Methodological concerns with the meta-analysis comparing insulin degludec/liraglutide and insulin glargine/lixisenatide

In a recent article published in Expert Opinion on Pharmacotherapy (Nov 5 2017), Cai and colleagues reported a systematic review and meta-analysis of randomized controlled trials of insulin degludec/liraglutide (IDegLira; Novo Nordisk) and insulin glargine/lixisenatide(IGlarLixi; Sanofi), both fixed ratio combinations (FRC) of a basal insulin and a glucagon-like peptide-1 receptor agonist (GLP-1RA) [1]. In the absence of head-to-head comparisons of drugs, meta-analyses can provide additional evidence not available from single trials; however, any conclusions must be based on proper methodology. Cai et al. [1] concluded that ‘Comparisons between the two treatment groups indicated no significant differences between groups in absolute HbA_1c changes or body weight changes relative to baseline.’ We question the validity of these overall conclusions, based on concerns about both the heterogeneity of the trials included and the statistical methods used.

1. Heterogeneity of the included studies

There were important differences between the trials in terms of the populations included, trial designs, and comparator arms, and we do not believe that these differences were adequately addressed.

First, the baseline medication differed between trials so that, depending upon the trial, some patients were previously on oral antidiabetic drugs (OADs) [2–4], some on basal insulin and OADs [5,6], and some on maximum tolerated dose of a GLP-1 RA [7]. These differences have not been accounted for in the analysis and can profoundly affect the possible impact of the FRC and therefore the conclusions of the analysis. For example, for the comparison ‘IDegLira vs. GLP-1 RA’, the authors reported combined effects based on a trial in which patients were previously on OADs [2] and a trial in which patients were already on maximum-dose GLP-1 RA therapy (cited as an abstract, now fully published) [7]. Since the patients came from different pretrial regimens, it is not meaningful to provide estimated outcomes based on combining effect sizes from such different populations. The change in body weight in the IDegLira arm was −0.5 kg in the trial where IDegLira was initiated in an OAD population while it was +2.0 kg in the trial where patients switched from previous GLP-1 RA therapy.

The authors stated that trial heterogeneity was dealt with by having the trial as a random effect in the model. However, a random effect cannot eliminate all between-trial heterogeneity. The problem of heterogeneity was addressed to some degree by grouping the trials according to comparator treatment in Table 3. However, the issue could have been further dealt with by also considering the different pretrial treatment regimens and attempting to more clearly control for these differences.

Secondly, the comparators in trials were subject to different dosing restrictions during the trials, therefore the end of trial difference between the treatment arms differed. We do not believe it is scientifically sound to compare the end-of-trial results between an FRC and an uncapped basal insulin comparator (i.e. no maximum dose specified, as was the case with the basal insulin in two of three IDegLira comparisons with insulin) with those between an FRC and a capped insulin comparator (as was the case with glargine in two of three IGlarLixi comparisons with insulin). As seen in the DUAL II (capped) [5] and DUAL V (uncapped) [8] trials, end-of-trial differences in HbA_1c and fasting plasma glucose (FPG) favoring the FRC were larger, and the weight benefit was smaller, when the insulin dose was capped in the comparator arm. As a consequence, comparing results across trials, not taking these differences into account, will not reflect the true similarity or difference between regimens.

Thirdly, two of the IGlarLixi trials included a run-in period, while the IDegLira trials did not. The FPG at baseline was 7.3/7.4 mmol/L in one trial [6], whereas baseline FPG values ranged from 8.9 to 9.9 mmol/L in the other two IGlarLixi trials [3,4] and all the IDegLira trials [2,5,7–9]. Furthermore, the target FPG was 4.4–5.6 mmol/L in all the IGlarLixi trials, whereas it was 4.0–5.0 mmol/L in all but one of the IDegLira trials. The authors do not discuss the fact that differences in baseline FPG and FPG target can have a significant impact on efficacy results.

2. Statistical methods

Conventional meta-analyses are designed to provide further evidence on the efficacy and/or safety of a particular drug or class versus comparators in trials. The conventional meta-analysis by Cai et al. [1] is followed by an attempt to make an indirect comparison of IDegLira and IGlarLixi. However, such a comparison needs even stronger assumptions than do direct comparisons regarding comparability of trials included, and requires a network meta-analysis (NMA) or similar indirect comparison methodology in order to draw appropriate statistical inference of any differences between the two treatments. NMA is the methodology accepted by health technology assessment groups around the world, and standard criteria for performing an NMA have been outlined by the International Society for Pharmacoeconomic and Outcomes Research [10]. Nowhere in the article do Cai et al. [1] state how the estimated mean changes from baseline for IDegLira vs. IGlarLixi were compared to arrive at the conclusion of ‘no significant differences’. It seems possible that the conclusion was based on whether the confidence intervals for the treatment contrasts were overlapping. But this method does not imply that there is no statistical difference.

In conclusion, we believe that this meta-analysis does not provide a balanced presentation of the differences between IGlarLixi and IDegLira, and we would like to ask the authors to reconsider their analyses and conclusions.

Declaration of interest

All authors are employees of Novo Nordisk A/S, while K Begtrup owns stock in Novo Nordisk A/S.