Study population and study design
In this single center, prospective, placebo-controlled, double blind, randomized, 2-arm parallel, interventional and exploratory pilot study 44 patients with T2D were randomized into 2 groups. The randomisation list was computer generated using a permuted block randomisation with block size of 4. The sequence generation method and the block size was concealed from the investigators. An independent pharmacist labelled the study medications according to the randomisation list. Study participants received empagliflozin 10 mg or placebo for a period of 3 months in addition to their concomitant medication. Non-invasive hemodynamic measurement, transthoracic echocardiography, blood pressure, blood- and urine-chemistry were performed at baseline (day 0), day 1, day 3 and after 3 months. Participants were recruited from the Department of Internal Medicine I at University Hospital Aachen, RWTH Aachen University, Germany. Inclusion criteria were as follows: type 2 diabetes, HbA1c ≥ 6.5% and age ≥ 18 years. The study protocol was approved by the local ethic committee and all subjects gave written informed consent. The trial was registered: EudraCT Number: 2016-000172-19.
Laboratory Measurement
Serum chemistry including haematology, lipid profile, glucose metabolism, eGFR (CKD-EPI formula), cystatin C, NT-proBNP, aldosterone were performed at every visit of the clinical trial. We collected 24 hrs urine at baseline, day 1, day 3 and after 3 months to measure renal excretion of glucose and sodium.
Hemodynamics
We used ClearSight System® (Edwards Lifesciences, Irvine, USA) as a validated 9 non-invasive tool to explore effects of empagliflozin on hemodynamic parameters including cardiac index (CI), stroke volume index (SVI), heart rate (HR), and systemic vascular resistance index (SVRI) at baseline, day 1, day 3 and after 3 months. ClearSight System® uses finger arterial pressure measurement based on the volume clamp method in combination with Physiocal calibration. Dividing the systolic area of the time integral of the pressure curve above the diastolic pressure by the estimated arterial impedance gives a beat-to-beat stroke volume which is multiplied with the heart rate to reach cardiac output, as has been described previously 9.
Transthoracic Echocardiography
Transthoracic and Doppler echocardiography were performed by technicians blinded to clinical information and treatment assignment with commercially available ultrasound systems (GE Healthcare, Chicago, USA). Standardized echocardiographic measurements were obtained in accordance with the guidelines of the EACI (European Association of Cardiovascular Imaging) and ASE (American Society of Echocardiography). Left ventricular systolic function (EF) was measured in 4 chamber and 2 chamber views by Simpson’s Biplane Method. Additionally we performed myocardial deformation analysis of the left ventricle to assess peak global longitudinal strain (GLS) of the endocardial layer by speckle-tracking echocardiography in 4 chamber, 2 chamber and apical 3 chamber views. For diastolic function we determined early (E) and late (A) diastolic mitral inflow velocities, deceleration time (DT), septal early diastolic mitral annular tissue velocity (septal e’) and lateral early diastolic mitral annular tissue velocity (lateral e’) by mitral pulse wave Doppler and tissue Doppler. We calculated E/e' ratio and E/A ratio by dividing E peak by average e’ calculated from septal e’ and lateral e’ respectively E peak by A. Additionally we performed myocardial deformation imaging as determined by 2D and 3D parameter global strain rate. Images were stored digitally for subsequent offline analysis. Interpretation of the echocardiograms was performed by two independent blinded investigators. Interobserver variability of the key echocardiographic endpoints E and e’ was 0.8 for E and 0.77 for e’.
Endpoints
The study was powered for primary study outcome of empagliflozin on systemic vascular resistance index (SVRI) in comparison to placebo after 1 day, 3 days and 3 months of treatment. Secondary endpoints included changes in the following parameters after 1 day, 3 days and 3 months: cardiac index (CI), stroke volume index (SVI), blood pressure, sodium excretion in 24 hrs urine collection, body weight, heart rate, serum levels of NT-proBNP, cystatin C, glucose, HbA1c and aldosterone.
Further secondary analysis included changes in left ventricular systolic function as determined by EF and GLS, and in left ventricular diastolic function as determined by standardized parameters.
Statistical analysis
Descriptive statistics of baseline characteristics were calculated as relative (%) and absolute frequencies for categorical variables. Quantitative variables were described as means and standard deviations, in case of non-normally distributed data, as median with 1st and 3rd quartiles. Data distributions were visualized using box-plots.
Outcome variables were analysed using linear mixed models with fixed effects for treatment, visits (day 1, day 3 and 3 months) and baseline measurement of the variable. For the primary endpoint analysis, randomisation blocks were also included as fixed effect. The random part of the models consisted of intercepts grouped by individuals. Restricted maximum likelihood estimation was used. For NT-proBNP the log transformed variable was used in the analyses. Treatment effects were estimated at each visit along with Wald type 95% confidence intervals. For the primary endpoint the null hypothesis that all treatment-visit interactions are zero was tested against the alternative that at least one of them is not zero using an F test. Kenward-Roger approximation of the degrees of freedom was used. As additional analyses, correlation between changes from baseline to 3 months were calculated for selected variables using the Pearson correlation coefficient, and changes from baseline were compared between treatment groups separately at each visit. Results were not adjusted for multiple comparison.
Four data points (2%) of the primary outcome measurements were missing. All observed data at each visit were used in the linear mixed models allowing inference under the assumption of missing at random missingness mechanism.