any such instances have been identified and reported and taken account of in our synthesis of results.

A narrative synthesis describing all included systematic reviews, meta-analyses, intervention and observational studies, participants and findings was carried out. The included studies were highly heterogeneous regarding type of DPP-4 inhibitors, comparison (form of control treatment or placebo), length of follow-up and outcome definition (e.g. types of adverse events included); therefore no additional meta-analyses were performed. The quality of the included studies is also reported.

During the search process, GS, and YVM identified and collected additional material relevant to the development of recommendations according to the methodology described by Martinez/Renom - Guiteras et al. (2017) [21].

Included studies and additional references were summarised in a document that was used in team meetings to develop recommendations on when the use of DPP-4 inhibitors could be safely discontinued or the dosage reduced in the management of type 2 diabetes in older people [21]. Each recommendation was given a rating for strength (weak or strong) and quality (low, moderate or high) of evidence following the GRADE methodology [26‐28].

We identified 1460 records through initial database searching (21 from search 1, 82 from search 2, 9 from search 3A and 1357 from search 3B). Additionally, we identified 988 records from reference lists of included studies, and one further study by snowballing. After removing duplicates, we screened 2009 records and excluded 1634 records after checking titles and abstracts. We assessed 375 full-texts for eligibility and excluded 341 records. We included 30 studies reported in 33 publications. The PRISMA flow diagram is presented in Fig. 1.

Table 1 shows details of included studies. Thirty studies met our inclusion criteria. These studies included more than 273,358 participants ≥65 years. The largest had 141,322 participants and the smallest 60 participants.

Seventeen of the included studies were interventional designs, one was an MA and 12 were observational in nature. None of the individual studies were also part of the MA. Length of follow-up varied from 12 weeks to 5 years. Data on outcomes was extracted at the end of follow-up for each included study. In 16 out of 30 studies information was given about the countries where studies had been conducted: the USA [29, 30], Australia, Canada, Denmark, the Netherlands and Sweden [31], Taiwan [32‐36], 38 countries [13], 13 countries [37], Japan [38], 12 European countries and Mexico [39], 14 European countries [40], 26 countries [12, 41‐43], 49 countries [44], UK [45‐47], Italy [48], France [49], Spain [50], and Greece [51].

Table 1 shows included studies involving older adults (at least 80% people ≥65 years: 16 studies) or presenting subgroup analyses in participants ≥65 years (11 studies including the meta-analysis), ≥75 years (2 studies), and ≥70 years (1 study). Additional file 3: Table S1 shows the characteristics of the participants in the included studies. Age is reported as mean or median years; for the whole sample where available, else for the different treatment groups. Mean age was reported in 27 studies and ranged from 53.1 to 80.2 years. Median age was reported in 3 studies and ranged from 58 to 77 years.

All included studies reported on participant sex (30 studies), though in some cases by treatment group only. The percentage of male participants ranged from 36.7% to 71.6%.

Fourteen studies reported ethnicity with the most common classification being white (range: 53.9 to 98.6%). Information about the care setting was reported by five studies: primary care in the USA [29], primary care in the UK [45], primary care in France [49], hospital department of internal medicine in Greece [51], and primary and hospital care in the UK [47]. Information about comorbidities was provided by 26 out of 30 studies. Concomitant diseases were frequent and hypertension and dyslipidaemia most commonly reported. Eighteen studies reported on concomitant medications with a majority of patients taking antihypertensive and lipid-lowering medications. Frailty status was reported by one study [52], with about 10% of patients assessed as frail. One study reported on disability after stroke as the main outcome, but no baseline data on disability were provided [51]. Cognitive status was not reported by any of the studies.

Most of the included studies addressed only our lower tier endpoints: adverse events and hypoglycaemia. A minority of studies investigated Tier 2 outcomes such as death, hospitalisation, cardiovascular events and, in one case, functional status. We found no studies in older people reporting on the clinically relevant endpoints of: quality of life, life expectancy, cognitive impairment or cognitive status.

