DIABETES TECHNOLOGY & THERAPEUTICS Volume 17, Number 6, 2015 ª Mary Ann Liebert, Inc. DOI: 10.1089/dia.2015.0155

COMMENTARY

Glycemic Variability: Challenges in Interpretation David Rodbard, MD

an Dijk et al.1 report in the current issue of Diabetes Technology & Therapeutics what they interpret to be a modest reduction in glycemic variability (GV) in subjects with type 1 diabetes mellitus receiving long-term continuous intraperitoneal insulin infusion (CIPII) compared with a control group of subjects with type 1 diabetes mellitus receiving subcutaneous (SC) insulin therapy (administered either using multiple daily injections or continuous SC insulin infusion [CSII]). There was a relative 5% reduction in the percentage coefficient of variation (%CV) (where %CV = 100 · [SD/mean glucose]) in the CIPII group compared with the SC group.1 However, there were no significant differences in terms of SD, mean amplitude of glycemic excursions (MAGE), or mean of the daily differences (MODD). It is important that the CIPII group had a significantly higher glycated hemoglobin (HbA1c) level and mean glucose level using either self-monitoring of blood glucose (SMBG) or continuous glucose monitoring (CGM) than the SC group. Table 1 of the present commentary focuses on a subset of the data in Table 2 of Van Dijk et al.1 We show the mean baseline HbA1c for each group and the mean glucose level, SD of glucose level, and %CV for each group using both the SMBG data and the CGM data. Because there were no meaningful changes in these parameters during the period of observation, the present author used the average value obtained at the beginning and end of the 26-week period. Unlike Table 2 of Van Dijk et al.,1 Table 1 in the present report does not use correction for baseline values because those data are not available to the present author. This may account for minor differences in the values reported. Table 1 is fully consistent with the presentation of Van Dijk et al.1: the HbA1c level is higher by a relative 8.1% in the CIPII group than in the CSII group, and the mean glucose level is relatively 10.2% and 14.6% higher, based on SMBG and CGM data, respectively. However, the %CV is smaller in the CIPII group by 6.8% and 13.5% compared with the CSII group, using SMBG and CGM data, respectively. In contrast, the average within-subject SD was virtually identical for the two groups. By definition, %CV is the ratio of the SD to the mean (using a factor of 100 to express the result as a percentage). Thus, if the mean glucose level is 10% higher in the CIPII group, then the %CV would be expected to be approximately 10% smaller in that group, even though the SD is the same as for the SC group. In view of the fact that the SD, MAGE, and

V

MODD were essentially identical for the two groups of subjects, an alternative interpretation of the data of Van Dijk et al.1 would be as follows: the mean glucose level was higher in the CIPII group, and this resulted in the lower %CV for that group despite the fact that GV in terms of SD, MAGE, and MODD was essentially unchanged. This would be a very different interpretation than the one provided by Van Dijk et al.1 The findings of Van Dijk et al.1 were also dramatically different from three previous studies2–4 in which use of CIPII resulted in a significantly lower mean glucose level, significantly lower SD, and smaller %CV than in control subjects using SC insulin delivery. The present author recalculated the %CV for the previous studies2–4 using a weighted slope for SD vs. mean glucose forced through the origin: 100 · [average SD]/[average mean glucose]. How can one reconcile the various measures of GV? The present author has been a proponent of the use of %CV.5 The %CV often will simplify the interpretation of GV data: in contrast to the SD, which usually has a nearly direct proportionality with the mean glucose level,5 %CV is relatively constant irrespective of mean glucose level and irrespective of HbA1c level5 for a defined, relatively homogeneous patient population. However, there is no guarantee that %CV will be absolutely constant, irrespective of HbA1c or mean glucose level, especially when dealing with two or more patient populations (as in the case of Van Dijk et al.1) or for markedly heterogeneous patient populations. The CIPII subjects studied by Van Dijk et al.1 were a highly selected group—previously selected for use of CIPII because of their refractory poor glycemic control. In contrast, the previous studies of CIPII2–4 assigned treatment using a crossover design2,3 or randomization to individuals from an identified population.4 Use of the %CV is based on two implicit assumptions: (1) that there is a direct linear relationship between SD and mean glucose level and (2) that the linear relationship passes through the origin (0,0) (i.e., the intercept is 0), such that there is a direct proportionality between SD and mean glucose level. If either of these assumptions is violated, then %CV will vary systematically with glucose level. Rather than assuming that %CV can be used as applied to the CIPII group and to both SC groups (multiple daily injections and CSII), one should examine the relationship of SD to mean glucose level for each of the three groups. That might indicate whether there is a larger SD in one group versus another for any

Biomedical Informatics Consultants LLC, Potomac, Maryland.

