Using Continuous Glucose Monitoring to Measure the Frequency of Low Glucose Values When Using Biphasic Insulin Aspart 30 Compared With Biphasic Human Insulin 30: A double-blind crossover study in individuals with type 2 diabetes

A total of 160 male and female patients with type 2 diabetes (BMI <40 kg/m², A1C <9.5%) pretreated with insulin for at least 6 months were recruited to this randomized, double-blind, two-period, crossover trial from 18 centers in the U.K. All oral antidiabetes drugs were stopped at study entry. Patients with a history of severe hypoglycemia or hypoglycemia unawareness were not specifically excluded. Baseline characteristics are shown in Table 1. The trial was performed in accordance with the Declaration of Helsinki and good clinical practice. Patients gave written informed consent before any trial-related activities began.

During an 8-week run-in, existing insulin therapy was optimized to achieve preprandial blood glucose levels of 5–7 mmol/l. Patients with A1C 6.5–8.5% were then randomized to receive BIAsp 30 (n = 70), 100 units/ml, or BHI 30 (n = 75), 100 units/ml, contained in 3-ml Penfill administered dose using NovoPen 3. Randomization was carried out by allocating the next available randomization number to each patient and allocating treatment sequence using a computer-generated code with a block size of four. Randomization was undertaken centrally with the randomization number containing information about the treatment for each patient sealed throughout the trial. To maintain the double-blinded trial design, both types of insulin were injected immediately before breakfast and evening meals (it was deemed unsafe to allow any injection-meal interval for BIAsp 30 due to the rapid action of insulin aspart). After the first 16-week treatment period, patients switched to the alternative insulin for a further 16 weeks, with no washout period. Throughout the trial, participants were free to adjust the dose (in discussion with research staff) according to individual needs once per week. No specific instructions were given with regard to evening exercise or snacks.

Interstitial glucose (IG) was recorded with the MiniMed Continuous Glucose Monitoring System (CGMS; Medtronic Diabetes, Northridge, CA) over four 72-h periods (two for each treatment period, halfway through, and at the end of each treatment period). CGMS was performed on normal weekdays, and patients were advised to not partake in strenuous exercise during the recording period.

The monitor recorded IG levels every 10 s then stored a smoothed average over 5 min. The range of IG detection was 2.2–22 mmol/l. The primary end point was the frequency of readings <3.5 mmol/l (IG_3.5). IG readings <2.5 mmol/l (IG_2.5) were also compared since studies suggest that glucose <2.8 mmol/l is likely to reflect clinically relevant hypoglycemia (3,4).

The 24-h data were categorized into total, daytime (0600–0000 h), and nighttime (0000–0600 h). Low IG readings were also grouped into episodes, defined as a set of continuous IG readings <3.5 mmol/l (or <2.5), allowing up to two consecutive readings above the threshold within the same episode. If a following episode started within 1 h of the start of the previous episode, they were combined and recorded as one episode.

Subjects recorded self-monitored blood glucose (SMBG) using a LifeScan One Touch meter (High Wycombe, U.K.), to allow regular insulin dose titration. Targets for fasting and preprandial blood glucose were 5–7 mmol/l. Measurements were recorded in patients’ diaries but only transferred to the case report form in the event of hypoglycemia. Episodes of hypoglycemia were classed as “minor ” if the patient was able to self-treat and blood glucose was <2.8 mmol/l (or as “symptoms only” if blood glucose was ≥2.8 mmol/l or not measured) or as “major” if patients were unable to self-treat.

Adverse events were defined as any undesirable medical event occurring during the trial period. Events were classed as serious if they resulted in death, were life threatening, or caused (or prolonged) hospitalization.

A1C was assessed at screening and randomization and at the end of the two treatment periods.

The World Health Organization Diabetes Treatment Satisfaction Questionnaire (WHO DTSQ) was completed by all subjects at visits 3, 7, and 11. The questionnaire consists of six items rated on a 7-point Likert scale, which can be summarized in an overall treatment satisfaction score from 0 to 36, where a higher score indicates greater satisfaction.