Vildagliptin (50 or 100 mg/daily) was examined in 9 out of 30 studies. Vildagliptin was compared with placebo in two trials [52, 53], with glimepiride in two trials [54, 55], with metformin in one trial [40] and with thiazolidinediones in one trial [29]. Also included were one uncontrolled trial on vildagliptin [56] and two observational studies [50, 57]. The outcomes for these studies were adverse events [29, 40, 52‐56], serious adverse events [52, 54‐56], hypoglycaemia [40, 50, 52‐57] and a list of other outcomes by Sicras-Mainar and Navarro-Artieda (2014) (macrovascular complications and cardiovascular events, cerebrovascular disease, all types of peripheral arterial disease and renal disease) [50].

Sitagliptin (25 to 100 mg/daily) was examined in five studies: two placebo-controlled [13, 30], one uncontrolled [32], one active controlled [37], and one cohort study [34]. Three of these studies included adverse events as one of their outcomes and their primary endpoint was change in HbA1c [30, 34, 37]. The other two studies reported a composite of cardiovascular events as their primary endpoints [13, 34].

Linagliptin (5 mg/daily) was compared with placebo in one trial [31]. The outcomes were adverse events, hypoglycaemia, and cardiovascular events [31]. Also, one meta-analysis investigated the cardiovascular safety of linagliptin [58].

Teneligliptin (20 mg/daily) was compared to placebo in one study [38], with adverse events and hypoglycaemia as outcomes.

Alogliptin (25 mg/daily) was compared to glipizide in one study [59] and to placebo in another study [44]. Adverse events and hypoglycaemia were the outcomes in one study [59]. The other study used a composite outcome of death from cardiovascular causes, nonfatal myocardial infarction, or nonfatal stroke [44].

Saxagliptin (5 mg/daily) was compared to placebo in one study reported by four publications with the following outcomes: a composite outcome of cardiovascular death, myocardial infarction, or ischemic stroke [41], hospitalisation for heart failure [12], a composite outcome of cardiovascular mortality, nonfatal myocardial infarction, or nonfatal ischemic stroke with and without hospitalisation for heart failure, coronary revascularization, or unstable angina as well as the individual components [42], and bone fractures [43]. One randomised trial compared saxagliptin (5 mg/daily) against glimepiride (1 mg/daily) with hypoglycaemia and adverse events as safety outcomes and glycaemic control as the primary outcome.

Nine observational studies compared patients treated with DPP-4 inhibitors with patients not receiving DPP-4 inhibitors [46]; other antidiabetic drugs [33, 35, 45, 49, 51]; or between cases and controls [36, 47, 48]. These studies reported on the following outcomes: hypoglycaemia, fractures, disability after stroke (with the modified Rankin scale), cardiovascular events, hospitalisation for heart failure, hospitalisation for sepsis, and mortality.

Additional file 1 provides the full list of reasons for exclusion of studies after full text analysis. The main reason for exclusion was that the study population did not match our age criteria for inclusion (n = 298).

Twenty-eight studies provided evidence on relevant outcomes comparing DPP-4 inhibitors against an alternative (i.e. non-DPP-4) drug regimen or placebo. For each study and outcome Table 2 summarises the results for the DPP-4 inhibitor and comparison groups, provides estimated risk ratios with 95% confidence intervals, and reports any statistical comparisons from the study itself. To help interpretation, Table 2 organises the results first by Tier of outcome (Tier 1 or Tier 2), and then by form of comparison within Tier (DPP-4 inhibitors versus placebo; versus other active treatments; and as an additional treatment). Two further studies (not tabulated) compared between different DPP-4 inhibitor based-treatments: 1) insulin plus 100 mg vildagliptin versus insulin plus 50 mg vildagliptin dose [56]; 2) vildagliptin plus metformin versus vildagliptin plus 2 oral antidiabetic agents (metformin, sulfonylureas and/or thiazolidinediones) [57]. Quality of studies is also reported in Table 2.

