The appraisal committee (section 6) considered evidence from a number of sources. See the committee papers for full details of the evidence.
Type 2 diabetes is a chronic metabolic disorder in which a lack of the hormone insulin or resistance to its action causes elevated blood glucose levels (hyperglycaemia). It is a progressive disease, gradually worsening over time. The UK Prospective Diabetes Study (UKPDS) estimated an increase in haemoglobin A1c (HbA1c), which identifies average plasma glucose concentration, of around 0.2% per year.
Approximately 2.7 million people in England of 17 and over had a diagnosis of diabetes in 2013, of whom 90% had type 2 diabetes. However, many people with type 2 diabetes are undiagnosed, and so the number of people with the condition may be higher than reported. The prevalence of type 2 diabetes in England is rising because of increased obesity, decreased physical activity and increased life expectancy after diagnosis because of better cardiovascular risk protection. Type 2 diabetes is particularly prevalent in people of African, South Asian and Caribbean family origin.
Type 2 diabetes is not easy to live with and has a big impact on the day‑to‑day lives of people with the condition, their families and their carers. People are often concerned about the disease developing further. They may have to inject insulin, or may develop complications such as deteriorating eye sight or neuropathy, which could make it difficult for them to take their medication, to manage their blood glucose levels or to stay active.
Lowering blood glucose levels and achieving good diabetes control minimises the risk of developing complications, reduces the likelihood that someone will need to inject insulin to manage their disease, and can help to reduce anxiety and depression caused by the stress of managing diabetes. Diabetes can sometimes be controlled by diet and exercise, otherwise, tablets or insulin are needed.
The assessment group (AG) did a systematic review of the literature to identify studies evaluating the clinical effectiveness and safety of canagliflozin, dapagliflozin and empagliflozin as monotherapies for adults with type 2 diabetes not controlled by diet and exercise alone. The AG noted that the target population as defined in the scope was also people with type 2 diabetes who were unable to take metformin, but because this was not a distinction made in the trials, this could not form part of the search criteria. The AG identified 7 relevant double‑blind randomised controlled trials (2 each for canagliflozin and empagliflozin [including both licensed doses] and 3 for dapagliflozin). Four of the trials were international, 2 were solely based in Japan, and 3 were based in 'Asian' countries (including Japan and China). The canagliflozin and dapagliflozin trials compared treatments with placebo, and the empagliflozin trials included comparisons (described as 'exploratory') with DPP4‑inibitors. The AG did not identify any additional trials relevant to the scope that were not identified in the companies' submissions.
The AG stated that most people in the trials:
had diabetes for less than 5 years
had an HbA1c of approximately 7.5% to 8.4% (in the main comparison groups) and 10.6% to 11.5% (in the high HbA1c subgroups)
had a BMI of 25 to 34 kg/m2
were women (34% to 59% in the main comparison groups).
The mean age was 50 to 60 years. The clinical trials also reported subgroups based on baseline HbA1c and weight.
The primary outcome in all trials was change in HbA1c from baseline to the end of the main intervention period (24 to 26 weeks). For the primary outcome, all active treatments reduced HbA1c by between -0.39% and -1.17% more than with placebo. The reduction for empagliflozin 25 mg was also greater than sitagliptin 100 mg in exploratory analyses, but there was no difference when sitagliptin 100 mg was compared with empagliflozin 10 mg.
Secondary outcomes included change in weight, systolic blood pressure, hypoglycaemia, and cholesterol (total cholesterol, high‑density lipoprotein [HDL] cholesterol and low‑density lipoprotein [LDL] cholesterol). All selective sodium‑glucose cotransporter 2 (SGLT‑2) inhibitors reduced weight, by between 0.97 kg and 3.9 kg more than placebo. Compared with placebo, all SGLT‑2 inhibitors reduced systolic blood pressure. The AG stated that given the infrequency of reported hypoglycaemia, the similar outcomes between active and placebo arms, and the cut‑off level used, it was reasonable to assume that the SGLT‑2 inhibitors did not cause hypoglycaemia. For cholesterol, not all trials reported all outcomes. Generally, the SGLT‑2 inhibitors led to increases in all types of cholesterol.
The AG reviewed outcomes related to adverse effects of treatment in the clinical trials. The SGLT‑2 inhibitors were generally associated with a higher incidence of urinary tract infections and genital tract infections, both of which were more common in women. Most of these infections were mild to moderate in severity and responded to standard treatment.
The companies reported that canagliflozin, dapagliflozin and empagliflozin were well tolerated. The AG noted that rates of stopping treatment across the studies ranged from 7% to 20%, with rates balanced across groups. It noted that in the study by Inagaki et al. (2014), the rate of stopping was 7% in the canagliflozin group and 20% in the placebo group.
