Newer agents for blood glucose control in type 2 diabetes: systematic review and economic evaluation

Why is there reluctance to use insulin?

In a previous technology assessment report (TAR) for NICE, on inhaled insulins, we pondered upon why there should be reluctance. 15 There seemed to be reluctance amongst both patients and physicians. What follows is based on that TAR. Time did not permit a systematic review.

The DAWN (Diabetes Attitude Wishes and Need) study found that 55% of patients who have never had insulin treatment are anxious about it being required. The authors, Peyrot et al. (2005)16 review previous studies of patient attitudes to insulin therapy. They note that these involve beliefs that ‘taking insulin … :

• leads to poor outcomes including hypoglycaemia, weight gain and complications

• means that the patient’s diabetes is worse and that the patient has failed

• means life will be more restricted and people will treat the patient differently

• will not make diabetes easier to manage’.

It is important to note that insulin treatment is not just about injections, but a whole package of care, including dietary adjustments, home blood glucose testing and self-adjustment of insulin doses. It is likely that, for most people, insulin injections are less troublesome than blood testing.

Changing to insulin does not mean that control will improve. Unpublished data from the Lothian audit show that the average HbA_1c in patients with type 2 diabetes on insulin is about 8.5% [J McKnight, presented at the Royal College of Physicians of Edinburgh (RCPE) conference, September 2005, personal communication]. The average for those with type 2 diabetes on OGLAs is 7.5%.

Similarly, a study from seven European countries17 found that only 9.5% of patients with type 2 diabetes, who were on insulin, had HbA_1c < 6.5%; another 44% had HbA_1c levels of 6.5–7.5%; and 47% had HbA_1c levels of over 7.6%.

One issue in insulin therapy is the provision of structured education programmes, such as DAFNE (Dose Adjustment For Normal Eating). Good education may reduce problems with insulin treatment.

What is the optimum treatment for people with type 2 diabetes inadequately controlled on oral agents?

It seems clear from the literature that there are differences of opinion on management of people with type 2 diabetes who are not adequately controlled on oral agents. A working group drawn from the American Diabetes Association (ADA) and the European Association for the Study of Diabetes (EASD) produced a consensus statement in 2006. 18 Some extracts from this statement give an impression of the problems:

The availability of the newer agents has provided an increased number of choices for practitioners and patients and heightened uncertainty regarding the most appropriate means of treating this widespread disease. Although numerous reviews on the management of type 2 diabetes have been published in recent years, practitioners are often left without a clear pathway of therapy to follow.

The most appropriate target levels for blood glucose, on a day-to-day basis, and HbA_1c, as an index of chronic glycaemia, have not been systematically studied.

They noted the different target levels proposed by the various bodies, and reached a consensus that: ‘an HbA_1c of over 7% should serve as a call to action to initiate or change therapy’.

They recommended that insulin should be initiated with either bedtime intermediate-acting insulin or once-daily long-acting insulin; metformin should be continued.

Goudswaard et al. (2004),19 in a Cochrane review, concluded that combinations of insulin and oral hypoglycaemic agents should be the starting point for people with type 2 diabetes who required insulin. Their review preceded the studies on long-acting analogues, such as glargine and detemir. The oral agents most commonly used in the trials they found were sulfonylureas; only 7% used metformin alone.

Douek et al. (2005)20 from the Metformin Trial Group carried out a randomised controlled trial (RCT) of adding metformin or placebo in people with type 2 diabetes who had been switched to insulin because of poor control. Continuation of metformin resulted in less weight gain, lowered insulin requirement and improved glycaemic control.

Aviles-Santa et al. (1999)21 also showed that adding metformin to an insulin regimen in people with type 2 diabetes reduced HbA_1c by 0.9% compared with placebo. Insulin requirement was 29% lower, and the weight gain of 3.2 kg, seen in the placebo group, was much more than in the metformin group (0.5 kg).

Strowig and Raskin (2005)22 carried out a review of combination therapy with insulin and either metformin or a glitazone, or both. Details of methods are not given and it was probably not systematic. They also concluded that it was worthwhile continuing an insulin sensitiser in patients with type 2 diabetes who were switched to insulin. Because metformin and glitazones have different balances of sites of preferential action (acting on glucose production and glucose disposal), they also made the case that triple therapy should also be considered. Bailey (2005)23 also supported combination therapy with metformin and a glitazone for reducing insulin resistance in type 2 diabetes.

Gerstein et al. (2006)24 randomised poorly controlled (HbA_1c level 7.5–11%) patients to continue oral agents or to switch to glargine, in the Canadian INSIGHT (International Nifedipine GITS Study: Intervention as a Goal in Hypertension Treatment) study. Those treated with glargine achieved lower HbA_1c and non-high-density lipoprotein (HDL) cholesterol, and greater satisfaction, but more weight gain. However, only 17.5% of patients on glargine reached the target of two or more consecutive HbA_1c levels of 6.5% or under. One weakness of the study was that at baseline about 17% of the patients had not been treated with any oral agent; another 40% were on oral monotherapy.

Hayward et al. (1997)25 noted that results from trials of insulin therapy in type 2 diabetes showed it to be efficacious, but thought that these results might not be replicated in routine care. In a very large study (8668 patients with type 2 diabetes) they found that ‘insulin therapy was rarely effective in achieving tight glycemic control’. Two years after starting insulin therapy, 60% still had HbA_1c levels of 8% or greater; 25% had levels between 8.0% and 8.9%, 20% between 9.0% and 9.9%, and 15% had levels over 10%. These are similar to the population-based audit from Lothian (J McKnight, personal communication).

The observation that starting insulin in routine care usually fails to give good control in people with type 2 diabetes failing on oral agents is presumably one reason why the physicians in the DAWN study16 showed considerable resistance to starting insulin therapy in type 2 diabetes – only about half of the physicians thought that insulin would be useful.

Yki-Järvinen et al. (2006)26 came to similar conclusions in people with type 2 diabetes who were obese (defined in this study as BMI of over 28.1 kg/m²) – insulin did not improve control. In many of these patients, poor control is associated with overweight or obesity.

Aas et al. (2005)27 tried another approach, randomising patients with poorly controlled type 2 diabetes to insulin or to a lifestyle intervention (exercise and diet counselling). Lifestyle intervention was as effective in glycaemic control but also resulted in weight loss. In a follow-up study in 2006, the authors also noted that lowering HbA_1c level by lifestyle measures had more beneficial effects on adipokine levels than when insulin therapy achieved the same lowering, which may result in a lower cardiovascular risk. 28 However, numbers in this study were small (38 in total), and the study needs to be replicated with larger numbers.

Beta-cell mass

As mentioned above, by the time of diagnosis of type 2 diabetes, beta-cell function is considerably impaired. An important issue is whether any treatments can preserve the remaining beta-cell function, or promote regeneration.

Conversely, it is important to know if any treatments might accelerate beta-cell decline. In the ADOPT (A Diabetes Outcome Progression Trial) trial, patients were randomised to monotherapy with glibenclamide, metformin or rosiglitazone. Outcomes included failure of monotherapy. By 5 years, 34% of the glibenclamide group had failed, compared with 21% on metformin and 15% on rosiglitazone. 29 Aston-Mourney et al. (2008)30 have argued, based on this trial and basic science studies, that it may be harmful to force the beta cell to produce more insulin, and that doing so may cause earlier beta-cell death. The implication might be that drugs that are insulin sensitisers, rather than insulin secretagogues, may help to preserve beta-cell function or mass, by reducing the pressure to produce more insulin. However, in the UKPDS4 the slopes of rises in blood glucose were similar for metformin and sulfonylureas, which does not support the sulfonylurea harm theory.

Meier (2008)31 has recently reviewed the evidence on beta-cell mass, and the hypothesis that ‘resting’ the beta cell would help, but concludes that: ‘as yet, there is no direct evidence for the induction of beta cell apoptosis (death) by sulfonylurea drugs or for the preservation of beta cell mass by either metformin, glitazones or exogenous insulin in patients with type 2 diabetes.’

Types of evidence

For efficacy, RCTs are the gold standard. Open-label extension studies are useful to see if the effects persist, and for the development (or sometimes waning) of side effects. The dropout rate may also be a useful guide to tolerability.

For our purposes, we are interested mainly in trials that use standard UK practice as the comparator. Standard practice is set out in the current NICE guideline for type 2 diabetes (NICE 2008)6 and is shown in the flow chart in Chapter 1.

Types of interventions

Intervention consists of treatment for a minimum of 12 weeks with exenatide, exenatide long-acting or liraglutide. Twelve weeks is chosen because of the time it takes for glycaemic control to be reflected in HbA_1c level, but should be regarded as the minimum acceptable rather than satisfactory. Longer-duration studies would be better.

The 2008 NICE guideline on management of type 2 diabetes (see flow chart) stated that for individuals with a BMI of over 25 kg/m², the first choice in addition to diet was metformin, and, if that was insufficient, an insulin secretagogue should be added. In practice that would be a sulfonylurea; the other secretagogues, the meglitinide agonists, are little used in the UK.

So the most relevant comparisons are:

1. The addition of a GLP-1 analogue to standard combination therapy when that is insufficient to achieve good control, i.e. metformin + a sulfonylurea versus metformin + sulfonylurea + a GLP-1 analogue. A variant might use two insulin sensitisers: metformin + glitazone versus metformin + glitazone + GLP-1.

2. In those who cannot tolerate metformin, a glitazone might be used in combination therapy instead: sulfonylurea + a glitazone versus sulfonylurea + glitazone + GLP-1 analogue. One outcome of interest will be progression to insulin treatment.

3. Another option suggested in the NICE guideline was to add a glitazone to the metformin and sulfonylurea combination, i.e. triple therapy. If that fails, insulin treatment is the next step, usually with a long-acting basal insulin, with metformin, and perhaps the other drugs, continued. So another possible comparison would be to try a GLP-1 agonist instead of insulin: metformin + sulfonylurea + glitazone + GLP-1 agonist versus basal insulin + metformin + sulfonylurea + glitazone.

4. In those who have started insulin recently there could be a case for stopping insulin and trying a GLP-1 analogue, so a further comparison is: insulin (with or without oral agents) versus oral agents + a GLP-1 analogue. This is not a licensed use. The Food and Drug Administration (FDA) patient information sheet49 states that ‘Byetta is not a substitute for insulin in patients whose diabetes requires insulin treatment’.

5. This comparison looks at adding exenatide to metformin monotherapy, and was included at the request of the GDG, which felt that there were some overweight patients in whom the further weight gain likely with the usual second-line combinations of adding a sulfonylurea (or a glitazone) was so undesirable that a GLP-1 agonist should be considered instead, given the likelihood of weight loss.

Ideally, the comparison would be of metformin + exenatide versus metformin + a gliptin but at the time of writing, no such trials had been done, so Comparison 5 is: metformin + exenatide versus metformin alone.

Excluded studies

The studies in the Table 2, below, were excluded for the reasons given. Some of these trials provided useful information, for example showing that the GLP-1 agonists were effective in lowering PG compared with placebo, or were early dose-ranging studies, but were not relevant to our key comparisons.

Included studies

Seven trials were relevant for our purposes, and are listed below, under the relevant comparisons. The quality of the trials seems reasonable, although some details were not reported, and insulin, when a comparator, may not have been optimally used. Table 3 gives the details.

HbA_1c results

These are shown in Tables 4A and 4B overleaf.

The trials show that in those whose control is not good enough on dual therapy, addition of exenatide improved HbA_1c by about 1% [Kendall (2005),59 Zinman (2007)60].

