Methods
The study design and predefined endpoints are described in detail elsewhere. Briefly, the study was a randomised, double blind, placebo controlled trial with a parallel group design conducted at 14 sites (13 hospital based outpatient clinics and one primary care unit) in Sweden. Written and verbal informed consent was obtained from all participants. The inclusion criteria were people with type 2 diabetes treated with multiple daily insulin injections, HbA1c concentrations ≥58 mmol/mol (7.5%) and ≤102 mmol/mol (11.5%), and a body mass index of 27.5–45 kg/m. Multiple daily insulin injections were defined as separate basal and mealtime insulin components, including at least two daily mealtime insulin doses. We excluded people using premixed insulin. Participants were required to have a fasting C peptide level of 0.1 nmol/L or higher. Some of the inclusion and exclusion criteria were subject to minor changes during the study, partly to enable the use of more general and conventional cut-offs for variables and partly to facilitate further recruitment. During the initial part of the study we changed the lower cut-off for HbA1c from 64 mmol/mol (8.0%) to 58 mmol/mol (7.5%) and body mass index from 28.0 to 27.5 kg/m. We also changed the cut-offs for C peptide, calcitonin, fasting glucose, creatinine, and age during the study, and clarified contraceptive methods and use of glucocorticoids (see supplementary file http://www.bmj.com/content/bmj/suppl/2015/10/28/bmj.h5364.DC1/linm027803.ww1_default.pdf for details). Other inclusion and exclusion criteria have been described previously.
After a maximum run-in period of eight weeks, we randomised participants to either subcutaneous liraglutide or placebo. The composition of the placebo was the same as for liraglutide but with the absence of the active pharmaceutical ingredient. The study was double blinded and we used minimisation allocation to randomise participants 1:1 to liraglutide or placebo (see supplementary file http://www.bmj.com/content/bmj/suppl/2015/10/28/bmj.h5364.DC1/linm027803.ww1_default.pdf for details). Using the Dexcom G4 PLATINUM (San Diego, CA) continuous glucose monitoring system, we carried out masked continuous glucose monitoring during one week of the eight week run-in period, week 12 of the trial, and one of the two final weeks of the follow-up period of 24 weeks. The device used for continuous glucose monitoring consists of a subcutaneous sensor, a wireless transmitter, and a receiver. The sensor continuously measures the glucose values in interstitial fluid over one week and sends data to the receiver. During masking the receiver does not display the values but rather stores them for downloading. We measured capillary glucose values before meals and 1.5 hours after meals during at least two days before baseline and at the end of follow-up. During the study we used the Contour XT blood glucose meter (Bayer) to measure capillary glucose levels. We used the diabetes treatment satisfaction questionnaire to estimate patient satisfaction with treatment. This validated questionnaire consists of eight questions and has been used in many clinical trials of diabetes treatment. Two versions are used, one for patients to record current satisfaction with treatment and one for patients to retrospectively compare satisfaction before and after study treatment. At each visit we asked participants to report any potential adverse events using an open ended question.
Liraglutide or placebo was administered at a dose of 0.6 mg during week 1, 1.2 mg during week 2, and 1.8 mg during week 3 and onwards. The patients chose when to administer the drug during the day, and they were supposed to use the same timing each day during the trial. The periods for increase of dose were extended based on individual tolerance to the trial product. During the trial, the highest tolerated dose of liraglutide or placebo was used.
Since the participants had inadequate glycaemic control at inclusion (HbA1c ≥58 mmol/mol (7.5%)), no general reduction in insulin doses were recommended when initiating or titrating liraglutide or placebo. Reductions in insulin were only considered in participants with normal or close to normal glucose levels at fasting or before meals (<7.0 mmol/L (126 mg/dL)) based on self measured blood glucose levels (7 point profile) for two days.
If glucose levels were not on target after titrating liraglutide or placebo we recommended increasing insulin doses to original levels. After the titration phase of liraglutide and placebo, we advised the participants to adjust their insulin doses throughout the study as performed daily before enrolling in the trial. The participants were advised to measure three or four glucose values each day before meals and at bedtime according to clinical guidelines for potentially adjusting insulin doses, diet, or physical activity.