370

GV: CHALLENGES IN INTERPRETATION

371

Table 1. Comparison of Baseline Glycated Hemoglobin and of Mean Glucose Level, SD of Glucose Level, and Percentage Coefficient of Variation of Glucose Using Self-Monitoring of Blood Glucose and Continuous Glucose Monitoring for Subjects Using Peritoneal (Continuous Intraperitoneal Insulin Infusion) and Subcutaneous (Either Multiple Daily Injections or Continuous Subcutaneous Insulin Infusion) Insulin Delivery

CIPII Baseline HbA1c (mmol/mol)b SMBGc Mean glucose (mmol/L) SD (mmol/L) %CVd CGMc Mean glucose (mmol/L) SD (mmol/L) %CVd

67

SC (MDI and CSII) 62

% difference in values for CIPII vs. SCa + 8.06

10.3 9.35 4.3 4.25 41.6% 44.65%

+ 10.16 + 1.18 - 6.83

10.6 9.25 3.8 3.85 36.6% 42.3%

+ 14.59 - 1.30 - 13.48

Data are from Table 2 of Van Dijk et al.1 a Percentage difference expressed using the values for the subcutaneous (SC) group as the denominator. b Baseline glycated hemoglobin (HbA1c) is considered relevant as a marker of mean glucose level throughout the study, in view of the fact that there was no significant change in mean glucose level during the study, as measured by self-monitoring of blood glucose (SMBG) and continuous glucose monitoring (CGM). c Values from SMBG and CGM are the averages of values at the beginning and end of the 26-week duration of the study, uncorrected for baseline characteristics. d Percentage coefficient of variation (%CV) is dimensionless. CIPII, continuous intraperitoneal insulin infusion; CSII, continuous SC insulin infusion; MDI, multiple daily injections.

specified level of mean glucose. One cannot assume that there will be a direct proportionality of SD to mean glucose level for the various groups. This kind of analysis might suggest that at any specified level of glucose there is a smaller SD for the CIPII group compared with the SC groups. It might also provide some additional insights. The %CV is a convenient measure that can potentially be used to compare the quality of glycemic control (in terms of GV) among different subjects within a defined patient population.5 One can use the quartiles for the %CV (25th, 50th, and 75th percentiles) to subdivide a patient population into four categories corresponding to excellent, good, fair, or poor GV.5 The %CV is also modestly correlated with the risk of hypoglycemia.6,7 In one series, there was no hypoglycemia observed if %CV was less than 20%.5 Monnier et al.8 have also examined the relationship among risk of hypoglycemia, mean glucose level, and SD of glucose level using multiple linear regression. Van Dijk et al.1 raise the question as to whether differences they observed in %CV for glucose level based on SMBG and CGM might be related to changes in postprandial patterns. It would indeed be interesting to evaluate whether changes in level and variability of glucose were related to meals (as might have been expected if there was accelerated absorption of insulin analogs from the intraperitoneal cavity) or to changes that are presumably related to the effect of basal insulin levels. It should be possible to examine this by anal-

