Objective

The objective of this appendix is to summarize and critically appraise the meta-analyses of randomized controlled trials (RCTs) comparing insulin degludec (IDeg) with insulin glargine (IGlar) that were used to inform the pharmacoeconomic analysis. Information on this meta-analysis was included in Appendices 1, 2, and 3 of the manufacturer’s pharmacoeconomic submission.¹²

Summary of Meta-Analyses

Three separate patient-level meta-analyses were reported: two examining hypoglycemia events and one analyzing total daily dose of insulin for IDeg versus IGlar. The analyses of hypoglycemia events (meta-analyses 1 and 2) used the same statistical approach. Five end points were analyzed in meta-analysis 1 and three end points were analyzed in meta-analysis 2 (Table 74), with different definitions of hypoglycemia, during day or night, or based on either the total treatment period (titration and maintenance phase) or just the maintenance period. In meta-analysis 1, there is overlap between different types of hypoglycemia events, whereas in meta-analysis 2, an event could be counted in only one category. The authors report that the meta-analyses were pre-planned, and the FDA provided input on the statistical plan.

A total of six RCTs were included in the meta-analyses. The data were pooled separately based on the patient populations: type 1 diabetes mellitus (T1DM) (Studies 3583 and 3770), insulin-naive type 2 diabetes mellitus (T2DM) on basal oral therapy (Studies 3579, 3672, and 3586), and T2DM receiving basal + bolus therapy (Study 3582).

Studies 3585 and 3580 were excluded because they did not include an IGlar treatment group, and Study 3668 was excluded because it enrolled a mixed treatment population that included insulin-naive patients and patients on basal insulin. Data from the IDeg flexible dosing group (IDeg-Flex) was also excluded from Study 3770.

The number of hypoglycemia episodes per patient was analyzed using a negative binomial regression model with log-link function. The logarithm of the time interval (i.e., extent of exposure plus seven days) was used as an offset. Differences in reporting patterns between patients were accounted for in the over-dispersion parameter included in the negative binomial model. The model included the following covariates: trial, treatment, antidiabetes treatment at screening, sex, and region as fixed factors, and age as a continuous covariate. Two alternative models were planned if the primary analysis did not converge. First, covariates were removed from the model one at a time; if that was unsuccessful, a Poisson model was estimated.

The final model was a negative binomial model, except for severe hypoglycemia, where a Poisson model was used. The results of meta-analysis 1 and meta-analysis 2 are summarized in Table 75 and Table 76 respectively.

The third meta-analysis analyzed the total daily insulin dose (U/kg), with separate analyses for total daily basal dose and total daily bolus dose for T1DM and T2DM patients on basal + bolus insulin. The total daily insulin dose (U/kg) was defined as the dose in units reported as actually taken by the patient at the end of the trial divided by the weight of the patient at baseline. The log of the dose was analyzed using an analysis of covariance model with trial, treatment, antidiabetes treatment at screening, sex, and region as fixed factors, and age at baseline as a continuous covariate. Missing data were imputed using last observation carried forward.

These analyses included two RCTs for patients with T1DM, one RCT for patients with T2DM on basal + bolus insulin, and three RCTs for patients with T2DM receiving basal insulin + oral antidiabetes drugs. The results of this analysis are summarized in Table 77, showing the dose ratio of IDeg versus IGlar.

Discussion

The meta-analysis included a select number of trials, and the criteria used to choose these studies were not stated. The manufacturer commented that all phase IIIa trials available at the time of publication were included in the meta-analysis. The meta-analysis does not include Study 3587 (insulin-naive patients with T2DM) or the crossover SWITCH studies, which the manufacturer states were not available at the time. No updated meta-analyses with these new data were published. It is unclear what impact, if any, the exclusion of these studies may have had on the results. All trials were open label; thus, subjective outcomes, such as hypoglycemia, may be prone to bias.

The analyses were conducted using accepted statistical methods, although the analyses did not account for within-trial clustering and correlation, which is likely to lead to a less conservative estimate. The authors state that no tests for heterogeneity between trials were made because the design and scope of the included studies were similar. No data were provided on study or patient characteristics. Had estimates of heterogeneity been conducted, they would have provided evidence to confirm that the rate ratios and insulin dose ratios were indeed similar between trials, and that no substantial between-study heterogeneity was present.

As the data were pooled for specific diabetes populations and hypoglycemia outcomes, it is difficult to compare these results (based on patient-level data) with the direct meta-analyses (analyzed based on trial-level data) that were reported in the review of indirect treatment comparisons. Dawoud et al.⁶² reported a severe hypoglycemia rate ratio of 1.03 (95% confidence interval [CI], 0.63 to 1.67) for IGlar versus IDeg in patients with T1DM. The meta-analysis by Freemantle et al.⁶³ reported a rate ratio of 0.79 (95% CI, 0.67 to 0.93) for IDeg versus IGlar for nocturnal hypoglycemia events in patients with T2DM on basal insulin–supported oral therapy. These results are similar to the data reported above. In Freemantle et al.,⁶³ the risk of symptomatic hypoglycemia was higher for IDeg versus IGlar (rate ratio 1.35; 95% CI, 1.27 to 1.44). In comparison, the patient-level meta-analysis generally suggested that IDeg had a lower rate of hypoglycemia than IGlar, although none of the analyses focused on symptomatic events only.