Standardization Versus Customization of Glucose Reporting

Introduction

B ergenstal et al. ¹ have made an important and admirable effort to provide leadership toward standardization of reporting of results from continuous glucose monitoring (CGM) and self-monitoring of blood glucose (SMBG). The ambulatory glucose profile (AGP) as applied to SMBG was developed by Mazze and the present author for SMBG ² and extended to CGM. ³ Kowalski and Dutta, ⁴ Kaufman, ⁵ Garg, ⁶ and the Food and Drug Administration ⁷ have provided valuable commentaries on this effort. The myriad of available reporting software and systems for reporting SMBG and CGM glucose data are so diverse that they create logistical problems and an undue cognitive burden on the physician.

Options for Presentation of the AGP

The AGP may be visualized in several different ways: (1) showing the percentiles (10^th, 25^th, 50^th, 75^th, and 90^th) ¹ ; (2) showing the data points superimposed on the percentiles ¹ ; or (3) showing only the glucose data points. If there are segments of time of day with insufficient data, the percentiles should be suppressed. ² When data are available from only a few days, one can show the glucose data with color coding for each day, with or without superimposition of the percentiles. ¹ In an interactive system, it would be desirable to be able to identify any data point by date, time of day, glucose value, and ancillary data by placing a cursor over the data point. If there are sufficient data (e.g., 4 weeks or more), the AGP can be generated separately for each day of the week; ¹¹ statistical tests can then evaluate which subsets of days of the week display similar patterns.

The color scheme shown in Figure 1 of Bergenstal et al. ¹ should be revised so that each glucose range is represented by a unique color. ^12,13 Color coding should become consistent for all CGM, SMBG, and continuous subcutaneous insulin infusion reports. Usability and cognitive laboratory testing of alternative color schemes ^8,9 may be the best way to achieve a consensus.

Synchronization with Respect to Meals

The AGP has much to offer. ^{1 –3,11,14,15} If the patient eats meals at nearly the same time of day and engages in similar physical activity profiles on all days, then the AGP will perform exceedingly well. Failure to synchronize with respect to meals and physical activity is the major limitation of the AGP by time of day. If time of meals varies considerably, then one may be better served by an “eight-point glucose profile,” where the relationship to meals, bedtime, and 3–4 a.m. is better defined—as often but not always utilized for SMBG data. ^11,14,15 Variation in mealtimes and other aspects of lifestyle can introduce considerable scatter, which leads to widening of the bands for glucose percentiles and results in blunting and blurring of the underlying glucose patterns. One can attempt to reduce the effects of this random scatter by analysis of a longer series of days, but the longer time span will introduce additional heterogeneity. ³ CareLink™ (Medtronic, Northridge, CA) software ¹⁶ provides an option to display glucose patterns synchronized with respect to the initiation of each type of meal.

Hypoglycemia

The hypoglycemia report of the “dashboard” ¹ presents 18 numerical entries describing frequency, severity, and duration. This report could potentially be suppressed if there were no problems with hypoglycemia. Several indices of hypoglycemia are available that provide a weighted or integrated assessment of hypoglycemia, combining the effects of frequency, duration, and severity. ^{17 –23} These indices are more succinct than the numerical values for the distribution shown in the AGP “dashboard” of Bergenstal et al. ¹

It would be desirable to show the risk of hypoglycemia by time of day, as observed empirically and as estimated from a model of the glucose distribution combined with the observed mean and standard deviation (SD). ²⁴

By use of a logarithmic scale for glucose, one can expand the scale in the hypoglycemic region and compress the scale in the hyperglycemic region, thereby facilitating detection of hypoglycemic events. ²⁵ Alternatively, one can use a variety of other nonlinear scales for glucose to reduce the risk that hypoglycemic episodes will be overlooked. For example, one could expand the glucose scale 40–100 mg/dL by a factor of 2 while compressing the scale in the range 200–400 mg/dL by a factor of 2, relative to the scale used for the range 100–200 mg/dL.

An expanded hypoglycemia report can also include the date, time, and glucose level for all major or severe hypoglycemic events and provide an opportunity for the user to record possible explanations when available. ¹⁴

Clarity and Brevity

We seek to achieve clarity and brevity. The Clinical View (line 1 of Fig. 3 of Bergenstal et al. ¹ ) presents 12 numerical entries. The Research View (lines 1 and 2 of Fig. 3 of Bergenstal et al. ¹ ) includes 35 numerical entries. Twenty-five more numerical entries present the normal ranges. These 60 entries may already be enough data to generate “information overload” and be intimidating for some potential users.

