3 The manufacturers' submission

3 The manufacturers' submission

The Appraisal Committee considered evidence from a number of sources. See the committee papers for full details of the evidence.

Clinical effectiveness

3.1 The manufacturers carried out a systematic literature search to identify all relevant trials of dapagliflozin and potential comparators in adults with type 2 diabetes. The manufacturers identified 5 randomised controlled trials of dapagliflozin (10 mg once daily): 3 in patients with type 2 diabetes inadequately controlled with metformin alone (studies 14, 12 and 4), and 2 in patients with type 2 diabetes inadequately controlled with insulin with or without oral antidiabetic drugs (studies 9 and 6).

3.2 Of the 3 trials of dapagliflozin as an add-on to metformin, 2 were placebo controlled with follow-up of 24 weeks (studies 14 and 12) and 1 compared dapagliflozin with a sulfonylurea for up to 52 weeks of follow-up (study 4). The primary outcomes assessed were change in HbA1c from baseline (studies 14 and 4) and changes in body weight from baseline (study 12). Secondary outcomes included change in fasting plasma glucose, the proportion of patients whose HbA1c levels reached a specific target, change in body weight, change in blood pressure, the proportion of patients reporting hypoglycaemia, adverse reactions and tolerability. Baseline patient characteristics in the 3 trials were broadly similar: mean age 52.7–60.8 years, HbA1c level 7.16–8.11%, body weight 86.1–92.1 kg and systolic blood pressure 126.0–135.9 mmHg.

3.3 The 2 trials of dapagliflozin as an add-on to insulin were both placebo controlled, with follow-up of 12 weeks (study 9) and 24 weeks (study 6). The primary outcome assessed was change in HbA1c from baseline. Secondary outcomes included change in fasting plasma glucose, the proportion of patients whose HbA1c reached a specific target, change in body weight, change in the daily dose of insulin, adverse reactions and tolerability. Baseline patient characteristics in the 2 trials were broadly similar: mean age 55.7–59.3 years, HbA1c level 8.40–8.57%, body weight 94.5–103.4 kg and systolic blood pressure 128.9–140.6 mmHg.

3.4 In the add-on to metformin trials (studies 12 and 14), dapagliflozin was associated with a statistically significant reduction in HbA1c compared with placebo at 24 weeks. In study 14 (n=272), reduction in HbA1c was −0.84% for dapagliflozin versus −0.30% for placebo (p<0.0001). In study 12 (n=182), reduction in HbA1c was −0.39% for dapagliflozin compared with −0.10% for placebo (p<0.0001). Dapagliflozin was associated with a statistically significant reduction in body weight compared with placebo at 24 weeks in both study 12 (−2.96 kg versus −0.88 kg, p<0.0001) and study 14 (−2.86 kg versus −0.89 kg, p<0.0001). Dapagliflozin was associated with a reduction in systolic blood pressure compared with placebo at 24 weeks in both study 14 (−5.1 mmHg versus −0.2 mmHg, p value not reported) and study 12 (−2.70 mmHg versus +0.10 mmHg, p=0.06). Dapagliflozin was not associated with a statistically significant increased risk of hypoglycaemia compared with placebo at 24 weeks in either study.

3.5 In study 4 (n=814), dapagliflozin was shown to be non-inferior (based on a non-inferiority margin of 0.35%) to a sulfonylurea with respect to HbA1c reduction at 52 weeks. Dapagliflozin was associated with a statistically significant change in body weight compared with a sulfonylurea at 52 weeks (−3.22 kg versus +1.44 kg, p<0.0001). Dapagliflozin was associated with a statistically significant change in systolic blood pressure compared with a sulfonylurea at 52 weeks in study 4 (−4.3 mmHg versus +0.8 mmHg, p<0.0001). Dapagliflozin also resulted in a statistically significantly lower proportion of patients experiencing at least 1 hypoglycaemic event (3.5% versus 40.8%, p<0.0001) compared with a sulfonylurea by 52 weeks.

3.6 In the add-on to insulin trials, dapagliflozin was associated with a reduction in HbA1c compared with placebo at 12 weeks (study 9) and 24 weeks (study 6). In the 12‑week study (n=47), the change in HbA1c was −0.61% for dapagliflozin versus +0.09% for placebo (p value not reported). In the 24‑week study (n=387), the reduction in HbA1c was −0.96 for dapagliflozin versus −0.39 for placebo (p<0.001). Dapagliflozin was associated with a statistically significant reduction in body weight (−1.67 kg versus +0.02 kg, p<0.0001) and systolic blood pressure (−6.9 mmHg versus −3.9 mmHg, p=0.02) compared with placebo at 24 weeks. A higher proportion of patients treated with dapagliflozin had experienced at least 1 hypoglycaemic event (42.3% versus 35.0%) compared with placebo by 24 weeks. Dapagliflozin was associated with a statistically significant reduction in the calculated mean daily insulin dose (−1.16 versus +5.08 international units per day, p<0.0001) compared with placebo at 24 weeks.

3.7 The manufacturers conducted pre-planned analyses to determine if there were any variations in the clinical effectiveness of dapagliflozin for the following subgroups (as defined by the manufacturers): race, ethnicity, baseline HbA1c, age, sex and baseline body mass index (BMI). Subgroup analyses were conducted on pooled data as well as some of the individual studies of dapagliflozin. The manufacturers reported that no statistically significant differences in clinical effectiveness across subgroups were observed, except for baseline HbA1c. Dapagliflozin treatment generally resulted in greater HbA1c reductions from baseline in people with higher baseline HbA1c.

3.8 The manufacturers conducted network meta-analyses to compare the clinical effectiveness of dapagliflozin as an add-on to metformin or insulin with comparator therapies listed in the scope. Four outcomes were assessed: mean change in HbA1c from baseline, mean change in weight from baseline, mean change in systolic blood pressure from baseline, and the proportion of patients experiencing at least 1 hypoglycaemic episode. Random-effects models were selected over fixed-effects models because of variations in the study characteristics. The manufacturers presented analyses that were both adjusted and unadjusted for the potential modifying effects of baseline HbA1c.

