We recruited 302 participants between March 13 and October 15, 2014; 224 were randomized (149 intervention, 75 controls) after completing the baseline phase (Fig. 1). Prior to randomization 78 participants discontinued, the primary reason for this was failure to meet screening HbA1c criterion. Participants’ baseline characteristics are summarized in Table 1, the full analysis set included 224 randomized participants, and there were no significant differences between groups.
Table 1 Baseline characteristics
There was no difference in HbA1c change at 6 months between intervention and control groups [−3.1 ± 0.75 mmol/mol (adjusted mean ± SE), (−0.29 ± 0.07%) and −3.4 ± 1.04, (−0.31 ± 0.09%), respectively; p = 0.8222]. A similar drop in HbA1c was detected in both groups comparing study end to baseline values.
In participants younger than 65 years, a prespecified subgroup, the drop in HbA1c was more pronounced in the intervention group compared with controls [−5.7 ± 0.96 mmol/mol, (adjusted mean ± SE) (−0.53 ± 0.09%) and −2.2 ± 1.31 mmol/mol (−0.20 ± 0.12%), respectively; p = 0.0301 (Supplementary Material p. 2)]. A significant interaction between treatment group and age was observed for change in HbA1c (p = 0.0017).
In participants aged 65 years or more, the drop in HbA1c was more pronounced for the controls compared to the intervention group [−5.4 ± 1.45 mmol/mol (−0.49 ± 0.13%)] and [−0.6 ± 1.09 mmol/mol (−0.05 ± 0.10%), respectively, p = 0.0081 (Supplementary Material p. 3)].
Significant reductions in all sensor measures of time spent in hypoglycemia, number of events, and area under the curve were observed for intervention participants compared with control (Table 2, Fig. 2, and Supplementary Material pp. 4–7).
Table 2 Glycemic and glucose variability measures
Time in hypoglycemia [<3.9 mmol/L (70 mg/dL)] reduced by 43% (−0.47 ± 0.13 h/day; mean ± SE) for intervention participants compared with control (p = 0.0006).
Time in hypoglycemia [<3.1 mmol/L (55 mg/dL)] reduced by 53% (−0.22 ± 0.068 h/day) for intervention participants compared with control (p = 0.0014).
Time in hypoglycemia [<2.5 mmol/L (45 mg/dL)] reduced by 64% (−0.14 ± 0.04 h/day) for intervention participants compared with control (p = 0.0013).
Nocturnal hypoglycemia [<3.9 mmol/L (70mg/dL), 23.00–06.00 h] reduced by 54% (−0.29 ± 0.08 h per 7 h) for intervention participants compared with control (p = 0.0001).
Daytime hypoglycemia [<3.9 mmol/L (<70 mg/dL), 06.00–23.00 h] reduced by 31% (−0.16 ± 0.08 h per 17 h) for intervention participants compared with control (p = 0.0374).
The frequency of events with glucose <3.9 mmol/L (70 mg/dL) reduced by 28% (−0.16 ± 0.065 per day mean ± SE) for intervention participants compared with controls (p = 0.0164). Events <3.1 mmol/L (55 mg/dL) reduced by 44% (−0.12 ± 0.037) for intervention participants compared with controls (p = 0.0017). Frequency of events <2.5 mmol/L (45 mg/dL) reduced by 49% (−0.06 ± 0.02) for intervention participants compared with controls (p = 0.0098).
A between-group difference for area under the curve of 51% (−7.80 ± 2.20 h/day × mg/dL mean ± SE) for sensor glucose level <3.9 mmol/L (70 mg/dL) was observed for intervention versus control participants (p = 0.0005). For sensor glucose levels <3.1 mmol/L (55 mg/dL), area under the curve reduced by 60% (−2.51 ± 0.76 h/day × mg/dL) for intervention participants compared with controls (p = 0.0012). Area under the curve was also significantly reduced by 67% (−0.70 ± 0.22 h/day × mg/dL) at glucose levels <2.5 mmol/L (45 mg/dL) for intervention compared with control participants (p = 0.0015).
For the prespecified subgroup aged less than 65 years, time in hypoglycemia [<3.9 mmol/L (70 mg/dL)] reduced by 35% for intervention participants compared to control (−0.37 ± 0.168 h/day, p = 0.0279) with 40% reduction in area under the curve (p = 0.0305) and no difference in the number of events. A trend towards reduced time and events for hypoglycemia at other thresholds was observed (Supplementary Material p. 2).
For participants 65 years or more, time in hypoglycemia (<3.9 mmol/L [70 mg/dL]) reduced by 56% for intervention participants compared to control (−0.60 ± 0.220, p = 0.0083) with 71% reduction in area under the curve (p = 0.0061). No difference was detected in number of events (p = 0.0513). Reduced time, events, and area under the curve for hypoglycemia at other thresholds was observed (Supplementary Material p. 3).
There was no difference in time in range (3.9–10.0 mmol/L (70–180 mg/dL)] between intervention and control participants [p = 0.7925, (Table 2)].
There was no difference in time in hyperglycemia [>10.0 mmol/L (180 mg/dL) and >13.3 mmol/L (240 mg/dL)] between the two groups (Table 2).
A number of glucose variability measures were explored and an improvement for intervention participants was observed (Table 2 and Supplementary Material pp. 8–9).
Glucose variability measured as coefficient of variation (CV) reduced by 2.26 ± 0.71% mean ± SE for intervention participants compared with controls (p = 0.0017). LBGI reduced by 0.3 ± 0.11 mean ± SE for intervention participants compared with controls (p = 0.0029). CONGA was reduced for intervention compared with controls by 3 ± 1.3 mg/dL mean ± SE at 2 h time interval (p = 0.0385), by 5 ± 2.2 at 4 h (p = 0.0133), and by 8 ± 3.0 at 6 h (p = 0.0046).
