The Appraisal Committee considered evidence submitted by the manufacturer of canagliflozin and a review of this submission by the Evidence Review Group (ERG).
The manufacturer's literature search identified 85 citations for use in the systematic literature review. It identified 11 clinical trials that evaluated canagliflozin in combination therapy for treating type 2 diabetes. The manufacturer indicated that 5 of these trials would not provide useful information for approaching the decision problem. No non-randomised clinical trials were included in the manufacturer's submission.
Of the 6 randomised controlled trials in the manufacturer's submission, 2 evaluated canagliflozin as part of dual therapy in combination with metformin (DIA3006 and DIA3009), 3 as part of triple therapy in combination with metformin and either a sulfonylurea or pioglitazone (DIA3012, DIA3002, DIA3015) and 1 with insulin with or without other antidiabetic treatments (DIA3008 insulin sub-study). All studies except DIA3015 evaluated 2 doses of canagliflozin (100 mg and 300 mg). DIA3015 used the higher dose only.
DIA3009 compared canagliflozin 100 mg (n=483) and canagliflozin 300 mg (n=485) with glimepiride (n=484) as part of dual therapy in combination with metformin for 104 weeks (52‑week core double blind and 52‑week extension double blind).
DIA3006 compared canagliflozin 100 mg (n=368) and canagliflozin 300 mg (n=367) with sitagliptin 100 mg (n=366) and with placebo/sitagliptin (n=183) as part of dual therapy in combination with metformin for 52 weeks (26‑week core double blind and 26‑week extension double blind).
DIA3015 compared canagliflozin 300 mg (n=378) with sitagliptin 100 mg (n=378) as part of triple therapy in combination with metformin and a sulfonylurea for 52 weeks.
DIA3002 compared canagliflozin 100 mg (n=157) and canagliflozin 300 mg (n=156) with placebo (n=156) as part of triple therapy in combination with metformin and a sulfonylurea for 52 weeks (26‑week core double blind and 26‑week extension double blind).
DIA3012 compared canagliflozin 100 mg (n=115) and canagliflozin 300 mg (n=114) with placebo (n=115; crossover to sitagliptin at 26 weeks) as part of triple therapy in combination with metformin and pioglitazone for 52 weeks (26‑week core double blind and 26‑week extension double blind).
The DIA3008 insulin sub-study compared canagliflozin 100 mg (n=566) and canagliflozin 300 mg (n=587) with placebo (n=565) as an add-on treatment to insulin with or without other antidiabetic drugs for 18 weeks. It was part of the ongoing CANVAS safety study in 4,330 patients with high risk or history of cardiovascular disease, which will report in 2017.
Patients were eligible for these trials if they had type 2 diabetes and inadequate glycaemic control on existing treatment (1 or 2 background therapies for all studies except the DIA3008 insulin sub-study, for which the maximum number of add‑on treatments was not stated). Inadequate glycaemic control was defined as an HbA1c level of 7.0% to 10.5%, except for DIA3009 in which the range was 7.0% to 9.5%. All patients enrolled into CANVAS had a history of, or were at high risk of, cardiovascular disease; those enrolled in the DIA3008 insulin sub‑study were on insulin alone or in combination with standard of care. The trials enrolled 12% to 28% patients in Europe, of whom 27 were in the UK.
The primary outcome for all studies was change in HbA1c from baseline to the end of the double-blind treatment period. Secondary outcomes included change in body weight, change in systolic blood pressure, and incidence of hypoglycaemia. Results for all trials reported in the manufacturer's submission were for the modified intention-to-treat populations (defined as randomised patients who received at least 1 dose of study drug using a last observation carried forward approach).
In DIA3009, mean change in HbA1c (minus glimepiride) at week 52 was -0.01% (95% confidence interval [CI] -0.109 to 0.085) for canagliflozin 100 mg plus metformin compared with glimepiride and metformin. At week 52, canagliflozin 300 mg and metformin produced a statistically superior mean reduction in HbA1c compared with glimepiride plus metformin, with a mean change (minus glimepiride) of -0.12% (95% CI -0.217 to -0.023, p<0.001). In DIA3006, mean change in HbA1c (minus placebo) was -0.627% (95% CI -0.758 to -0.481) for canagliflozin 100 mg plus metformin and -0.77% for canagliflozin 300 mg plus metformin (95% CI -0.914 to -0.636), compared with -0.66% (95% CI -0.795 to -0.516) for sitagliptin plus metformin (p<0.001 compared with placebo for the canagliflozin arms).
In DIA3009, there was a greater improvement in systolic blood pressure at 52 weeks with both doses of canagliflozin compared with glimepiride (mean difference in systolic blood pressure reduction [minus glimepiride] -3.5 mmHg [95% CI -4.9 to -2.1] with canagliflozin 100 mg and -4.8 mmHg [95% CI -6.2 to -3.40 with canagliflozin 300 mg). In DIA3006, canagliflozin 100 mg and 300 mg decreased systolic blood pressure from baseline at 26 weeks with a difference in mean systolic blood pressure (minus placebo) of -5.36 mmHg (95% CI -7.280 to -3.439) and -6.58 mmHg (95% CI -8.504 to -4.653) respectively compared with -3.54 mmHg (95% CI -5.273 to -1.413) in the sitagliptin arm (p<0.001 for both canagliflozin doses compared with placebo).
In DIA3009 at week 52, both doses of canagliflozin were associated with a statistically greater change in body weight compared with glimepiride. Weight loss was -4.2 kg (standard error 0.2) for canagliflozin 100 mg and -4.7 kg (standard error 0.2) for canagliflozin 300 mg compared with a slight increase in weight with glimepiride of +1.0 kg (standard error 0.2). In DIA3006 at week 26, a weight loss of -3.7 kg (standard error 0.2) was observed with canagliflozin 100 mg and -4.2 kg (standard error: 0.2) with canagliflozin 300 mg compared with -1.2 kg (standard error 0.2) with sitagliptin. Both doses of canagliflozin demonstrated statistical superiority to sitagliptin up to week 52 with a difference in mean weight reduction (minus sitagliptin) of -2.1 kg for canagliflozin 100 mg and -2.5 kg for canagliflozin 300 mg (p<0.001 for both doses).