In the Kendall (2005)59 trial, the changes in HbA_1c at 30 weeks were greater in those whose baseline level was higher.

When exenatide is compared with various insulin regimens, the results are similar, suggesting non-inferiority, although the issue of non-optimisation of the insulin treatment remains an issue.

Hypoglycaemia

Table 5 shows the frequency of hypoglycaemia.

Weight

Most studies reported weight loss with exenatide treatment. Results are shown in Table 6.

Does nausea cause the weight loss?

Maggs et al. (2005)65 carried out an analysis of patients in three trials [Buse et al. (2004),56 DeFronzo et al. (2005),58 Kendall et al. (2005)59] to see if the weight loss with exenatide was related to the nausea. Severe nausea was found in only 4%. They found little correlation between nausea and weight loss (or HbA_1c level). In the extension studies (to 52 weeks) the majority of patients had very little nausea, but lost the same amount of weight as the more nauseated subgroups.

Heine et al. (2005)53 found that although the magnitude of weight reduction tended to be greater in patients taking exenatide who experienced longer durations of nausea, patients who did not report any episodes of nausea during the trial (n = 120) still demonstrated a mean weight change of –1.9 kg (CI –2.5 to –1.4).

Adverse events other than hypoglycaemia

Table 7 shows the most frequent side effects.

The most striking finding is the high frequency of nausea with exenatide, with vomiting not uncommon. However, the number who had to stop exenatide because of side effects was much lower. Most nausea was mild, and the frequency decreased over time. For example, Heine et al. (2005)53 reported that 55% of patients reported nausea in the first 8 weeks, but only 13% did so in the last 8 weeks. However, 18 patients from the exenatide group withdrew because of nausea (compared with one patient in the insulin group). Heine et al. reported the frequencies of mild, moderate and severe nausea to be 33%, 20% and 5%, respectively.

Kendall et al. (2005)59 also reported that the frequency of nausea diminished over time, and only 4% had to withdraw because of it.

Zinman et al. (2007)60 reported that 9% of the exenatide group withdrew because of nausea, but that most nausea was mild (44%) or moderate (41%), and that it declined over time.

Cardiovascular risk factors

Three trials reported lipid and blood pressure data.

• DeFronzo (2005)58 reported that exenatide treatment was not associated with an increased incidence of cardiovascular, hepatic, or renal adverse events. Also no changes in plasma lipids, laboratory safety parameters, heart rate, blood pressure, or electrocardiogram variables were observed between treatment arms.

• Nauck et al. (2007)54 reported that HDL was higher by 0.04mmol/l with insulins, but that blood pressure fell with exenatide (systolic by 5 mmHg and diastolic by 2 mmHg), but did not change with insulin.

• Zinman et al. (2007)60 found no significant differences in lipids and blood pressure.

Other outcomes

Patient-reported outcomes from the Barnett (2007)38 trial were reported by Secnik et al. (2006)66 in a poster presented at the International Diabetes Federation (IDF) in 2006. Responses to the following health outcome instruments were examined: the Psychological General Well-Being Index (PGWB), Diabetes Symptom Checklist-Revised (DSC-R), European Quality of Life-5 Dimensions (EQ-5D), Treatment Flexibility Scale (TFS) and Hypoglycaemia Fear Survey (HFS). No statistically significant between-group differences between twice-daily exenatide and glargine were found on any of these measured health outcomes.

Secnik et al. (2006)67 reported some patient-reported outcomes from the Heine trial, including EQ-5D, the vitality scale of the Short Form questionnaire-36 items (SF-36) health survey, the Diabetes Symptom Checklist, and the Diabetes Treatment Satisfaction Questionnaire. No differences were found, suggesting that the greater number of injections with exenatide (twice daily versus once for glargine), and the frequent (at least initially) nausea was not enough to affect overall satisfaction, perhaps because those were balanced by weight loss on exenatide (on average, 2.3 kg) versus gain on insulin (mean 1.8 kg).

An abstract from the Nauck (2006) trial by Yurgin et al. (2006)68 also reported EQ-5D and SF-36 data, stating that the exenatide group showed some improvement, whereas the biphasic aspart group showed no change.

Follow-up studies: open-label extensions

Klonoff et al. (2008)42 report results in people who had been on exenatide for at least 3 years. The participants were from the three 30-week studies [Buse et al. (2004),56 DeFronzo et al. (2005),58 Kendall et al. (2005)59], only one of which met our inclusion criteria for this review. However, the pooled open-label follow-up can provide useful data on duration of efficacy and side effects.

The withdrawal rate was high. Of 527 eligible patients, 310 withdrew. The reasons for withdrawal included adverse events (11%), poor control (3%), and patient or investigator decision (41% – reasons not given).

Weight loss was maintained amongst the 41% (217) who stayed in the follow-up study. The mean weight loss at 3 years was 5.3 kg. Overall, 84% of patients lost weight. Reductions in HbA_1c level were also sustained (but this may be because those in whom it rose again left the study). Total cholesterol fell by 5% and triglycerides by 12%, presumably because of the weight loss, because there was a correlation between weight loss and cardiovascular risk factors.

The most frequent adverse effect (in 59%) was nausea, usually mild. Next came hypoglycaemia, but only in those treated with a sulfonylurea. Upper respiratory infections were common (36%) but the significance of that cannot be assessed without a control group. There were no serious side effects other than a few severe hypoglycaemic episodes. So exenatide appears safe, but the high dropout rate reduces the value of the study.

Results from routine care

Rather different results were found in routine care by Wolfe and King (2007). 69 Two hundred consecutive exenatide-treated patients included 56 treated for 12 months. The nadir of weight occurred at 6 months. Few details are given of later weight loss in this ADA conference abstract, but the suggestion is that there was a plateau after 6 months.

Loh and Clement (ADA poster 2007)70 reported a small follow-up study of 30 patients with type 2 diabetes who were treated with exenatide, some in addition of oral antidiabetic (OAD) drugs, others in addition to insulin. At 1 year, there was weight loss (mean 2 kg, p = 0.0033) but no significant reduction in HbA_1c level overall. Maximum weight loss occurred by 7 months, with most patients regaining weight over months 7–12. Half the patients had stopped exenatide by 12 months, because of therapeutic failure or side effects. Loh and Clement conclude that in the ‘real world’ exenatide may not give as good results as seen in trials.

Yoon et al. (2008)71 in a conference abstract (ADA 2008), reported use of exenatide added to insulin. In a case series of 226 patients who started exenatide, 34 (15%) stopped within 3 months due to adverse effects. 71 Another 78 discontinued it later, mainly due to side effects or lack of efficacy. The final analysis of those who had used it for more than a year (116) showed weight loss of 6 kg, and a 20% reduction in insulin dosage. Eleven patients with an initial mean insulin dose of 17 units per day were able to stop insulin.

Another study from routine care, reported by Bhushan et al. 72 at the ADA 2008 conference, followed 201 patients for 16 weeks; all received exenatide in addition to previous treatment (details of which not given). Weight loss was seen in 69%, and averaged about 2 kg. Total cholesterol fell by 6 mg/dl. Blood pressure was unchanged.

It seems logical that exenatide be combined with insulin, although this is not a currently licensed indication. In an abstract from the recent EASD, Govindan et al. 73 presented a small case series from Wolverhampton, of 27 obese patients (mean BMI 43 at baseline), who were already on insulin but poorly controlled (mean HbA_1c level 8.8%). About half had nausea on exenatide, but only three had to stop it. The mean weight fell from 128 kg to 115 kg after 3 months; BMI from 43 to 40; and insulin dose from a mean of 170 to 36 units/day. The average insulin dose reduction comes about because 10 patients could stop it altogether, although mean HbA_1c level did not improve much (by only 0.3%; NS). Longer follow-up might show greater benefit, and it suggests that trials of combined exenatide and insulin therapy are justified.

Also from the EASD conference, Wintle et al. (from Amylin and Lilly)74 presented data from diabetic care records from the General Electric database, on 2086 patients treated with exenatide for 6 months or more. Patients had previously been on metformin, sulfonylurea or glitazone monotherapy (about 30%), or on dual therapy (38%) or triple therapy (34%), but were not well controlled (mean HbA_1c level 8.4% and BMI 38.5).

Exenatide reduced HbA_1c level by 0.9% in those who had been on monotherapy, but by less (0.5–0.8%) in those who had been on combination treatment.

Kendall et al. (Amylin and Lilly)75 reported a pooled analysis of 2 years of exenatide treatment. Patients were split into three groups according to pattern of weight loss: one group that lost none (they gained about 1 kg, but as their HbA_1c level fell by over 1%, they were presumably taking the exenatide, suggesting that compliance was not the issue); a second group (34%), which lost weight quite quickly (about 4 kg by week 12); and a third group (46%), which lost as much weight as the second group, but who did so more slowly. Groups 2 and 3 lost on average 6 kg by 2 years.

In the group that did not lose weight, HbA_1c level fell by about 1.2% but started rising again in the second year, to a drop of about 0.7% (from graph). In groups 2 and 3, the fall in HbA_1c level of about 1.5% was more sustained – about 1.5% reduction at 52 weeks and 1.3% at 104 weeks.

This finding might have implications if NICE recommended a stopping rule for exenatide, as it could be stopped in those in whom it was least effective (no weight loss), thereby improving the cost-effectiveness.

Exenatide LAR

The exenatide LAR formulation has been studied in a 15-week Phase II trial [Kim (2007)]47 in patients with type 2 diabetes. The trial reported that a 2-mg dose of exenatide LAR showed a reduction in HbA_1c level of 1.4% (relative to placebo), which the authors say is approximately twice as great as that seen with twice-daily injections of conventional exenatide. Preliminary results have suggested that the LAR formulation is also better tolerated than the original formulation, with less nausea, and (in the 2-mg form) is associated with greater weight loss; however, patient numbers were small. Results from other trials are awaited. The Amylin website45 reports an unpublished 30-week RCT of long-acting exenatide versus twice-daily Byetta, and states that ‘results showed that exenatide once weekly demonstrated powerful glucose efficacy, complemented by striking weight loss’.

This trial is presumably the DURATION trial, recently described in two abstracts. The ADA abstract, Drucker et al. (2008),76 reported the 30-week results in brief. They showed that once-weekly exenatide reduced HbA_1c level slightly more than twice daily: 1.9% versus 1.5%. Seventy-seven per cent of the once-weekly group achieved an HbA_1c level of less than 7.0%, compared with 61% for the twice daily. The trial recruited 295 patients who were poorly controlled (mean HbA_1c level 8.3%), but most were on no oral drugs (15%) or monotherapy (45%). Only the 40% on two oral agents are relevant to this review. However, the trial clearly suggests that the future lies with once-weekly exenatide. No details on cost are yet available, but some economies would be expected compared with twice-daily injections.

The second abstract is from EASD77 and is a 22-week open-label follow-up of 241 of the DURATION patients by Buse et al. (2008)77 [the same team as Drucker et al. (2008)76]. Much of the abstract is about the patients who switched from twice-daily to weekly, but the 52-week HbA_1c level results in the original once-weekly group are reported in brief as being sustained – reduction at 52 weeks of 2% (1.9% at 30 weeks).

GLP-1 agonists and beta-cell function

Rodent studies have reported that liraglutide can increase beta-cell mass.

Gallwitz (2006)78 has reviewed some of the animal and in vitro studies. The animal studies are mainly in rats, with a couple in mice. The evidence suggests that beta-cell growth is stimulated and that apoptosis is reduced. In isolated human islets, GLP-1 expands beta-cell mass. However, Gallwitz found no evidence regarding beta-cell mass in humans.