3.9 For dapagliflozin as an add-on to metformin, the manufacturers created separate networks for the outcome of systolic blood pressure at 24 weeks (±6 weeks) and for the other 3 outcomes at 24 weeks (±6 weeks) and 52 weeks (±6 weeks). For the 24‑week analysis of systolic blood pressure, the network included dapagliflozin, dipeptidyl peptidase-4 (DPP‑4) inhibitors, glucagon-like peptide-1 (GLP‑1) analogues, sulfonylureas, thiazolidinediones and placebo in 8 studies. For the 24‑week analysis of outcomes other than systolic blood pressure, the network included dapagliflozin, DPP‑4 inhibitors, GLP‑1 analogues, thiazolidinediones and placebo in 15 studies. For the 52‑week analysis, the network included dapagliflozin, DPP‑4 inhibitors, thiazolidinediones and sulfonylureas in 6 studies.

3.10 The numerical results of the 24‑week network meta-analyses for the add-on to metformin comparisons were provided as academic in confidence. After adjusting for baseline HbA1c, dapagliflozin was associated with a statistically significant reduction in HbA1c compared with placebo. No statistically significant differences in the change in HbA1c were reported between dapagliflozin and other therapies. Dapagliflozin was associated with a statistically significant reduction in body weight compared with placebo, DPP‑4 inhibitors and thiazolidinediones, but not compared with GLP‑1 analogues. Dapagliflozin was associated with a statistically significant reduction in systolic blood pressure compared with placebo and sulfonylureas. However, no statistically significant differences in change in systolic blood pressure were reported between dapagliflozin and the other 3 drug therapies. No statistically significant differences in the risk of hypoglycaemia were reported between dapagliflozin and other drug therapies.

3.11 For dapagliflozin as an add-on to insulin, the manufacturers conducted a single network meta-analysis for all outcomes except systolic blood pressure at 24 weeks (±8 weeks). The network included dapagliflozin, DPP‑4 inhibitors, thiazolidinediones and placebo in 4 studies. The 12‑week study of dapagliflozin (study 9) and 3 other studies comparing thiazolidinediones with placebo were excluded from this analysis because they allowed up-titration of insulin to maintain glycaemic control. One of the studies identified, a study comparing thiazolidinediones with placebo, was excluded from the main analysis of mean change in HbA1c at 24 weeks because of the higher reported baseline HbA1c values compared with the other 3 studies. The outcome of change in systolic blood pressure at 24 weeks could not be analysed because, of the 4 identified studies, 3 either did not report changes in systolic blood pressure or involved up-titration of insulin.

3.12 Results of the 24‑week network meta-analyses for the add-on to insulin comparisons were provided as academic in confidence. Dapagliflozin was associated with a statistically significant reduction in HbA1c compared with placebo. No statistically significant differences in changes in HbA1c were reported between dapagliflozin and DPP‑4 inhibitors. When the study comparing thiazolidinediones with placebo was included as a sensitivity analysis, dapagliflozin was less effective in reducing HbA1c compared with thiazolidinediones. Dapagliflozin was associated with a statistically significant reduction in body weight compared with placebo and DPP‑4 inhibitors, and changes were reported to be similar to thiazolidinediones. Dapagliflozin was associated with a statistically significantly lower risk of experiencing a hypoglycaemic event compared with thiazolidinediones. However, no statistically significant differences were reported for the comparison of dapagliflozin with DPP‑4 inhibitors and placebo.

3.13 Data on the risks of adverse reactions associated with dapagliflozin were presented using pooled results from the placebo-controlled randomised controlled trials, including dapagliflozin as monotherapy and add-on therapy. Most results presented were based on short-term studies (24 weeks). The manufacturers reported that dapagliflozin was associated with a higher incidence of genital and urinary tract infections and a slightly higher incidence of volume depletion events (hypotension, hypovolaemia or dehydration) compared with placebo. Renal impairment or failure events were reported for a small proportion of patients (less than 1.5%) with no apparent difference between treatment groups. The manufacturers reported that the incidence of cancer was similar between patients who received dapagliflozin (1.47%) and patients who received placebo (1.35%). However, rates of bladder cancer (0.16% versus 0.03%), prostate cancer (0.34% versus 0.16%) and breast cancer (0.40% versus 0.22%) were higher in patients treated with dapagliflozin than in those treated with placebo respectively. In terms of cardiovascular safety, a meta-analysis of 14 randomised controlled trials did not find any evidence that dapagliflozin is associated with increased cardiovascular risk for a composite end point of cardiovascular death, myocardial infarction and stroke (hazard ratio [HR] 0.79, 95% CI 0.54 to 1.17).

3.14 Evidence on the clinical and cost effectiveness of dapagliflozin in triple therapy for people with type 2 diabetes that is inadequately controlled with metformin and a sulfonylurea was submitted in an addendum to address the comparisons specified in the scope. The manufacturers stated that dapagliflozin is currently being studied in an ongoing trial as a triple therapy add-on to 2 other oral agents. Therefore, data were pooled from a subset of people who were given metformin and a sulfonylurea at baseline from 2 placebo-controlled trials (studies 18 and 19), which were designed to assess the efficacy and safety of dapagliflozin in older people (average age 63–64 years) with type 2 diabetes and cardiovascular disease. A post-hoc analysis of this subset was conducted for changes from baseline in HbA1c, weight, systolic blood pressure and hypoglycaemic events at 24 weeks (results provided as academic in confidence).

3.15 No trials of dapagliflozin compared with active comparators in triple therapy were reported by the manufacturers. Therefore, the assessment of the clinical effectiveness of dapagliflozin compared with DPP‑4 inhibitors, GLP‑1 analogues and thiazolidinediones was based on indirect evidence. The manufacturers did not conduct a systematic review of triple therapy for people with type 2 diabetes that is inadequately controlled with metformin and a sulfonylurea. However, they referred to a literature review of add-on therapy to metformin and sulfonylureas for type 2 diabetes produced in 2009 by the Canadian Agency for Drugs and Technologies in Health. A summary of the results of this review suggested that DPP‑4 inhibitors, GLP‑1 analogues and thiazolidinediones were associated with statistically significant reductions in HbA1c compared with continued therapy with metformin and sulfonylureas. No statistically significant differences in HbA1c reduction were reported between DPP‑4 inhibitors, GLP‑1 analogues and thiazolidinediones. Thiazolidinediones, but not DPP‑4 inhibitors or GLP‑1 analogues, were associated with statistically significant weight gain compared with metformin and sulfonylureas. The manufacturers noted that since 2009, new data have become available including studies of the DPP‑4 inhibitors linagliptin and saxagliptin.