In DIA3015 at week 52, canagliflozin 300 mg produced a statistically superior reduction in HbA1c compared with sitagliptin, with a difference in mean change in HbA1c for canagliflozin 300 mg (minus sitagliptin) of -0.37% (95% CI -0.50 to -0.25). DIA3002 and DIA3012 showed a statistically significant reduction in HbA1c at 26 weeks for both doses of canagliflozin compared with placebo. Mean reduction in HbA1c at week 26 with canagliflozin 100 mg (minus placebo) was -0.71% (95% CI -0.904 to -0.524, p<0.001) in DIA3002 and -0.62% (95% CI -0.811 to -0.437, p<0.001) in DIA3012. Mean reduction with canagliflozin 300 mg was -0.92% (95% CI -1.114 to -0.732, p<0.001) in DIA3012 and -0.76% (95% CI -0.951 to -0.575, p<0.001) in DIA3002.
Both doses of canagliflozin, when used in triple therapy, resulted in a statistically significant greater reduction in systolic blood pressure compared with sitagliptin or placebo in DIA3015 and DIA3012 but not DIA3002. In DIA3015, canagliflozin 300 mg statistically significantly decreased systolic blood pressure compared with sitagliptin with a difference in mean change (minus sitagliptin) in systolic blood pressure at week 52 of -5.9 mmHg (95% CI -7.642 to -4.175, p<0.001). Similarly, a statistically significant difference was seen in DIA3012 at 26 weeks, with a mean difference in systolic blood pressure (minus placebo) of -4.07 (95% CI -6.879 to -1.251, p=0.005) with canagliflozin 100 mg and -3.46 (95% CI -6.281 to -0.643, p=0.016) with canagliflozin 300 mg. No statistically significant difference in mean systolic blood pressure (minus placebo) was seen at 26 weeks in DIA3002 with canagliflozin 100 mg (-2.24 [95% CI -4.719 to 0.241, p=0.077]) or canagliflozin 300 mg (-1.62 [95% CI -4.111 to 0.866, p=0.201]).
Both doses of canagliflozin, when used in triple therapy, lowered body weight. In DIA3015 at week 52, canagliflozin 300 mg treatment resulted in statistically significant reductions in body weight relative to sitagliptin (-2.8 kg [95% CI -3.3 to -2], p<0.001). In DIA3002 at week 26, canagliflozin 100 mg and 300 mg (in combination with metformin and a sulfonylurea) resulted in a reduction in weight of -2.1 kg (standard error 0.3) and -2.6 kg (standard error 0.3) respectively compared with a weight change of -0.7 kg (standard error 0.3) in the placebo group. Changes in body weight were also evident at week 52 with a difference in mean weight change (minus placebo) of -1.0 kg (95% CI -1.8 to -1.2) for canagliflozin 100 mg and -2.1 kg (95% CI -2.9 to -1.2) for canagliflozin 300 mg. In DIA3012 at week 26, a change in weight of -2.7 kg (standard error 0.3) and -3.8 kg (standard error 0.3) was observed for canagliflozin 100 mg and 300 mg respectively (in combination with metformin and a thiazolidinedione).
At week 18 in the DIA3008 insulin sub-study, the difference in mean change in HbA1c compared with placebo was -0.65% (95% CI -0.73 to -0.56) in the canagliflozin 100 mg group and -0.73% (95% CI -0.82 to -0.65) in canagliflozin 300 mg arm (p<0.001 both comparisons). Both doses of canagliflozin resulted in a significant reduction in systolic blood pressure (p<0.001) compared with placebo. The difference in mean systolic blood pressure change (minus placebo) at week 18 was -2.58 mmHg (95% CI -4.060 to 1.091; p<0.001) with canagliflozin 100 mg and -4.38 mmHg (95% CI -5.850 to -2.903; p<0.001) with canagliflozin 300 mg. Change in body weight (minus placebo) at week 18 was -1.8 kg (95% CI -2.2 to -1.6) in the canagliflozin 100 mg group and -2.3 kg (95% CI -2.7 to -2.1) in the canagliflozin 300 mg group (p<0.001 for both comparisons).
The manufacturer conducted a systematic literature review to identify randomised controlled trials that evaluated treatments relevant to the NICE scope for this appraisal in patients with type 2 diabetes. The manufacturer's base-case network meta-analyses included 38 studies comparing treatments given in combination with metformin (metformin background), 10 studies comparing treatments given in combination with metformin and a sulfonylurea (metformin and sulfonylurea background), 2 studies comparing treatments given in combination with metformin and pioglitazone (metformin and pioglitazone background) and 14 studies comparing treatments given in combination with insulin (insulin background). The manufacturer used a Bayesian hierarchical model for the network meta-analyses. After assessing relative goodness of fit of fixed-effects and random-effects models using the deviance information criterion, the model associated with the lowest score was selected (with a difference of at least 3 points). All analyses were conducted by background therapy. The manufacturer conducted sensitivity analyses to determine the robustness of results. The manufacturer's submission reported outcomes that were relevant to its economic model including change in HbA1c, weight, BMI, systolic blood pressure and incidence of hypoglycaemia.
