The study had a multicentre, randomised, controlled, crossover design.
Participants were recruited at four adult and four paediatric sites in Europe with experience in the use of insulin pumps and CGM. Participants were included if they were aged between 6 and 70 years, had a type 1 diabetes duration of more than 1 year and a HbA1c level between 7.5% and 9.5% (58.5 and 80.3 mmol/mol). Adult participants were aged 19−70 years; paediatric participants were aged 6−18 years. In addition, eligible participants had been using CSII with rapid-acting insulin analogues for more than 6 months, were naive to CGM and had successfully completed a five-question multiple choice test concerning pump therapy and general understanding of diabetes . Exclusion criteria included ≥3 incidents of severe hypoglycaemia in the last 12 months, a history of hypoglycaemia unawareness (i.e. hypoglycemia without symptoms), concomitant chronic disease known to affect diabetes control and any pharmacological treatment that might modify glycaemic values.
Participants were randomised to either a Sensor On/Sensor Off (On/Off) or a Sensor Off/Sensor On (Off/On) treatment sequence, in a 1:1 ratio. Randomisation was generated electronically through the electronic case report form according to a predefined randomisation sequence (implemented by a statistician). Paediatric and adult participants were randomised separately. Participants, study staff and investigators were not blinded to the HbA1c data.
During a 1-month run-in phase, participants used a glucometer (Bayer Ascensia Contour; Bayer Diabetes Care, Basel, Switzerland) and an insulin pump system (MiniMed Paradigm REAL-Time System; Medtronic, Tolochenaz, Switzerland) able to integrate CGM in the study phase. All participants received structured training on diabetes management and device use and had their knowledge assessed, as previously described .
Each treatment period was 6 months long, with a 4-month washout phase between the two periods. Study visits occurred every 6 weeks during the treatment periods, when the data from all devices were uploaded onto a computer system using diabetes management software (CareLink Therapy Management System for Diabetes-Clinical; Medtronic, Tolochenaz, Switzerland). There were no study visits during the washout period.
All participants wore a continuous glucose monitor (Guardian REAL-Time Clinical; Medtronic, Tolochenaz, Switzerland), which they were blinded to (the device screen was turned off), for 2 weeks prior to randomisation and prior to crossover. Participants in the Sensor Off arm wore the device for 2 weeks prior to each study visit.
No common treatment protocols or fixed algorithms were provided to the centres , and therapy adjustments were made in consultation with participants at clinic visits. Participants were individually encouraged to make self-adjustments to their treatment using real-time CGM values, hyper- and hypoglycaemic alerts and trends, or to incorporate self-monitoring of blood glucose (SMBG) results into treatment adjustments, with written examples of therapy changes provided in the optional patient diary. Participants completed a ten-question test to demonstrate technical knowledge on the pump (4 weeks before randomisation) and a 12-question test on CGM (at visit 1 of the On/Off sequence or visit 6 of the Off/On sequence).
The primary endpoint was the difference in HbA1c levels between the Sensor On and Sensor Off arms after 6 months of follow-up, adjusting for baseline levels. Prespecified secondary endpoints included changes in glycaemic patterns, as expressed by mean 24 h glucose and 24 h AUC values, and changes in the time spent in hypoglycaemia (<3.9 mmol/l), hyperglycaemia (>10 mmol/l) and euglycaemia (3.9–10 mmol/l).
Sensor data for the secondary endpoints were extracted from CareLink Clinical (CareLink Therapy Management System for Diabetes-Clinical, Medtronic, Tolochenaz, Switzerland) during the 15-day period prior to the end-of-period (6-month) visit. For the Sensor On arm, 100% sensor use was calculated as the number of days in the Sensor On period multiplied by 288, the maximum number of sensor readings per day. HbA1c was analysed by a central laboratory at screening, randomisation and at 3, 6, 10, 13 and 16 months using a Diabetes Control and Complication Trial-aligned HPLC method (Laboratorium für Klinische Forschung, Kiel, Germany).
Data on adverse events were collected at each visit and the number of SMBG values <3.9 mmol/l were calculated from the glucose meter downloads for 15 days prior to the end of each period. Severe hypoglycaemia was defined as an episode requiring assistance from another person or neurological recovery in response to restoration of plasma glucose to normal. Ketoacidosis events were defined as episodes of hyperglycaemia (blood glucose >13.9 mmol/l) with low serum bicarbonate (<15 mmol/l), low pH (<7.3) or both, together with either ketonaemia or ketonuria, that required treatment in a healthcare facility. Patterns of insulin use (total insulin dose, relative proportions of basal and bolus doses, number of boluses and temporary basal rates, and use of basal suspend and bolus wizard features) during the 15 days prior to the end of each study period were analysed from the data uploaded from the insulin pumps.
A sample size of 124 participants (62 randomised to sequence On/Off and 62 randomised to sequence Off/On) provided 90% power to detect a mean difference of 0.3% in the primary endpoint, assuming an SD of 1.0%, a two-tailed matched-pairs t test and a type I error of 5%. To account for the duration of the study and possible dropouts, the sample size was increased by approximately 20%, to 160 participants.
The two groups were compared using an ANOVA with adjustment for period effect and subject as random effect. Period was included in the model regardless of statistical significance. The mean difference in HbA1c between the Sensor On and Sensor Off arms, with the corresponding 95% CI and p value, were estimated. Possible carry-over effects were minimised in the study design by using a 4-month washout period . The intention-to-treat (ITT) population consisted of all randomised participants, irrespective of their compliance with the planned course of treatment or deviations from the protocol. In the case of missing data, measurements from the same study period were carried forward to substitute for missing end-of-period values. If no measurements could be carried forward, the end-of-period data that were available for only one period were used to impute missing data in the other period.
Secondary endpoints were compared using an ANOVA model similar to the one used for the analysis of the primary endpoint. Analyses were performed in the ITT population. All statistical analyses were performed using Statistical Analysis System (SAS), version 9.2 (SAS Institute, Cary, NC, USA), and p values <0.05 were considered statistically significant.
The study protocol received institutional or national Ethics Committee approval at each of the study centres and the study was conducted in line with the Good Clinical Practice provisions of the Declaration of Helsinki with all amendments and local regulatory requirements. Written informed consent or assent was obtained from all participants before enrolment; parental consent was obtained for participants aged <18 years.