Continuous Glucose Monitoring and the Reality of Metabolic Control in Preschool Children With Type 1 Diabetes Free

After obtaining institutional review board approval for the protocol and parental consent, 10 children <6 years of age with type 1 diabetes for >6 months and followed at The Texas Children’s Hospital Diabetes Care Center were recruited. Subjects were on a twice-daily injection regimen consisting of mixed NPH as the intermediate-acting insulin and lispro (Eli Lilly, Indianapolis, IN) as the rapid-acting insulin. No insulin was administered with lunch; however, two subjects were using an additional morning dose of glargine insulin (Aventis Pharmaceuticals, Bridgewater, NJ) given separately. Subjects were excluded if they were on continuous subcutaneous insulin infusion therapy or medications that alter glucose metabolism.

Before study entry the parents and/or caregivers underwent a review of SMBG, insulin adjustment, and carbohydrate counting. Every subject underwent two 72-h periods of monitoring 1 month apart using the CGMS. HbA_1c was measured during each monitoring visit using the Bayer (Tarrytown, NY) DCA 2000 instrument with a nondiabetic range of 4.3–6.3%. The DCA 2000 is certified by the National Glycohemoglobin Standardization Program with HbA_1c results comparable with those reported in the DCCT, where relationships to mean blood glucose and risk for vascular complications have been established. Insulin management was done based on the values obtained from self-monitoring, reflecting the standard practice of our clinic.

The CGMS consists of a glucose oxidase–based sensor inserted subcutaneously in an abdominal site and attached via a cable to a monitor that can be worn much like an insulin pump. The chemical reaction creates a current measurable by the device that in turn generates a corresponding glucose value. The monitor takes a reading every 10 s and displays an average every 5 min for a total of 288 readings per day. CGMS is calibrated by entering SMBG readings four times per day. The device allows patients to record their meals and insulin doses. After completion of the monitoring period the device is downloaded and analyzed. In the present study the analysis was done using CGMS Solutions 3.0b software. Parents and investigators were blinded to the results until the child had completed the entire study.

The CGMS was inserted in each child by a certified diabetes educator study nurse at Texas Children’s Diabetes Care Center. Parents and/or caregivers were instructed to document at least four daily meter blood glucose measurements for calibration. All subjects used the FreeStyle glucometer (TheraSense, Alameda, CA) for SMBG. Caregivers were instructed not to use alternate sites for testing. They also kept records of mealtime and insulin doses.

The data from the CGMS were downloaded and analyzed to identify the number of hypoglycemic episodes and their duration and distribution (daytime versus nighttime). Hypoglycemia was defined as a CGMS or a SMBG glucose value <60 mg/dl. Hyperglycemia was defined as a glucose value >150 mg/dl. We used 80–150 mg/dl as an acceptable range for adequate control because achieving a target glucose range of 80–120 mg/dl is difficult in this very young age-group. Comparison was made between the two sets of data generated by the CGMS and SMBG.

Analysis to evaluate the agreement between the CGMS and SMBG was done using the following criteria defined by Medtronic MiniMed (20). For optimal accuracy, MiniMed recommends a minimal correlation of 0.79 between the two sets of data with a mean absolute error <28%. The company cautions the user to employ clinical judgment in interpreting any result that does not satisfy the above criteria.

To evaluate the ability of the CGMS to correctly detect hypoglycemic episodes, we assigned one of four categories to each paired CGMS measurement with hypoglycemia (blood glucose <60 mg/dl) set as the positive event: 1) true positive (TP) when both CGMS and meter report a value <60 mg/dl; 2) false-positive (FP) when the CGMS reports a value <60 mg/dl, while the meter reports a value >60 mg/dl; 3) true negative (TN) when both modalities report a value >60 mg/dl; or 4) false negative (FN) when the CGMS reports a value >60 mg/dl when the meter reads <60 mg/dl. Sensitivity [TP/(TP + FN)], specificity [TN/(TN + FP)], positive predictive value [TP/(TP + FP)], and negative predictive value [TN/(TN + FN)] were calculated.

We also analyzed the data using the Clarke error grid (21). This method evaluates whether the difference between glucose values obtained through two different methods is clinically relevant. For example, a difference of 30 mg/dl between two methods is more relevant when the blood glucose is 50 mg/dl versus when it is 300 mg/dl. The analysis is done as follows. The two sets of values generated by the CGMS and SMBG are plotted against each other. Then the graph is divided into five zones. Zone A includes values that differ by no more than 20% or are in the hypoglycemic range (<70 mg/dl) for both methods. Zone B includes values that differ by >20% but do not lead to significantly different treatment decisions. Zone C includes values that would lead to overcorrection of acceptable blood glucose. Zone D represents values that lead to failure to respond to hypo- or hyperglycemia. Zone E includes dangerous treatment decisions, i.e., giving insulin when the patient is hypoglycemic or vice versa. Each zone is further subdivided into an upper and a lower zone. Clinically acceptable correlation between two methods occurs when >95% of the values fall within zones A and B.