The manufacturer presented differences in HbA1c change at 52 weeks from its meta-analysis (complete with 95% credible intervals [95% CrI]) for canagliflozin 100 mg and 300 mg compared with the different comparators as dual therapy with a metformin background. Canagliflozin 100 mg produced a numerically greater reduction in HbA1c than sitagliptin 100 mg (-0.01%, 95% CrI -0.48 to 0.44) and dapagliflozin 10 mg (-0.14%, 95% CrI -0.81 to 0.47), but not liraglutide 1.2 mg (0.40%, 95% CrI -0.33 to 1.11), canagliflozin 300 mg (0.13%, 95% CrI -0.25 to 0.52), pioglitazone 30 mg (0.11%, 95% CrI -0.44 to 0.84), exenatide 10 micrograms (0.02%, 95% CrI -0.65 to 0.55) or glimepiride (0.00%, 95% CrI -0.45 to 0.46). Canagliflozin 300 mg was associated with a numerically greater reduction in HbA1c than pioglitazone 30 mg (-0.02%, 95% CrI -0.57 to 0.72), exenatide 10 micrograms (-0.11%, 95% CrI -0.78 to 0.42), glimepiride (-0.13%, 95% CrI -0.58 to 0.33), canagliflozin 100 mg (-0.13%, 95% CrI -0.52 to 0.25), sitagliptin 100 mg (-0.14%, 95% CrI -0.61 to 0.31) and dapagliflozin 10 mg (-0.27%, 95% CrI -0.94 to 0.34). However, it was not associated with a numerically greater reduction in HbA1c compared with liraglutide 1.2 mg (0.27%, 95% CrI -0.46 to 0.98).
When given as part of dual therapy with a metformin background, canagliflozin 100 mg and 300 mg were associated with greater weight reductions at 52 weeks compared with sitagliptin 100 mg (differences of -2.12 kg [95% CrI -2.66 to -1.57] and -2.48 kg [95% CrI 3.03 to -1.93] respectively), glimepiride (differences of -3.97 kg [95% CrI -5.54 to -2.48] and -4.33 kg [95% CrI -5.89 to -2.85] respectively) and pioglitazone 30 mg (differences of -4.57 kg [95% CrI -6.28 to -2.93] and -4.93 kg [95% CrI -6.64 to -3.29] respectively). The reduction in weight was at least similar for canagliflozin 100 mg and 300 mg compared with dapagliflozin 10 mg (differences of -0.11 kg [95% CrI -1.10 to 0.89] and -0.48 kg [95% CrI -1.47 to 0.53] respectively). When comparing canagliflozin with the glucagon-like peptide-1 (GLP‑1) analogues, canagliflozin 100 mg and 300 mg gave a lesser weight reduction than exenatide 10 micrograms (differences of 1.47 kg [95% CrI -1.48 to 4.41] and 1.11 kg [95% CrI -1.84 to 4.05] respectively) but a greater weight reduction than liraglutide 1.2 mg (-0.49 [95% CrI -1.37 to 0.38] and -0.85 kg [-1.73 to 0.02] respectively).
The manufacturer reported that canagliflozin 100 mg and 300 mg were associated with greater reductions in systolic blood pressure compared with glimepiride (differences of -3.52 mmHg [95% CrI -5.02 to -2.05] and -4.73 mmHg [95% CrI -6.22 to -3.26] respectively), liraglutide 1.2 mg (differences of -3.50 mmHg [95% CrI -6.55 to -0.43] and -4.71 mmHg [95% CrI -7.73 to -1.66] respectively) and sitagliptin 100 mg (differences of -2.84 mmHg [95% CrI -4.44 to -1.22] and -4.04 mmHg [95% CrI -5.64 to -2.44] respectively), and smaller reductions in systolic blood pressure than pioglitazone 30 mg (differences of 2.05 mmHg [95% CrI -3.22 to 7.37] and 0.83 mmHg [95% CrI -4.44 to 6.15] respectively).
The manufacturer reported that canagliflozin 100 mg and 300 mg were associated with a lower risk of hypoglycaemia compared with glimepiride (odds ratios 0.11 [95% CrI 0.07 to 0.16] and 0.10 [95% CrI 0.07 to 0.15] respectively). It described a higher risk of hypoglycaemia for canagliflozin 100 mg and 300 mg compared with dapagliflozin 10 mg (odds ratios 3.65 [95% CrI 1.44 to 9.93] and 3.43 [95% CrI 1.34 to 9.23] respectively) and sitagliptin (odds ratios 1.77 [95% CrI 0.97 to 3.44) and 1.66 [95% CrI 0.90 to 3.23] respectively). The manufacturer advised that the difference in risk of hypoglycaemia compared with dapagliflozin in this network meta-analysis was likely to be because of differences in how hypoglycaemic events were reported in the clinical trials, but did not provide any further explanation for this.
Comparisons for canagliflozin and comparators with a metformin and sulfonylurea background were provided by the manufacturer. Compared with sitagliptin 100 mg, canagliflozin 100 mg produced a similar reduction in HbA1c (0.07%, 95% CrI -1.48 to 1.64), whereas canagliflozin 300 mg was associated with a slightly greater HbA1c reduction (-0.17%, 95% CrI -1.30 to 0.97). Canagliflozin 100 mg gave a similar HbA1c reduction (difference of 0.03%, 95% CrI -1.78 to 1.79) compared with exenatide 10 micrograms, and canagliflozin 300 mg gave a higher HbA1c reduction (difference of -0.21%, 95% CrI -1.87 to 1.42). Canagliflozin 100 mg and 300 mg were associated with greater weight reductions than sitagliptin 100 mg (differences of -2.03 kg [95% CrI -7.78 to 3.76] and -2.64 kg [95% CrI -6.83 to 1.56] respectively) and similar weight reductions to exenatide 10 micrograms (differences of 0.47 kg [95% CrI -6.09 to 7.24] and -0.14 kg [95% CrI -6.15 to 6.08] respectively). Canagliflozin 100 mg and 300 mg were associated with a higher reduction in systolic blood pressure than sitagliptin 100 mg (differences of -5.76 mmHg [95% CrI -9.02 to -2.53] and -5.16 mmHg [95% CrI -6.94 to -3.38] respectively). Canagliflozin 100 mg and 300 mg were associated with a similar risk of hypoglycaemia as sitagliptin 100 mg (odds ratios of 0.75 [95% CrI 0.43 to 1.29] and 0.96 [95% CrI 0.72 to 1.29]) respectively) and exenatide 10 micrograms (odds ratios of 0.96 [95% CrI 0.46 to 2.00] and 1.23 [95% CrI 0.62 to 2.48] respectively).