Because there was no direct evidence to compare the SGLT‑2 inhibitors with all the comparators in the scope, the companies and the AG did network meta‑analyses comparing SGLT‑2 inhibitors with dipeptidyl peptidase‑4 (DPP‑4) inhibitors, sulfonylureas, pioglitazone and repaglinide for people with type 2 diabetes not controlled by diet and exercise alone. Not all network meta‑analyses included repaglinide; submissions noted a lack of evidence and infrequent use in clinical practice. The AG noted that the eligibility criteria for the trials did not include metformin contraindication or intolerance, therefore not all of the patients in the trials were in line with the scope for this appraisal.
All companies and the AG presented network meta‑analysis results for outcomes including mean change in HbA1c, mean change in weight or BMI, mean change in systolic blood pressure, and hypoglycaemia incidence.
In its network meta‑analyses Janssen presented outcomes for: SGLT‑2 inhibitors (canagliflozin 100 mg and 300 mg; dapagliflozin 5 mg and 10 mg; empagliflozin 10 mg and 25 mg), DPP‑4 inhibitors (linagliptin, saxagliptin, sitagliptin, vildagliptin); pioglitazone (15 mg, 30 mg and 45 mg); sulfonylureas (glibenclamide, gliclazide, glimepiride, glipizide). The company presented both fixed‑effects and random‑effects models and did analyses at 26 weeks (plus or minus 4 weeks) to match the assessment times in its trials. Trials reporting results at 16 to 21 weeks and 31 to 36 weeks, trials published in conference abstracts only, and trials assessing repaglinide were included in sensitivity analyses. The company also did sensitivity analyses excluding non‑double‑blind trials.
Results for canagliflozin 100 mg were as follows:
compared with SGLT‑2 inhibitors, dapagliflozin, and empagliflozin 10 mg it resulted in a greater reduction in weight, and there were no differences for HbA1c and systolic blood pressure
compared with SGLT‑2 inhibitors, canagliflozin 300 mg, and empagliflozin 25 mg there were no differences (other than weight change, where canagliflozin 300 mg was associated with a greater weight loss)
compared with DPP‑4 inhibitors, it resulted in a greater reduction in HbA1c, weight and systolic blood pressure (other than when compared with sitagliptin for HbA1c, where there was no difference)
compared with sulfonylureas, only results for the HbA1c outcome were presented; there were no differences
compared with pioglitazone (all doses) it was more effective for change in weight and systolic blood pressure, and there was no difference for HbA1c
in all comparisons, there were no differences for hypoglycaemia.
The company stated that most sensitivity analyses had a minor effect on the results.
AstraZeneca presented outcomes for interventions as classes of treatment, rather than for specific drugs. The company stated this approach was relatively common in meta‑analyses of antidiabetic agents because of the large number of drugs and similar levels of effectiveness within most drug classes. Classes of drug considered were SGLT‑2 inhibitors, DPP‑4 inhibitors, sulfonylureas, and pioglitazone. The company only included trials reporting data at 24 weeks (plus or minus 6 weeks). It did sensitivity analyses using the alternative model to that presented in the base case (fixed or random effects); adjustment of HbA1c using a meta‑regression; and exclusion of 9 trials including only people described as 'Asian'. For SGLT‑2 inhibitors:
compared with DPP‑4 inhibitors and pioglitazone, there were no statistically significant differences for HbA1c and hypoglycaemia, and SGLT‑2 inhibitors were statistically significantly more effective for weight and systolic blood pressure reduction
compared with sulfonylureas, there were no statistically significant differences for HbA1c or systolic blood pressure; and SGLT‑2 inhibitors demonstrated statistically significantly greater weight loss and fewer hypoglycaemic events.
The company presented results for sensitivity analyses. It stated that there were only small differences between the base case and sensitivity analyses.
The company and the AG noted that some people in some of the dapagliflozin trials had a response to treatment with placebo, which was not seen in trials for other SGLT‑2 inhibitors. The company stated this may have been because of the short duration of the trials, and a motivated placebo group having diet and exercise interventions for the first time.