Xu et al. (1999)79 reported that exenatide treatment improved diabetic control in rats that were made diabetic by partial pancreatectomy, and that this was related to an increase in beta-cell mass (assessed histologically). Interestingly, the improved control was seen even after exenatide was stopped after 10 days. Gedulin et al. (2005)80 also reported an increase in beta-cell mass in rats after exenatide treatment.

Tourrel et al. (2001)81 treated newborn rats, made diabetic with streptozotocin, with exenatide and, again, noted an increase in beta-cell mass, which persisted (although the beta cells were less responsive to glucose).

If these findings are confirmed in humans, it would be of great importance, because it would suggest that the progressive nature of type 2 diabetes4,5 could be halted. Barnett (2007)38 and Holst et al. (2008)82 both note that if the GLP-1 analogues could increase beta-cell mass there would be an argument for treatment early in the disease, before too many beta cells are lost.

However, there are few data on the effect in humans – some very short experiments on islet cells in vitro, reviewed by Wajchenberg (2007),83 who concludes that there is, as yet, no clinical evidence that the GLP-1 analogues protect beta cells.

Bunck et al. (2008),84 in an ADA abstract from their RCT of exenatide versus glargine,85 reported that the beneficial effects seen on exenatide were not sustained – 5 weeks after stopping exenatide all of the improvements had gone, which may suggest that beta-cell function was not improved.

Further research is required, ideally with some means of determining at an early stage (2–3 years?) whether beta-cell mass is maintained in humans with type 2 diabetes.

Postprandial hyperglycaemia

The slowing of gastric emptying by the incretin mimetics could, in theory, reduce postprandial hyperglycaemia.

Acute pancreatitis

There have been recent concerns about acute pancreatitis in people who have been treated with exenatide. 88 The FDA had (as at end of 2006) reviewed 30 reports of acute pancreatitis in patients on exenatide. Nearly all had other possible reasons for pancreatitis, including gallstones and alcohol use. Nearly all improved after exenatide was stopped, and a few in whom it was started again had a recurrence of symptoms. However, the improvement after the drug was stopped may be coincidental. The FDA has asked for a warning to be added to patient information and arranged enhanced monitoring, but has not restricted use. 89

The Medicines and Healthcare Products Regulatory Agency (MHRA) (Drug Safety Update May 2008)90 has called for vigilance. It notes that by September 2007 there had been 89 reports of acute pancreatitis, with, curiously, 87 in the USA and two in Germany. One case has since been reported in the UK, after only 5 µg of the drug.

Types of interventions

Treatment for a minimum of 12 weeks with DPP-4 inhibitors (sitagliptin or vildagliptin) in combination with meglitinide analogues, metformin, a sulfonylurea or a TZD.

As with the GLP-1 analogues, the comparisons of interest for this review are based on the licensed indications, and on the standard treatment of type 2 diabetes, as set out in the NICE guideline (2008),6 the algorithm that was reproduced in Chapter 1.

The licensed indications are as follows. 94

What do the excluded studies tell us?

Compared to placebo, sitagliptin and vildagliptin reduced HbA_1c level by around 0.7% and 0.6%, respectively. The sitagliptin versus placebo trials demonstrated substantial heterogeneity. (However, after eliminating a single study of Japanese patients only, Cochrane review noted that the heterogeneity decreased to an I²-value of 25%.) There was no weight loss advantage with the DPP-4 inhibitors.

Compared to monotherapy with other agents, neither drug showed any advantage.

There are no data on diabetic complications or mortality, but that is to be expected because of the short duration. Most included trials were 24 weeks’ duration; three were for 52 weeks.

The trials gave no data on costs or quality of life (QoL).

Both drugs were well tolerated. No severe hypoglycaemia was reported.

Comparison 1a

Comparison 1d

Comparison 1e

Quality of included trials

The quality of the included trials, as shown in Table 10, was good.

HbA_1c results

The results for HbA_1c in Table 11 show that sitagliptin and vildagliptin have similar effects to a glitazone, but an improvement of 0.9% compared with placebo.

Weight change

Table 12 shows there was less weight gain than with the glitazones [Bolli (2008),123 Scott (2007)120] but that is mainly because people on glitazones gained weight – not because those on a DPP-4 inhibitor lost any. In the comparison with glipizide, the sitagliptin arm ended 2.6 kg lighter. In the Hermansen et al. trial (2007)124 there was more weight gain with the DPP-4 inhibitor than in the placebo control arm.

The Cochrane review91 concluded (see tables 13 and 14 under ‘Additional tables’ for details) that in trials against placebo there was greater weight loss after placebo treatment than with sitagliptin and vildagliptin. The pooled estimate for sitagliptin studies was a weighted mean difference (WMD) of 0.7 kg (95% CI 0.3 to 1.1, p = 0.0002) in favour of placebo and 0.8 kg (95% CI 0.2 to 1.3, p = 0.009), for vildagliptin studies in favour of placebo. Most active hypoglycaemic comparators also resulted in more pronounced weight losses than sitagliptin or vildagliptin treatment.

So the DPP-4 drugs do not seem to have as great a weight reduction effect as exenatide, but in most cases there is no weight gain, which, compared with sulfonylureas and glitazones, is an advantage.

Adverse events

Table 13 shows selected adverse events.

For full details, see tables 15–27 of the Cochrane review by Richter et al. (2008). 91 As mentioned above, the drugs were well tolerated. Discontinuation due to adverse effects did not differ significantly between sitagliptin or vildagliptin intervention and control arms. The risk ratios of serious adverse events also did not show statistically significant differences between groups.

In the Cochrane review, headache was reported more often with DPP-4 inhibitors, especially following vildagliptin therapy.

Hypoglycaemia

Bolli et al. (2008)123 defined hypoglycaemia as symptoms that are suggestive of low blood glucose, confirmed by a self-monitored PG level of under 3.1 mmol/l. Hypoglycaemia was reported in only one patient in the Bolli study – in the vildagliptin group – and it was mild.

In the Hermansen et al. trial (2007),124 any hypoglycaemia was reported in 16% of the sitagliptin group versus 0.9% of the control group. In the Scott et al. study (2007),120 hypoglycaemia was reported in 1% of both groups. No severe hypoglycaemia was reported.

Nauck et al. (2007)54 defined severe hypoglycaemia as requiring medical assistance, and had another category where non-medical assistance was sufficient. Any hypoglycaemia was reported in 32% in the glipizide arm and in 5% in the sitagliptin arm; severe hypoglycaemic attacks were reported in 1.2% and 0.2%, respectively. Hypoglycaemia of the ‘non-medical assistance needed’ category was reported in 1.4% (eight patients) and 0.2% (one patient).

In the wider Cochrane review by Richter et al. (2008),91 severe hypoglycaemia was not reported in patients taking sitagliptin or vildagliptin, and there were no statistically significant differences in hypoglycaemic episodes between sitagliptin/vildagliptin and comparator groups.

Infections

The Cochrane review by Richter et al. (2008)91 reported an increase in all-cause infections.

The Merck & Co responses to the consultation mentioned the analysis by Williams-Herman et al. (2008) (who are from Merck & Co),125 and stated that this did not find any increase in infections.

There are three reviews that report infection rates in DPP-4 inhibitor trials:

• The Cochrane review by Richter et al. (2008)91 included all RCTs in adults with type 2 diabetes, with trial duration of at least 12 weeks. It included 25 trials: 11 sitagliptin, 14 vildagliptin. Study duration ranged from 12 to 52 weeks. Searches were carried out until January 2008. All-cause infections [for example, nasopharyngitis, upper respiratory tract infection, urinary tract infection (UTI)] showed a statistically significant increase after sitagliptin treatment [relative risk (RR) 1.29, 95% CI 1.09 to 1.52, p = 0.003] but did not reach statistical significance following vildagliptin therapy (RR 1.04, 95% CI 0.87 to 1.24, p = 0.7).

• A review by Amori et al. (2007)32 also included RCTs of at least 12 weeks’ duration. Searches were until 20 May 2007. They found eight sitagliptin studies and 12 vildagliptin studies. They found a slightly increased risk of nasopharyngitis (6.4% for DPP-4 inhibitor versus 6.1% for comparator; RR 1.17; 95% CI 0.98 to 1.40), which was significant only for sitagliptin (RR 1.38, CI 1.06 to 1.81). The risk of UTI was increased by about 50% (3.2% for DPP-4 inhibitor versus 2.4% for comparator; RR 1.5; 95% CI 1.0 to 2.2), and this was seen with both DPP-4 inhibitors, although in the individual comparisons, confidence intervals (CIs) on risk ratios were wide and overlapped with unity. Amori et al. accepted that the relative risks were small, but commented: ‘there are more than 20 million patients with diabetes in the United States who are both more likely to develop a urinary tract infection and are at higher risk of complications, including death from urosepsis. A relative risk of 1.5 increases the number of UTIs by 1 million new cases per year, placing a significant burden on the individual patient and the health-care system. Until more safety data are available, it may be prudent to avoid use of these agents in patients with a history of recurrent UTIs’.

• The analysis by Williams-Herman et al. (2008)125 included only sitagliptin (100-mg dose). It pooled 12 large Phase IIb and Phase III RCTs, with duration at least 18 weeks (up to 2 years), based on data available in the industry database at November 2007. They reported that the only infection more common in the sitagliptin group was nasopharyngitis, with 7.1% in the sitagliptin group versus 5.9% in the comparators, but that the 95% CI for the difference overlapped with no difference (95% CI –0.1 to +2.4). They found no difference in the frequency of UTIs.

So we have two independent reports suggesting an increase in UTIs, and the manufacturer’s analysis reporting no increase.

Table 14 shows the trials included in these reviews.

Quality of life

No publication provided data on health-related QoL.

Hypothetical adverse effects

In addition to reducing the breakdown of the incretins, GLP-1 and gastric inhibitory peptide, DPP-4 inhibitors also prolong the action of a number of neuropeptides, such as neuropeptide Y, growth hormone-releasing hormone and chemokinines, such as stromal cell-derived factor 1 and macrophage-derived chemokine. Potential side effects include neurogenic inflammation, increase in blood pressure, enhanced inflammation and allergic reactions. DPP-4 contributes to T-cell activation, raising the possibility that these compounds compromise immune function. 130 Levels of tissue DPP-4 are reduced in nasal tissue of people with chronic rhinosinusitis, and DPP-4 inhibition seems to aggravate nasopharyngitis, as could be observed in clinical studies.

Therefore, the long-term safety of DPP-4 inhibitors merits further investigation, and it seems to be important to monitor DPP-4-treated patients for the development of inflammatory conditions, such as angioedema, rhinitis and urticaria.

Costs

Both of the sitagliptin trials used 100 mg daily, which, at a cost (BNF 55131) of £33.26 for 28 tablets, comes to £432 per year.

No cost is available for vildagliptin yet. The dose used by Bolli et al. (2008)123 was also 100 mg daily.

Beta-cell function

A progressive reduction in beta-cell mass contributes significantly to gradual loss of glycaemic control in individuals with type 2 diabetes. A major goal of diabetes research is to restore the beta-cell mass typically lost during the natural progression of type 2 diabetes. Current treatments not only show no ability to reduce beta-cell loss, but also some, such as the sulfonylureas, have been shown to induce beta-cell apoptosis in cultured human islets. 132 If the DPP-4 inhibitors can enhance beta-cell survival and stimulate beta-cell growth, they may provide a means to preserve or restore functional beta-cell mass in individuals with type 2 diabetes.