Cost effectiveness

3.16 The manufacturers submitted an economic model to evaluate the cost effectiveness of dapagliflozin for use:

  • in dual therapy as an add-on to metformin in adults with type 2 diabetes for whom metformin alone (with diet and exercise) does not provide adequate glycaemic control

  • as an add-on to insulin (with or without other oral antidiabetic therapies) when the underlying treatment regimen including insulin does not provide adequate glycaemic control and

  • in triple therapy for people with type 2 diabetes that is inadequately controlled with metformin and a sulfonylurea.

    For the add-on to metformin analysis, the comparator treatments were sulfonylureas, DPP‑4 inhibitors and thiazolidinediones (pioglitazone). For the add-on to insulin analysis, the comparator treatments were DPP‑4 inhibitors. For the triple therapy analysis, the comparator treatments were DPP‑4 inhibitors, thiazolidinediones and GLP‑1 analogues.

3.17 The manufacturers developed a simulation model run within an Excel front end but with the main calculations performed using C++ programming. The patient cohort entered the model with a set of baseline patient characteristics and modifiable risk factors that included HbA1c, total body weight, total cholesterol to high-density lipoprotein cholesterol ratio and systolic blood pressure. The value of these variables changed as the model simulation progressed, as a result of the effects of antidiabetic treatment and through natural progression, calculated from the UK Prospective Diabetes Study (UKPDS number 68) risk factor equations. The model then predicted the incidence of 7 specific macro- and microvascular events on the basis of the UKPDS 68 event risk equations. Macrovascular events predicted in the model included ischaemic heart disease, myocardial infarction, congestive heart failure and stroke. Microvascular events included amputation, nephropathy (end-stage renal failure) and blindness. The model also calculated the probability of drug-related hypoglycaemic events (non-severe and severe), other adverse events including urinary tract infections and genital infections, and treatment discontinuation caused by adverse events.

3.18 Simulated patients moved through the model in 6‑month cycles over a 40-year time horizon. At the start of the model, patients were assumed to have no complications associated with type 2 diabetes. At the end of the first 6‑month cycle, the UKPDS risk equations determined the probability of fatal and non-fatal complications in addition to diabetes-related deaths (myocardial infarction, congestive heart failure, stroke and amputation) and deaths from other causes (estimated separately from UK life tables). If a patient survived beyond the first cycle, they moved to the next cycle in which they remained at risk of treatment-related adverse events and long-term macro- or microvascular events. Once a diabetes-related death or death from other causes occurred, then costs, life years and quality-adjusted life years (QALYs) were updated and the simulation ended for that patient.

3.19 The model simulated a cohort of patients who received dapagliflozin (the 'treatment' cohort), and a cohort with the same baseline characteristics who received comparator treatments (the 'comparator' cohort). Simulated patients in each cohort received a particular therapy until their HbA1c increased up to a specified threshold (representing inadequate glycaemic control), at which point they stopped therapy and moved on to the second-line therapy (assumed to be the same in both cohorts). For the metformin and insulin add-on analyses, the model included up to 2 additional therapy lines after dapagliflozin and the comparator. The manufacturers assumed that second-line therapy was metformin and insulin, and third-line therapy for the remainder of the patients' simulated lifetime was intensified insulin (assumed to be a 50% increase from the starting dose). For the insulin add-on analysis, second-line therapy was intensified insulin for the remainder of the simulation. For the triple therapy analysis, all comparator triple therapies were assumed to be preceded by dual therapy with metformin and a sulfonylurea. The manufacturers assumed that after triple therapy, all patients would receive metformin and insulin. An NHS and personal social services perspective was taken and costs and benefits were discounted at 3.5%.

3.20 For the metformin add-on analyses, baseline patient characteristics, clinical-effectiveness data and adverse event rates were taken from study 4 for the comparison of dapagliflozin and a sulfonylurea and from the manufacturers' network meta-analysis (at 24 weeks) for all of the other comparisons. For the insulin add-on analysis, baseline patient characteristics, clinical-effectiveness data and adverse event rates were taken from the network meta-analysis (at 24 weeks). For the triple therapy analysis, clinical-effectiveness data were drawn from a pooled analysis of a subset of patients treated with dapagliflozin in 2 clinical trials (studies 18 and 19) and the Canadian Agency for Drugs and Technologies in Health's review of oral antidiabetic drugs as triple therapy. The manufacturers commented that the baseline patient characteristics from studies 18 and 19 were not representative of the triple therapy patient population. Therefore, baseline patient characteristics were taken from study 4 comparing dapagliflozin with a sulfonylurea in patients with type 2 diabetes inadequately controlled with metformin alone.

3.21 The HbA1c thresholds for switching treatment were based on baseline HbA1c values taken from the same sources. In the metformin add-on analyses, a threshold value of 7.72% taken from study 4 was used for the comparison of dapagliflozin and a sulfonylurea and a value of 8.17% from the metformin add-on network meta-analyses was used for the comparison of dapagliflozin with DPP‑4 inhibitors and thiazolidinediones. In the insulin add-on analysis, a threshold value of 8.90% was used based on the insulin add-on network meta-analyses. In the triple therapy analysis, the HbA1c threshold for switching treatment was 7.72%, taken from study 4.

3.22 The economic model included changes in weight associated with treatment. UKPDS risk equations based on BMI were included in the model. Therefore, changes in patient weight over time were converted to a BMI value based on baseline weight and height characteristics. If a treatment was associated with weight loss, this involved assumptions about how long the weight loss was maintained for along with the subsequent time until the loss of effect and return to the baseline body weight. In the dapagliflozin therapy group for the add-on to metformin and insulin analyses, weight reduction was assumed to be maintained for 2 years in the model based on 2-year extension data from the trial of dapagliflozin compared with a sulfonylurea. After year 2, weight was assumed to return to its baseline value until treatment was switched in a linear trend for the dapagliflozin therapy group. After this, a natural progression in weight gain of 0.1 kg per year was assumed. Because no data were available for DPP‑4 inhibitors, the same assumptions were applied. All other treatments were associated with a weight gain, which was applied in the first year, after which a natural progression in weight gain of 0.1 kg per year was assumed.