Boehringer Ingelheim presented outcomes for the following interventions in its network meta‑analyses: SGLT‑2 inhibitors (canagliflozin 100 mg and 300 mg, dapagliflozin 5 mg and 10 mg, and empagliflozin 10 mg and 25 mg), sulfonylureas (as a class), DPP‑4 inhibitors (alogliptin, linagliptin, saxagliptin, sitagliptin and vildagliptin), pioglitazone and repaglinide. The company noted that its economic model only considered sitagliptin 100 mg as a proxy for all DPP‑4 inhibitors. The company considered 3 time points in its network meta‑analysis: 24 weeks, 52 weeks and more than 52 weeks (only the results for 24 and 52 weeks are included here because these are the results used in the economic model). The company also presented results for a meta‑regression analysis, in which results were adjusted for baseline HbA1c.
For change in HbA1c, all results including empagliflozin 10 mg and 25 mg and other SGLT‑2 inhibitors showed greater reductions in HbA1c compared with placebo at 24 weeks and 52 weeks. For hypoglycaemia and urinary tract infection outcomes, the company found no differences for any treatment compared with placebo at any time point. However it noted that studies reported low numbers or zero events, therefore results were unreliable with wide credible intervals. For weight change, greater reductions in weight were seen with all SGLT‑2 inhibitors compared with placebo at 24 weeks. The company noted this was maintained for empagliflozin at 52 weeks (results at 52 weeks were not presented for other SGLT‑2 inhibitors). People taking pioglitazone, sulfonylureas and DPP‑4 inhibitors had increases in weight. For systolic blood pressure, all SGLT‑2 inhibitors showed decreases compared with placebo.
The AG considered the following interventions in its network meta‑analysis: canagliflozin (100 mg and 300 mg), dapagliflozin (10 mg), empagliflozin (10 mg and 25 mg), sulfonylureas, DPP‑4 inhibitors (linagliptin, sitagliptin and vildagliptin) and pioglitazone. It used trials of 24 to 26 weeks in which placebo was the comparator.
All SGLT‑2 inhibitors were more effective than placebo for HbA1c and weight change.
Compared with sitagliptin, SGLT‑2 inhibitors were either more effective for HbA1c (canagliflozin 100 mg and 300 mg, and dapagliflozin) or there was no difference (empagliflozin 10 mg and 25 mg), and all SGLT‑2 inhibitors were more effective for weight change.
Compared with sulfonylureas, there were no differences for HbA1c, and SGLT‑2 inhibitors were more effective for weight change.
Compared with pioglitazone, dapagliflozin and empagliflozin 10 mg were less effective for HbA1c (no differences compared with other SGLT‑2 inhibitors), and SGLT‑2 inhibitors were more effective for weight change.
The AG considered the effectiveness of the SGLT‑2 inhibitors compared with each other. It noted that both doses of canagliflozin lowered HbA1c more than dapagliflozin and both doses of empagliflozin. It stated that some of this reduction may be because studies suggested that canagliflozin, unlike other SGLT‑2 inhibitors, may also have an effect on the SGLT‑1 receptor (which reduces absorption of glucose in the gut). However, it could not be certain whether this dual mechanism of action was clinically significant.
The AG stated that there were several issues to consider when interpreting the results of the network meta‑analyses:
The higher doses of canagliflozin and empagliflozin were more effective than the starting doses. However in the clinical trials, people were randomised to the larger dose, rather than have to titrate up to it if the starting dose was insufficiently effective. Therefore it was not clear if the results seen for people starting on larger doses would be seen in clinical practice.
In the dapagliflozin clinical trials in the network, people in the placebo arm had a reduction in their HbA1c levels. This could be because of better access to lifestyle advice, but this was unlikely.
Many trials included in the network provided data on only some of the variables that are used in the UK Prospective Diabetes Study (UKPDS) outcomes model.
There was a lack of data in the trials to calculate the cholesterol ratio (ratio of total cholesterol to HDL cholesterol [TC:HDL ratio]) for use in the economic models and, when it was reported, it was often high. These high results were not likely to reflect clinical practice because of the use of statins.
Some of the trial evidence included the intervention given as combination therapy. For example, most available evidence for sulfonylureas for HbA1c and weight gain was from studies in which it was given with metformin. This may not represent their effectiveness when used as monotherapy.
Several trials noted issues with the durability of the effect of sulfonylureas (that is, the initial response was followed by a relatively rapid deterioration). In 1 trial the AG noted that 34% of people taking sulfonylureas needed additional treatment within 5 years compared with 15% of those taking rosiglitazone.
Comments from the patient organisation were that people with diabetes reported advantages of taking dapagliflozin (when used as combination therapy, as currently recommended by NICE). These were lowered blood glucose levels leading to increased self‑confidence in overall diabetes management, ease of administration, and no need to take the tablets with food. A concern about the treatment was the risk of genital fungal infection. It was noted that dapagliflozin has been shown to have positive effects on weight management, so may be of increased benefit to people with type 2 diabetes who are overweight.