The Cochrane reviewers91 found few data on measurements of beta-cell function, especially for vildagliptin. The variety of methods used also made definite conclusions on the effects of DPP-4 inhibitors on beta-cell function difficult. Inspection of the sitagliptin homeostasis model assessment beta (HOMA-beta) data seems to indicate that sitagliptin, when compared with placebo, results in increased values of beta-cell function measurements, but the effect in comparison with other hypoglycaemic agents does not seem to be clear cut.

Most studies are quite short. An exception is the 2-year extension study by Scherbaum et al. (2008). 117 This study (funded by Novartis, with the corresponding author from the company) was one of our exclusions because it compared vildagliptin with only placebo, but it does provide some data on a measure of beta-cell function, the insulin secretory rate relative to glucose level after meals. This measure reflects the responsiveness of the beta cell to glucose, rather than absolute insulin production or plasma insulin level. The extension study was undertaken in under half of those who completed the original study (108 compared with 264). All of the original recruits had HbA_1c levels in the range of 6.2–7.5%. At recruitment the mean duration was 2 years.

Scherbaum et al. found that their measure of insulin secretory rate/glucose ‘tended to increase’ from the end of year 1 to the end of year 2, by which they mean that there was an increase that did not reach statistical significance. The implication is that there may be a steady improvement in beta-cell function. However, the mean HbA_1c level in the vildagliptin group fell after initiation, reached a nadir of about 6.2% by around 32 weeks, and then slowly rose to about 6.4% by 110 weeks. That rise suggests that vildagliptin is not having a dramatic effect on beta-cell function. It may be slowing the progression of the disease, which has been reported by the UKPDS (16 or 17);4,5 it is worth noting that the graph shows that mean HbA_1c level rose a little more steeply in the placebo group, whereas in UKPDS the lines were roughly parallel.

So far, no definite conclusions can be drawn on the effects of sitagliptin and vildagliptin on long-term beta-cell function. If beta-cell function does improve, and if that improvement is sustained over the long term (say 10–20 years), then that would be very important and there would be a case for early use, perhaps as the first drug to be used when diet fails. Or, given that diet usually fails, perhaps from the diagnosis of type 2 diabetes.

There could be an issue about the duration of diabetes at which any beta-cell preservation effect might be seen. The UKPDS study2 reported that at diagnosis, about half of beta-cell function had been lost. If patients are then treated with incretin enhancers or mimetics after they had had diabetes for many years, it may be too late to see much effect. It would be interesting to assess effects on beta-cell function by duration of known diabetes, and perhaps also in people with impaired glucose tolerance (there is one trial of the effects of a DPP-4 inhibitor on people with impaired glucose tolerance116).

Emerging studies

Another third-line trial was reported at the ADA 2008 conference, in abstract only, by Dobs et al. (2008)126 It was an 18-week RCT of adding sitagliptin to metformin and rosiglitazone. HbA_1c level fell by 0.7% overall, but by 1.3% in those whose baseline HbA_1c level was over 9%.

At the same meeting, Krobot et al. (2008)133 had an abstract from a second-line trial, of metformin and sitagliptin versus metformin and glipizide. The effects on HbA_1c level were similar, but hypoglycaemia was less frequent with sitagliptin (any hypoglycaemic event 5%) than the sulfonylurea (32%).

At the September 2008 conference of the EASD three new gliptin trials were presented. One by Goodman et al. (2008)134 was of vildagliptin versus placebo as an add-on to metformin, and would be an exclusion under our criteria. The other two are of interest. Arjona-Ferreira et al. (2008)135 describe a Merck & Co-funded trial of adding sitagliptin in patients with inadequate control (HbA_1c level 7.5–11%) on metformin and rosiglitazone. Adding sitagliptin reduced HbA_1c level by 0.7% overall, but by 1.3% in those with baseline HbA_1c level of over 9%.

Braceras et al. (2008)136 presented a Novartis trial comparing vildagliptin to a glitazone in patients not adequately controlled (initial HbA_1c level of over 7%) on metformin, and found them to be roughly equivalent. HbA_1c level fell by 0.68% on vildagliptin + metformin, and by 0.57% on glitazone + metformin.

Another new trial118 was published in full in August 2008, but is not relevant for our purposes. It compared vildagliptin and placebo in patients who had not previously had drug treatment. Their hyperglycaemia was mild (baseline HbA_1c level 6.2–7.5%). After 1 year on treatment, HbA_1c level fell by 0.3% in those receiving vildagliptin and rose by 0.15% in those on placebo, which was statistically significant, if not clinically so. It does provide a useful reminder that the size of reduction in HbA1c level depends on baseline level.

Background

An ideal basal insulin would have a flat action profile (i.e. the same level at all times of day) with no day-to-day variability in the same patient. Older long-acting insulins use a crystalline or amorphous suspension, that forms a slowly dissolving depot after subcutaneous injection. The newer long-acting analogues have adopted different approaches. Both have structural changes.

In glargine, these changes mean that it is soluble in the acidic (pH 4.0) solvent in which it is provided, but once injected into the neutral pH of the subcutaneous tissues, it forms stable hexamers which slowly release the insulin into the bloodstream. 137 In detemir, one amino acid is omitted and a long-chain (14-carbon) fatty acid, myristoyl, is attached. This facilitates binding of detemir to serum albumin. It has been suggested138 that albumin binding may facilitate transport into the brain, and that this might cause slightly less weight gain than is seen with other insulins. 139

Inclusion criteria

Search strategy

Relevant literature was identified, and comprehensiveness checked, by:

• searches of bibliographic databases, MEDLINE, Cochrane Library and EMBASE

• checking reference lists of retrieved studies

• obtaining lists of published studies from manufacturers

• our peer review process.

Searches were also carried out to identify emerging evidence, from conference abstracts and trial registers. Studies available only in abstract were included in the assessment of clinical effectiveness if there was a paucity of studies published in full in peer-reviewed journals, but they were reported with appropriate caution. Our default position is for studies available only in abstract not to be used.

Authors of previous studies were not contacted.

Quality assessment of studies

The quality of systematic reviews was assessed using the following quality criteria, based on the NICE guidelines manual:

• appropriate and clearly focused question

• inclusion and exclusion criteria described

• literature search sufficiently rigorous to identify all relevant studies

• study selection described

• data extraction described

• study quality assessed and taken into account

• study flow shown

• study characteristics of individual studies described

• quality of individual studies given

• results of individual studies shown

• enough similarities between studies selected to make combining them reasonable.

Each of the items was rated as: well covered/adequately addressed/poorly addressed/not addressed/not reported/not applicable.

The overall quality of the review was rated as (++), (+) or (–).

Randomised controlled trials were assessed on the following criteria, based on the NICE guidelines manual:

• appropriate and clearly focused question

• method of randomisation

• allocation concealed

• participants blinded

• outcome assessors blinded

• all relevant outcomes measured in standard, valid, reliable way

• proportion of participants excluded/lost to follow-up

• handling of missing data

• intention-to-treat (ITT) analysis performed

• statistical analysis appropriate

• only difference between groups is treatment under investigation

• results in multicentre studies comparable for all sites

• groups comparable at baseline.

Again, overall quality of the trials was classified as (++), (+) or (–).

Data extraction

Data extraction was carried out by one researcher and checked by another. Any disagreements were resolved through discussion, involving a third person if necessary.

Data analysis

The clinical effectiveness, relative to the key comparators, was assessed, in terms of difference in effect size. (The key question for the cost-effectiveness analysis is not whether a drug is better than the comparator, but how much better.)

Data were summarised using tables and text. In addition, we performed a meta-analysis of all relevant trial data, combining data from the previous meta-analyses with newly identified trials. Data were summarised for continuous variables (for example, HbA_1c, weight change) as WMDs with 95% CIs using the inverse variance method and a random effects model. For dichotomous variables (hypoglycaemia), data were expressed as relative risks with 95% CIs (for patients with or without hypoglycaemia) and summarised using the Mantel-Haenzsel method and a random effects model. For data already used in previous meta-analyses, data were generally used as given in the meta-analyses, although some double-checking was undertaken with the original papers. Where not given directly, standard deviations (SDs) were either calculated from standard errors (SEs) or CIs, or in case of no measure of variability reported, the average of the SDs for the other studies for that outcome measure was used. If the SDs were missing for more than half of the studies, meta-analysis was considered not to be reliable and a statistical summary was not presented. Meta-analyses were generally done for end-of-study values except for weight change, as most studies reported data for weight change without giving baseline values. Heterogeneity was assessed using the chi-squared statistic.

We had set out to conduct sensitivity analyses to explore uncertainties in important parameters, and of the impact of hypoglycaemic episodes and the fear of hypoglycaemic episodes on QoL.

We did not include any indirect comparisons, for two main reasons. Firstly, such comparisons are prone to bias due to confounding variables, which may not all be apparent. Secondly, they are used mainly in technology appraisals, when seeking to decide which of two or more options is better or best. We do not expect the guideline development group will wish to make any recommendations of whether glargine should be used in preference to detemir, or vice versa, because such comparisons would be based partly on cost, which may change. Having two drugs available in each group encourages competition on price.

Search results

Fourteen papers were identified as potentially relevant systematic reviews. Of these, five fulfilled the inclusion criteria. 140–144 Most of the remainder did not use systematic review methodology, and one was only a protocol for a systematic review (Table 15). Two further systematic reviews were identified after the completion of the present analysis and these will only be summarised briefly. 145,146

Description of reviews

The characteristics of the included reviews are shown in Appendix 4. Of the five included reviews, the reviews by Duckworth et al. (2007)140 and Wang (2003)143 had only a very limited description of methodology, the review by Horvath et al. (2007)141 was a Cochrane review, and the reviews by Warren et al. (2004)144 and Tran et al. (2007)142 were Health Technology Assessments (one from the UK and one from Canada). Four of the reviews had non-industrial funding, whereas the review by Duckworth et al. (2007)140 was funded by Sanofi-aventis (New Jersey, NJ, USA).

Inclusion criteria

Only four out of the five reviews specified the study design of the studies to be included. The others included RCTs (or just ‘clinical trials’), where Warren et al. (2004)144 specified a minimum duration of 4 weeks, Horvath et al. (2007)141 24 weeks and Wang (2003)143 specified a minimum number of participants of 100. Wang also included other designs to answer different parts of their review question, but only the clinical efficacy trials are considered here.

Both Health Technology Assessments and the review by Wang included both participants with type 1 and type 2 diabetes. The remaining two reviews were concerned only with participants with type 2 diabetes. The present review only summarises parts of the included reviews that describe patients with type 2 diabetes.

The reviews by Duckworth and Davis (2007),140 Warren et al. (2004)144 and Wang (2003)143 focused on insulin glargine, whereas the reviews by Horvath et al. (2007)141 and Tran et al. (2007)142 reviewed the effects of both insulin glargine and insulin detemir. Comparison treatments were NPH insulin in the study by Duckworth and Davis, and Horvath et al. , another long-acting basal insulin in the review by Warren et al. , conventional human insulin or OAD agents in the review by Tran et al. , and comparison treatments were not specified in the review by Wang.

Outcomes that reviews set out to assess included glycaemic control (HbA_1c, FPG), hypoglycaemia (overall, severe and nocturnal), other adverse events, mortality, cardiovascular morbidity, diabetic late complications and health-related QoL.