Nineteen studies provided evidence on Tier 1 outcomes. Studies varied in what they classified as adverse events, and as serious, severe, or significant adverse events. Hypoglycaemia was defined by 3 studies as hypoglycaemic symptoms confirmed by self-monitoring of blood glucose <3.1 mmol/l [29, 40, 52]; another study defined symptomatic hypoglycaemia as an episode with clinical symptoms without regard to glucose level, asymptomatic hypoglycaemia was defined as an episode of glucose level ≤ 70 mg/dL without symptoms [37]; another study defined confirmed hypoglycaemia as a symptomatic or asymptomatic event with plasma glucose <3.0 mmol/l without requiring external assistance, severe hypoglycaemia was defined as symptomatic event requiring external assistance without regard to plasma glucose level [39]; the other 3 studies did not provide a definition of hypoglycaemia. In 10 studies [37, 39, 40, 46, 49, 50, 53‐55, 59] hypoglycaemia was considerably less frequent in older people treated with DPP-4 inhibitors than in older people on other treatments, on placebo, or when used as an additional medication.

Eleven studies [29‐32, 37‐40, 52, 53, 59] reported on adverse events other than hypoglycaemia. All showed only small, non-significant, differences. Two studies reported on fractures, one an RCT comparing saxagliptin to placebo [43] and the other a retrospective cohort study comparing DPP-4 inhibitors to other non-insulin anti-diabetic drugs [45]; both finding no significant difference. A nested case-control study reported that hospitalisation for sepsis was not significantly different between cases and controls using DPP-4 inhibitors [36].

Thirteen studies considered Tier 2 endpoints. The results for these generally more impactful outcomes were much more variable. The meta-analysis by Johansen et al. (2012) found that major cardiovascular events (fatal or non-fatal myocardial infarction or stroke, or hospitalisation for unstable angina pectoris) were significantly reduced by around 70% with linagliptin compared to comparators (mostly patients on placebo, but including a minority on anti-diabetic drugs) [58]. However, the large-scale RCTs [12, 13, 42, 44] reported no significant difference between DPP4 inhibitors (sitagliptin, saxagliptin, and alogliptin, respectively) and placebo using a similar endpoint; while one of them found a statistically significant 47% higher risk of hospitalisation for heart failure in the saxagliptin group [41]. Four observational studies reported no significant differences between DPP-4 inhibitors and other active treatments for various cardiovascular outcomes such as myocardial infarction, heart failure, ischaemic stroke, and hospitalisation for heart failure [33, 34, 47, 48], although in one study all-cause mortality was significantly lower in users of DPP-4 inhibitors than in controls [48]. One retrospective observational study reported significantly lower percentages of cardiovascular events, ischemic heart disease, and cerebrovascular accident with vildagliptin plus metformin compared to sulfonylureas plus metformin [50]; but Schweizer et al. (2009) found no difference between vildagliptin and metformin in rates of cardiovascular and cerebrovascular events [40], and Rosenstock et al. (2013) observed that alogliptin and glipizide did not differ in major cardiac events, though in both of these latter randomised trials samples sizes were small and events rare [59]. A cohort study reported significantly less mortality, myocardial infarction, ischemic stroke and hospitalisation for heart failure with DPP-4 inhibitors plus metformin compared to sulfonylureas plus metformin [35]. A small observational study reported significantly lower in-hospital mortality in people admitted with acute ischemic stroke and better scores on the modified Rankin scale of disability with DPP-4 inhibitors compared to other antidiabetics [51].

We deemed study synthesis by meta-analysis inappropriate, due to high heterogeneity of treatments, outcome definitions and follow-up periods. However, to gain a global overview and aid interpretation, Table 2 indicates the treatment arm favoured on each outcome in each study, based purely on the reported direction of effect or (where missing) risk ratio point estimate and ignoring statistical significance. Under this “vote counting” method [19], for the Tier 1 outcomes 8 studies fully favoured the DPP-4 inhibitor, 4 fully favoured the comparator, and 7 were mixed or neutral. For the Tier 2 outcomes, 7 studies favoured DPP-4 inhibitors, 2 favoured comparators, and 4 were mixed or neutral. However, only 2 studies received high overall ratings for quality (see below); both reporting Tier 1 outcomes favouring placebo and one reporting Tier 2 outcomes with mixed results. There was no suggestion for either tier of outcomes of the pattern of results differing according to whether the comparison was a placebo, other active treatment, or DPP-4 inhibitors as an additional treatment.

Both studies reported that hypoglycaemic events were similar between the groups: 1) 2.1 events per patient-year with insulin plus vildagliptin 100 mg versus 2.3 events per patient-year with insulin plus vildagliptin 50 mg [56]; 2) no events with vildagliptin plus metformin versus 1 event with vildagliptin plus 2 oral antidiabetic agents [57].

