From June 2018 to May 2020, we enrolled consecutive T2DM patients aged between 30 and 75 years, on MDI with or without additional anti-diabetic drugs, referred to our Diabetic Centre with poorly controlled glycaemia, defined as HbA1c levels between 8.0% and 12.0% (64 to 108 mmol/mol), despite a daily dose of insulin of 0.7–1.8 units/kg or maximum dose of 220 units and >2 blood glucose self-assessments per day.

Patients with any of the following clinical conditions were excluded: 1) overt cardiovascular disease, i.e., any documented evidence of known cardiac, cerebral or peripheral artery disease, as well as any symptoms or abnormal findings at physical examination or non-invasive diagnostic tests (including electrocardiogram and cardiac and arterial ultrasound studies) indicating cardiovascular disease; 2) presence of serious medical conditions, including liver cirrhosis, malignancies, acute or chronic inflammatory disease; 3) pregnancy; 4) psychological conditions that might have hampered patient’s compliance; 5) two or more symptomatic hypoglycaemia episodes in the past 6 months; 6) presence of significant diabetic microvascular complications (i.e., diabetic retinopathy or neuropathy, or chronic kidney disease, as indicated by an estimated glomerular filtrate rate lower than 15 ml/min/1.73 m2); 7) any other condition that, in the judgment of the investigators, precluded the participation of the patient in the study. Furthermore, in all patients included in the study, autoimmune diabetes was excluded by measuring anti-glutamic acid decarboxylase and anti-tyrosine phosphatase (IA-2) antibodies.

The study protocol was drawn up in accordance with the requirements of Good Clinical Practice of the European Union and the current revision of the Declaration of Helsinki and was approved by the Ethics Committee of our Institution. Aims and methods of the study were clearly discussed with potentially eligible patients, who provided formal written informed consent to participate in the study. A flow chart of the study is shown in Fig. 1.

Flowchart of the trial protocol. CGM continuous glucose monitoring, CKD chronic kidney disease, CSII continuous subcutaneous insulin injection, DM diabetes mellitus, ECG electrocardiogram, eGFR estimated glomerular filtration rate, FMD flow-mediated dilation, GAD glutamic acid decarboxylase, HM Holter monitoring, IA-2 islet antigen-2, MDI multiple daily insulin injection, NMD nitrate-mediated dilation

Participants were enrolled at our Diabetes Care Unit. Before randomisation, patients underwent a 2-month run-in phase consisting of three clinical visits, designed to achieve optimal insulin injection treatment. During this period, insulin treatment was progressively increased according to a standardised titration protocol to achieve pre-prandial and post-prandial target ranges (fasting and pre-prandial 70–110 mg/dl and 2-h post prandial less than 180 mg/dl) of blood glucose levels, allowing adjustments of both basal and bolus insulin. The total dose increase was targeted at 10–40% above the basal dose of insulin. Patients were treated with both long-acting (Glargine or Detemir) and rapid-acting (Lispro, Aspart, or Glulisine) analogues of insulin. Patients were allowed to use insulin pens during the study for the administration of either rapid-acting or long-acting insulin analogues and injection preference was left to the Investigator’s standard clinical practice.

After the 2-month run-in period, patients were randomized to continue with multi-injection insulin treatment or pump treatment. Randomization was performed according to a computer-generated table of casual numbers, with even and odd figures assigned to CSII and MDI, respectively, without any stratification/block due to the low number of subjects expected. Since the study was open and based on results of objective glucose level data (obtained by CGM), diabetologists who enrolled and randomized patients could also be involved in the assessment of CGM data.

After randomization, patients assigned to pump treatment underwent a training period of up to 3 weeks after the end of the run-in phase to allow an optimal management of therapy. In both groups, treatment could be further adjusted according to glycemic data. Other types of anti-diabetic drugs were continued, but no adjustment of the dose was allowed.

Pump treatment was performed with the Medtronic 640 system (Medtronic Inc. Minneapolis, MN, USA).

All patients underwent the following tests at baseline and after 6 months of treatment: 1) clinical cardiological and diabetes visit; 2) standard clinical chemistry tests and glycated haemoglobin (HbA1c) measurements; 3) resting 12-lead electrocardiogram (ECG); 4) seven-day continuous glucose monitor (CGM); 5) 24-h ECG Holter monitoring (HM); 6) flow-mediated dilation (FMD) and nitrate-mediated dilation (NMD) of the brachial artery.

Continuous glucose monitoring (CGM) was performed for 7 days using the Medtronic iPro2 system (Medtronic, Inc. Minneapolis, MN, USA). CGM measures interstitial glucose at 5-min intervals, and patients were instructed to perform the required sensor calibration procedure four times a day according to the Manufacturer’s instructions. Data from the pump and blood glucose meter were uploaded with the Medtronic CareLink Therapy Management Software. Masked CGM data were obtained with Medtronic iPro2, with glucose data recorded over 7 days before randomisation and on completion of 6-month randomised treatment.