Results are expressed as means ± SD. Due to the extreme variability in the frequency of hypoglycemic episodes, these particular results were expressed as median (range). The analysis was carried out using Wilcoxon’s rank test to compare the 6-month average HbA_1c before study entry to the HbA_1c at study entry as well as the number and duration of nighttime versus daytime hypoglycemic episodes. Statistical significance was defined as P < 0.05. The analysis was carried out using SPSS software version 12.0 (Chicago, IL).

Our study group consisted of 10 patients (8 females and 2 males) with type 1 diabetes. The average age at study entry was 3.65 ± 1.34 years (range 1.75–5.67 years) with disease duration of 1.88 ± 1.38 years (1.00–4.08 years). The average HbA_1c at study entry was 8.1 ± 0.8%. There was no difference between the HbA_1c (8.6 ± 0.8 vs. 8.5 ± 0.6%, P = 0.62), insulin dose (0.7 ± 0.1 vs. 0.7 ± 0.2 units · kg^–1 · day^–1, P = 0.57), or BMI (16.4 ± 1.6 vs.16.1 ± 1.3 kg/m², P = 0.14) recorded during the first and second monitoring episodes.

A total of 22 sensors were inserted; one patient was excluded from the study after his first sensing episode due to failure to comply with the study protocol. Three monitoring periods were repeated due to several episodes of disconnection between the sensor and the cable, making the total fraction of nonsatisfactory periods 14%. The total number of monitoring periods analyzed was thus 19, with a median duration of 70 h (range 49–90 h). The total number of CGMS readings was 15,102, with 379 paired CGMS/SMBG readings. The average number of measurements was 7 ± 1 per subject per day. The overall tolerance to the device was excellent, with no reported local irritation or infection. The device did not interfere with the care of the child and was well accepted by all children and families.

We found good agreement between the two methods using Medtronic criteria. The mean correlation coefficient was 0.94 ± 0.04. The mean absolute error for all monitoring episodes was 12.4 ± 5.1%. Accuracy analysis revealed that the sensitivity of the CGMS to detect hypoglycemia was 70% and the specificity 99%. The positive predictive value of a hypoglycemic reading obtained via CGMS was 90%, while the negative predictive value was 97% (Fig. 1A). There was good agreement between the two methods as evaluated by the Clarke error grid, with 97.4% of the values falling in the A and B regions (Fig. 1B).

The total number of hypoglycemic episodes was 82, with a median number of 3 episodes per subject per monitoring period (0–12). We also analyzed the night-day distribution of hypoglycemia. Nighttime was defined as the period from 2200 to 0700, a period during which most preschool children are expected to be asleep (Table 1).

We found that 51% of the hypoglycemic episodes (n = 42) out of the total of 82 detected by the CGMS did not have a concomitant finger-stick value. Assuming that the CGMS did not detect all of the hypoglycemic episodes and mislabeled some normal values as hypoglycemia and by using the accuracy data generated above, we calculated an estimate of the number of true episodes as being 97 rather than 82. The number of episodes that both CGMS and SMBG would detect or the true positives would be 64. Even with this reduced number, the average number of episodes is 1.1 episode · subject^–1 · 24 h^–1. Because the median duration of an episode was 1 h, each subject would still have a total of 60 min of hypoglycemia per 24 h.

There was no difference in the mean CGMS (198 ± 38 vs. 200 ± 45 mg/dl, P = 0.8) or SMBG (181 ± 27 vs. 187 ± 38 mg/dl, P = 0.6) glucose values between the first and second monitoring episodes. We looked for a possible effect that the CGMS itself might have on glycemic control. However, when we compared the first and second monitoring periods, we found no difference between the total duration of hypoglycemia (median 4.7 vs. 4.2 h/monitoring period, P = 0.4) or number of episodes (median 3 vs. 3 episodes/monitoring period, P = 0.6). The correlation between the average HbA_1c and the average glucose value recorded by the CGMS for each subject was 0.93 (P < 0.0001). However, we found no correlation between either HbA_1c or average glucose on one hand and the number or duration of either hypo- or hyperglycemic episodes on the other hand. This was true whether looking at nighttime, daytime, or overall results.

The median number of hyperglycemic episodes was 3 episodes · subject^–1 · 24 h^–1 (2–4). We found that subjects spent 7% (range 0–22%) of a 24-h period with a glucose value <60 mg/dl, 64% (35–83%) with a glucose value >150 mg/dl, and only 24% (10–49%) of their time with acceptable values (80–150 mg/dl) (Fig. 2).

