Patients and Study Procedures
The study design and the main outcomes of the trial have previously been described in detail [9, 11, 12]. The study was approved by the ethics committee at the University of Gothenburg in accordance with the principles of the Declaration of Helsinki. All participants gave their written informed consent. Recruitment occurred between February 2013 and February 2014. The study was registered in the EudraCT database before it commenced (EudraCT No 2012-001941-42). The proposed design was a randomized, double-blind trial lasting 24 weeks that included patients with T2D and poor glycemic control despite being treated with MDI for at least six months.
Patients were randomized to liraglutide or placebo, which were titrated from 0.6 mg during the first week to 1.2 mg during the second week to 1.8 mg from the third week onwards. When initiating and increasing the dose of liraglutide/placebo, the patients were advised to temporarily reduce the insulin dose if the blood glucose level was already on target in order to avoid the potential risk of hypoglycaemia. After the titration of liraglutide/placebo to 1.8 mg, the insulin dose was increased again to the original dose. Patients were advised to continue to adjust their insulin dose throughout the study as they had done previously.
The primary endpoint was the change in HbA1c from baseline to week 24. Secondary endpoints included the effects of treatment on body weight and total insulin dose.
In brief, inclusion criteria were T2D patients treated with MDI with or without metformin (patients with premixed insulin were excluded), HbA1c ≥ 58 mmol/mol and ≤ 102 mmol/mol, BMI 27.5–45.0 kg/m2. Overall, 124 participants were randomized 1:1 to subcutaneous liraglutide or placebo [7, 9].
Patients had subcutaneous sensors (Dexcom G4 PLATINUM system, San Diego, CA, USA) that measured glucose values every fifth minute. In the current study (MDI-Liraglutide 6), masked CGM data were analyzed to compare the effects of liraglutide and placebo on blood glucose levels. Thus, masked CGM was performed one week before randomization (run-in period), in week 12, and before week 24. When masked, the receiver did not display glucose values and instead stored them for downloading. In total, 57 patients in the liraglutide group and 50 patients in the placebo group had masked CGM data at week 12. Masked CGM data at week 24 were available for 52 patients in the liraglutide group and 47 patients in the placebo group. Liraglutide was added to insulin therapy according to an algorithm described previously [9, 11].
We compared liraglutide treatment with placebo in terms of their effects on time in hypoglycaemia (< 3.9 mmol/l and < 3.0 mmol/l), time in target (4–7 mmol/l), time in range (4–10 mmol/l), and time in hyperglycaemia (> 10 mmol/l and > 14 mmol/l). Estimations were performed for weeks 12 and 24; 24 weeks was considered the main follow-up period. For each glycaemic metric, estimations were performed per 24 h, but they were also stratified into daytime (06:00–22:00 and 06:00–24:00) and nighttime (22:00–06:00 and 24:00–06:00) estimations. We also estimated the mean amplitude of glycaemic excursions (MAGE ) as a glucose variability measure.
Variables Associated with Hypoglycaemia
We estimated how time in hypoglycaemia at weeks 12 and 24 was related to mean glucose level as estimated by masked CGM during the corresponding time period. In addition, we evaluated possible baseline predictors for time in hypoglycaemia (< 3.0 mmol/l and < 3.9 mmol/l, respectively) at week 24. The following baseline variables were evaluated as potential predictors: age, sex, diabetes duration, HbA1c, mean glucose level and standard deviation (SD) measured by masked CGM, BMI, waist circumference, abdominal sagittal diameter, waist-hip ratio, fasting adiponectin level, fasting C-peptide level, fasting proinsulin level, metformin use, total daily insulin dose, percent meal insulin of total insulin dose, time in hypoglycaemia (< 3.0 mmol/l), and diabetes treatment satisfaction status scale (DTSQ). For the predictors that were statistically significant in the liraglutide group at the 5% level, we also evaluated whether their predictive abilities were significantly stronger in the liraglutide group compared with the placebo group through interaction analysis.
All laboratory analyses were performed at the Karolinska University Hospital, Stockholm, Sweden. HbA1c was measured according to the International Federation of Clinical Chemistry (IFCC) method, with all values converted according to the National Glycosylation Standard Program (NGSP) for dual reporting . A commercially available ELISA kit from Mercodia AB was used to measure proinsulin.
For descriptive purposes, data are presented here as mean (SD), median (minimum, maximum) for continuous variables and as number (%) for categorical variables. For comparisons between groups, Fisher’s nonparametric permutation test was used for continuous variables, Fisher’s exact test for dichotomous variables, the Mantel–Haenszel chi-square test for ordered categorical variables, and the Pearson chi-square test for non-ordered categorical variables. For all tests, p < 0.05 was considered statistically significant.
Analysis of the percent of time in hypoglycaemia in relation to the CGM mean and various baseline predictors was performed using a log-linear quasi-binomial model with robust standard errors, accounting for the positive, skewed, and heteroscedastic distribution of percent of time in hypoglycemia. In prediction analyses, the baseline variables were analyzed one at a time and further adjusted for time in hypoglycaemia at baseline and CGM mean. An interaction between the predictor and treatment group was included to test whether the effect of the predictor differed between treatment groups, in order to eliminate spurious correlations due to study-related causes and phenomena such as regression to the mean.