ysis of the postprandial excursions based on the SMBG data, assuming that the glucose measurements were obtained in a systematic manner with respect to meals. For the CGM data, one might consider the following approaches: 1. Evaluate the SD and %CV for the nocturnal period (e.g., 11:00 p.m.–5:00 a.m.) as representative of variability unrelated or minimally related to meals. Evaluate SD and %CV for additional time segments that can be coordinated with meals (e.g., 5:00 a.m.–11:00 a.m., 11:00 a.m.–5:00 p.m., and 5:00 p.m.–11:00 p.m., as roughly indicative of variability related to breakfast, lunch, and dinner, respectively). This could provide a measure of meal-related variability even if the exact timing of meals had not been recorded. Alternatively, one can calculate the ‘‘within series’’ SD of glucose for a time segment of any specified number of hours on a rolling or running basis and thus evaluate in relationship to time of day. This type of measure of GV has been designated as SDws h, where h indicates the duration of the time segment (e.g., h = 3).9 2. With CIPII or CSII, we can usually make inferences regarding the timing of meals based on time of use of a bolus calculator, time of administration of the presumptive premeal boluses, and (when available) information regarding anticipated carbohydrate consumption. In turn, one can calculate the average glycemic profile above baseline, synchronized with respect to time of initiation of the meal (more precisely, with respect to time of insulin bolus). Van Dijk et al.,1 and others, might consider alternative study designs. For subjects previously using CIPII, one could suspend the intraperitoneal insulin pump and initiate use of CSII and then monitor the SMBG and CGM data, using a crossover study design similar to that of Haardt et al.3 By making comparisons within subjects (as opposed to between groups), the sensitivity of the analysis can be greatly enhanced because it eliminates the large contribution of between-subject variability (e.g., Haardt et al.3). In conclusion, the relative 5% reduction in GV (%CV) observed after corrections1 and the uncorrected 6.8% (SMBG) to 13.5% (CGM) reductions in %CV recalculated here (Table 1) might conceivably be due to reduced GV or might simply be due to the higher mean glucose level in the CIPII group in the study population of Van Dijk et al.,1 combined with the use of %CV as the sole parameter for GV. The lack of significant change in SD, MAGE, or MODD suggests that the principal effect is simply related to the difference in mean glucose level (and mean HbA1c level). Additional randomized studies using each subject as their own control would be desirable, with subanalyses of mealrelated and non–meal-related portions of the day. Author Disclosure Statement

D.R. is Chief Scientific Officer of Biomedical Informatics Consultants LLC. References

1. Van Dijk PR, Groenier KH, DeVries JH, Gans ROB, Kleefstra N, Bilo HJG, Logtenberg SJJ: Continuous intraperitoneal insulin infusion versus subcutaneous insulin

372

2.

3.

4.

5. 6.

therapy in the treatment of type 1 diabetes: effects on glycemic variability. Diabetes Technol Ther 2015;17:379–384. Catargi B, Meyer L, Melki V, Renard E, Jeandidier N; EVADIAC Study Group: Comparison of blood glucose stability and HbA1C between implantable insulin pumps using U400 HOE 21PH insulin and external pumps using lispro in type 1 diabetic patients: a pilot study. Diabetes Metab 2002;28:133–137. Haardt MJ, Selam JL, Slama G, Bethoux JP, Dorange C, Mace B, Ramaniche ML, Bruzzo F: A cost-benefit comparison of intensive diabetes management with implantable pumps versus multiple subcutaneous injections in patients with type I diabetes. Diabetes Care 1994;17:847–851. Selam JL, Raccah D, Jean-Didier N, Lozano JL, Waxman K, Charles MA: Randomized comparison of metabolic control achieved by intraperitoneal insulin infusion with implantable pumps versus intensive subcutaneous insulin therapy in type I diabetic patients. Diabetes Care 1992;15:53–58. Rodbard D: Clinical interpretation of indices of quality of glycemic control and glycemic variability. Postgrad Med 2011;123:107–118. Rodbard D: Hypo- and hyperglycemia in relation to the mean, standard deviation, coefficient of variation, and nature

RODBARD

of the glucose distribution. Diabetes Technol Ther 2012;14: 868–876. 7. Qu Y, Jacober SJ, Zhang Q, Wolka LL, DeVries JH: Rate of hypoglycemia in insulin-treated patients with type 2 diabetes can be predicted from glycemic variability data. Diabetes Technol Ther 2012;14:1008–1012. 8. Monnier L, Wojtusciszyn A, Colette C, Owens D: The contribution of glucose variability to asymptomatic hypoglycemia in persons with type 2 diabetes. Diabetes Technol Ther 2011;13:813–818. 9. Rodbard D: New and improved methods to characterize glycemic variability using continuous glucose monitoring. Diabetes Technol Ther 2009;11:551–65.

Address correspondence to: David Rodbard, MD Biomedical Informatics Consultants LLC 10113 Bentcross Drive Potomac MD 20854 E-mail: [email protected]

Glycemic variability: challenges in interpretation.

Glycemic variability: challenges in interpretation. - PDF Download Free
94KB Sizes 3 Downloads 11 Views