The manufacturer's submission presented comparisons for canagliflozin and comparators with a metformin and thiazolidinedione background. Compared with sitagliptin 100 mg, canagliflozin 100 mg was associated with a smaller reduction in HbA1c at 26 weeks (difference of 0.07%, 95% CrI -0.19 to 0.33) and canagliflozin 300 mg was associated with a greater reduction (difference of -0.07%, 95% CrI -0.33 to 0.19). Canagliflozin 100 mg and 300 mg were associated with weight reductions compared with sitagliptin 100 mg (differences of -2.70 kg [95% CrI -10.10 to 4.62] and -3.65 kg [-11.07 to 3.67] respectively). No statistical analysis was conducted for changes in systolic blood pressure because of limitations in the evidence base. Canagliflozin 100 mg was associated with a similar risk of hypoglycaemic events as sitagliptin 100 mg (odds ratio of 0.86, 95% CrI 0.10 to 7.09), whereas canagliflozin 300 mg was associated with a higher risk (odds ratio of 1.89, 95% CrI 0.30 to 13.72).
When given as an add-on treatment to insulin, the reduction in HbA1c at 26 weeks with canagliflozin 100 mg and 300 mg was greater than with sitagliptin 100 mg (differences of -0.01% [95% CrI -0.17 to 0.16] and -0.15% [95% CrI -0.31 to 0.02] respectively) and dapagliflozin 10 mg (differences of -0.01% [95% CrI -0.17 to 0.15] and -0.15% [95% CrI -0.31 to 0.01] respectively). The higher dose of canagliflozin gave a greater HbA1c reduction than exenatide 10 micrograms (difference of -0.05%, 95% CrI -0.31 to 0.21) but the lower dose did not (difference of 0.09%, 95% CrI -0.17 to 0.35). A similar pattern was observed for the comparison of canagliflozin and pioglitazone 30 mg (difference of -0.10% [95% CrI -0.37 to 0.17] for canagliflozin 300 mg and 0.04% [95% CrI -0.23 to 0.30] for canagliflozin 100 mg). Pioglitazone 45 mg produced a greater HbA1c reduction than either canagliflozin 100 mg or 300 mg.
When given as an add-on treatment to insulin, canagliflozin 100 mg and 300 mg were associated with greater weight reduction than sitagliptin 100 mg (differences of -2.10 kg [95% CrI -2.67 to -1.53] and -2.67 kg [95% CrI -3.24 to -2.11] respectively), pioglitazone 30 mg (differences of -3.28 kg [95% CrI -11.72 to 5.14] and -3.85 kg [95% CrI -12.31 to 4.57] respectively) and dapagliflozin 10 mg (differences of -0.41 kg [95% CI -1.01 to 0.20] and -0.98 kg [95% CrI -1.59 to -0.38] respectively) but not exenatide 10 micrograms (differences of 0.64 kg [95% CrI -0.44 to 1.71] and 0.07 kg [95% CrI -1.01 to 1.14] respectively). Canagliflozin 100 mg and 300 mg were associated with higher reductions in systolic blood pressure compared with dapagliflozin 10 mg (differences of -0.92 mmHg [95% CrI -3.93 to 2.08] and -2.62 mmHg [95% CrI -5.64 to 0.40] respectively). When comparing canagliflozin and exenatide 10 micrograms, canagliflozin 300 mg was associated with a lower reduction in systolic blood pressure (difference of -1.33 mmHg, 95% CrI -5.19 to 2.52) but canagliflozin 100 mg was not (0.38 mmHg, 95% CrI -3.46 to 4.23). No analysis of BMI results or all hypoglycaemic events was conducted because of lack of data at 26 weeks for the canagliflozin trial.
The canagliflozin clinical trials included in the submission evaluated the safety of canagliflozin in 10,285 people with type 2 diabetes. The manufacturer provided data from 3 pre-specified pooled safety datasets, focusing on a broad dataset that included data for canagliflozin 100 mg (n=3,092), canagliflozin 300 mg (n=3,085) and all non-canagliflozin arms (placebo, glimepiride and sitagliptin [n=3,262]).
In the broad dataset, the incidence of any adverse event was similar across groups (76.6% for canagliflozin 100 mg, 77.0% for canagliflozin 300 mg and 75.8% in the non-canagliflozin group). The incidence of adverse events leading to discontinuation in the broad dataset was higher in the canagliflozin 300 mg group (7.3%) than the canagliflozin 100 mg (5.6%) and non-canagliflozin (5.0%) groups. Adverse events that led to discontinuation of more than 0.2% of patients in the canagliflozin 300 mg group were decreased glomerular filtration rate, renal impairment and increased blood creatinine. The incidence of adverse events considered related to the study drug by the investigator was slightly higher in the canagliflozin 100 mg and 300 mg groups (33.6% and 29.4% respectively) than the non-canagliflozin group (21.8%). The incidence of serious adverse events and deaths was similar in the canagliflozin and non-canagliflozin groups. Adverse events occurring in at least 5% of subjects in a canagliflozin group were nasopharyngitis, hypoglycaemia, upper respiratory tract infection, urinary tract infection, diarrhoea, arthralgia, back pain and headache.