3.23 The model estimated the impact of macro- and microvascular complications of diabetes, changes in body weight and other adverse events on health-related quality of life. An age-dependent baseline utility function was derived from the Department of Health Survey for England (2003) which collected EQ‑5D data from patients with no major complications. Data on the impact on health-related quality of life of diabetes complications were taken from UKPDS (number 62) except for end-stage renal disease. In the UKPDS 62, the EQ‑5D questionnaire was completed by 3667 UK patients. This resulted in the following utility decrements: −0.09 (ischaemic heart disease), −0.055 (myocardial infarction), −0.108 (congestive heart failure), −0.164 (stroke), −0.28 (amputation) and −0.074 (blindness). The impact of end-stage renal disease on health-related quality of life was taken from the Health Outcomes Data Repository, a database of diabetic inpatients treated at Cardiff and Vale National Health Service Hospitals Trust, resulting in a loss in utility of −0.263. The impact of change in body weight on health-related quality of life was taken from a study of 100 Canadian patients with type 2 diabetes who completed a time trade-off exercise, which was commissioned by the manufacturers. Separate values were calculated for the changes in utility caused by a 1-unit decrease (+0.0171) or increase (−0.0472) in BMI. The impact of hypoglycaemic events on health-related quality of life was taken from a study by Currie et al. (2006) that estimated separate EQ‑5D utility decrements for symptomatic, nocturnal and severe events in UK patients with type 2 diabetes. The resulting utility decrements reported in the manufacturers' submissions were −0.042, −0.0084 and −0.047 respectively. The impact of urinary tract infections on health-related quality of life was taken from a study of urinary tract infections in ambulatory women, resulting in a utility decrement of −0.00283. In the absence of any other available data, the same utility values were used for genital infections.

3.24 The economic model included the acquisition costs of antidiabetic drugs taken from the England and Wales drug tariff (February 2012). The cost of insulin in the model was applied as a cost per kilogram of body weight per day, and therefore, varied in line with changes in patient body weight in the model simulation. The manufacturers assumed that insulin used as second- or third-line treatment in the model (with or without an oral antidiabetic) involved a 50% increase in dose over the initial starting dose in the add-on to metformin analysis, and a 25% increase in the add-on to insulin analysis.

3.25 The annual costs of macro- and microvascular diabetic complications, except for end-stage renal failure, were taken from UKPDS 65, which calculated the healthcare resource use of 3488 patients with type 2 diabetes. The UKPDS 65 study provided estimates of the first year event costs and the subsequent annual maintenance costs for patients who survived until the end of the simulation. The annual cost of end-stage renal failure (£34,806) was based on the weighted average cost of automated peritoneal dialysis, continuous ambulatory peritoneal dialysis, hospital haemodialysis and satellite unit-based haemodialysis, taken from a separate UK-based study. The cost of a severe hypoglycaemic event (£390) was taken from a study that measured health service costs incurred by 320 patients with type 2 diabetes in Germany, Spain and the UK who had experienced at least 1 hypoglycaemic event in the previous year. It was assumed that symptomatic and nocturnal hypoglycaemic events were not associated with any treatment costs. Urinary tract infections and genital infections were associated with the cost of a GP visit (£36). The costs of renal monitoring (£39), based on a GP visit and urine sample, were also included in the first year of the model only for the dapagliflozin treatment group. Treatment discontinuation was also assumed to incur the cost of a GP visit.

3.26 The manufacturers' base-case deterministic cost-effectiveness results for the add-on to metformin analyses found that the comparison between dapagliflozin and a sulfonylurea resulted in an incremental cost-effectiveness ratio (ICER) of £2671 per QALY gained (incremental costs £1246, incremental QALYs 0.467). The comparisons between dapagliflozin and DPP‑4 inhibitors and between dapagliflozin and thiazolidinediones found that dapagliflozin resulted in higher QALYs (incremental gains of 0.02 and 0.42 respectively) and lower costs (−£149 and −£60 respectively). Dapagliflozin therefore dominated both comparator treatments. For the add-on to insulin analysis, the comparison between dapagliflozin and DPP‑4 inhibitors resulted in an ICER of £4358 per QALY gained (incremental costs £517, incremental QALYs 0.119). The manufacturers' base-case deterministic cost-effectiveness results for the triple therapy analyses as add-on to metformin and a sulfonylurea found that dapagliflozin dominated DPP‑4 inhibitors, thiazolidinediones and GLP‑1 analogues, resulting in lower costs and higher QALYs.

3.27 The manufacturers also presented 2 scenario analyses that included alternative BMI-related utility values. The scenarios applied utilities of ±0.0061 and ±0.0038 respectively for a ±1 unit change in BMI. Both values were taken from a study by Bagust et al. (2005) evaluating the impact of BMI on EQ‑5D utility in patients with type 2 diabetes, and had been used in NICE's guideline on type 2 diabetes and technology appraisal guidance on exenatide in combination with oral antidiabetic therapy. For the metformin add-on comparisons, the ICERs for dapagliflozin compared with a sulfonylurea were £8863 and £10,514 per QALY gained respectively. Dapagliflozin remained dominant for the comparison of dapagliflozin with DPP‑4 inhibitors and thiazolidinediones. For the comparison of dapagliflozin with DPP‑4 inhibitors as add-on to insulin, the ICERs were also sensitive to changes to the BMI-related utility values. When changes in utility of ±0.0061 and ±0.0038 were applied, the ICERs increased to £21,171 and £32,409 per QALY gained respectively.

Evidence Review Group comments

3.28 The ERG commented on the scope of the appraisal and how the manufacturers addressed it in their submission. The ERG noted that the manufacturers did not include adults with type 2 diabetes that is inadequately controlled with sulfonylurea monotherapy in their submission. The ERG commented that the standard first-line monotherapy in type 2 diabetes is metformin, which is usually tolerated. The ERG noted that GLP‑1 analogues were not included as a comparator in the dual therapy setting, but considered that this was appropriate because their use in dual therapy is restricted. The ERG stated that NICE's guideline on type 2 diabetes recommends the use of pioglitazone as an alternative add-on treatment to a sulfonylurea in people with type 2 diabetes that is inadequately controlled by metformin. However, it also noted that there are increasing concerns about the adverse reactions associated with pioglitazone. The ERG commented that, in the triple therapy setting, DPP‑4 inhibitors would be expected to be given to patients before GLP‑1 analogues because they are cheaper and are administered orally. Overall, the ERG considered that DPP‑4 inhibitors are the key comparators for dapagliflozin in both the dual therapy and triple therapy settings.

3.29 The ERG stated that the manufacturers' approach to the systematic review of clinical evidence for dapagliflozin, which involved separate network meta-analyses for dapagliflozin as add-on therapy to metformin and as an add-on to insulin, was appropriate. The ERG noted that analyses were conducted for outcomes at 24 weeks and at 52 weeks and that studies reporting outcomes at less than 18 weeks, between 30 and 46 weeks, or greater than 58 weeks were excluded from the review. The ERG commented that it was not clear whether studies of between 31 and 45 weeks or greater than 58 weeks were also identified in the review. However, in response to a request for clarification, the manufacturers provided a full list of identified trials, none of which were between 31 and 45 weeks' duration. The ERG also noted that, for the network meta-analysis of insulin add-on therapies, a post-hoc amendment to the protocol was made to include studies in the range of 24 weeks ±8 weeks instead of ±6 weeks, to allow more studies to be included in the analysis.