The clinical experts stated that the SGLT‑2 inhibitors have an insulin independent mode of action, unlike other oral diabetes treatments used when metformin cannot be tolerated. This makes the risk of hypoglycaemia extremely low. They stated that the SGLT‑2 inhibitors were effective in improving HbA1c, and also had additional benefits of reducing weight and blood pressure. The clinical experts stated there were no data to confirm whether any SGLT‑2 inhibitor was most effective. For adverse events, the clinical experts stated that genital fungal infection was a concern, but this was usually mild and not repeated. There were no data to suggest an increase in more serious adverse events such as malignancies, but more long‑term data would be needed to confirm this. The patient expert stated that she had not had any adverse events while taking SGLT‑2 inhibitors.
The AG carried out a systematic review of the literature to identify studies of the cost effectiveness of SGLT‑2 inhibitor monotherapy compared with sulfonylureas, DPP‑4 inhibitors, pioglitazone and repaglinide for people with type 2 diabetes for whom metformin was not appropriate. No studies were found to be relevant to all SGLT‑2 inhibitors, and the AG and all the companies used existing economic models for diabetes to consider the cost effectiveness of SGLT‑2 inhibitor monotherapy.
The AG noted that the UKPDS had been used for many assumptions in the cost‑effectiveness analyses. It explained that UKPDS68 included a number of equations for estimating the progression of HbA1c, systolic blood pressure, ratio of total cholesterol to HDL cholesterol and smoking status over time, and the annual risk of micro‑ and macrovascular events associated with diabetes, for example stroke and blindness. It also predicts the annual risk of death and provides costs associated with adverse events. UKPDS68 was used by Oxford University to derive the OM1 cost‑effectiveness model. It has recently been updated by UKPDS82, which provides an alternative set of equations based on longer follow‑up data to those used in UKPDS68. The latest version is UKPDS84.
In all the models, people entered having had 1 of the scope interventions. The intervention determined the initial change from baseline in outcomes HbA1c, systolic blood pressure, weight change, and cholesterol levels. These outcomes progressed over time, with HbA1c worsening until it rose above 7.5%, triggering the start of another treatment (which improved outcomes, followed by another progressive worsening of HbA1c). Throughout the model, people received a pre‑specified treatment sequence depending on their initial treatment.
All models included micro‑ and macrovascular health states for morbidities and increased mortality associated with diabetes. Microvascular health states included retinopathy (including macular oedema and blindness), chronic kidney disease (ranging from stage 1 to end‑stage renal disease), and neuropathy (including peripheral vascular disease and amputation). Macrovascular health states included ischaemic heart disease, myocardial infarction, stroke, and congestive heart failure. The models also accounted for weight change, hypoglycaemia, urinary tract infections, genital tract infections, peripheral oedema, and stopping treatment. In addition, they included a health state in which modelled patients were free from complications. Health states were associated with costs, utility values, and in some cases a possible treatment contraindication or with excess risk of death (for example, through stroke or myocardial infarction).
The AG stated that the assumptions used in the Janssen model differed from those of the other 2 submissions. The main difference was the assumption used to model the change (or 'drift') in HbA1c over time. AstraZeneca, Boehringer Ingelheim and the AG all used the UKPDS68, whereas Janssen assumed a treatment‑specific drift in HbA1c that it described as 'linear in segment but inherently non‑linear'; that is, Janssen had assumed a linear drift in HbA1c, but downward pressure from rescue medication led to concave mean-HbA1c curves over time. All the models submitted were done from the perspective of the NHS and personal social services, discounted costs and health effects at 3.5% annually, and had a time horizon of 40 years. The cycle length was either 6 months (AstraZeneca) or 12 months (all other models).
The companies and the AG took most of their clinical effectiveness values from their own network meta‑analyses. Some data were also taken from the literature or trial data, and in some instances assumptions were used for missing values.
The AG, AstraZeneca and Boehringer Ingelheim all based their quality‑of‑life values on data from the UKPDS, and Janssen used the CODE‑2 study (an observational study of 4,000 people with type 2 diabetes in Europe, including the UK, based on the EQ‑5D health survey and using a UK tariff) dataset as its main source of quality‑of‑life values. The AG stated that all sources used to derive quality‑of‑life values by the companies were appropriate.
For costs, the AG stated there was variation in the models:
Direct drug costs in the models were similar (based on list prices), but the AG added additional costs of £72.26 for brain natriuretic peptide (BNP) monitoring (£26.26 for the test and £46.00 for a dedicated GP appointment) to the costs of pioglitazone in its model.