The manufacturer noted that sodium–glucose cotransporter‑2 (SGLT‑2) inhibitors are associated with a higher incidence of genital mycotic infections because of urinary glucose excretion. In the broad dataset, the incidence of genital mycotic infection adverse events in women was higher in those receiving canagliflozin 100 mg (14.7%) and canagliflozin 300 mg (13.9%) than in those taking placebo (3.1%). In men, the incidence was 7.3% in the canagliflozin 100 mg group and 9.3% in the canagliflozin 300 mg groups, compared with 1.6% of men in the non‑canagliflozin group.
The ERG considered that only the trials with active comparator arms were relevant to the NICE scope (DIA3009 and DIA3006 [dual therapy] and DIA3015 [triple therapy]) and judged all 3 to be of generally good methodological quality. The ERG considered the manufacturer's submission to provide a generally unbiased estimate of canagliflozin's treatment effect within the stated scope of the decision problem, and that the manufacturer's interpretation of the evidence was largely appropriate and justified.
However, the ERG identified some exceptions and uncertainties:
A lack of direct evidence for comparisons with treatments other than sitagliptin and glimepiride.
No discussion of the implications of differences in results between the modified intention-to-treat analyses (reported in the manufacturer's submission) and per protocol analyses (reported in the trial journal publications and clinical study reports).
A low volume of direct evidence for some of the loops in the network meta‑analyses, which led the ERG to conclude that the results should be interpreted with caution.
It was not convinced by the manufacturer's justification for network meta-analysis for triple therapy assessing effects at 26 weeks, rather than at 52 weeks.
The exclusion of background treatments that did not include metformin (because including these would better reflect all the potential uses of canagliflozin within its European marketing authorisation).
Lack of evidence for the longer-term efficacy and safety of canagliflozin (because most outcomes were measured up to 52 weeks in the clinical trials).
It did not have confidence in the manufacturer's favourable interpretation of adverse events related to canagliflozin compared with other treatments, and the discontinuation rates compared with sitagliptin and glimepiride.
The manufacturer's submission included de novo economic analyses of canagliflozin in combination therapy for type 2 diabetes using the following regimens:
dual therapy in combination with metformin
triple therapy in combination with metformin and a sulfonylurea
triple therapy in combination with metformin and a thiazolidinedione
add-on treatment to insulin (with or without other antihyperglycaemic agents).
The manufacturer advised that these populations reflected the licensed indications for canagliflozin combination treatment and patients in the clinical trials, and it considered these patients to be representative of the population most likely to receive canagliflozin in clinical practice in England. The 100 mg and 300 mg doses of canagliflozin were considered separately in the base-case analyses (increasing the dose from 100 mg to 300 mg according to the summary of product characteristics (see section 2.2) was explored in a scenario analysis). Comparator treatments used at the different points in the treatment pathway were sulfonylureas, thiazolidinediones, dipeptidyl peptidase-4 (DPP‑4) inhibitors, GLP‑1 analogues, dapagliflozin and insulin. The manufacturer selected a comparator to represent each treatment class in its economic model (based on frequency of use in clinical practice, comparators in the canagliflozin clinical trials and available data).
The manufacturer's systematic literature review of potentially relevant published cost-effectiveness evidence identified 52 analyses, and a similar systematic literature review found 21 economic evaluation models of type 2 diabetes. The manufacturer chose the ECHO‑T2DM model, which was a stochastic micro‑simulation model in which cohorts of individual hypothetical patients were created and simulated over time using Monte Carlo (first-order uncertainty) techniques. Second-order (parameter) uncertainty was captured by hypothetical cohorts of 1,000 patients defined by baseline characteristics sourced from the canagliflozin clinical trials including demographics (for example, age and sex), biomarker values (for example, HbA1c, systolic blood pressure and BMI, which are short-term outcomes used to predict the likelihood of longer-term events in the model) and disease indicators (for example, disease duration and history of complications). Individual patient outcomes were simulated over time through health states capturing micro- and macrovascular complications and death. There were 1,000 cohorts of 1,000 patients for each model run for the base-case comparisons, and patient cohorts were simulated until death or the end of the designated follow-up period. Results of key parameter values (including treatment effects, risk equation coefficients, and quality-adjusted life year [QALY] disutility weights) were then aggregated. The model used a lifetime time horizon (40 years), the cycle length was 1 year, health benefits and costs were each discounted at 3.5% and the analysis was from an NHS perspective.
The manufacturer's model included 3 parallel sets of microvascular complications (to reflect increasing severity of retinopathy, chronic kidney disease and neuropathy) and 4 types of macrovascular complications (ischaemic heart disease, myocardial infarction, stroke and congestive heart failure; because of interdependence with neuropathic outcomes, peripheral vascular disease was classified as a microvascular neuropathy complication). These were represented by Markov health states and were associated with costs, utility values and, in some cases, a possible treatment contraindication (for example, end-stage renal disease for pioglitazone) or excess risk of death (for example, myocardial infarction or stroke). Annual probabilities of experiencing worsening microvascular complications were derived primarily from the Wisconsin Epidemiologic Study of Diabetic Retinopathy, the Rochester Epidemiology Project, and the Centers for Disease Control model of chronic kidney disease. Risk equations from the original UK Prospective Diabetes Study (UKPDS) Outcomes Model were used to simulate macrovascular complications in the base case. The manufacturer mainly used published utility values derived from CODE‑2, which was a non-interventional, observational study. Utility decrements were applied to the baseline quality-of-life value (estimated at 1.027 using multivariate regression techniques) for patient characteristics (for example, age and duration of disease), microvascular and macrovascular complications, hypoglycaemic events, obesity and adverse events.
The manufacturer explained that diabetes treatments have an initial effect followed by annual 'drift', where the effect lessens over time. Biomarker values after the first cycle were estimated using annual drift values for each treatment. Based on published values, HbA1c drift was assumed to be 0.14% for canagliflozin, dapagliflozin, DPP‑4 inhibitors and GLP‑1 analogues, 0.07% for thiazolidinediones, 0.24% for sulfonylureas and 0.15% for insulin. The drift values for systolic blood pressure, lipids, weight and eGFR were assumed to be the same for all treatments. The manufacturer's model allowed patients to take anti‑dyslipidaemia and anti-hypertension medications, and applied rescue treatments when biomarker thresholds for dyslipidaemia and hypertension were exceeded.