3.30 The ERG commented that the manufacturers' approach to presenting the clinical effectiveness of dapagliflozin as a triple therapy add-on to metformin and a sulfonylurea was not very clear. Overall, the ERG considered that the methodology for the review of dapagliflozin in triple therapy (submitted as an addendum) was less robust than the main submission. However, the ERG acknowledged that the manufacturers had not intended to provide clinical-effectiveness data on dapagliflozin in triple therapy because of ongoing trial-based research due to report in 2013.

3.31 The ERG noted that the decision to switch or intensify treatment in the manufacturers' economic model was based on HbA1c levels above the thresholds currently recommended in the NICE guideline on type 2 diabetes. The ERG also noted that, when the manufacturers changed the HbA1c threshold levels in scenario analyses, along with changes to other input parameters, the ICERs for dapagliflozin increased. Overall, the ERG considered that the HbA1c threshold levels for switching treatment applied in the model reduced its relevance to UK clinical practice.

3.32 The ERG commented that the loss in utility associated with hypoglycaemic events, taken from Currie et al. (2006), may have been too large when applied within the model. The ERG noted from this study that a severe hypoglycaemic event in the previous 3 months was interpreted by the authors as causing a 4.7% loss in utility (−0.047). The ERG considered that the loss in utility associated with hypoglycaemic events should have been applied for 3 months rather than 12 months, resulting in QALY losses of −0.012 and −0.004 for severe and symptomatic hypoglycaemic events respectively.

3.33 The ERG commented on the appropriateness of the utility values applied to weight change in the model. It noted that the majority of QALY gains associated with dapagliflozin arose from direct impact of weight change on health-related quality of life rather than diabetic complications or adverse events. The ERG noted that in study 12, the dapagliflozin treatment group experienced a lower gain in utility (0.018 versus 0.047) compared with placebo at 24 weeks. However, when the utility estimates associated with changes in BMI were applied to the observed weight changes in study 12, the dapagliflozin treatment group experienced a higher gain in utility (0.016 versus 0.000) compared with placebo at 24 weeks. The ERG also noted that the study by Bagust et al. involved a multivariate analysis of EQ‑5D utility values that controlled for the complications of diabetes and estimated a smaller change in utility (±0.0061) associated with a unit increase or decrease in BMI. The ERG considered these alternative utility values, which were applied in the manufacturers' scenario analyses, to be more reasonable.

3.34 The ERG noted that the weighted average annual costs of pioglitazone (£414.07), based on the England and Wales NHS drug tariff for February 2012, were substantially higher than those estimated from the November 2012 tariff (£139.16). The ERG also estimated different annual costs of DPP‑4 inhibitors as add-on to metformin (£450.51 as opposed to £433.57) and GLP‑1 analogues as add-on to metformin and a sulfonylurea (£946.26 as opposed to £886.90). With regard to the costs of macro- and microvascular diabetic complications, the ERG noted that the UKPDS 65 study also included annual inpatient (£157) and non-inpatient (£159) costs for patients who did not experience a complication. The ERG commented that these annual costs of £483 (after inflating from 1999 to 2011 prices) should have been applied in the model for patients who did not experience a diabetic complication.

3.35 The ERG noted that, although the model cycle length was 6 months, the probabilities of macro- and microvascular events estimated from the UKPDS 68 study appeared to be for a 12‑month period and that no adjustment was made for this in the model. Further, the ERG noted from the DSU report on the economic model that the annual costs of macro- and microvascular events were not halved to correspond with the 6‑month cycle length used in the model but were applied in full immediately on the event occurring. The ERG commented that this would increase the annual costs of these events by half of the annual maintenance costs associated with the event.

3.36 The ERG noted that not all of the risk equations derived from the UKPDS 68 study were implemented in the model. From this study, the model implemented the risk of mortality in the year after a diabetic complication but not the risk of mortality in subsequent years after the event. Furthermore, risk equations for fatal myocardial infarction and fatal stroke were derived from a separate UKPDS study (number 66). This resulted in the risk of fatal myocardial infarction being a function of HbA1c and systolic blood pressure and the risk of fatal stroke being a function of systolic blood pressure only. The ERG considered that there was no obvious justification made by the manufacturers to include risk equations from this separate study. It also noted that this may have reduced the impact of HbA1c levels and increased the impact of systolic blood pressure in the model.

3.37 The ERG noted that, in the UKPDS 68 risk equations, baseline HbA1c was based on patients with newly diagnosed type 2 diabetes. However, the baseline HbA1c values implemented in the model were the trial baseline value minus the treatment-specific effect on HbA1c and therefore baseline HbA1c values differed between treatment groups. The ERG considered that the baseline HbA1c should have been the same for both treatment groups in the model. It noted that using different treatment-specific baseline HbA1c values resulted in the risk factor curves for both treatment groups not converging over time, whereas if the baseline HbA1c values had been the same for both treatment groups, the curves would have converged after the initial treatment effects. Similar considerations would apply to the other risk factors used in the UKPDS equations. Overall, the ERG concluded that the implementation of the UKPDS risk factor equations in the manufacturers' economic model may have been incorrect.

3.38 Similarly, the ERG noted that the event equation from UKPDS 68 used to estimate congestive heart failure included BMI at diagnosis. The ERG again noted that the baseline BMI values implemented in the model were the trial baseline value minus the treatment-specific effect on BMI and therefore that baseline BMI values differed between treatment groups. Because dapagliflozin was associated with a greater reduction in body weight compared with comparator drug therapies, the ERG considered that this may have biased the risk of congestive heart failure in favour of dapagliflozin. Furthermore, because the risk of congestive heart failure was associated with an increased risk of myocardial infarction and stroke, any overestimate of the rate of congestive heart failure would also result in an overestimate of the rate of myocardial infarction and stroke, along with the associated risk of fatality.

3.39 In the triple therapy analyses, the ERG considered that it was unnecessary for the model to include dual therapy with metformin and a sulfonylurea before switching to triple therapy. Because the model structure only permitted 3 lines of treatment, this resulted in patients switching to insulin and metformin after triple therapy. Therefore, unlike the dual therapy analyses, the triple therapy analysis did not enable patients to receive intensified insulin, which is associated with higher costs and additional weight gain.