At treatment intensification (see table 1), the AG model assumed that people stayed on their initial monotherapy, whereas all the companies assumed that people switched treatments (at either the first or second intensification). This increased the total costs for all treatments, and also increased any initial cost variation between the starting monotherapy.
The price of canagliflozin 300 mg reduced after the company submissions were received (from approximately £608 to the same price as the 100 mg dose, approximately £477). All the companies used the higher price of canagliflozin 300 mg, whereas the AG was able to use the lower price.
The first year costs in the Janssen model were similar to the AG model, but costs for those with a history of adverse events were lower. The AG stated that this may be because the costs in the Janssen model did not include outpatient costs.
The costs in the AstraZeneca model were higher than those assumed by the AG; the AG was not sure why there was a discrepancy.
Boehringer Ingelheim applied the inpatient costs of the UKPDS84, but not the outpatient costs.
|
Company/group |
First intensification |
Second intensification |
Third intensification |
|---|---|---|---|
|
Janssen |
+gliclazide Other than:
|
Switch to NPH insulin |
+Insulin aspart |
|
AstraZeneca |
Switch to NPH insulin |
Intensify NPH insulin |
None |
|
Boehringer Ingelheim |
+Sulfonylurea; or +DPP‑4 inhibitor |
Switch to NPH insulin |
None |
|
Assessment group |
+gliclazide Other than:
|
+NPH insulin |
−gliclazide, +bolus |
Please note that in table 1, Janssen repaglinide intensifications differed and are not described in detail for the second intensification onwards.
Janssen used the ECHO‑T2DM model, using data from the CODE‑2 trial for most health‑related quality of life values. It did not identify any sources to determine disutility rates associated with adverse events, therefore it did a time trade‑off study of participants in the UK to determine the effect on quality of life from urinary tract and genital tract infections.
The company presented incremental cost‑effectiveness results (ICERs) for all treatments. The results for canagliflozin were presented for 3 arms: 100 mg, 300 mg, and 100 mg increased to 300 mg. The company presented results with and without pioglitazone, because it stated that use of pioglitazone was declining in the UK. In response to the appraisal consultation document, the company provided updated cost‑effectiveness results which corrected 2 errors found in the model (the updated results used a corrected reduction in HbA1c for sulfonylurea and pioglitazone, and a corrected stopping rule associated with eGFR for empagliflozin 10 mg) and used the updated lower price of canagliflozin 300 mg. This document only presents the updated base case results, however all sensitivity and scenario analyses were based on the original base‑case. ICERs compared with pioglitazone and sulfonylureas are presented in table 2. Compared with pioglitazone, sulfonylureas and DPP‑4 inhibitors were dominated. The ICER for canagliflozin 300 mg compared with pioglitazone was £42,782 per quality‑adjusted life year (QALY) gained, and canagliflozin 300 mg dominated all other treatments. In pairwise analyses canagliflozin 100 mg had ICERs of £1,987 per QALY gained compared with DPP‑4 inhibitors and £7,875 per QALY gained compared with a sulfonylurea, and it dominated (that is, was cheaper and more effective than) dapagliflozin and empagliflozin 10 mg.
The company did deterministic sensitivity analyses, all of which used canagliflozin 100 mg as the intervention arm. The company stated that canagliflozin 100 mg dominated dapagliflozin and empagliflozin in most analyses and results were relatively stable compared with all comparators.
The company did scenario analyses on 17 key drivers of cost effectiveness in the economic model. The assumption of HbA1c progression had the biggest impact on results. When HbA1c progression was based on equations taken from the UKPDS (instead of the 'linear in segment' assumption of progression used in the base case; see section 3.30), the ICERs for canagliflozin 100 mg were:
£71,395 per QALY gained compared with dapagliflozin
£50,826 per QALY gained compared with empagliflozin 10 mg
£133,274 per QALY gained compared with sulfonylureas.
The company presented probabilistic analyses for canagliflozin 100 mg compared with all comparators. Pioglitazone had the highest probability of being cost effective at ICERs of £20,000 and £30,000 per QALY gained, with probabilities of approximately 70% and 40% respectively. The probabilities for all other treatments were less than 20%.
The AG reviewed the model submitted by Janssen. It noted that the modelling was sensitive to the annual rate of HbA1c progression assumed for canagliflozin (changing the annual rate of drift in the base case from 0.14% to 0.112% [20% decrease] and to 0.168% [20% increase]). The AG stated that the changes are likely more because of the time spent on therapy and its immediate effects on treatment cost, weight, adverse events and hypoglycaemia than because of any changes in the modelled complications of diabetes. The ICERs compared with canagliflozin 100 mg were presented for a decrease and increase in HbA1c drift for canagliflozin:
pioglitazone: £45,862 and £211,446 per QALY gained
sulfonylureas: £593 and £8,751 per QALY gained
DPP‑4 inhibitors: canagliflozin dominant and £8,528 per QALY gained.