After randomisation, follow-up visits took place at weeks 6, 12, 18, and 24. At all visits we checked HbA1c levels, insulin doses, hypoglycaemic events, adverse events, blood pressure, and weight. Additionally, extensive clinical examinations, assessment of treatment satisfaction, and laboratory testing, including of biobank samples, were carried out at weeks 12 and 24.
If fasting self measured blood glucose values taken on three separate days or any fasting plasma glucose samples analysed by the central laboratory exceeded 15.5 mmol/L (279 mg/dL) from baseline to week 12, or 13.5 mmol/L (245 mg/dL) from week 12 to week 24, the participant was contacted for an unscheduled visit about rescue treatment. If an increased fasting plasma glucose level was confirmed by the central laboratory and no treatable intercurrent cause for hyperglycaemia was diagnosed, rescue treatment was initiated, which primarily consisted of the investigator assisting in increasing insulin doses.
According to guidelines for trials we defined hypoglycaemias as non-severe symptomatic, non-severe asymptomatic, and severe requiring the assistance of another person. The number of non-severe hypoglycaemias were predefined to be evaluated both below 4.0 mmol/L (72 mg/dL) and 3.0 mmol/L (54 mg/dL). The participants were counselled on typical symptoms of hypoglycaemia and provided with a diabetes diary to record glucose values, signs of hypoglycaemia, and recovery from symptoms after intake of carbohydrates. If symptoms occurred we instructed the participants to perform a finger stick glucose measurement immediately, but to avoid delay in treating symptoms.
The primary endpoint was change in HbA1c level between baseline and week 24. Novo Nordisk provided the study drug and treatment codes. Apoteket (National Pharmacy), Sweden, handled study treatment logistics. Gothia Forum (Gothenburg, Sweden) monitored the study. Laboratory tests were measured at the Research Centre for Laboratory Medicine at Karolinska University Hospital, Stockholm, Sweden.
Patient Involvement
No patients were involved in setting the research question or the outcome measures, nor were they involved in recruitment or the design and implementation of the study. There are no plans to involve patients in dissemination of the results.
Statistical Analysis
The main statistical analyses for predefined endpoints were presented in the original protocol and are published elsewhere. A detailed statistical analysis plan was signed before the database was locked. The full analysis set, used for the primary efficacy analysis, consisted of all randomised participants who received at least one dose of study drug and had at least one follow-up measurement. The per protocol population consisted of all participants who had at least one visit at week 18 or 24 during follow-up, no major protocol deviations, and was determined before the database was locked. The primary efficacy analysis was the change in HbA1c level from baseline to 24 weeks between the two treatment groups using analysis of covariance, with HbA1c at baseline as the covariate on the full analysis set. The last observation was carried forward from six weeks. In all efficacy analyses we excluded measurements obtained after rescue treatment.
For comparison between the two study groups we used Fisher's exact test for dichotomous variables, Mantel-Haenszel's χ test for ordered categorical variables, and Fisher's non-parametric permutation test for continuous baseline variables. For efficacy variables of change between baseline and 24 weeks, we used analysis of covariance with baseline value as a covariate. A sensitivity analysis was performed on all predefined endpoints, including all randomised participants irrespective of whether they took the study drugs or had any valid follow-up measurements. The methodology was the same as for the primary efficacy analysis except that we used the last observation carried forward principle from baseline values onwards. All significance tests were two sided and conducted at the 5% significance level. For all analyses we used SAS System version 9 (SAS Institute, Cary, NC).
The study was powered to detect a difference in HbA1c concentration of 7 mmol/mol (0.7%) between liraglutide and placebo. For both groups we assumed a standard deviation of 1.2% for a change in HbA1c level, thus we needed 57 participants in each group. We assumed a drop-out rate of 5%, thus we needed 60 participants in each group.