Trial design

The trial was designed as an open-label, crossover study (Fig. 1). A total of 13 individuals with type 1 diabetes were randomised to receive either 10 mg of dapagliflozin or placebo as an adjunct to insulin therapy for an initial 7 day period, followed by the alternative treatment for another 7 days. A 14 day washout period was implemented between the two intervention periods to avoid carry-over effects.

Fig. 1Sample size

A priori sample size calculation was based on the primary outcome: the median GLP-1 concentration during the OGTTC. A former pilot study performed by our research group showed an SD of 6.32 pmol/l for GLP-1 during OGTTC [7]. A difference of 10 pmol/l between the two treatment phases was deemed clinically relevant, and a conservative SD of 10 pmol/l was assumed. Based on a paired-means test, we calculated a required number of 13 individuals to detect this difference at a two-sided alpha level of 0.05 with a power of 90%. This sample size calculation was conservative, as repeated measures of GLP-1 over 120 min may provide statistical power higher than 90%.

Participants

A total of 1177 outpatients of the Department of Diabetes, Endocrinology, Clinical Nutrition & Metabolism, Inselspital, Bern University Hospital, University of Bern, were assessed for eligibility. In total, 982 candidates were excluded based on the inclusion or exclusion criteria. The inclusion criteria were a duration of type 1 diabetes >5 years, a BMI of 20–29 kg/m2 and C-peptide concentrations <0.1 nmol/l. The exclusion criteria were a diagnosis of renal or hepatic dysfunction, a history of any type of malignancy, the intake of drugs influencing glucose homeostasis during the last 3 months, alcohol or drug abuse, active smoking (five or more cigarettes per day) and pregnancy. Thirteen individuals gave consent after detailed explanation of the trial (Fig. 2). The study sample was broadly representative of the general Swiss population in terms of sex, gender, ethnicity and socioeconomic status. However, the limited sample size constrained the ability to demonstrate age representativeness. Aside from gender, these demographic characteristics were not systematically assessed. Gender was determined via self-report and was neither used as an inclusion or exclusion criterion nor explicitly considered in the trial design or recruitment process. The study was approved by the Cantonal Ethics Committee Bern prior to conducting any study-related actions involving potentially eligible participants. All procedures performed involving human participants were in accordance with the ethical standards of the institutional and/or national research committees and with the 1964 Declaration of Helsinki and its later amendments.

Fig. 2

CONSORT flow diagram of study recruitment and randomisation. Following recruitment, 13 individuals with type 1 diabetes were randomised to receive either 10 mg of dapagliflozin or placebo as an adjunct to insulin therapy for an initial 7 day period, followed by the alternative treatment for another 7 days

Randomisation

Randomisation in a 1:1 ratio was conducted electronically via REDCap, an electronic database that was distributed and maintained by the clinical trial unit of the University of Bern. As the study was unblinded, both investigators and participants were informed of the scheduled intervention sequence.

Examinations: HEC and OGTTC

During both intervention periods, participants were asked to ensure that their macronutrient intake was similar in both treatment periods and to abstain from alcohol. On day 7 of each intervention period, a HEC was performed, followed by an OGTTC. Blood sampling was conducted after an 8 h overnight fasting period. In brief, during the HEC, insulin was infused at a rate of 40 mU/m2 body surface/min for 120 min. The body surface was calculated using the formula of DuBois et al [8]:

$$\left[}^\right]=0.007184\times \text[\text]}^\times \text[\text]}^$$

Blood glucose concentrations were measured every 5 min and titrated at 7.0 mmol/l (±2.0 mmol/l as the predefined target range) by intravenous infusion of 20% glucose. After completion of the HEC, blood glucose concentrations were maintained within the target range for at least 60 min. Subsequently, participants ingested 50 g of oral d-glucose dissolved in 300 ml of tap water for the OGTTC (time point 0). Blood glucose concentrations were measured every 5 min. Insulin requirements for 50 g of d-glucose were calculated as follows:

$$\left(1\right)\, \text=\frac[\text]}$$

$$(2)\, \text[\text]=\frac[\text]}}$$

The calculated total insulin dose was administered via three decreasing infusion rates over a total of 60 min (50% from 0 to 20 min, followed by 30% from 21 to 40 min and 20% from 41 to 60 min of the OGTTC). Blood sampling for primary and secondary endpoints was performed every 15 min during 120 min of the HEC and 120 min of the OGTTC.

