Sensitivity Analyses

A number of sensitivity analyses were conducted to identify key drivers of outcomes in the base case analysis and the robustness of this. The influence of time horizon on the model outcomes was assessed by running the analysis over 20 and 10 years. It should be noted that these analyses do not capture all late-stage, long-term outcomes, as some patients were still alive at the end of these simulations. The effect of discount rates on future costs and clinical outcomes were investigated through analyses in which they were set (symmetrically) to 0% and 6% per annum. The key drivers of clinical outcomes were assessed by abolishing the differences in individual clinical parameters between the IDegLira arm and the insulin glargine U100 arm in turn. An additional analysis with only the statistically significant differences between IDegLira and insulin glargine U100 was conducted (Table 2).

Two alternative approaches to HbA1c progression were explored (see electronic supplementary material). In the first, no HbA1c changes were applied following the treatment effects applied in the first year of the analysis. This attempts to capture the legacy effect, where an early improvement in HbA1c has a benefit in the later years of life, even if the HbA1c difference no longer persists. In the second, the United Kingdom Prospective Diabetes Study (UKPDS) HbA1c progression equation was applied in both arms of the simulation. HbA1c increases over time in both arms of the analysis, with the HbA1c benefit in the IDegLira arm gradually reduced. Analyses were run with the upper and lower 95% confidence interval of the HbA1c change seen in the IDegLira arm of DUAL V applied, with all other parameters in the IDegLira and insulin glargine U100 arms remaining unchanged. The base case analyses assumed that the BMI difference between the treatment arms was abolished on treatment switching, and an alternative to this was explored in a sensitivity analysis with the difference maintained for the duration of the analysis. The influence of treatment switching was assessed in analyses with treatment switching brought forward to 3 years in both arms, pushed back to 7 years in both arms, and no treatment switching.

The effect of over- or underestimating the direct cost of treating diabetes-related complications was investigated in two scenarios. In the first, the cost of treating complications was increased by 10%, and in the second the cost was reduced by 10%. The impact of applying alternative disutilities for severe and non-severe hypoglycemic events was assessed by using the values published by Currie et al. (−0.0118 per severe hypoglycemic event and −0.0035 per non-severe hypoglycemic event) [²⁰]. A scenario was investigated in which 28% of patients in the insulin glargine U100 arm were assumed to require twice-daily insulin, incurring the cost of a further needle for subcutaneous injection, based on a 5-year parallel group study of insulin glargine U100 versus neutral protamine Hagedorn (NPH) insulin [²¹]. To investigate the impact of consumables on cost-effectiveness outcomes, a scenario was evaluated with costs of needles and SMBG testing excluded. The effect of the cost of basal insulin was investigated in an analysis with the cost of insulin glargine U100 replaced with the cost of NPH insulin.

In February 2014, an update to the IMS CORE Diabetes Model incorporating data from the UKPDS 82 was released, and an analysis using this version of the model has been conducted. While a validation study of the revised model has been published, the model proprietors suggest that the update is used in a sensitivity analysis, with the previous version being used in the base case [²²]. Similarly, version 9.0 of the IMS CORE Diabetes Model was released in summer 2015, and this was used in a sensitivity analysis. To date, no validation studies or user guide for the updated version of the model have been released and therefore it was not considered appropriate to use version 9.0 of the model for the base case analysis.

Probabilistic sensitivity analysis (PSA) was performed using the predefined function in the IMS CORE Diabetes Model. Cohort characteristics, treatment effects, complication costs, and utilities were sampled from distributions and the simulation was run using a second-order Monte Carlo approach. Cohorts of 1000 patients were run through the model 1000 times for the PSA, as results were not subject to random statistical variation with these settings.