Trials included in systematic reviews

Review quality

The review by Duckworth and Davis (2007)140 was of poor quality. Its search strategy was restricted to a PubMed search and English articles only, and no information was given on other methodological procedures such as study selection, quality assessment of trials, data extraction or data analysis. Inclusion criteria were briefly specified, but only for participants, interventions and outcomes, not for study design.

Both the Cochrane review by Horvath et al. (2007)41 and the Canadian HTA Assessment by Tran et al. (2007)142 were of good quality. Inclusion criteria were well described, as was study selection, quality assessment of trials, data extraction, and data analysis. A comprehensive search was carried out and described in detail. Study flow was shown. Both reviews included a meta-analysis.

The UK HTA Assessment by Warren et al. (2004)144 appears good but had some reporting omissions. Inclusion criteria were well described and the search strategy was very comprehensive. However, it is unclear whether study selection and quality assessment were done in duplicate and data extraction was only done by one reviewer. Study flow was not shown.

The review by Wang (2003)143 was of poor quality. Inclusion criteria were described and the search strategy was adequate. However, study selection, quality assessment, data extraction and data analysis were not described, nor was study flow shown. Although no details of quality assessment methodology were given, some comments on study quality were made.

Results

Main results are shown in Table 17 and subgroup analyses in Table 18.

Glycaemic control

Trials generally showed a reduction in HbA_1c level from baseline to end of study, but without any difference between comparison groups. Horvath et al. (2007)141 carried out two meta-analyses regarding HbA_1c results for insulin glargine versus NPH insulin, one including only the four for which SDs were available or could be calculated, and the other including studies where this was not the case and where a pooled SD was used (two extra studies, i.e. six in the meta-analysis). In both analyses, there was no significant difference between glargine and NPH in end-of-study level of HbA_1c, with a WMD between groups of around 0 (for all six studies WMD 0.0, 95% CI –0.01, 0.1). Similarly, Tran et al. (2007)142 in their meta-analysis of seven studies found no significant difference in HbA_1c values between glargine and NPH (WMD 0.05, 95% CI –0.07 to 0.16). For the remaining reviews, results were presented according to whether patients had received previous insulin treatment without oral treatment or were previously insulin naive with concomitant oral treatment. For the two trials (or rather one with subgroup analysis) of people with previous insulin treatment, HbA_1c level at the end of study was similar between the glargine and NPH groups (reduction from baseline between –0.35 and –0.6%). For the trials in insulin-naive patients using concomitant oral therapy, most trials showed no significant difference between glargine and NPH at the end of the study either, except in the study by Fritsche (2003),162 where, after 24 weeks of treatment HbA_1c level was significantly more reduced with morning glargine than with evening glargine or evening NPH (–1.24% versus –0.96% and –0.84%, respectively). Subgroup analyses in two trials, one of insulin-naive patients and one of patients applying insulin once rather than twice daily also showed no difference in HbA_1c values between groups. However, one study found a significant effect for HbA_1c in favour of glargine in a subgroup analysis of patients with a BMI of more than 28 kg/m² (HbA_1c change from baseline –0.42% with glargine and –0.11% with NPH, p = 0.024). There was no significant difference in end-of-study HbA_1c values in the two studies of insulin detemir versus NPH, irrespective of previous treatment and cointerventions.

Where reported, the percentages of patients reaching the FPG or HbA_1c targets were also similar between insulin glargine and NPH insulin, except in the study by Fritsche (2003),162 where significantly more patients reached the target with morning glargine than with evening glargine or evening NPH.

Hypoglycaemia

Glycaemic excursions

Data on glycaemic excursions were only systematically summarised by the review by Tran et al. (2007),142 who reported data from three studies that had measured eight-point glucose profiles. There was generally no statistically significant difference between glucose profiles for glargine versus NPH with the exception of two trials. One study showed significantly lower pre-dinner glucose levels for glargine, and the other reported significant values for morning (but not evening) glargine in comparison with evening NPH. For insulin detemir, eight-point glucose profiles were similar in comparison to NPH, irrespective of the cointervention.

Total daily insulin dose

Total daily insulin dose was not systematically reported by the systematic reviews. Warren et al. (2004)144 reported for one trial of patients with previous insulin use, that patients on pre-trial once-daily NPH used slightly more insulin at trial end than at baseline, and patients on more than once-daily NPH pre-trial used slightly less insulin in the glargine group at the end of the trial (reduced by 4.4 U/day) than patients treated with NPH, who used about the same (no more data presented). For one trial of previously insulin-naive patients on oral therapy, Warren et al. (2004)144 reported similar insulin consumption of 23 U/day for glargine and 21 U/day for NPH, but statistical information was not provided. Insulin daily doses were not provided for the trials using insulin detemir.

Weight change

Weight change was not systematically reported by the systematic reviews. Wang (2003)143 reported a significant change in weight gain for a trial of patients previously treated with insulin, with patients receiving insulin glargine gaining significantly less weight than patients on NPH insulin (+0.4 kg versus +1.4 kg, p < 0.001). In two other trials of previously insulin-naive patients on oral therapy, no significant difference in weight change was seen between the glargine and NPH insulin groups (total changes between no change and +2.6 kg). Horvath et al. (2007)141 reported significantly less weight gain with insulin detemir than NPH insulin with a weight difference of between 0.8 and 1.6 kg between the comparison groups (p < 0.05).

Diabetic complications

Data on diabetic complications were not systematically reported by the reviews – and were generally not available in the trials (and trials were underpowered for assessing such outcome parameters). Several reviews – and trials – reported mortality data, but numbers were generally small and deaths were considered to be unrelated to the trial interventions. No data on diabetic late complications were included in any of the reviews, but Horvath et al. (2007)141 found some information on diabetic retinopathy for one trial of patients with previous insulin treatment and for one trial of patients on oral therapy (some of whom had been insulin pretreated). In the trial including oral therapy, 8.4% of patients in the insulin glargine group and 14% of patients in the NPH insulin group who had had no retinopathy at baseline developed non-proliferative retinopathy, and 1.8% and 2.4%, respectively, developed clinically significant macular oedema. Progression of retinopathy by more than three stages was seen in 5.9% of patients on glargine and 9.1% of patients on NPH (no significance values reported). In the study of patients on previous insulin treatment without oral therapy, significantly more patients on glargine had a progression of retinopathy by three or more stages than with NPH (7.5 versus 2.7%, p = 0.028). In the study of patients on concomitant oral therapy, no significant difference in development of clinically significant macular oedema was seen between glargine and NPH (11.2% with glargine, 6.5% with NPH, p = NS). However, there was a marked difference in this outcome between previously insulin-naive patients and patients pretreated with insulin. In insulin-naive patients, the development of clinically significant macular oedema in 14% in the glargine group and 4% in the NPH group. In contrast, patients previously treated with insulin had incidences of 1.9% and 12.7% (no significance reported). Numbers of diabetic late complications occurring in one of the trials of insulin detemir were too small to draw any conclusions.

Adverse events

No significant differences in adverse events, number of patients with adverse events, severe adverse events, or withdrawals because of adverse events were generally seen between insulin glargine or detemir and NPH insulin. There was some indication in some trials that a greater number of patients on insulin glargine reported injection site pain than patients on NPH insulin, but pain was usually mild and did not result in discontinuation of treatment. Where reported, no differences in insulin antibodies were seen between study groups. None of the studies was long enough to assess any longer-term effects.

Health-related QoL

No data were reported on health-related QoL. Wang (2003)143 and Horvath et al. (2007)141 reported on one study each, and it was suggested that there was a significantly greater improvement of treatment satisfaction with insulin glargine than with NPH insulin.

Additional reviews identified after completion of this review

Two systematic reviews, both including meta-analyses, were identified after completion of the main analyses for this review. The review by Bazzano et al. (2008)145 focused on the safety and efficacy of glargine compared with NPH insulin in type 2 diabetes, whereas the review by Monami et al. (2008)146 considered both glargine and detemir compared with NPH insulin in type 2 diabetes. Bazzano et al. included 12 RCTs and Monami et al. included 11 RCTs of glargine versus NPH insulin and three RCTs of detemir versus NPH insulin. All of the RCTs included in the two reviews have been considered by the present review.

The review by Bazzano et al. was of good quality. The search strategy was thorough, inclusion criteria were described, as was data extraction, quality assessment and data analysis. Study flow was shown. Descriptive and quality data were given for each included RCT. The review by Monami et al. was also of good quality. Inclusion criteria, search strategies, data extraction, quality assessment and data analysis were described. Study flow was shown and descriptive and quality data were shown for each trial.

Both reviews suggested that there was no significant difference between glargine or detemir and NPH insulin for glycaemic control. Bazzano et al. reported slightly less patient-reported hypoglycaemia with glargine than with NPH insulin, and Monami et al. reported less symptomatic and nocturnal hypoglycaemia with glargine or detemir versus NPH. Bazzano et al. reported slightly less weight gain with NPH than with glargine, whereas Monami et al. reported no differences in BMI when comparing glargine and NPH, but a lower BMI with detemir than with NPH insulin.

Review conclusions and research recommendations

Review conclusions and recommendations are shown in Table 19.

Although there were some differences in assessment of the data between reviews, all reviews essentially came to the same conclusions. All reviews concluded that insulin glargine – and insulin detemir where assessed – led to a glycaemic control equivalent to that when using NPH insulin.

Regarding the occurrence of hypoglycaemia, all reviews concluded that insulin glargine – and where assessed, probably also insulin detemir – were more effective at reducing nocturnal hypoglycaemias than NPH insulin. In addition, there was no between group differences for severe hypoglycaemias, and the evidence was inconclusive regarding overall/symptomatic hypoglycaemias (with some reviews being more optimistic than others). However, the review by Horvath et al. (2007)141 suggested that even the effect on nocturnal hypoglycaemias was only minor. Only Tran et al. (2007)142 systematically assessed glycaemic excursions and concluded that, overall, there was no significant difference in glucose profiles between glargine or detemir and NPH insulin. None of the studies came to any firm conclusions regarding total insulin dose or weight change. Not enough trial information was available to make any conclusions about diabetic secondary complications or health-related QoL. Overall, reviews concluded that there were no significant differences in adverse events between glargine or detemir than with NPH insulin (although there may be slightly more injection site pain with glargine, as reported by some reviews).

In some of the reviews, it was suggested that the clinical relevance of the findings was unclear: trials were thought to have major design flaws (for example, all being open label, giving limited information on important factors such as cointerventions, etc.). In addition, Warren et al. (2004)144 suggested that trial patients may have used different regimens than patients in usual clinical practice.

Not all of the reviews included clear recommendations for research; where given, research recommendations included the need for:

• long-term follow-up data to assess effectiveness in terms of diabetes late complications and safety issues

• studies in young and old patients (i.e. younger and older than the age range of 55–62 years in the included studies)

• more uniform and rigorous study design and reporting of results; including definitions of different types of hypoglycaemia

• studies of QoL, focusing on assessing both the short-term immediate impact of acute episodes of hypoglycaemia and the longer-term impact of living with a reduced fear of hypoglycaemia; and other aspects of the impact of the different insulin on patients’ QoL.

Search results

Fourteen papers were identified as potentially relevant RCTs. Of these, six fulfilled the inclusion criteria, but one175 turned out to refer to a trial [Hermansen et al. (2006)172], already included in the review by Horvath et al. (2007). 141 One abstract and one full publication referred to the same trial176,177 of insulin glargine versus insulin detemir. Full data extraction was undertaken for five trials. 143,177–180 Table 20 shows the excluded trials. Trials were excluded because they did not include the comparisons of interest (for example, no comparison with another basal insulin), because data were inadequate or because no outcomes of interest were investigated.