Twenty-two studies were funded by pharmaceutical companies and authored or co-authored by employees of the sponsor (22 out of 30, 73%). In the remaining eight studies, one study gave no information about funding although authors stated globally they had no conflict of interest [32]; two studies reported no funding and no conflict of interest [33, 51]; two studies reported funding outside pharmaceuticals and no conflict of interest [36, 48]; two studies reported funding outside pharmaceuticals and conflict of interest from some of the authors receiving fees from pharmaceuticals [34, 47]; one study reported no funding but one author declared receiving fees from pharmaceuticals [35]. The 30 included studies were authored by 219 authors: 29% (63 out of 219) were employees of the pharmaceutical sponsor, and an additional 27% declared conflicts of interest (60 out of 219, often consulting fees by the sponsor). Sixty-one authors of 11 publications (61 out of 219, 28%) declared no conflicts of interest. Support by professional medical writers was given in at least 8 publications (8 out of 30, 27%).

We found four additional references that were taken into consideration for the development of the recommendation: 1 meta-analysis, 1 pooled analysis, 1 observational study, and a report from the FDA. These are shown in Additional file 4: Table S2. These references did not meet our age or study design criteria for inclusion. They were counted as being relevant to recommendations principally because they provided information about clinically relevant endpoints not adequately addressed by the 30 included studies, albeit for younger populations.

One of the additional references reported that there were no statistically significant differences between vildagliptin compared to other anti-diabetic treatments or placebo for long-term outcomes including acute coronary syndrome, transient ischaemic attack, stroke, myocardial infarction, cardiovascular and cerebrovascular death [60]. Two of the additional references reported an increase in the risk of hospitalisations for heart failure and an increase in heart failure outcomes in people under DDP-4 inhibitors compared to people under other anti-diabetic treatments or placebo [61, 62]. These studies concur with that of Scirica et al. (2013) [12] for patients age 75 and over, included in the present SR.

Furthermore, the U.S. Food and Drug Administration (FDA) and the European Medicines Agency (EMA) reported that there are still some uncertainties with respect to long-term pancreatic safety with DPP-4 inhibitors and evaluation of these outcomes is ongoing [63]. Although the currently available data are reassuring, pancreatitis will continue to be considered a risk associated with these drugs until more data are available. These additional references suggest that certain risks like heart failure and related hospitalisation, and pancreatitis, may be increased with the use of DPP-4 inhibitors compared to other anti-diabetic treatments, independently of age group.

Recommendations were developed following a standardized schema and reflecting the strength and the quality of the evidence. Two meetings were held by GS (researcher and clinician), YVM (researcher) and ARG (researcher and geriatrician), with AS participating in one of these as a senior clinician and researcher. Subsequent to these meetings we agreed a recommendation which was later confirmed with IK and MMV for its inclusion in the Comprehensive Medication Review (CMR) tool developed within the PRIMA-eDS project.

From the results of our SR and the additional references of interest we developed one recommendation in relation to DPP-4 inhibitors use in older people with type 2 diabetes. The recommendation is that the clinician should consider discontinuing gliptins where the patient has HbA1c < 8.5%, principally because of the sparse and inconsistent evidence for clinically relevant benefits, but taking the patient’s symptoms into account (Table 6). The recommendation was considered as a weak recommendation. The quality was downgraded from high to moderate for indirectness.

We considered glycaemic control in the recommendation although it was not one of our study endpoints. The aim was that clinicians would focus on those patients who may benefit more from the recommendation as they could be already having acceptable glycaemic control. In older people, rigid glycaemic control (<HbA1c 8.0%) has been found to be associated with a higher risk of hypoglycaemia and undesirable long-term outcomes like increased mortality [64]. The target population in the PRIMA-eDS trial were people 75 years or older with at least eight prescribed medications reflecting a high comorbidity burden. An HbA1c <8.5% has been recommended in guidelines as a target goal in older people who have comorbidities, poor health, dementia, frailty or limited life expectancy [8, 65‐67]. As a general rule, PRIMA-eDS recommends clinicians to take symptoms into consideration as well as the individual participant characteristics such as frailty level and comorbidities.