The AG stated that when comparing canagliflozin with dapagliflozin and empagliflozin, the main scenario analyses of interest were: using patient characteristics from the same database that was used in NICE's guideline on type 2 diabetes (but collected from a separate analysis conducted by Janssen); using UKPDS68 HbA1c progression; and using UKPDS68 HbA1c progression and quality of life (while also assuming that people can intensify their treatment to NPH insulin but not to basal‑bolus insulin). These scenarios changed the ICERs to between £5,000 to £10,000 per QALY gained.
AstraZeneca used the Cardiff diabetes model. The company did analyses for all drugs as a class, including the SGLT‑2 inhibitors, because they have similar safety and effectiveness and there is a limited amount of evidence for the individual treatments as monotherapy. The company stated that its primary analyses compared SGLT‑2 inhibitors with DPP‑4 inhibitors, because it expects SGLT‑2 inhibitors to displace DPP‑4 inhibitors in clinical practice.
In response to the assessment report, the company stated that it had found an error in its network meta‑analysis for the results for hypoglycaemic events. The resulting base case ICERs were £6,125 per QALY gained compared with DPP‑4 inhibitors, £20,639 per QALY gained compared with pioglitazone and £59,013 per QALY gained compared with sulfonylureas.
The company presented results of one‑way sensitivity analyses, including varying HbA1c and weight change outcomes using 95% credible intervals:
Compared with DPP‑4 inhibitors, the ICER was less than £10,000 per QALY gained in all sensitivity analyses.
Compared with pioglitazone, the ICER was most sensitive to the disutility associated with BMI increase, with a range of £14,626 to £32,065 per QALY gained.
Compared with sulfonylureas, the company noted that the ICER was sensitive to uncertainty about the relative efficacy of SGLT‑2 inhibitors and sulfonylureas for HbA1c (£42,724 to £165,409 per QALY gained) and weight change (£28,422 to £68,366 per QALY gained); and in utility value for decrease in BMI (£4,434 to £62,810 per QALY gained). The company stated these ICERs reflected the greater relative uncertainty in the network meta‑analysis for the comparison of SGLT‑2 inhibitors with sulfonylureas.
The company presented a range of scenario analyses for SGLT‑2 inhibitors compared with the comparators, including varying the HbA1c values at baseline and varying the HbA1c thresholds for intensifying treatment, altering the assumptions around maintenance of weight effects and the drug costs that were applied:
Compared with DPP‑4 inhibitors, the ICER was most sensitive to using the lowest priced DPP‑4 inhibitor (£22,756 per QALY gained).
Compared with pioglitazone, assuming weight convergence between SGLT‑2 inhibitors and DPP‑4 inhibitors at the second treatment switch increased the ICER to £38,199 per QALY gained (although the company stated that weight convergence was unlikely to occur in reality).
Compared with sulfonylureas, the ICER remained above £40,000 per QALY gained. The company stated that the base‑case ICER and scenario analyses compared with sulfonylureas were likely to be overestimates because sulfonylureas had an initially high clinical‑effectiveness estimate, but with a faster Hb1Ac progression than other treatments.
The company did probabilistic sensitivity analyses. At an ICER of £20,000 per QALY gained the probability that the SGLT‑2 inhibitors were cost effective compared with DPP‑4 inhibitors was 66%. Compared with pioglitazone and sulfonylureas the probabilities were 51% and 13% respectively.
The AG stated that it had concerns about the calculation of costs in the company model. This was because it appeared that the model only included inpatient and outpatient costs for patients who experienced a complication; inpatient and outpatient costs appeared to be completely omitted if the patient did not experience a complication. It stated that if this was the case, it would be a serious omission, and would bias the analysis in favour of the more effective treatment. It also noted that the company had used the same source for the costs of complications of diabetes (blindness and amputation; UKPDS84) as the AG, but that the AG had derived lower values, and it could not identify why.
The company presented 2 economic models based on OM1, which used patient‑level data from the UKPDS to extrapolate diabetes risk and predict long‑term costs and outcomes. Both models were similar, with patients initially treated for 1 year. In model A, people then entered the OM1 model with these treatment effects (for hypoglycaemia, urinary tract infection and weight change). Progression of disease was informed by UKPDS, with no further direct treatment effects, discontinuations, switches or intensifications. In the first year, people in the model could not die, and costs, quality of life and adverse events not related to treatment were not considered. The company stated that this accounted for the short‑term nature of treatment effectiveness evidence. In model B, the more complex model, people could stop treatment, switch and intensify treatment.