Furthermore, 72% of all the glucose values exceeding 150 mg/dl were >200 mg/dl, and 55% of the glucose values below 100 mg/dl were <60 mg/dl. We also found that of the 136 documented meals times, 94 ± 7% of them had a glucose value >200 mg/dl in the 2-h period following the meal. This was distributed as breakfast (93%), lunch (98%), and dinner (92%).

Glucose monitoring remains the cornerstone of evaluating the efficacy of therapy in subjects with diabetes (22). Even if faithfully done, SMBG is a snapshot of the real picture with potential to miss fluctuations. With this in mind, we designed our study to assess other means of evaluating current therapeutic regimens in preschool children.

There are several limitations to our study, including the number of subjects. Larger studies of preschool children are needed to confirm our results. Although lacking a control group with normal patients wearing the device, we performed accuracy analysis to detect the rate of false-positives and false negatives. We did not obtain venous blood glucose to compare SMBG and CGMS readings to a laboratory reference method for the following reasons. 1) Obtaining frequent venous samples is difficult to do at home and generally requires a hospital admission. We felt that the hospital stay would not be a true reflection of what really happens at home, particularly for this very young age-group. 2) The FreeStyle meter used in our study has a very good correlation with venous samples, with 96.7% of the values falling within a 10% difference of laboratory reference methods (23).

In a study of children on continuous subcutaneous insulin therapy, Boland et al. (10) reported similar results with frequent hypoglycemic events and significant postprandial hyperglycemia; however, their study subjects were older children and adolescents (mean age 11.6 ± 4.6 years). They reported the same number of hypoglycemic episodes (0.9 vs. 1.0 episodes · subject^–1 · day^–1) but a higher duration of nighttime hypoglycemia (117 vs. 72 min · subject^–1 · night^–1) than in our study. This is somewhat surprising, as we would have expected our younger subjects on twice-daily injections to have longer and more frequent hypoglycemic episodes than older children on continuous subcutaneous insulin infusion. We found a good correlation between the CGMS and SMBG when MiniMed criteria were used. However, correlation is expected to be good in any two methods measuring the same entity (i.e., glucose); we therefore used the Clarke error grid to test the clinical relevance of the difference when present. The grid revealed differences in 2.6% of the time. Almost all children had several hypoglycemic episodes during each of the 72-h periods. The CGMS had a low sensitivity and mislabeled 30% of hypoglycemic episodes as normal values; however, only 18% of those mislabeled episodes had CGMS values >80 mg/dl. We did not find a difference between the frequencies of nighttime versus daytime hypoglycemia, although nighttime episodes lasted longer. This is likely due to the fact that nighttime episodes are frequently undetected. We also found that most children spend significant portions of their time with unacceptable glucose values (either hypo- or hyperglycemia). Adequate control was achieved only 24% of the time. Even after correction by using the accuracy data, each subject still had a median of 1 h of hypoglycemia in a 24-h period.

Our results show that twice-daily subcutaneous insulin injection in preschool children with type 1 diabetes fails to achieve glycemic control. In fact, treatment often overshoots, with a high risk of hypoglycemia (55%) when glucose concentrations fall to <100 mg/dl, or undershoots, with hyperglycemia (>200 mg/dl) occurring 72% of the time when concentrations reach 150 mg/dl. Hyperglycemia is another dominant feature of glucose control in this age-group. In fact, 64% of their time was spent with a glucose value >150 mg/dl, and postprandial hyperglycemia was almost a universal feature (94%). This is a surprising fact because all of our subjects were on lispro insulin.

In conclusion, our study shows that preschool children with type 1 diabetes have suboptimal control on twice-daily insulin injection therapy, with frequent and prolonged hypoglycemia, especially at night, lasting up to 1 h per day. The CGMS reveals trends in blood glucose that current means of evaluating the adequacy of therapy (including HbA_1c and SMBG) are not able to discern. We also conclude that the HbA_1c is not an ideal indicator of adequate control in this age-group but merely reflects an average of extreme highs and lows. In light of the frequency of postprandial hyperglycemia, it might be beneficial to measure postmeal glucose values in this age-group. Further studies are needed to investigate whether basing clinical decision making on the CGMS results in better control in preschool children. Given the serious complications associated with hyperglycemia and neurological risks of hypoglycemia, especially in younger patients, failure of current therapeutic regimens using twice-daily insulin injections to achieve reasonable control is an area for concern. It is necessary to study newer types of insulin or insulin pump therapy in preschool children with type 1 diabetes in order to reduce daily glycemic excursions.

G.S.J. is supported through a clinical research grant from the Glaser Pediatric Research Network, part of the Elizabeth Glaser Pediatric AIDS Foundation. This study was funded through an unrestricted research grant from Medtronic MiniMed and Novo Nordisk and was presented in the form of an abstract at the Endocrine Society meeting, Philadephia, Pennsylvania, 19–22 June 2003.

The authors are thankful for Dr. William L. Clark’s assistance in providing the software for the Clarke error grid.