A1C and CGMS data were analyzed using a mixed-model ANOVA appropriate to the crossover design using the PROC MIXED procedure in SAS (SAS, Cary, NC). This adjusts for differences in characteristics (such as age and weight) between the groups assigned to each type of insulin. Where the CGMS data were not normally distributed, data were transformed by adding one and using a logarithmic transformation or, alternatively, using a rank-based transformation or the Wilcoxon's signed-rank test. An intention-to-treat (ITT) analysis was undertaken with the ITT population, defined as all randomized patients who took study medication and had at least one follow-up value.

Self-reported hypoglycemia was tested using Wilcoxon's signed-rank test; the timing of low IG episodes was tested using the Kruskal-Wallis test, and proportions were tested using Fisher's exact test.

Missing A1C values at visit 2 were replaced by the visit 1 value for the same patient, using the last-observation-carried-forward method. No other missing data were replaced in the study.

The trial profile is shown in Fig. 1. A1C at follow-up was not obtained for 13 patients, leaving 147 in the ITT population. The CGMS population was based on data collected from 145 patients (CGMS data were incomplete for 15 patients). Some minor differences in mean height, weight, and BMI were seen between the two groups based on receiving BIAsp 30 or BHI 30 first because of the higher proportion of men in the latter group (Table 1).

Mean ± SD total daily insulin doses at the end of the treatment periods were 68.8 ± 37.8 units for BIAsp 30 (median 59.0 [range 6.0–238.7]) and 66.6 ± 34.6 units for BHI 30 (58.0 [11.3–240.0]).

Morning doses were larger than evening doses for both treatments: morning dose 36.0 ± 20.2 units for BIAsp 30 (median 32.0 [range 4–120]) vs. 34.6 ± 18.9 units for BHI 30 (32.0 [5–130]) and evening dose 33.0 ± 19.0 units (28.0 [2.0–118.7]) vs. 32.0 ± 17.4 units (28.0 [4.0–110.0]) for BIAsp 30 and BHI 30, respectively.

The mean total number of IG readings recorded per subject was 1,477 during BIAsp 30 treatment and 1,468 with BHI 30 treatment.

At least one episode of IG_3.5 was recorded (at any time) by 82% of patients during each treatment period (BIAsp 30 and BHI 30; P = 1.0). For IG_2.5, the occurrence was 46 and 54%, respectively (P = 0.28). The percentages of patients that recorded at least one low IG episode (IG_3.5 or IG_2.5) during daytime and nighttime periods were as follows: 1) daytime occurrence: IG_3.5 73% with BIAsp 30 and 70% with BHI 30 (P = 0.60); IG_2.5 41% during each treatment (P = 0.10); and 2) nighttime occurrence: IG_3.5 51% with BIAsp 30 and 66% with BHI 30 (P = 0.015); IG_2.5 25 and 37%, respectively (P = 0.039).

There were no significant differences in the total or daytime number of IG_3.5 episodes per patient per week between the two treatments arms, but nighttime IG_3.5 episodes were significantly less frequent with BIAsp 30 than with BHI 30.

Total median values were 3.0 (range 0–16) vs. 3.0 (0–20) for BIAsp 30 vs. BHI 30, respectively, and means were 3.76 ± 3.60 vs. 3.93 ± 3.64, respectively (P = 0.62). Daytime median values were 2.0 (0–13) vs. 2.0 (0–18) for BIAsp 30 vs. BHI 30, respectively, and means were 2.58 ± 2.79 vs. 2.36 ± 2.67 (P = 0.32). Nighttime median values were 1.0 (0–8) vs. 1.0 (0–7) for BIAsp 30 and BHI 30, respectively, and means were 1.18 ± 1.56 vs. 1.62 ± 1.71 (P = 0.011).