The manufacturer presented pairwise comparisons using cohorts of simulated patients from the probability distribution of patient characteristics for canagliflozin 100 mg and 300 mg in dual therapy, triple therapy and as an add-on to insulin. At the clarification stage, the manufacturer provided some updated cost‑effectiveness results after identifying that, where BMI was not available in the network meta-analysis, the change in actual BMI had been used instead of being calculated using a weighted average of height in the UK general population. The updated incremental cost-effectiveness ratios (ICERs) were generally more favourable for canagliflozin than the ones they replaced.
Canagliflozin 100 mg as part of a dual therapy regimen showed QALY gains and increased costs compared with a sulfonylurea (ICER of £1,537 per QALY gained [incremental costs £288; incremental QALYs 0.188]), a DPP‑4 inhibitor (ICER of £97 per QALY gained [incremental costs £1; incremental QALYs 0.013]), and dapagliflozin (£8,674 per QALY gained [incremental costs £63; incremental QALYs 0.007]). The ICER provided by the manufacturer at the clarification stage for the comparison with dapagliflozin was £2,993 per QALY gained (incremental costs £33; incremental QALYs 0.011). Compared with a GLP‑1 analogue, canagliflozin 100 mg was less effective and less costly (incremental costs -£2,424; incremental QALYs -0.048). Canagliflozin 100 mg continued to be less effective and less costly than a GLP‑1 analogue after correcting the BMI data (incremental costs -£2,414; incremental QALYs -0.034). Canagliflozin 100 mg was dominated by pioglitazone (that is, canagliflozin was more costly and less effective [incremental costs £2,833; incremental QALYs -0.159]).
Canagliflozin 300 mg as part of a dual therapy regimen, when compared with a sulfonylurea, produced higher costs and greater QALY gains (ICER of £4,899 per QALY gained [incremental costs £976; incremental QALYs 0.199]), a DPP‑4 inhibitor (ICER of £18,349 per QALY gained [incremental costs £576; incremental QALYs 0.031]), and dapagliflozin (ICER of £27,419 per QALY gained [incremental costs £625; incremental QALYs 0.023]. The ICER provided at the clarification stage for the comparison with dapagliflozin was £21,626 per QALY gained (incremental costs £616; incremental QALYs 0.029). Similar to the 100 mg dose, canagliflozin 300 mg was less costly and less effective than a GLP‑1 analogue, giving an ICER of £76,214 per QALY gained for a GLP‑1 analogue to replace canagliflozin 300 mg (incremental costs -£1,892; incremental QALYs -0.025). Canagliflozin 300 mg continued to be less effective and less costly than a GLP‑1 analogue after correcting the BMI data (incremental costs -£1,879; incremental QALYs -0.018). Like the 100 mg dose, canagliflozin 300 mg was also dominated by pioglitazone (incremental costs £3,353; incremental QALYs -0.141).
The manufacturer also presented cost-effectiveness results for triple therapy regimens containing canagliflozin 100 mg and 300 mg. In a triple therapy combination with metformin and a sulfonylurea, canagliflozin 100 mg dominated a DPP‑4 inhibitor (incremental costs -£42; incremental QALYs 0.016), and also dominated a GLP‑1 analogue (incremental costs -£1,297; incremental QALYs 0.001). Adjusting the change in BMI at the clarification stage caused canagliflozin to become less costly and less effective than a GLP‑1 analogue, giving an updated ICER of £265,928 per QALY gained for a GLP‑1 analogue to replace canagliflozin 100 mg. Canagliflozin 100 mg was associated with greater costs and QALYs than long-acting insulin, with an ICER of £263 per QALY gained (incremental costs £135; incremental QALYs 0.514). An updated ICER provided at the clarification stage was £183 per QALY gained. In a triple therapy combination with metformin and a thiazolidinedione, canagliflozin 100 mg had higher costs and QALY gains than a DPP‑4 inhibitor, producing an ICER of £1,095 per QALY gained (incremental costs £7; incremental QALYs 0.007).
In combination with metformin and a sulfonylurea, canagliflozin 300 mg compared with a DPP‑4 inhibitor had an ICER of £13,287 per QALY gained (incremental costs £461; incremental QALYs 0.035) and dominated a GLP‑1 analogue (incremental costs -£685; incremental QALYs 0.004). Canagliflozin 300 mg continued to dominate a GLP‑1 analogue after updating the BMI data at the clarification stage. In combination with metformin and a thiazolidinedione, the ICER for canagliflozin 300 mg compared with a DPP‑4 inhibitor was £21,430 per QALY gained (incremental costs £691; incremental QALYs 0.032). Canagliflozin 300 mg compared with long-acting insulin gave an ICER of £607 per QALY gained (incremental costs £379; incremental QALYs 0.624). After updating the BMI data at the clarification stage, the ICER increased slightly to £671 per QALY gained.
Canagliflozin 100 mg as an add-on treatment to insulin dominated dapagliflozin (incremental costs -£72; incremental QALYs 0.003). Canagliflozin 100 mg was associated with lower costs and lower QALY gains compared with a DPP‑4 inhibitor (incremental costs -£13; incremental QALYs -0.010), with an ICER of £1,340 per QALY gained for canagliflozin 100 mg to be replaced by a DPP-4 inhibitor. Compared with a GLP‑1 analogue, canagliflozin 100 mg was also associated with lower costs and lower QALY gains (incremental costs -£836; incremental QALYs -0.065) with an ICER of £12,915 per QALY gained for canagliflozin 100 mg to be replaced by a GLP‑1 analogue.