Decision Support Unit comments

3.40 The DSU was commissioned by NICE to examine the economic model submitted by the manufacturers. The DSU was asked to report on whether the model functioned as described in the manufacturers' submission, to report any important aspects of the model that were not described in the submission, to examine whether the C++ programming code followed the steps described by the manufacturers and used the data described in the submission, and to check that the economic model produced the results described in the submission.

3.41 The DSU identified several differences between the economic model described in the submission and the executable model provided by the manufacturers. There were some differences between the macro- and microvascular event equations and risk factor equations in the model and those described in the manufacturers' submission. The effect of treatment on body weight was applied immediately in the model rather than gradually over the first year of treatment. All-cause mortality was not adjusted for fatal stroke and myocardial infarction events. The model did not apply the cost of renal monitoring to all patients who started treatment with dapagliflozin, although the DSU noted that this was unlikely to have a significant impact on the ICERs. There were some differences between the written submission and the model in regard to the time periods over which some of the costs and changes in utility were applied. The DSU also noted that the process used to sample from the relevant distributions in the probabilistic sensitivity analysis did not produce appropriately distributed samples, which may have underestimated the uncertainty around the QALYs estimated in the model.

3.42 The DSU identified several aspects of the executable model that were not described in the manufacturers' submission. In the manufacturer's model, the probability of an event occurring in a 6‑month cycle was calculated as the difference between the output of the event equation for the current cycle and the output of the event equation for the previous cycle. Treatment discontinuations applied in the first cycle of the model resulted in the patient switching treatment immediately without incurring costs or QALYs from the initial treatment except for the cost of discontinuation. The impact of treatment-related changes to BMI on health-related quality of life in the probabilistic sensitivity analysis was based on mean parameter values, which may have resulted in an underestimate of the uncertainty around the QALY differences estimated in the model.

3.43 The DSU commented that it was unable to reproduce the results of the probabilistic sensitivity analyses reported in the manufacturers' submission on the basis of the C++ programming code provided. However, the ICERs generated by the DSU did not vary substantially from those reported in the submission and it was noted that these differences may have arisen because of differences in the steps taken by the DSU to set up the probabilistic sensitivity analyses. When the DSU ran the model using the C++ programming code provided for the mean parameter values (deterministic analysis), it was also unable to reproduce the results of the deterministic analyses reported in the manufacturers' submission. Furthermore, when the DSU ran this code, it did not appear to have produced a stable estimate of the incremental QALYs after 100 runs. Finally, the DSU commented that the results generated by the programming code for the probabilistic sensitivity analyses when all parameters were set to their mean values did not match the results generated by the programming code that used mean parameter values. The DSU considered that similar results should have been produced and that this affected the confidence that could be placed on the results from the model.

Manufacturers' response to the appraisal consultation document

3.44 The manufacturers provided a response to the concerns raised by the DSU in its report on the economic model. The manufacturers stated that the economic model produced a stable estimate of the incremental costs and QALYs after 1000 rather than 100 simulations. The manufacturers implemented changes to the risk factor progression and event equations, and to the gamma and beta distributions applied to the cost and utility parameters in the probabilistic sensitivity analysis. The manufacturers also amended the model source code to correct for errors in the calculation of transition probabilities and the adjustment of all-cause mortality.

3.45 The manufacturers presented revised network meta-analyses for the dual therapy and add-on to insulin therapy comparisons, based on the WinBUGs programme code included in the technical support documents published by the DSU (Technical support document 2: a generalised linear modelling framework for pairwise and network meta-analysis of randomised controlled trials). The manufacturers also presented a validation exercise, which compared the results of the revised network meta-analyses with those presented in its original submission. The manufacturers commented that the revised analyses, which were provided as academic in confidence, produced similar results compared with the original analyses. The results of the revised 52‑week network meta-analysis were applied for the revised dual therapy analyses because these data enabled the same set of baseline characteristics and risk factors to be used for each comparator in the dual therapy analyses. The revised network meta-analysis at 24 weeks was applied for the add-on to insulin analysis in the manufacturers' revised economic model.

3.46 The manufacturers provided further clarification about how changes in body weight were modelled over time for the different treatments. In addition, the manufacturers provided unpublished follow-up data from study 4 which, they stated, showed that patients who remained on dual therapy of dapagliflozin and metformin maintained their weight loss for up to 4 years. The manufacturers therefore suggested that, for treatments associated with weight loss, the assumption in the model that this weight loss was maintained for 2 years may have been conservative.

3.47 The manufacturers made a number of revisions to the economic model to address the ERG's concerns. The revised economic model applied the same baseline risk factors for all treatment groups, which were taken from the revised network meta-analyses for the dual therapy and add-on to insulin analyses. The manufacturers applied an HbA1c threshold level of 7.5%, as currently recommended NICE's guideline on type 2 diabetes, for switching treatment for the dual therapy analyses. However, the manufacturers commented that this threshold may not reflect UK clinical practice because patients with type 2 diabetes are reviewed by their clinicians only once or twice a year and are therefore likely to have HbA1c levels that exceed 7.5% at the time of review. For the triple therapy and add-on to insulin analyses, the manufacturers applied HbA1c thresholds of 8.61% and 9.04% respectively for switching treatment. For the triple therapy analyses, the manufacturers also revised the sequence of treatments in the revised model so that the starting treatment was triple therapy rather than dual therapy.

3.48 In their revised model, the manufacturers applied utility values of ±0.0061 per unit increase or decrease in BMI taken from the study by Bagust et al. The manufacturers commented that the ERG had misinterpreted how the loss in utility associated with hypoglycaemic events was applied over a 6‑month cycle in the economic model. Therefore, the manufacturers did not reduce the loss in QALYs associated with hypoglycaemia to −0.012 for a severe event and −0.004 for a symptomatic event in their revised base-case analyses (instead, retaining the original utility values). In scenario analyses, the manufacturers applied a range of upper (−0.0104) and lower (−0.000657) estimates of the loss in utility associated with urinary tract and genital infections taken from a systematic literature review as requested by the Committee. The manufacturers also reduced the average annual cost of pioglitazone from £414.07 to £112.18 and included an annual cost of £483 for people not experiencing diabetic complications in the revised economic model.