During the trial participants in the CSII arm were not allowed to use personal CGM but they were instructed to check blood glucose levels as those in the MDI arm.

The following parameters were obtained from the glucose recordings: 1) mean 24-h glucose concentration; 2) the standard deviation (SD) and coefficient of variation (CV) of interstitial blood glucose measurements; 3) time in range (TIR, i.e. the percentage of time with glucose levels between 70 and 180 mg/dL); 4) time above range 1 and 2 (TAR-1 and TAR-2), as a measure of hyperglycemic burden, defined as the percentage of time with glucose levels between 181–250 mg/dL and >250 mg/dL, respectively; 5) time below range 1 and 2 (TBR-1 and TBR-2), as a measure of hypoglycemic burden, defined as the percentage of time with glucose levels between 54–69 mg/dL and <54 mg/dL, respectively [20].

Peripheral vasodilator function was assessed by the same expert operator using methods previously described in detail [13, 14, 21,22,23,24]. Patients were invited to refrain from exercise, smoking, alcoholic drinks and caffeine use in the 3 h preceding the test. Usual drugs were permitted as they remained constant throughout the study.

Endothelium-dependent vasodilatation was assessed by measuring flow-mediated dilatation (FMD). Briefly, subjects rested in the supine position for at least 10 min in a warm, quiet room (22 °C to 24 °C) before testing. A 10-MHz multi-frequency linear array probe and a high-resolution ultrasound machine were used to acquire images of the right brachial artery. Brachial artery diameter was measured throughout the whole test using a totally automated system that automatically identifies the internal edges of the vessel and tracks the walls of the artery via the brightness intensity of the walls vs. the lumen of the vessel (Cardiovascular Suite, Quipu srl, Pisa, Italy). The software provides a diameter measurement every second throughout the test. A mechanical support keeps the probe in a fixed position throughout the whole examination. After acquisition of baseline images of the brachial artery, a forearm cuff, positioned 1 cm under the antecubital fossa, was inflated to 250 mmHg and released after 5 min, thus resulting in forearm reactive hyperaemia. FMD was calculated as the maximum percent change of the brachial artery diameter during hyperaemia compared to baseline.

After recovery of brachial artery diameter to basal values, endothelium-independent vasodilatation was assessed by measuring nitrate-mediated dilatation (NMD). To this aim, 25 μg of sublingual glyceryl trinitrate was administrated and NMD was measured as the maximum percent change of the brachial artery diameter compared to the basal diameter.

Twenty-four-hour ECG Holter recordings were performed using 3-channel recorders (Schiller Medilog AR4, Milan, Italy) and monitoring the ECG leads CM5-CM1 and modified aVF [25]. The recordings were analyzed by an expert operator using the Medilog Darwin‐2 system (Schiller Medilog).

Cardiac autonomic function was assessed by measuring HRV both in the time- and frequency-domain, using previously described methods [26, 27].

Time-domain HRV parameters included: SDNN (standard deviation of all RR intervals); SDNNi (mean of the standard deviations of RR intervals of all 5 min segments in the recording), RMSSD (the square root of the mean of the sum of the squares of differences between adjacent RR intervals), pNN50 (percent of consecutive pairs of adjacent RR intervals differing by more than 50 ms in the entire recording) and the triangular index (the ratio between the total number of RR intervals in the 24 h and the number of RR intervals with the modal value).

Frequency-domain HRV parameters, derived from power spectrum analysis of RR intervals by fast Fourier transform, included the amplitudes of RR oscillations in the range of very low frequency (VLF; 0.033–0.04 Hz), low frequency (LF; 0.04–0.15 Hz) and high frequency (HF; 0.15–0.40 Hz).

The primary end-point of the study was endothelium-dependent vascular dilator function, i.e., FMD. Based on our previous experience, FMD in T2DM patients is around 4.5%, with a standard deviation of 1.2% [13]. Thus, we calculated that, to have an 80% power to detect as significant (at a two-tailed p value < 0.05) a between-group difference in FMD of 1.25%, we needed to enrol 15 patients per group.

Continuous variables were compared by independent t-test or Mann-Whitney test, as indicated, whereas Fisher exact test was applied to compare proportions. A repeated measure analysis of variance was applied to assess whether changes of variables at follow-up compared baseline differed between the 2 groups.

Correlations between percent changes in the main GV parameters (SD, CV and TIR) and percent changes in vascular function (FMD and NMD) at follow-up compared to baseline were assessed by Spearman correlation test.

The SPSS 28.0 statistical software (SPS, Florence, Italy) was used to perform statistical analyses. A p value < 0.05 was required for statistical significance.