Canagliflozin 300 mg as an add-on treatment to insulin produced greater costs and QALY gains compared with a DPP‑4 inhibitor (ICER of £7,975 per QALY gained [incremental costs £322; incremental QALYs 0.040]), and dapagliflozin (ICER of £5,992 per QALY gained [incremental costs £327; incremental QALYs 0.055]). Compared with a GLP‑1 analogue, canagliflozin 300 mg was associated with lower costs and lower QALY gains (incremental costs -£526; incremental QALYs -0.015), with an ICER of £35,575 per QALY gained for canagliflozin to be replaced by a GLP‑1 analogue.
The manufacturer conducted deterministic and probabilistic sensitivity analyses for canagliflozin compared with a DPP‑4 inhibitor (in dual therapy, triple therapy and as an add-on to insulin therapy) and with a sulfonylurea in dual therapy, because it perceived these to be the key comparators for canagliflozin. The manufacturer did not report results for other comparators used in its base case, including dapagliflozin.
The manufacturer's deterministic sensitivity analyses showed that the results from the model were most sensitive to varying the metabolic drift of HbA1c associated with canagliflozin and its comparators. The manufacturer commented that, except for HbA1c drift, the ICERs for the comparisons of both doses of canagliflozin with a sulfonylurea in dual therapy were largely insensitive to parameter changes. However, the manufacturer noted the ICERs for canagliflozin compared with a DPP‑4 inhibitor were generally less stable.
The manufacturer presented results of probabilistic sensitivity analyses for canagliflozin compared with some of the comparators used in its base case and commented that there was greater uncertainty where the incremental changes in costs and QALYs were small. It noted that, at a maximum acceptable ICER of £20,000 per QALY gained, the probability of canagliflozin being cost effective compared with a DPP‑4 inhibitor for the different comparisons was 45% to 56% for the 100 mg dose and 52% to 61% for the 300 mg dose. The manufacturer stated that there was less uncertainty in the dual therapy comparisons of canagliflozin (100 mg and 300 mg) with a sulfonylurea, with the probability of cost effectiveness being around 90% for both doses (at maximum acceptable ICERs of £20,000 or £30,000 per QALY gained).
The manufacturer undertook scenario analyses that explored dose escalation from the recommended starting dose of 100 mg to 300 mg (see section 2.2). Modelling techniques were used because this had not been studied in clinical trials. If patients in the simulated cohort had tolerated the 100 mg dose of canagliflozin but had not reached an HbA1c of less than 7.5% at 6 months, then the canagliflozin dose was increased to 300 mg in cycle 2. The manufacturer based this assumption on clinical specialist opinion. Switching to the higher dose was only permitted at the end of the first cycle: patients who tolerated treatment and had satisfactory glycaemic control continued the 100 mg dose and switched to standard rescue therapy when needed. It was assumed that patients who switched to the higher dose experienced the same treatment effects as patients treated with 300 mg for the full 12 months. The manufacturer stated that the dose escalation scenario improved the cost effectiveness of canagliflozin for 11 out of 12 comparisons made. Only the ICER comparing canagliflozin with a sulfonylurea in dual therapy increased (from £1,537 per QALY gained to £1,721 per QALY gained) and canagliflozin was dominant in 7 of the scenarios.
The ERG concluded that the methods and inputs in the manufacturer's economic evaluation were generally in line with the NICE reference case, but noted:
Not all comparators in the NICE scope had been included in the manufacturer's decision problem.
Comparators used were not always the most widely prescribed.
The NICE reference case states that data from head-to-head trials should be presented in the reference-case analysis if possible, but the manufacturer had instead sometimes used results from the meta-analyses. The ERG indicated that although the manufacturer had stated this was for consistency, this was not properly justified because the manufacturer had not provided a fully incremental cost-effectiveness analysis (that is, only pairwise comparisons had been generated).
It was not clear if the preference data for utility values wholly represented the population of England because they were derived from a European study (CODE‑2).
The ERG considered the model to be internally consistent and well validated.
Overall, the ERG considered the clinical-effectiveness data in the model to be appropriate (although sometimes network meta-analysis estimates were used instead of head-to-head trial data). It found the equations used for the extrapolation of biomarker outcomes to be well-established and appropriate for the patient population in England, and the drift assumptions for the biomarkers to be reasonable. However, the ERG was concerned that using only 1,000 patients per cohort in the base case might not robustly capture the ICERs.
The ERG commented that health benefits had been measured and evaluated in line with the NICE reference case and that utility values had been appropriately incorporated into the manufacturer's model. The ERG generally agreed with the costs used by the manufacturer for drugs, adverse events and health states (complications and comorbidities), although there was some uncertainty in the costs used for insulin.
Although the ERG agreed with the manufacturer's conclusions from the results of the deterministic analyses as presented, the ERG considered that the uncertainty in the decision problem had not been fully explored, and that the manufacturer did not present the results in the most informative way. The ERG was unclear why second-order uncertainty was not switched off in the deterministic sensitivity analyses (that is, by specifying only 1 patient cohort). Consequently, the ERG believed that the manufacturer's results were partly confounded by stochasticity in other parameters (that is, they included uncertainty from more than 1 source and did not truly reflect the uncertainty associated with varying 1 parameter).
The ERG reviewed the manufacturer's scenario analyses, focusing on the dose escalation scenario. It noted that, like the deterministic sensitivity analyses, second-order uncertainty was not switched off in the scenario analyses, meaning that the results were again partly confounded by parameter stochasticity. The ERG noted that the manufacturer's submission concluded from the dose escalation scenario analysis that the dose escalation schedule from 100 mg to 300 mg was cost effective, but the ERG did not consider that this conclusion applied to all treatment comparisons, noting that:
Canagliflozin 100 mg compared with a thiazolidinedione in dual therapy remained dominated in the scenario analyses as well as in the base case.
Canagliflozin 100 mg as an add-on to insulin was cheaper and less effective than a DPP‑4 inhibitor in the base case, but associated with an ICER of £503 per QALY gained in the scenario analysis.