Self-monitoring blood glucose frequency for intervention participants fell from 3.8 ± 1.4 tests/day mean ± SD (3.8 tests/day median) at baseline to 0.5 ± 1.1 (0.1 median) from the first unblinded sensor wear with full access to sensor glucose data (day 15–31), reducing further to 0.4 ± 1.0 tests/day (0.0 median) by study end (day 208). The overall blood glucose monitoring rate over 6 months was 0.3 ± 0.7, median 0.1 (Fig. 3).
During the treatment phase (day 15 onwards) average sensor-scanning frequency was 8.3 ± 4.4 (mean ± SD) times/day (median 6.8), i.e., double the frequency of blood glucose testing (Fig. 3). There was no significant difference in the number of scans performed by those <65 years and ≥65 years of age [8.1 ± 4.6 (median 6.8) and 8.5 ± 4.1 (median 6.9), respectively, p = 0.6627].
There was no correlation between frequency of sensor scanning and reduced time in hypoglycemia or change to HbA1c. Device use for the intervention group (n = 138) was 88.7 ± 9.2% (defined as the percentage of data collected, assuming continuous device wear for 6 months).
Self-monitoring of blood glucose frequency for control participants was 3.9 ± 1.5 test/day (median 3.9) at baseline and this rate was maintained until study end [3.8 ± 1.9 (median 3.9), Fig. 3]. Control group participants <65 years performed less blood glucose monitoring tests (2.78 ± 1.08 test/day) than those ≥65 years (3.46 ± 0.94), p = 0.0247.
At baseline, 95% of participants used an insulin pen device or syringe for intensive insulin therapy, with the remainder (5%) on CSII (Table 1); 78% used analogue insulin, seven participants from each group (n = 14) utilized human insulin, and 35 participants used both human and analogue insulin (intervention n = 22, control n = 13).
There was no difference detected in total daily dose of insulin, basal, or bolus insulin doses between the two groups. None of the changes in insulin were correlated with the treatment effect on HbA1c or time in hypoglycemia (<3.9 mmol/L [70 mg/dL]).
There was no difference in total daily dose of insulin by study end for intervention participants (from 87.6 ± 44.0 (mean ± SD) to 85.2 ± 39.7 units) compared with controls (from 90.1 ± 40.6 to 87.8 ± 41.5), −0.4 ± 3.75 units mean ± SE (p = 0.9059). Basal insulin was similar for intervention and control participants (−2.3 ± 1.96 units mean ± SE, p = 0.2498). Bolus insulin was similar for intervention and control participants (1.4 ± 2.53 units mean ± SE, p = 0.5856). Similarly, for participants above or below 65 years, there was no difference detected in the total daily dose of insulin (0.7 ± 4.86, p = 0.8871; and −3.3 ± 5.40, p = 0.5403, respectively).
There were no changes in body weight (p = 0.2496) or BMI (p = 0.2668) from baseline for either group.
Total treatment satisfaction score for DTSQ (status versus change) was significantly improved for intervention group participants (13.1 ± 0.50, mean ± SE) compared with controls (9.0 ± 0.72), p < 0.0001. Satisfaction with treatment results using DQoL demonstrated significant improvement for the intervention group (−0.2 ± 0.04, mean ± SE) versus the control group (0.0 ± 0.06), p = 0.0259, for this element of the questionnaire. There were no other significant differences observed in other aspects of DTSQ and DQoL or for the DDS scales (Fig. 4, Supplementary Material p. 10).
User questionnaire results showed intervention participants agreed with positive aspects of the device including use, comfort, and utilization of sensor glucose information (Supplementary Material p. 11).
The system was used for 6 months by intervention participants and worn (blinded) for 4 weeks by control participants (n = 224). In total, serious adverse or adverse events (n = 515) were experienced by 114 (76.5%) intervention and 47 (62.7%) control participants.
There were no serious adverse events related to the device or study procedure. Forty-two serious events were experienced by 16 (10.7%) intervention and 12 (16.0%) control participants.
Four hypoglycemia serious adverse events were experienced by four participants (three intervention and one control) and 57 hypoglycemia adverse events by 10 (7%) intervention and seven (9%) control participants.
None of the severe hypoglycemic episodes [13] or hypoglycemic adverse events were associated with the device.
Three participants (one intervention, two controls) experienced an adverse event leading to withdrawal from the study; none were associated with the device.
Six (4.0%) intervention participants reported nine device-related adverse events (two severe, six moderate, and one mild). These were sensor-adhesive reactions, primarily treated with topical preparations. All were resolved at study exit.
There were no reported events of diabetic ketoacidosis or hyperosmolar hyperglycemic state. Seven cardiac events were reported for four (2.7%) intervention and three (4.0%) control participants (none were considered to be related to study procedures or the device).
Anticipated symptoms refer to those typically expected using a sensor device and equate to symptoms normally experienced with blood glucose finger-stick testing, e.g., pain, bleeding, bruising. There were 158 anticipated sensor insertion site symptoms observed for 41 (27.5%) intervention and 9 (12.0%) control participants. These symptoms were primarily (63%) due to the sensor adhesive (erythema, itching, and rash) and resolved without medical intervention. Adverse events and anticipated symptoms associated with the insertion of the sensor and sensor wear are summarized in Table 3 and Supplementary Material p. 12.