Description of trials

Characteristics of the included trials are shown in Appendix 5.

Trial quality

Details of the quality assessment of the trials are shown in Table 21.

The LEAD trial (2007)179 and the trial by Wang et al. (2007)185 had a number of quality deficits, whereas the trials by Philis-Tsimikas et al. (2006)180 and Rosenstock et al. (2008)177 and the PREDICTIVE-BMI trial (2007)178 were of better quality.

In the LEAD trial (2007),179 the method of randomisation was not described, nor was allocation concealment. Participants and outcome assessors were not blinded. As far as reported, all relevant outcomes were measured in a standard, valid, reliable way. The proportion of participants excluded/lost to follow-up was only reported for the whole study group, but not for comparison groups separately, with five patients (1%) withdrawing before receiving treatment or not providing any outcomes, and 49 excluded due to major protocol violations (11%). ITT analysis was performed, but handling of missing data was not reported. The comparison groups were comparable at baseline. The study population was 100% Asian.

The trial by Wang et al. (2007)185 was underpowered (only 24 participants); randomisation and allocation concealment were not described, and neither was blinding. As far as reported, all relevant outcomes were measured in a standard, valid, reliable way. Withdrawals or dropouts were not mentioned, and handing of missing data and ITT analysis were not reported. The study groups were comparable at baseline.

The PREDICTIVE-BMI (2007)178 trial had adequate randomisation and allocation concealment. Participants were not blinded, blinding of outcome assessors was not reported. As far as reported, all relevant outcomes were measured in a standard, valid, reliable way. The proportion of participants excluded/lost to follow-up was reported with reasons for each comparison group separately, with no significant differences between study groups (7% withdrawals/losses to follow-up). ITT analysis was performed, and handling of missing data was by last observation carried forward. The comparison groups were comparable at baseline.

In the trial by Philis-Tsimikas et al. (2006),180 the method of randomisation was described and adequate, but allocation concealment was uncertain. Participants and outcome assessors were not blinded. As far as reported, all relevant outcomes were measured in a standard, valid, reliable way. The proportion of participants excluded/lost to follow-up was reported with reasons for each comparison group separately, with no significant differences between study groups (11% withdrawals/losses to follow-up). ITT analysis was performed, and handling of missing data was by last observation carried forward. The comparison groups were comparable at baseline.

The trial by Rosenstock et al. (2008)177 had adequate randomisation and allocation concealment. Participants were not blinded, blinding of outcome assessors was not reported. As far as reported, all relevant outcomes were measured in a standard, valid, reliable way. The proportion of participants excluded/lost to follow-up was reported with reasons for each comparison group separately, with no significant differences between study groups (10% withdrawals/losses to follow-up). ITT analysis was performed, and handling of missing data was by last observation carried forward. The data were analysed in a non-inferiority analysis. The comparison groups were comparable at baseline.

Results

Results for the five trials are shown in Table 22.

Glycaemic control

None of the trials found any significant difference in HbA_1c values between insulins glargine or detemir and NPH insulin at study end. Levels of HbA_1c decreased by between 0.92% and 1.74% from baseline to study end. No significant difference between glargine and NPH was seen in the LEAD trial (2007)179 for patients reaching the HbA_1c target (< 7.5%: 38% for glargine, 30% for NPH) or the FBG target (≤ 6.7 mmol/l: 62% for glargine, 59% for NPH). There was a significant difference in the proportion of patients reaching the HbA_1c target (< 7.5%) without nocturnal hypoglycaemia in favour of glargine (23% for glargine, 14% for NPH, p = 0.017). There was no significant difference between detemir and NPH for patients reaching HbA_1c ≤ 7.0% in the PREDICTIVE-BMI trial (2007)178 (27% of patients in each group).

The results of the meta-analysis are shown in Figure 2 for insulin glargine and in Figure 3 for insulin detemir. Baseline HbA_1c values in the trials included in the meta-analysis were between 8.5% and 9.7% in the glargine versus NPH trials, and between 7.8% and 9.2% in the detemir versus NPH trials. None of the meta-analyses showed a significant effect for insulin glargine (nine studies) or insulin detemir (four studies) versus NPH for HbA_1c level. The WMD was 0.00% (95% CI –0.11 to 0.10) for glargine and 0.07% (95% CI –0.03 to 0.18) for detemir. There was significant heterogeneity for the results for insulin glargine which disappeared when the only study of patients on previous insulin therapy [Rosenstock et al. (2001)]159 was excluded.

FIGURE 2.

HbA_1c glargine versus Neutral Protamine Hagedorn.

FIGURE 3.

HbA_1c detemir versus Neutral Protamine Hagedorn.

Hypoglycaemia

The LEAD trial (2007)179 and the PREDICTIVE-BMI trial (2007)178 found significant results in favour of glargine and detemir respectively in comparison with NPH for all hypoglycaemia-related outcomes reported. The trial by Wang et al. (2007)185 found significantly fewer episodes of nocturnal hypoglycaemia with glargine compared with NPH, but no significant difference for all hypoglycaemic events. The trial by Philis-Tsimikas et al. (2006)180 found significant effects in favour of detemir for all comparisons of evening detemir versus evening NPH, but not for some of the other comparisons.

In the LEAD trial (2007)179 there were 682 hypoglycaemic episodes in the glargine group compared with 1019 in the NPH group (p < 0.004). There were 515 episodes of symptomatic hypoglycaemia in the glargine group compared with 908 in the NPH group (p < 0.0003), five of severe hypoglycaemia in the glargine group compared with 28 in the NPH group (p < 0.03), and 221 episodes of nocturnal hypoglycaemia in the glargine group compared with 620 in the NPH group (p < 0.001).

In the trial by Wang et al. (2007),185 there were two hypoglycaemic events in two patients in the glargine group and six hypoglycaemic events in four patients in the NPH group (p = NS). There was one nocturnal hypoglycaemic event in one patient in the glargine group and four nocturnal hypoglycaemic events in four patients in the NPH group (p = 0.028).

The PREDICTIVE-BMI trial (2007)178 reported significantly fewer hypoglycaemic events with detemir than with NPH (256 versus 481, RR 0.62; p < 0.0001) and also significantly less nocturnal hypoglycaemia (46 versus 107, RR 0.43, p < 0.0001).

In the trial by Philis-Tsimikas et al. (2006)180 there were too few major hypoglycaemic episodes for statistical analysis (only two events in the evening detemir group). For all confirmed hypoglycaemic episodes, there were 91 events in 32 patients on morning detemir, 82 events in 27 patients on evening detemir, and 153 events in 53 patients on evening NPH, with a significant difference in favour or evening detemir versus evening NPH, but not of morning detemir versus evening detemir or NPH. For nocturnal hypoglycaemia, there were six events in four patients on morning detemir, 19 events in eight patients on evening detemir, and 47 events in 22 patients on evening NPH, with a significant difference in favour or either detemir group versus evening NPH, but not of morning detemir versus evening detemir.

The meta-analyses for severe hypoglycaemia (Figures 4 and 5) included six studies (reporting the number of patients with severe hypoglycaemia) for insulin glargine versus NPH and four studies for insulin detemir versus NPH. There was no significant difference in the number of patients with severe hypoglycaemia in the glargine or detemir groups compared with NPH insulin [RR 0.82 (95% CI 0.45 to 1.49) for glargine and RR 0.59 (95% CI 0.15 to 2.24) for detemir]. There was no significant heterogeneity.

FIGURE 4.

Severe hypoglycaemia glargine versus Neutral Protamine Hagedorn.

FIGURE 5.

Severe hypoglycaemia detemir versus Neutral Protamine Hagedorn.

The meta-analysis for overall hypoglycaemia (Figures 6 and 7) included seven studies (reporting the number of patients with any hypoglycaemia) for insulin glargine versus NPH, and four studies for insulin detemir versus NPH. There was a significant difference in the number of patients reporting any hypoglycaemia in favour of the glargine and detemir groups compared with NPH insulin [RR 0.89 (95% CI 0.83 to 0.96, p = 0.002) for glargine, and RR 0.68 (95% CI 0.54 to 0.86, p = 0.001) for detemir]. There was no significant heterogeneity for glargine versus NPH but there was for detemir versus NPH (p = 0.002).

FIGURE 6.

Overall hypoglycaemia glargine versus Neutral Protamine Hagedorn.

FIGURE 7.

Overall hypoglycaemia detemir versus Neutral Protamine Hagedorn.

The meta-analysis for symptomatic hypoglycaemia (Figure 8) included four studies (reporting the number of patients with symptomatic hypoglycaemia) for insulin glargine versus NPH. There was a significant difference in the number of patients reporting symptomatic hypoglycaemia in favour of the glargine groups compared with NPH insulin [RR 0.80 (95% CI 0.68, 0.93, p < 0.004)]. There was significant heterogeneity (p = 0.04).

FIGURE 8.

Symptomatic hypoglycaemia glargine versus Neutral Protamine Hagedorn.

The meta-analysis for nocturnal hypoglycaemia (Figures 9 and 10) included seven studies (reporting the number of patients with nocturnal hypoglycaemia) for insulin glargine versus NPH and four studies for insulin detemir versus NPH. There was a significant difference in the number of patients reporting nocturnal hypoglycaemia in favour of the glargine and detemir groups compared with NPH insulin [RR 0.54 (95% CI 0.43 to 0.69, p < 0.00001) for glargine and RR 0.54 (95% CI 0.24 to 0.68, p < 0.00001) for detemir]. There was significant heterogeneity for glargine versus NPH (p = 0.03) but not for detemir versus NPH. The heterogeneity disappeared when the only study of patients on previous insulin therapy [Rosenstock et al. (2001)]159 was excluded.

FIGURE 9.

Nocturnal hypoglycaemia glargine versus Neutral Protamine Hagedorn.

FIGURE 10.

Nocturnal hypoglycaemia detemir versus Neutral Protamine Hagedorn.

Glucose excursions

The LEAD trial (2007)179 found eight-point blood glucose profiles to be similar between groups at study end, except for postdinner values, where blood glucose concentration in the glargine group was significantly lower than in the NPH group (236 mg/dl versus 249 mg/dl, p = 0.044).

In the Wang et al. (2007)185 trial, a continuous glucose monitoring system was used. No differences between glargine and NPH were found in average blood glucose values, pre-breakfast, 2 hours post breakfast, pre-lunch, 2 hours post lunch, and 2 hours post supper blood glucose values, but the SDs of blood glucose, FPG and bedtime PG were significantly smaller with glargine than NPH, pre-supper and bedtime blood glucose values were significantly lower with glargine than NPH, and 3 a.m. blood glucose values were significantly larger with glargine than NPH.

In the trial by Philis-Tsimikas et al. (2006),180 nine-point blood glucose profiles were similar for the two evening insulin groups, whereas the mean profile of the morning insulin detemir group was characterised by lower glycaemic values in the daytime and higher values overnight (p < 0.001). Pre-breakfast PG values were between 1.19 and 1.47 mmol/l higher (p < 0.001) in the morning detemir group, and pre-dinner PG values between 0.65 and 0.84 mmol/l lower (p ≤ 0.01) in the morning detemir group than in the evening groups.

Total daily insulin dose

No significant differences in mean daily insulin doses between treatment groups were reported in the LEAD trial (2007),179 the trial by Wang et al. (2007),185 the PREDICTIVE-BMI trial (2007),178 or the trial by Philis-Tsimikas et al. (2006). 180

Weight change

In the LEAD trial (2007),179 BMI increased both in the glargine and in the NPH group to a similar extent during the course of the trial (+1.4 and +1.3 kg/m²). Similarly, in the trial by Wang et al. (2007)185 body weight increased to a similar extent in both groups (+1.47 kg with glargine and +1.20 kg with NPH).