For IG_2.5 episodes, there were no statistically significant differences between treatments (total, day, or night; data not shown). There were also no significant differences in duration of IG_3.5 or IG_2.5 episodes (total, day, or night) between BIAsp 30 and BHI 30, although nighttime episodes tended to be longer than daytime episodes (data not shown).

The hourly frequency of IG_3.5 over 24 h is shown in Fig. 2A. The nocturnal frequency was higher than the daytime frequency for both treatments. BHI 30 treatment resulted in a peak frequency of IG_3.5 at 0500–0600 h compared with 0200–0300 h for BIAsp 30. Both treatments showed a small peak at lunchtime. The overall difference in timing of IG_3.5 was statistically significant (P < 0.001). A similar pattern was seen for IG_2.5, but data were 70% less frequent.

The total percentage of time spent with IG <3.5 or <2.5 mmol/l tended to be lower for BIAsp 30 than for BHI 30, but differences were not statistically significant. In the daytime there were no significant differences between treatments for IG_3.5 or IG_2.5. At nighttime, the percentage of time spent with IG_3.5 or IG_2.5 was significantly lower for BIAsp 30 than for BHI 30.

The percentages of patients who reported any episodes of minor hypoglycemia were 90 and 84% for BIAsp 30 and BHI 30 therapy, respectively. Two episodes of major hypoglycemia were reported during BIAsp 30 treatment (one each for two patients) compared with seven episodes (from five patients) during BHI 30 treatment (too few episodes to allow statistical analysis).

Mean total and daytime rates of hypoglycemia were similar for BIAsp 30 and BHI 30. Rates of nocturnal hypoglycemia were significantly lower with BIAsp 30 (1.5 ± 4.54) compared with BHI 30 treatment (3.8 ± 8.0) episodes per patient per year, (P = 0.002).

The hourly frequency of self-reported hypoglycemia over 24 h is shown in Fig. 2B. The highest peak for both treatments occurred at lunchtime, with a second peak observed later in the day, although this tended to be earlier with BHI 30 than with BIAsp 30. Most symptomatic nighttime episodes during treatment with BHI 30 occurred between 0300 and 0700 h and were more frequent than those experienced during treatment with BIAsp 30, which tended to occur slightly earlier.

While receiving BIAsp 30, 58% of patients reported a total of 189 adverse events, while 56% of patients reported 217 events while receiving BHI 30. Only 4 and 6% of these were serious for BIAsp 30 and BHI 30 treatment, respectively.

Following the run-in period, the overall mean A1C was 7.46%. After BIAsp 30 treatment, patients achieved a mean A1C of 7.28% compared with 7.22% after BHI 30. The treatment difference of 0.06% (BIAsp 30 − BHI 30) was not statistically significant (95% CI [−0.04 to 0.17], P = 0.21).

There were no between-treatment differences in overall Diabetes Treatment Satisfaction Questionnaire scores (mean 30.60 ± 5.84 vs. 30.95 ± 5.01 for BIAsp 30 and BHI 30, respectively; treatment difference −0.46, P = 0.25).

CGMS is a useful tool for assessing daily glucose fluctuations (5) but has certain limitations: It measures glucose concentrations in the extracellular interstitium rather than in the intravascular space. The relationship between glucose concentrations in these two compartments is not straightforward and may alter according to physiological variation in insulin concentration and glucose uptake, utilization, and elimination (6,7). These limitations are countered by the ability to record continuous glucose data and detect unrecognized hypoglycemic episodes (8–13). Most studies to date have involved patients with type 1 diabetes, and there is relatively little information describing CGMS in type 2 diabetic patients.