3.49 The manufacturers presented ICERs for the revised dual therapy analyses, which included clinical-effectiveness data from the revised 52‑week network meta-analyses, changes to the model in response to the DSU report, the same baseline patient characteristics and risk factors for all treatment groups, and an HbA1c switch threshold of 7.5%. As a result of these changes, the ICER for the comparison between dapagliflozin and sulfonylureas was £1498 per QALY gained. For the comparisons between dapagliflozin and DPP‑4 inhibitors and thiazolidinediones, the ICERs were £689 and £5342 per QALY gained respectively. A scenario analysis which applied the upper and lower estimates of the loss in utility associated with urinary tract and genital infections resulted in very small changes to the ICERs for all comparisons.

3.50 The manufacturers also presented ICERs for the revised dual therapy analyses which included the changes described in section 3.49 and additional changes, which included reduced costs of pioglitazone, adjusted costs of diabetic complications and utility values of ±0.0061 per unit increase or decrease in BMI. As a result of these additional changes, the ICER for the comparison between dapagliflozin and sulfonylureas was £7735 per QALY gained. For the comparisons between dapagliflozin and DPP‑4 inhibitors and between dapagliflozin and thiazolidinediones, the ICERs were £3337 and £77,615 per QALY gained respectively.

3.51 The manufacturers presented ICERs for the revised add-on to insulin analyses, which included clinical-effectiveness data from the revised 24‑week network meta-analyses, changes to the model in response to the DSU report and an HbA1c switch threshold of 9.04%. As a result of these changes, the ICER for the comparison between dapagliflozin and DPP‑4 inhibitors was £2509 per QALY gained. A scenario analysis that applied the upper and lower estimates of the loss in utility associated with urinary tract and genital infections resulted in very small changes to the ICER. The manufacturers also presented an ICER that included adjusted costs of diabetic complications and utility values of ±0.0061 per unit increase or decrease in BMI. As a result of these additional changes, the ICER increased to £5634 per QALY gained.

3.52 The manufacturers also presented ICERs for the revised triple therapy analyses, which included altering the treatment sequences in the model so that patients in the model started treatment with triple add-on therapy to metformin and a sulfonylurea, incorporating model structural changes and applying an HbA1c switch threshold of 8.61%. As a result of these changes, dapagliflozin continued to dominate DPP‑4 inhibitors, thiazolidinediones and GLP‑1 analogues. The manufacturers did not present any additional scenario analyses for the relevant comparisons in triple therapy.

3.53 The manufacturers presented the results of a validation exercise, which compared the results from the revised model with the results that would have been obtained from using the CORE diabetes model for all relevant comparisons in dual therapy, insulin add-on therapy, and triple therapy. For the dual therapy analyses, the CORE model produced an ICER of £8879 per QALY gained for the comparison of dapagliflozin with sulfonylureas and ICERs of £2014 and £7093 per QALY gained for the comparisons of dapagliflozin with DPP‑4 inhibitors and with thiazolidinediones. For the insulin add-on analyses, the CORE model resulted in an ICER of £1675 per QALY gained for dapagliflozin compared with DPP‑4 inhibitors. For the triple therapy analyses, the CORE model produced ICERs of £1759 per QALY gained for the comparison of dapagliflozin with DPP‑4 inhibitors and £16,054 per QALY gained for the comparison of dapagliflozin with thiazolidinediones. The CORE model also produced an ICER of £32,243 per QALY lost for the comparison of dapagliflozin with GLP‑1 analogues.

3.54 Both the ERG and the DSU reviewed the manufacturers' revised economic model and analyses provided in response to the appraisal consultation document. Overall, the DSU considered that the manufacturers had adequately addressed all of the significant areas of concern about the model. The ERG noted that the revised dual therapy analyses used clinical-effectiveness data from the revised 52‑week network meta-analyses rather than the revised 24‑week network meta-analyses, which resulted in significant changes to the model input parameters. The ERG noted that, as a result of applying a lower HbA1c threshold for switching treatment, the revised model resulted in switching treatment earlier and thus reducing the costs of first-line dapagliflozin treatment whilst maintaining any long-term weight loss. The ERG also noted that the manufacturers' revised economic model had incorrectly amended the costs for people who did not experience diabetic complications.

3.55 The ERG highlighted a number of concerns about how changes in body weight were modelled in the manufacturers' revised analyses. The ERG noted that the manufacturers stated that, in order to simulate a linear, gradual regain of weight, the time to loss of weight effect was set such that weight was regained by the time of switch to next treatment. However, the ERG noted that in the manufacturers' comparisons of dapagliflozin with sulfonylureas and with DPP‑4 inhibitors in the revised dual therapy analyses, weight loss associated with dapagliflozin was largely maintained and not reversed at the time of switching to next treatment. The ERG also noted that the manufacturers' revised economic model and analyses did not address the Committee's concerns about the duration over which differences in weight change were maintained between treatments.

Additional DSU analysis in response to the revised manufacturers' model

3.56 In response to the concerns about the manufacturers' revised economic model raised by the ERG, the DSU was asked to review the manufacturers' revised economic analyses and to assess further how changes in weight were modelled over time for different treatments in the revised model. The DSU was also asked to conduct a range of further exploratory analyses for dapagliflozin in dual therapy and add-on to insulin therapy.

3.57 The DSU noted that, in the manufacturers' revised economic model, the assumptions about the duration over which any treatment-related weight change was reversed for the comparison of dapagliflozin with thiazolidinediones as add-on to metformin and the add-on to insulin analyses were consistent with those used in the original model. Therefore, for treatments associated with weight loss, weight was regained before first treatment switch. However, the DSU noted that for the comparisons of dapagliflozin as add-on to metformin with sulfonylureas and DPP‑4 inhibitors, treatment-related weight loss was not reversed at treatment switch in the revised model. The DSU suggested that the weight profiles for dapagliflozin and DPP‑4 inhibitors may have been incorrectly amended in the model.

3.58 The DSU noted that, for second- and third-line treatments, the weight at the start of treatment in the revised model was based on the weight at the time of switching from the previous treatment. The DSU noted that this was problematic if the treatment switch occurred before the treatment-related weight loss was regained. The DSU stated that where this happened this resulted in a weight difference between treatment groups that is maintained throughout the duration of the model. The DSU amended the manufacturers' revised model to ensure that, if a treatment switch occurred before the weight loss was fully regained, the starting weight at the next line of treatment was set equal to the weight that would have been achieved after the weight regain for the previous treatment. This resulted in a convergence of weight profiles over time for treatments associated with weight loss.