The company presented results for model B relative to the cheapest treatment (compared with pioglitazone in 52‑week data, and dapagliflozin in 24‑week data; see tables 3 and 4). In pairwise comparisons using 52‑week data, empagliflozin 10 mg had ICERs of approximately £30,000, £50,000 and £70,000 per QALY gained compared with sulfonylureas, pioglitazone and repaglinide respectively. When using 24‑week data, empagliflozin 10 mg had an ICER of £9,834 per QALY gained compared with dapagliflozin; was cheaper but less effective than canagliflozin 300 mg (when using the outdated higher price of canagliflozin 300 mg, see section 3.33); and was dominated by canagliflozin 100 mg and empagliflozin 25 mg.
|
Incremental costs |
Incremental QALYs |
ICERs (£/QALY) |
|
|---|---|---|---|
|
Empa 25 mg once daily |
£2,834 |
0.06 |
£46,480 |
|
Empa 10 mg once daily |
£2,837 |
0.06 |
£50,892 |
|
Pio 45 mg once daily |
Baseline |
Baseline |
Baseline |
|
Repa 1 mg once daily |
£635 |
0.03 |
£25,349 |
|
Sita 100 mg once daily |
£2,504 |
0.02 |
£163,917 |
|
Sulfonylureas |
£1,527 |
0.01 |
£121,660 |
|
Treatment |
Incremental costs |
Incremental QALYs |
ICERs (£/QALY) |
|---|---|---|---|
|
Empa 25 mg once daily |
£46 |
0.02 |
£2,172 |
|
Empa 10 mg once daily |
£68 |
0.01 |
£9,834 |
|
Cana 300 mg once daily |
£970 |
0.06 |
£17,363 |
|
Cana 100 mg once daily |
£1 |
0.03 |
£39 |
|
Dapa 10 mg once daily |
Baseline |
Baseline |
Baseline |
The company did not present any sensitivity or scenario analyses for model B. In response to the appraisal consultation document, the company provided a summary of one‑way sensitivity analyses not previously presented, which showed that results were most sensitive to the incidence of hypoglycaemic events, weight loss and the incidence of urinary tract infection. Other variables, including cost and utility decrements associated with adverse events, had less impact.
The AG stated that based on a comparison of the written submission with model B it appeared that the effects of placebo had not been included in the model (apart from hypoglycaemia and urinary tract infection rates), which could have underestimated the absolute treatment effects from baseline to 24 or 52 weeks.
The AG, in common with Boehringer Ingelheim, used the OM1 model for its submission. The AG assumed the use of the larger doses of canagliflozin and empagliflozin rather than the starting doses because it assumed that people would be at the maximum tolerated dose of each monotherapy drug before moving to dual therapy.
Table 5 presents the results of the model. Note that after consultation on the assessment report, the AG noted that the baseline assumption for ischaemic heart disease prevalence had been incorrectly set to zero. It therefore presented a revised base case (setting baseline ischaemic heart disease to 2.7%), which had a minor effect on the cost‑effectiveness results. This document presents the revised base case figures only, however all sensitivity and scenario analyses are based on the original base case (the AG did not have time to update the sensitivity and scenario analyses).
The AG noted that the SGLT‑2 inhibitors were of similar cost, but the canagliflozin overall costs were cheaper. This was because the greater effect of canagliflozin on HbA1c meant that treatment was intensified to the more expensive subsequent lines of treatment slightly later. The AG noted that because people remain on initial treatment for the duration of the model, the initial expense of the SGLT‑2 inhibitors and the DPP‑4 inhibitor sitagliptin compared with other treatments is maintained over the time horizon of the model. The AG noted that a key difference between the AG modelling and that of the companies was that the AG assumed that people remained on monotherapy and added treatments to it. Retaining the original monotherapy increased the total costs, and in particular increased the total cost for the SGLT‑2 inhibitors, and also sitagliptin.
The AG assumed an increase in weight of 0.1 kg per year. However it stated that there was debate about the effects of treatment on weight, because initial weight loss may be transient, and weight gain more permanent. Therefore it modelled 5 different scenarios for BMI, with a decrement of 0.0061 for each point above a BMI of 25 kg/m2 (as well as a scenario which assumed that BMI has no impact on quality of life, 'No BMI'). Scenarios were presented in which:
weight changes are maintained with no rebound to natural history (BMI‑1)
weight gains are maintained, and weight losses rebound to natural history after 1 year (BMI‑2)
weight gains are maintained, and weight losses rebound to natural history at intensification (BMI‑3)
weight changes rebound to natural history after 1 year (BMI‑4)
weight changes rebound to natural history at intensification (BMI‑5).