In the present study, CGMS demonstrated that rates of IG <3.5 and <2.5 mmol/l at night were approximately double those during the day in patients with type 2 diabetes treated with premixed insulins BIAsp 30 and BHI 30. This is contrary to the general perception, based on self-reported data (14,15), that nocturnal hypoglycemia is uncommon in type 2 diabetic patients. Indeed, we also found that in contrast to the CGMS data, self-reported episodes were highest during the day. The majority of nocturnal episodes were therefore unrecognized, although it is unsurprising that patients fail to wake when mildly hypoglycemic at night. There is evidence in type 1 diabetic patients that asymptomatic nocturnal hypoglycemia may induce hypoglycemia unawareness (16), but the clinical relevance of our findings in type 2 diabetic subjects is uncertain.

Generally, the distribution of self-reported episodes of hypoglycemia corroborates the CGMS data, increasing our confidence that these findings are robust. The shapes of the two frequency profiles were similar, with both identifying a peak in occurrence of low glucose at lunchtime. However, self-reported episodes at lunchtime were higher during BIAsp 30 treatment than during BHI 30 treatment (opposite to the pattern at night). Thus, there was no overall difference in the frequency of self-reported hypoglycemia.

Our CGMS data show that twice-daily BIAsp 30 led to significantly fewer nocturnal episodes of IG <3.5 mmol/l than twice-daily BHI 30. Furthermore, the percentage of time patients spent with IG <3.5 and <2.5 mmol/l at night was lower during treatment with BIAsp 30. Since total daily insulin doses were similar for both insulins, differences in the frequencies of low IG readings may reflect differences in insulin kinetics. The faster onset and shorter duration of the rapid-acting analog, insulin aspart, contained in BIAsp 30 (17,18) may explain the difference in the time of the peak frequency of nocturnal low IG readings, which occurred at 0300 h with BIAsp 30 compared with 0600 h for BHI 30. Self-reported, symptomatic, nocturnal hypoglycemia was also significantly lower with BIAsp 30 than with BHI 30, suggesting that these differences are clinically relevant.

The data on the timing of low glucose values in this study may help to guide the dose of premixed insulins containing rapid-acting insulin analogs. Our data suggest that when moving from BHI 30 to BIAsp 30, the prebreakfast dose should be reduced to lower the risk of prelunch hypoglycemia and the predinner dose increased (keeping the overall total insulin dose the same). This should effectively move the ratio of doses toward a 50/50 split, as observed in other recent studies (19–21). Furthermore, because of the observed peak in low glucose values at lunchtime, prelunch rather than pre-dinner blood glucose levels may be safer targets for adjusting the prebreakfast dose of a premixed insulin.

In conclusion, nocturnal low glucose levels in patients with type 2 diabetes using twice-daily insulin may be more frequent than supposed. The use of BIAsp 30 was associated with similar overall rates of low IG and symptomatic hypoglycemia than BHI 30, but with fewer episodes at night, probably reflecting the longer duration of action of regular human insulin relative to insulin aspart. These data suggest that a more aggressive approach to achieving glucose targets, particularly lower fasting glucose, may be safer when using preparations containing fast-acting insulin analogs.

The following investigators participated in the REACH Study: Dr. J. Alcolado, University Hospital of Wales; Dr. J. Dean, Bolton PCT/Bolton Hospital NHS Trust; Dr. M. Fisher, Glasgow Royal Infirmary; Prof. A. Hattersley, University of Exeter; Dr. S. Heller, Northern General Hospital; Prof. P. Home, University of Newcastle; Dr. D. Hopkins, Central Middlesex Hospital; Dr. R. Jones, St. Thomas's Hospital; Dr. J. Lorains, Clatterbridge Hospital; Dr. P. McNally, Leicester Royal Infirmary; Prof. A. Morris, Ninewells Hospital & Medical School; Prof. T. O'Brien, University College Hospital, Galway; Dr. S. Page, QMC, Nottingham; Prof. R. Donnelly and Dr. A. Scott, Derby City General Hospital; Dr. J. Thow, York District Hospital; Dr. J. Vora, Royal Liverpool University Hospital; Dr. S. Gray, St. John's Hospital; and Dr. D. Bowen-Jones, Arrowe Park Hospital.

For editorial assistance, the authors thank Scott Gouveia.