Canagliflozin 100 mg as an add-on to insulin was cheaper and less effective than a GLP‑1 analogue in the base case and remained so in the scenario analysis.
The ERG concluded that although all relevant variables appeared to have been included in the manufacturer's probabilistic sensitivity analyses, it was unclear if uncertainty in the decision problem had been sufficiently explored because the distributional assumptions were not well described in the manufacturer's submission and were not transparent in the model. The ERG also commented that standard errors used for the HbA1c treatment effect parameters were too small and that cost-effectiveness acceptability data had not been presented for all base-case comparisons.
The ERG did exploratory work to:
examine the variation in base-case ICERs by re-running some of the manufacturer's analyses
examine the variation in final ICERs for dual therapy with no parameter uncertainty and 100,000 patients
produce incremental analyses for dual therapy
calculate the probability that canagliflozin is cost effective compared with dapagliflozin
determine the effect of varying the efficacy estimates for glimepiride and exenatide.
The ERG found there was minimal variation in the dual-therapy ICERs when it re‑ran the manufacturer's base-case analyses. However, it was unsure if the manufacturer's use of 1,000 cohorts of 1,000 patients estimated the ICERs robustly, and how the base-case results might change if parameter uncertainty was removed. Its exploratory analyses with a single cohort of 100,000 patients and no parameter uncertainty had a small impact on the different cost and QALY outcomes compared with the manufacturer's base-case analyses (typically hundreds of pounds or less, and thousandths of a QALY). The ERG noted, however, that the small differences in incremental QALYs could in some cases drive an apparently large variation in the ICER compared with the manufacturer's ICERs described in sections 3.32 and 3.33:
Canagliflozin 100 mg was dominated by a thiazolidinedione (incremental costs £2,929; incremental QALYs -0.166).
Canagliflozin 100 mg compared with a sulfonylurea had an ICER of £1,579 per QALY gained (incremental costs £274; incremental QALYs 0.174).
Canagliflozin 100 mg compared with dapagliflozin had an ICER of £100,719 per QALY gained (incremental costs £193; incremental QALYs 0.002).
Canagliflozin 100 mg compared with a DPP‑4 inhibitor had an ICER of £12,938 per QALY gained (incremental costs £211; incremental QALYs 0.016).
Canagliflozin 100 mg was less costly and less effective than a GLP‑1 analogue, with an ICER of £67,414 per QALY lost (incremental costs -£2,381; incremental QALYs -0.035).
Canagliflozin 300 mg was dominated by a thiazolidinedione (incremental costs £3,245; incremental QALYs -0.14).
Canagliflozin 300 mg compared with a sulfonylurea had an ICER of £5,368 per QALY gained (incremental costs £1,121; incremental QALYs 0.21).
Canagliflozin 300 mg compared with a DPP‑4 inhibitor had an ICER of £9,246 per QALY gained (incremental costs £412; incremental QALYs 0.04).
Canagliflozin 300 mg compared with dapagliflozin had an ICER of £17,161 per QALY gained (incremental costs £751; incremental QALYs 0.04).
Canagliflozin 300 mg was less costly and less effective than a GLP‑1 analogue, with an ICER of £86,412 per QALY lost (incremental costs -£1,867; incremental QALYs -0.0216).
The ERG did a fully incremental exploratory analysis for canagliflozin 100 mg in dual therapy in combination with metformin (excluding thiazolidinediones because their use in clinical practice in England is declining). In the ERG's exploratory analysis using the manufacturer's base-case inputs, canagliflozin 100 mg had an ICER of £507 per QALY gained (incremental costs £8; incremental QALYs 0.015) compared with the next best option, which was dapagliflozin. When using the ERG's preferred inputs (excluding parameter uncertainty and with 100,000 patients), canagliflozin 100 mg had an ICER of £84,800 per QALY gained (incremental costs £245; incremental QALYs 0.003) compared with the next best option, a DPP‑4 inhibitor. The ERG noted that dapagliflozin, canagliflozin 100 mg and a DPP‑4 inhibitor were associated with similar total costs and similar overall total QALYs.
The ERG did a fully incremental exploratory analysis (excluding thiazolidinediones) for canagliflozin 300 mg in dual therapy in combination with metformin. The ICER for canagliflozin 300 mg compared with the next best option, dapagliflozin, was £17,639 per QALY gained (incremental costs £587; incremental QALYs 0.033) for the manufacturer's base-case inputs of 1,000 cohorts of 1,000 patients, and £17,903 per QALY gained (incremental costs £585; incremental QALYs 0.033) compared with the next best option, a DPP‑4 inhibitor, using data from the ERG's preferred input of a single cohort of 100,000 patients.
The ERG noted that the manufacturer's submission did not report the probability of canagliflozin being cost effective compared with dapagliflozin. The ERG's exploratory analyses for dual therapy with metformin and an SGLT‑2 inhibitor showed that the probability of canagliflozin 100 mg being cost effective compared with dapagliflozin was 52.5% at maximum acceptable ICERs of £20,000 and £30,000 per QALY gained. For canagliflozin 300 mg, the probability of cost effectiveness was 46.7% at a maximum acceptable ICER of £20,000 per QALY gained and 50% at £30,000 per QALY gained.
The ERG was aware that the manufacturer had chosen a single comparator to be representative of each treatment class, but considered that there was variation in specific parameters between drugs within each treatment class that might not be fully captured by the manufacturer's deterministic sensitivity analyses. The ERG explored greater variation in hypoglycaemia event rate for sulfonylureas and agreed with the manufacturer that this parameter was not very influential. Similarly, the ERG found that using cost and efficacy data for liraglutide instead of exenatide 10 micrograms gave results that were consistent with the manufacturer's deterministic sensitivity analyses.
Full details of all the evidence are in the manufacturer's submission and the ERG report.