Outcomes Primary outcome

The primary outcome was the median differences in GLP-1 concentrations observed during the OGTTC after the intake of dapagliflozin and placebo. GLP-1 was measured every 15 min during 120 min of the OGTTC.

Secondary outcomes

The secondary outcomes were the median differences in GLP-1 concentrations measured every 15 min during 120 min of the HEC, the median differences in glucagon and somatostatin concentrations measured every 15 min during 120 min of the HEC and OGTTC, and the median concentrations of total ketone body measured at baseline and every 15 min during 120 min of the HEC and OGTTC after dapagliflozin vs placebo intake.

Exploratory outcome

The exploratory outcome was the plasma glucose concentrations during the OGTTC measured after dapagliflozin vs placebo intake.

Changes from protocol

In the original study protocol, the AUC for GLP-1 was defined as the primary outcome. However, to facilitate the interpretation of results, we decided to report unit-based outcomes of a repeated-measures mixed-model analysis, leveraging all time points instead of the AUC. AUC analysis is a way of aggregating results, while a repeated-measures mixed-model analysis provides greater statistical robustness together with increased readability. AUC values are provided in the electronic supplementary material (ESM Table 1).

Biochemical analysis

For pre-analytic conservation of GLP-1 and glucagon, a P800 blood collection tube holding a proprietary mixture of protease inhibitors (BD Biosciences, San Jose, CA, USA) was used. For measurement of total ketone bodies and somatostatin, EDTA-coated plasma samples were collected. Active (7-36) GLP-1 concentrations were determined using a human GLP-1 ELISA kit (cat. no. EHGLP; Thermo Fisher Scientific, Basel, Switzerland) with a detection range of 47.4–6066.0 pmol/l. Glucagon concentrations were measured using a human glucagon ELISA kit (cat. no. EHGCG; Thermo Fisher Scientific, Basel, Switzerland) with a detection range of 2.03–130.00 ng/l. Somatostatin concentrations were measured using a human somatostatin ELISA kit (cat. no. abx153137; Abbexa, Cambridge, UK) with a detection range of 3.8–305.5 pmol/l. Concentrations of β-hydroxybutyrate and acetoacetate were determined via LC-MS at the central laboratory unit of Bern University Hospital.

Statistical analysis Primary analysis

The primary analysis was based on the full analysis set (FAS). We depicted the time course of the primary and secondary outcomes in line plots separately for the active and placebo treatments. We presented the median outcome concentrations with lower and upper quartiles separately per treatment. Mean differences in log-transformed primary and secondary outcome concentrations were calculated using a mixed-effects repeated-measures linear model. The model included fixed effects for the treatment (dapagliflozin vs placebo), treatment period (first vs second) and time points (categorical), as well as a random intercept for participants. We used restricted maximum likelihood (REML) and the Satterthwaite method for calculating 95% CIs and p values. We used an autoregressive correlation structure determined as best-fitting by the Akaike’s information criterion (AIC). We back-transformed the model coefficients and confidence limits to display a geometric mean ratio with a 95% CI between the two treatments that can be interpreted on the original scale.

Secondary analyses

A secondary analysis was performed based on the per-protocol patient set (ESM Table 2). We presented the median AUC and quartiles of the primary and secondary outcomes separated per treatment period. The AUC was calculated using the linear trapezoidal method. We analysed differences in the log-transformed AUC using a mixed-effects linear model based on the FAS. The model included fixed effects for the treatment (dapagliflozin vs placebo) and treatment period (first vs second), as well as a random intercept for participants. We used REML and the Satterthwaite method for calculating 95% CIs and p values. We used an unstructured correlation matrix that fitted the data best as assessed by the AIC. We back-transformed the model coefficients and confidence limits to display a geometric mean ratio with a 95% CI between the two treatments that can be interpreted on the original scale.

Additional analyses

In order to exclude a carry-over effect and/or treatment-by-period interaction, we tested the primary outcome model for an interaction between period and intervention; no evidence of such an interaction was found. Additionally, we calculated the within-participant intraclass correlation (ESM Table 3) of primary and secondary outcomes from mixed-effects repeated-measures linear models. Normality of model residuals was assessed using Q–Q plots. Due to the log-normal distribution of the outcome measures, data were log-transformed before analysis.

Post hoc exploratory analysis

OGTTC plasma glucose concentrations were added as exploratory outcomes post hoc and were analysed like the primary outcome, except that no log-transformation was necessary.