In the PREDICTIVE-BMI trial (2007),178 significantly less weight gain was seen with insulin detemir than with NPH insulin over the course of the trial (+0.4 kg versus +1.9 kg, p < 0.0001). Similarly, patients in the detemir group had a significantly smaller increase in BMI (+0.17 kg/m² versus +0.77 kg/m², p < 0.0001).

In the trial by Philis-Tsimikas et al. (2006),180 patients in the morning detemir group gained a mean of 1.2 kg, patients in the evening detemir group gained a mean of 0.7 kg, and patients in the evening NPH group gained a mean of 1.6 kg, with weight gain being significantly less in the evening detemir group than in the evening NPH group (p = 0.005, no other significant differences).

Overall (eight studies), the glargine groups gained 0.23 kg less weight than the NPH groups (range –1.10 to +0.23 kg). However, a meta-analysis could not be carried out for this outcome because of too many missing SDs. The detemir groups (four studies) gained 1.20 kg less weight than the NPH groups (range –0.8 to –1.6 kg) but, again, a meta-analysis could not be carried out due to too many missing SDs.

Diabetic complications

These were not reported by any of the trials.

Adverse events

The LEAD study (2007)179 reported 66 adverse events in 45 patients that were possibly treatment related (22 patients in the glargine group and 23 patients in the NPH group). The majority was related to injection-site reactions, and, although p-values were not reported, there does not seem to have been a significant difference between groups. There was no significant difference in serious adverse events between groups, and none of the events were considered unusual for the demographic group studied (i.e. not related to the treatment).

The trial by Wang et al. (2007)185 did not report adverse events.

In the PREDICTIVE-BMI trial (2007)178 there were 91 adverse events in the detemir group and 73 in the NPH group, six of these in the detemir group and four in the NPH group were serious (but thought to be unlikely to be related to basal insulin). There were three withdrawals because of adverse events in the detemir group and none in the NPH group.

In the trial by Philis-Tsimikas et al. (2006),180 there was no significant difference in overall adverse events between comparison groups (123–144 events in 67–82 patients in each group). No serious adverse events were considered to be related to the insulins. There was no significant difference in potential allergic reactions (1–5 events in 1–5 patients per group) or injection site reactions (2–7 events in 2–6 patients per group) between the groups.

Health-related QoL

This was not reported by any of the trials.

Glargine vs detemir

The main results of the included glargine versus detemir trial are shown Table 23.

The results of the trial by Rosenstock et al. (2008)177 suggest that the effects of glargine and detemir are similar. After 52 weeks of treatment, there were no significant differences in HbA_1c level, percentage of patients reaching HbA_1c value of ≤ 7.0% (with or without hypoglycaemia), overall hypoglycaemic events or nocturnal hypoglycaemic events. There was statistically significantly less weight gain with detemir overall than with glargine (+2.7 versus 3.5 kg, p = 0.03), but the difference of 0.8 kg is of doubtful clinical significance. However, when analysing use of detemir once or twice daily, only the once-daily detemir group was at an advantage for weight gain (+2.3 kg), whereas the weight gain in the twice-daily detemir group was similar to that of the glargine group (+3.7 kg). The mean daily dose was higher for detemir (0.52 U/kg with once-daily dosing, 1.00 U/kg with twice-daily dosing) than for glargine (0.44 U/kg). Injection site reactions were slightly more common with detemir than with glargine (4.5% versus 1.4%, p-value not reported).

Another short study, available in abstract only186 compared the effect of once daily glargine and detemir on blood glucose profiles over the course of a week, and found no significant difference.

Recent evidence

No new trials of glitazones with hard clinical outcomes, which were not known to the previous guideline group, were found.

We did find a trial that reported proxy outcomes. In the PERISCOPE trial, Nissen et al. (2008)202 compared pioglitazone with glimepiride (a sulfonylurea) to see if there were any differences in progression of coronary artery disease. A total of 543 patients had coronary intravascular ultrasonography to measure the extent of coronary atherosclerosis, were randomised to pioglitazone or glimepiride, and had their coronary investigation repeated 18 months later. The investigators were asked to try to achieve an HbA_1c level of < 7%. Baseline HbA_1c levels were identical in the two groups (7.4%), but, over time the glimepiride group developed slightly higher levels – HbA_1c 7.0% versus 6.9% (from the text: ‘figure 2 suggests that by study end the difference was about 0.3%’).

The main outcome measure was the mean atheroma volume. This increased by 0.73% (95% CI 0.33 to 1.12) in the glimepiride group but decreased by 0.16% (–0.57% to + 0.25%) in the pioglitazone group. The clinical significance of this small difference is uncertain, and, if the effect was due to the insulin-sensitising pioglitazone having advantages over the insulin secretagogue glimepiride, then, as the accompanying editorial points out, the more cost-effective approach would have been to compare metformin with a sulfonylurea. 203

A claim has been made recently that similar results have been obtained with rosiglitazone. These come from an unpublished trial, called VICTORY (Vein-Coronary Atherosclerosis and Rosiglitazone after bypass surgery). The results were presented at the American College of Cardiology 2008 conference, and the claim is reported in a newsletter, Heartwire (10 April 2008). 204 The data reported are of atheroma plaque volume, with a smaller percentage increase in those on rosiglitazone compared with those on placebo. Two comments are necessary. First, atheroma increased in both groups. Second, the different was not statistically significant (the p-value was 0.22). Further assessment must await full publication, but the details available at present do not justify the claim that the effect of rosiglitazone is similar to those seen with pioglitazone in PERISCOPE.

As reported in the recent guideline, a meta-analysis of the risk of cardiovascular events with pioglitazone was carried out by Lincoff et al. (2007)205 (who include Nissen and Wolski, who undertook the similar meta-analysis for rosiglitazone). Based on 19 trials with 16,930 participants, they concluded that pioglitazone was associated with a reduced risk of death, MI or stroke. They speculate that the differences in cardiovascular risk between rosiglitazone and pioglitazone are related to different effects on blood lipids (pioglitazone having a greater reduction in triglycerides and an increase in HDL cholesterol).

This meta-analysis included only trials funded by the manufacturer, because the authors used patient level data obtained from Takeda. Because most trials were short term and had relatively small numbers, around 80% of the events came from the PROactive trial.

Fractures

In the PERISCOPE trial, fractures occurred in 3% of the pioglitazone group but in none of the sulfonylurea group (p = 0.004).

Fracture risk has been reported in other studies. Kahn et al. (2006)29 in the ‘durability’ study (ADOPT) reported that 9.3% of women on rosiglitazone had fractures compared with 5.1% on metformin and 3.5% on glibenclamide. The increases were in fractures of upper limb and foot, rather than in the classical osteoporosis-associated neck of femur and vertebrae. There was no difference in men.

A case–control study by Meier et al. (2008),206 using British general practice data from General Practice Research Database, also found that use of glitazones was associated with increased fracture rates. No such increase was seen with other oral diabetes drugs.

A letter to physicians issued by Takeda Pharmaceuticals, and posted on the US FDA website207 reported an analysis of its clinical trials database on pioglitazone. They compared the incidence of fractures in over 8100 patients treated with pioglitazone compared to over 7400 patients treated with a comparator.

The fracture incidence calculated was 1.9 fractures per 100 patient-years in women treated with pioglitazone and 1.1 fractures per 100 patient-years in women treated with a comparator. The observed excess risk of fractures for women in this data set on pioglitazone is therefore 0.8 fractures per 100 patient-years of use. There was no increased risk of fracture identified in men.

The letter stated ‘the risk of fracture should be considered in the care of female patients with type 2 diabetes mellitus who are currently being treated with pioglitazone, or when initiation of pioglitazone treatment is being considered’.

More on the safety of glitazones

As new trials are reported, they are being added to new meta-analyses. Dahabreh (2008)221 updated the Nissen and Wolski (2007)194 meta-analysis with the results from the DREAM (Diabetes REduction Assessment with ramipril and rosiglitazone Medication)222 and ADOPT29 trials, and the interim report from the RECORD (Rosiglitazone Evaluated for Cardiac Outcomes and Regulation of glycaemia in Diabetes) trial. 196 [Note: DREAM included patients with impaired glucose tolerance (IGT) or impaired fasting glucose (IFG), not diabetes.] He noted the debate about the methods for doing meta-analysis when some trials had no events, and carried out the analyses using methods that allowed inclusion of such trials, as well as the Peto method, which was used in the original Nissen and Wolski meta-analysis.

The results were consistent with the previous finding of an increase in MI with rosiglitazone but ORs were slightly less, and in two of the five meta-analyses their CIs sometimes just overlapped with no increase (95% CIs of 0.97 to 1.59 and 0.96 to 1.57).

It is curious that rosiglitazone appears to increase non-fatal MI but not cardiovascular death. It may simply be a function of numbers, because the CV death ORs have much wider CIs.

Another meta-analysis by Mannucci et al. (2008)174 included 84 published and 10 unpublished trials of pioglitazone compared with placebo or active comparators, but excluded the PROactive trial. They reported a reduction of all-cause mortality with pioglitazone (OR 0.30; 95% CI 0.14 to 0.63, p < 0.05), but no significant effect on non-fatal coronary events.

Several new studies have asked why rosiglitazone should increase cardiovascular events but pioglitazone does not. Most have concluded that the likely reason is that while the two glitazones have the same effects on glycaemic control, and the same side effects of fluid retention and heart failure, they have different effects on blood lipids. Berneis et al. (2008)223 (based on data from the abstract only) carried out a very small crossover trial in 9 patients, giving them all 12 weeks on pioglitazone and 12 weeks on rosiglitazone. Total cholesterol increased more on rosiglitazone (need absolute levels) than on pioglitazone (p = 0.04), and triglycerides increased on rosiglitazone but decreased on pioglitazone (p = 0.004).

Chappuis et al. (2007)224 also studied patients on both glitazones, this time with 17 patients having 12 weeks on each. The effects of HbA_1c level were similar, but triglyceride and cholesterol levels were lower with pioglitazone.

Deeg et al. (2007)225 carried out a much larger comparison, with 369 randomised to pioglitazone and 366 to rosiglitazone. The two drugs had differing effects on lipids, with rosiglitazone having the more atherogenic pattern, including higher low-density lipoprotein (LDL) cholesterol levels.

Norris et al. (2007)226 carried out a systematic review of the comparative effectiveness and safety of pioglitazone and rosiglitazone. They concluded that effects of glycaemic control, weight and most adverse events were similar, but that rosiglitazone may increase total cholesterol compared with pioglitazone. However, they concluded they had insufficient evidence with which to compare cardiovascular event rates.

Data from the VADT (Veterans Affairs Diabetes Trial) have been used to assert that rosiglitazone does not cause cardiovascular harm [Duckworth (2009)227]. However, this evidence seems dubious, given that most patients in both arms were taking rosiglitazone.