3.59 The DSU applied a number of changes and assumptions to the manufacturers' revised model, in addition to the amendment described in section 3.58. These included:

  • for the dual therapy analyses, using clinical-effectiveness data from the revised 24‑week network meta-analyses for the comparisons of dapagliflozin with DPP‑4 inhibitors and thiazolidinediones and from study 4 for the comparison of dapagliflozin with a sulfonylurea

  • applying an HbA1c threshold of 7.5% for switching to second-line and third-line treatment in the dual therapy analysis and for switching to second-line treatment in the add-on to insulin analysis

  • for any treatments associated with weight loss, assuming weight regain during year 3 to the level expected in a patient who experiences a natural weight gain of 0.1 kg per year from the start of treatment

  • assuming no diabetic complications at the start of treatment

  • reducing the loss in QALYs associated with hypoglycaemia to –0.012 for a severe event and –0.004 for a symptomatic event

  • using utility values associated with weight change of ±0.0061 per unit of BMI

  • reducing the annual cost of pioglitazone to £69.09 based on the latest NHS drug tariff

  • using an annual cost of £483 for people not experiencing diabetic complications.

3.60 For the dual therapy analyses, using data from the 24‑week network meta-analysis, the DSU base-case deterministic pair-wise analysis resulted in ICERs of £13,338 per QALY gained for the comparison of dapagliflozin with thiazolidinediones and £13,947 per QALY gained for the comparison of DPP‑4 inhibitors with thiazolidinediones. An incremental analysis resulted in ICERs of £13,338 per QALY gained for the comparison of dapagliflozin with thiazolidinediones and £16,847 per QALY gained for the comparison of DPP‑4 inhibitors with dapagliflozin (based on incremental costs of £136 and incremental QALYs of 0.008). Using data from study 4, the pair-wise comparison of dapagliflozin and sulfonylureas resulted in an ICER of £12,405 per QALY gained.

3.61 The DSU also conducted a probabilistic sensitivity analysis based on a mean of 1000 samples. Using data from the 24‑week network meta-analysis, the analysis resulted in pair-wise ICERs of £15,257 per QALY gained for the comparison of dapagliflozin with thiazolidinediones and £15,511 per QALY gained for the comparison of DPP‑4 inhibitors with thiazolidinediones. An incremental analysis resulted in ICERs of £15,257 per QALY gained for the comparison of dapagliflozin with thiazolidinediones and £41,654 per QALY gained for the comparison of DPP‑4 inhibitors with dapagliflozin (based on incremental costs of £17 and incremental QALYs of less than 0.001). Using data from study 4, the comparison of dapagliflozin and sulfonylureas resulted in an ICER of £15,148 per QALY gained. The DSU noted that in the probabilistic sensitivity analysis, people spent longer on first-line treatment because of the interaction between baseline HbA1c values, treatment switching threshold and effectiveness data, thus resulting in higher incremental costs and ICERs than the deterministic analysis. The results of these probabilistic sensitivity analyses also showed that, at £20,000 per QALY gained, dapagliflozin had the highest probability (40.4%) of being cost effective compared with DPP‑4 inhibitors (35.5%) and thiazolidinediones (24.1%) and also the highest probability (61.0%) of being cost effective compared with sulfonylureas.

3.62 The DSU conducted a scenario analysis that applied the manufacturers' original utility values associated with hypoglycaemia (–0.047 for a severe event and –0.042 for a symptomatic event). As a result of this change, dapagliflozin was extendedly dominated by DPP‑4 inhibitors and thiazolidinediones, because the ICER of dapagliflozin compared with thiazolidinediones was higher than that of the next most effective alternative (DPP‑4 inhibitors). The comparison of dapagliflozin and sulfonylureas resulted in an ICER of £10,317 per QALY gained. The DSU also conducted a scenario analysis which used the same clinical-effectiveness data from the 52‑week network meta-analysis as those used in the manufacturers' revised model, thus allowing all treatments to be compared with each other in a single analysis. On the basis of a full incremental analysis, DPP‑4 inhibitors were dominated by thiazolidinediones. The comparison of thiazolidinediones and sulfonylureas resulted in an ICER of £12,108 per QALY gained and the comparison of dapagliflozin and thiazolidinediones resulted in an ICER of £94,466 per QALY gained.

3.63 The DSU conducted an additional scenario analysis to explore the impact of weight convergence between treatment groups at the time of switching to the last line of treatment. In the manufacturers' revised model for the dual therapy analyses, the DSU modelled weight convergence between dapagliflozin (associated with weight loss) and a sulfonylurea (associated with weight gain) by increasing the weight gain for the last treatment in the sequence (insulin treatment). For this scenario analysis the DSU presented pair-wise ICERs using the data from the 24‑week network meta-analysis and separately the data from study 4. Applying the 24‑week meta-analysis data resulted in a higher ICER of £60,965 per QALY gained for the pair-wise comparison of dapagliflozin with thiazolidinediones and an ICER of £16,847 per QALY gained for the comparison of DPP‑4 inhibitors with dapagliflozin. The ERG noted that the latter ICER was largely unchanged from its base-case analysis because the weight profiles at last treatment switch were very similar across the 2 treatment groups. The pair-wise comparison of dapagliflozin and sulfonylureas using study 4 data resulted in an ICER of £21,200 per QALY gained.

3.64 The DSU noted that in the manufacturers' revised add-on to insulin analysis, the time to weight regain was set to occur before first treatment switch based on an HbA1c threshold of 9.04%, resulting in a switch to second-line treatment at 8 years. The DSU explored the impact of setting a time to weight regain of 1 year and an HbA1c switching threshold of 7.5% in line with the dual therapy analyses. The DSU also applied all other changes as described in section 3.59. The DSU base-case deterministic pair-wise analysis of dapagliflozin compared with DPP‑4 inhibitors resulted in an ICER of £3706 per QALY gained. The probabilistic sensitivity analysis resulted in a longer duration of first-line treatment and incremental costs for dapagliflozin, and consequently in a higher ICER of £7402 per QALY gained. When the DSU applied the manufacturers' original utility values associated with hypoglycaemia, the ICER was reduced to £2959 per QALY gained. When the DSU applied the assumption of weight convergence at last treatment switch, the ICER increased to £12,879 per QALY gained. The DSU noted that this scenario resulted in longer first-line treatment duration for people before switching to insulin treatment in both treatment groups and consequently, higher incremental costs for dapagliflozin.

  • National Institute for Health and Care Excellence (NICE)