QALY gains for SGLT‑2 inhibitors were lowest when it was assumed that BMI had no impact on quality of life, with higher lifetime QALY gains for gliclazide, repaglinide and pioglitazone than SGLT‑2 inhibitors. However, if QALY gains for BMI were taken into account, the lifetime QALY gain was highest for the SGLT‑2 inhibitors. These gains were reduced if it was assumed that weight losses rebound after 1 year, and if it was assumed that weight losses rebound at treatment change.
The AG presented their results relative to the next least costly treatment that was not dominated (see table 6), and also compared with DPP‑4 inhibitors, sulfonylureas, and pioglitazone (see tables 7, 8 and 9, respectively). The AG stated that the SGLT‑2 inhibitors and DPP‑4 inhibitors were considerably more expensive than the other comparators, and if there were no direct quality‑of‑life effects from weight changes, the SGLT‑2 inhibitors were estimated to be dominated.
|
Treatment |
No BMI |
BMI‑1 |
BMI‑2 |
BMI‑3 |
BMI‑4 |
BMI‑5 |
|---|---|---|---|---|---|---|
|
Sulfonylureas |
– |
– |
– |
– |
– |
– |
|
Repaglinide |
Dom |
£3,388 |
£3,388 |
£3,388 |
£434,000 |
£16,413 |
|
Pioglitazone |
Dom |
Dom |
Dom |
Dom |
Dom |
Dom |
|
DPP‑4 inhibitor |
Dom |
Dom |
Dom |
Dom |
Dom |
Dom |
|
Cana 300 mg |
Dom |
£45,641 |
£207,000 |
£124,000 |
Dom |
£259,000 |
|
Empa 25 mg |
Dom |
Dom |
Dom |
Dom |
Dom |
Dom |
|
Dapa 10 mg |
Dom |
Dom |
Dom |
Dom |
Dom |
Dom |
Following consultation on the appraisal consultation document, the AG identified an error in the calculation of systolic blood pressure reduction for canagliflozin 300 mg. This reduced the base case ICERs for canagliflozin 300 mg (see table 10), but had no impact on the ICERs for dapagliflozin or empagliflozin. The updated ICERs shown in table 10 are based on the AG's original base case, and do not include a correction for the previously‑identified error for ischaemic heart disease (see section 3.52).
|
Treatment |
No BMI |
BMI‑1 |
BMI‑2 |
BMI‑3 |
BMI‑4 |
BMI‑5 |
|---|---|---|---|---|---|---|
|
versus DPP‑4 |
£4,401 |
£1,341 |
£3,586 |
£2,729 |
£3,929 |
£2,896 |
|
versus sulfonylureas |
£1.1 mn |
£32,015 |
£71,038 |
£57,865 |
£330,000 |
£118,000 |
|
versus pioglitazone |
£385,000 |
£27,003 |
£52,432 |
£44,590 |
£201,000 |
£90,653 |
The AG presented several scenario analyses, including urinary and genital tract infection rate applied to all cycles and assuming linear progression of HbA1c. When compared with the cheaper treatments, most scenarios did not have a substantial effect on the results. When compared with sitagliptin and assuming weight changes were maintained with no rebound to natural history (best‑case scenario for SGLT‑2 inhibitors), the ICERs remained under £10,000 per QALY gained.
The AG presented probabilistic ICERs, which were similar to the deterministic ICERs:
In probabilistic analyses when assuming no utility gain from the impact of BMI:
Including all comparators, SGLT‑2 inhibitors and sitagliptin had a 0% chance of cost effectiveness even at ICERs of £50,000 per QALY gained.
Compared with DPP‑4 inhibitors only, the probability of being cost effective was canagliflozin 45%, dapagliflozin 4%, empagliflozin 26%, and sitagliptin 26%, assuming an ICER of £20,000 per QALY gained.
In probabilistic analyses, assuming weight changes were maintained indefinitely:
Including all comparators, the probabilities were canagliflozin 6%, repaglinide 74%, and sulfonylureas 20%, when assuming an ICER of £30,000 per QALY gained.
Compared with DPP‑4 inhibitors only, the probability of being cost effective was canagliflozin 93%, dapagliflozin 0%, empagliflozin 6%, and sitagliptin 0%, assuming an ICER of £20,000 per QALY gained.