The effect of all this has been that sales of rosiglitazone have fallen. A report in the newsletter Endocrine Today228 states that sales fell from US$617M worldwide in the first quarter of 2007 to US$327M in the fourth quarter (although it does not say whether the price was reduced). A Canadian report notes that there was a sudden decline in the use of rosiglitazone after the publication of the Nissen and Wolski meta-analysis,194 accompanied by an increase in the use of pioglitazone. 229

Points raised in the consultation process

In their responses to the draft guideline, GlaxoSmithKline referred to new studies that provided safety data. The studies cited were the ACCORD (Action to Control Cardiovascular Risk in Diabetes Study)230 and the VADT. 227

The ACCORD study was a trial of intensive versus standard therapy, aiming at a separation in HbA_1c level. The intensive group did worse with higher mortality and no reduction in cardiovascular events. In the intensive group, 2.6% of patients died from cardiovascular causes, versus 1.8% in the standard group (p = 0.02); 91% of the intensive group were treated with rosiglitazone versus 58% of the standard group. So the ACCORD study did not provide new data on the safety of rosiglitazone.

Inclusion criteria

Search results

Eleven papers were identified as potentially relevant RCTs. Of these, eight fulfilled the inclusion criteria and compared pioglitazone plus insulin with insulin. 231–238 One compared pioglitazone plus insulin with pioglitazone. The remaining trials were excluded because they did not examine the comparison of interest, and one was the uncontrolled extension of a trial that seemed relevant but could not be identified (shown in Table 24).

Description of studies – insulin + pioglitazone vs insulin

Characteristics of the included trials are shown in Appendix 6.

Quality of studies – insulin + pioglitazone vs insulin

Details of the quality of included trials are shown in Table 25.

For four231,233,234,238 of the eight trials, randomisation was adequate, whereas for the remaining four trials the randomisation procedure was not reported or unclear. Three trials231,233,238 had adequate allocation concealment, whereas the rest of the trials did not report on allocation concealment. All but one trial234 were described as double blind. Five trials used ITT analysis. 231,233–236 Five trials reported on follow-up rates231,233–235,238 and in those trials, between 77% and 92% of participants completed the trial, without any significant differences between comparison groups. Six of the eight trials reported that they had carried out a power calculation. 231–233,235,236,238 Two trials (reported as abstracts)236,237 did not report relevant baseline characteristics, five trials reported that there comparison groups were similar at baseline,232,233–235,238 while Berhanu et al. (2007)231 stated that participants in the placebo group had a slightly higher BMI at baseline and longer diabetes duration, but it was unclear whether these differences were significant. All but one trial237 reported on sources of funding, and all funding included industry funding.

Results – insulin + pioglitazone vs insulin

Details of the results for included trials for insulin plus pioglitzone versus insulin are shown in Table 26.

HbA_1c

All studies reported HbA_1c values and could be included in the meta-analysis (Figure 11). Baseline HbA_1c values were between 7.6% and 10% in the pioglitazone-plus-insulin groups, and between 7.8% and 9.8% in the insulin-without-pioglitazone groups. End-of-study HbA_1c values were significantly lower in the groups taking pioglitazone plus insulin than in the groups taking insulin without pioglitazone (WMD –0.58%, 95% CI –0.70 to –0.46, p < 0.00001). There was no significant heterogeneity. In the study by Mattoo et al. (2005),233 18% of patients on pioglitazone plus insulin and 6.9% of patients on insulin without pioglitazone attained HbA_1c values of below 7.0%. There was no significant difference between patients using two or fewer daily injections and patients using three or more. Similarly, there was no significant difference between patients who had previously been on OAD agents and those who had not. In the study by Rosenstock (2002),235 no significant difference in HbA_1c level was reported for the group using 15 mg/day of pioglitazone and the group using 30 mg/day.

FIGURE 11.

Forest plot of HbA_1c results – pioglitazone and insulin.

Hypoglycaemia

Six studies reported on hypoglycaemia outcomes and could be summarised in a meta-analysis (Figure 12). There were marginally more patients with hypoglycaemic episodes in the pioglitazone-plus-insulin groups than with insulin without pioglitazone (RR 1.27, 95% CI 0.99 to 1.63, p = 0.06). The results showed significant heterogeneity (p = 0.001). The study by Raz et al. (2005),234 which used biphasic insulin aspart 30 (BIAsp 30) rather than other insulin regimens contributed most to the heterogeneity. There is evidence to suggest that BIAsp 30 is associated with a reduced rate of nocturnal and major episodes of hypoglycaemia compared with other types of insulin. 241 After eliminating this study from the analysis, there remained moderate heterogeneity (I² = 57%, p = 0.05) and there was significantly more hypoglycaemia in the pioglitazone-plus-insulin groups (RR 1.40, 95% CI 1.14 to 1.73, p = 0.002).

FIGURE 12.

Forest plot of frequency of hypoglycaemia – pioglitazone and insulin.

Description of studies – insulin + pioglitazone vs pioglitazone

There was only one trial, published as an abstract, comparing pioglitazone with pioglitazone plus insulin. Characteristics of the included trial are shown in Appendix 7.

The focus of the study by Raskin (2006)242,243 was on the safety and efficacy of BIAsp 30 (30% soluble and 70% protaminated insulin aspart) in insulin-naive patients with type 2 diabetes, who are taking any two OAD agents. The study was a randomised parallel-group trial with a duration of 34 weeks and was carried out in the USA.

Results – insulin + pioglitazone vs pioglitazone

The trial by Raskin et al. (2006)241,242 found a significantly greater reduction of HbA_1c level at study end in the BIAsp 30-plus-metformin-plus-pioglitazone group than in the metformin-plus-pioglitazone group (–1.5% versus –0.2%, p < 0.0001)(Table 27). There were also larger proportions of patients reaching HbA_1c values of less than 7% in the BIAsp 30-plus-metformin-plus-pioglitazone group (76.3% versus 24.1% in the metformin-plus-pioglitazone group), as well as values less than or equal to 6.5% (59.1 versus 11.5%), values less than or equal to 6% (33.3 versus 2.3%) and values less than or equal to 5.5% (14.0 versus 0%). However, the BIAsp 30-plus-metformin-plus-pioglitazone group had significantly more minor hypoglycaemic events than the metformin-plus-pioglitazone group (8.3 versus 0.1 events/year, p < 0.001). The patients in the BIAsp 30-plus-metformin-plus-pioglitazone group also gained significantly more weight than the patients in the metformin-plus-pioglitazone group (4.6 versus 0.8 kg, p < 0.05). Peripheral oedema occurred in 10% of patients in the BIAsp 30-plus-metformin-plus-pioglitazone group and in 12% of patients in the metformin-plus-pioglitazone group.

Summary

Eight RCTs were identified comparing combinations of insulin and pioglitazone with insulin without pioglitazone regimes (two published only as abstracts). One trial (published only as abstract) was identified comparing a pioglitazone-plus-insulin regime with a pioglitazone-without-insulin regime. Compared with the insulin regimes, the pioglitazone-plus-insulin regimes reduced HbA_1c level by a mean of –0.54% (95% CI –0.70 to –0.38, p < 0.00001). However, hypoglycaemic events were marginally increased with the pioglitazone regimes (RR 1.27, 95% CI 0.99 to 1.63, p = 0.06). Where reported, studies tended to find reduced insulin doses in the pioglitazone groups, as well as increased HDL cholesterol values. None of the other lipid parameters reported (triglycerides, total cholesterol, LDL cholesterol) showed any systematic differences between the comparison groups. The studies tended to show increased weight (mean difference 2.91 kg) and more peripheral oedema with pioglitazone. The one trial comparing a pioglitazone-plus-insulin (plus metformin) regime with a pioglitazone (plus metformin) regime found significantly lower HbA_1c values in the groups taking insulin, but also more minor hypoglycaemic events and more weight gain. The rates of peripheral oedema appeared to have been similar between the groups.

Search strategy

The databases MEDLINE, EMBASE, Science Citation Index, ISI Proceedings and the NHS Economic Evaluation Database (NHS EED) were searched, as described in Appendix 1 (see Economic searches). Articles for inclusion were retrieved and initially screened by one author and then further screened selected by the health economist for inclusion.

QoL studies

Secnik et al. (2006)67 summarised the QoL effect of exenatide 10 µg twice daily and glargine once daily as observed in a 26-week Phase III trial among 455 per-protocol patients with type 2 diabetes. These were added to patients’ existing regimes of metformin and a sulfonylurea. Both the addition of exenatide and the addition of glargine demonstrated statistically significant improvements in the SF-36 vitality subscale score: from 53.18 to 56.30 for exenatide and from 55.18 to 57.62 for glargine. They were also associated with statistically significant improvements in the DSC-R (range 0–5) total score, with exenatide recording an improvement from 1.07 to 0.90, and glargine an improvement from 0.99 to 0.84. Both exenatide and glargine were reported as showing statistically significant improvements in the psychology: fatigue, psychology: cognitive, ophthalmology, hypoglycaemia and hyperglycaemia subscales of the DSC-R. Statistically significant improvements in the Diabetes Treatment Satisfaction Questionnaire were also observed: from 26.41 to 29.48 for exenatide and from 26.31 to 30.04 for glargine, with the perceived frequency of both hypoglycaemia and hyperglycaemia recording improvements for both groups. However, while the change in EQ-5D was of similar size between the two groups, for exenatide the change from 0.82 to 0.85 was not statistically significant, with p = 0.08, whereas for glargine the change from 0.84 to 0.87 was with p = 0.05.

A study by Yurgin et al. (2006),68 available only as an abstract, reported the effects of exenatide compared with biphasic insulin when added to existing regimes of metformin and a sulfonylurea from a 52-week non-inferiority trial among 505 patients with type 2 diabetes. The HbA_1c effects were similar, –0.98% for exenatide and –0.88% for biphasic insulin. Exenatide led to statistically significant improvements in EQ-5D visual analogue scale (VAS) of 3.39, the SF-36 vitality scale of 3.89 and the DSC-R of –0.13. No significant effect was observed in the TFS. There were no statistically significant changes in these for biphasic insulin, although it should be noted that there was also no statistically significant difference between exenatide and biphasic insulin with the exception of the DSC-R, which recorded an increase under biphasic insulin of +0.05.

Weight, nausea, QoL and cost of treatment

Ratner et al. (2006)244 reported a progressive reduction in weight of an average around 2.4 kg by week 30 within a placebo-controlled trial of exenatide among 150 patients with type 2 diabetes. From these, 92 patients also completed a 52-week follow-up study to give a total time horizon of 82 weeks. The average weight loss at 30 weeks was –3.0 kg, this increasing to –5.3 kg by week 82.

Blonde et al. (2006)245 report similar results from a somewhat larger placebo-controlled trial in 1446 patients, of whom 1125 (or 78%) completed the initial 30-week trial. In total, 974 of these patients entered the open-label phase, 668 of these having been originally randomised to receive exenatide within the placebo-controlled trial. Only 551 of these patients could be evaluated at the 82-week point due to enrolment dates, 314 of these completing the 52-week follow-up study. The ITT group and the completer cohort had similar weights and BMIs: 98 kg and 34 kg/m² and 99 kg and 34 kg/m², respectively. For this 82-week completer cohort, the average change at 30 weeks was –2.1 kg, which was reportedly similar to the range of –1.6 kg to –2.8 kg reported for the 10-µg arm of the placebo-controlled trial. Unfortunately the mean change for the 10-µg arm was not stated, and it should also be noted that the placebo control group also experienced weight loss of between –0.3 kg and –0.6 kg at week 30. Among the 82-week completer cohort, at week 82 the average weight loss was 4.4 kg, with 81% of patients having lost weight. The average change in weight among the 82-week completer cohort showed a generally increasing trend with BMI: for patients of less than 25 kg/m² the average weight loss was 2.9%, whereas for patients in increasing BMI increments of 5 kg/m² the average weight loss was 3.6%, 4.6%, 4.3%, until those with a BMI of more than 40 kg/m², for whom the average weight loss was 5.5%.