Other features of the “dashboard” ¹ may be redundant and could potentially be eliminated or displayed in a more compact fashion. The SD and interquartile range (IQR) are so highly correlated, typically with r=0.95 ^{26 –28} (see supplementary material for Rodbard et al. ²⁶ at http://online.liebertpub.com/doi/suppl/10.1089/dia.2009.0077/suppl_file/Supp_Data.pdf and for Rodbard et al. ²⁷ at http://online.liebertpub.com/doi/suppl/10.1089/dia.2009.0078), that little additional information is provided by display of both parameters. (For a Gaussian [normal] or nearly Gaussian distribution, the IQR=1.35×SD.)

A simple graphical depiction, the so-called “stacked bar chart” of the glucose frequency distribution, ^12,29 may be superior to a tabular display of the glucose frequency distribution. ¹ The “stacked bar chart” is decided superior to the use of the pie charts ¹² and may also be preferable to the frequency histogram because it shows glucose in clinically relevant ranges.

Additional Graphical Displays

As noted above, it is desirable to display a separate “composite AGP” for each of the days of the week. ¹¹ “Stacked bar charts” can be used to portray glucose by date, by time of day, in relationship to meals, and by day of the week. ^12,29

An extremely compact view of glucose data can be generated for both SMBG and CGM data by assigning narrow ranges of glucose values to categories, assigning a unique color to each category—a subtle difference from that shown in Figure 1 of Bergenstal et al. ¹ —and displaying glucose categories simultaneously by date (vertical axis) versus time of day (horizontal axis). ¹³ This type of display can also be constructed separately for each of the days of the week. ¹³ The clinician also needs access to several other types of graphs (e.g., glucose by date) to facilitate analysis of the response to therapeutic interventions.

Integrated Measures of Glycemic Control

Kowalski and Dutta ⁴ and Garg ⁶ emphasize the need for measures of glycemic control that provide information not attainable from the hemoglobin A1c (HbA_1c) level. Several such measures are available. These measures assign greater weight to hypoglycemia than to hyperglycemia. Furthermore, they provide a greater weight or penalty to values that are further outside the target range: a glucose value of 40 mg/dL generates a much greater penalty score than a value of 69 mg/dL. Similarly, glucose values of 400 mg/dL should receive a much greater penalty than a glucose level of 181 mg/dL. The Low Blood Glucose Index, High Blood Glucose Index, average daily risk range, and Blood Glucose Risk Index, ^17,18 the Hypoglycemia Index, Hyperglycemia Index, and Index of Glycemic Control ^{19 –21} (see supplementary material for Rodbard ²⁰ at http://online.liebertpub.com/doi/suppl/10.1089/dia.2009.0015/suppl_file/Appendix-2.xls), and the various Glycemic Response Assessment Diabetes Equations GRADE _hypoglycemia , GRADE _{hyperglycemia} , and overall GRADE score ^22,23,30 provide such measures. These measures are highly correlated among themselves and with the percentage of time that the glucose falls in the target range ^{26 –28} (see supplementary material for Rodbard et al. ²⁶ at http://online.liebertpub.com/doi/suppl/10.1089/dia.2009.0077/suppl_file/Supp_Data.pdf and for Rodbard et al. ²⁷ at http://online.liebertpub.com/doi/suppl/10.1089/dia.2009.0078). Each of these measures has its own pros and cons and its own normal ranges, ^17,23 but there is remarkable consistency among their results. ^{25 –28} Using data from an appropriate reference population, these measures of quality of glycemic control can be expressed on a percentile scale. ³¹ This avoids the need for the clinician to remember or look up the “normal range” for each parameter and enables one to obtain an average of glycemic control and glycemic variability from different parameters when those parameters have different magnitudes and different normal or reference ranges. ^{31 –33}

Thomas et al. ³⁴ proposed a graphical approach, the “glucose pentagon,” to combine the multiple aspects of glycemic control. This pentagon primarily reflects several aspects of hyperglycemia. (The glucose pentagon includes mean glucose, HbA_1c, SD, area under the curve above a threshold for hyperglycemia, and percentage of time in the hyperglycemic range. These parameters are highly correlated with hyperglycemia, but only weakly—or negatively—correlated with hypoglycemia.) The glucose pentagon should be modified to include one or more measures of hypoglycemia. It would be desirable to scale each of the parameters in terms of percentiles of a defined reference population as in Rodbard and co-workers ^{26 –28,31} or nearly equivalently, by use of a normalized “z score,” also calculated based on a reference population. A z score is calculated by expressing the observed value in terms of SD units from the mean for a reference population: z=(observed value – reference mean)/σ, where σ is the standard deviation of values for a reference population.

Analysis of Glycemic Variability

Just as it is desirable to standardize the reporting of CGM and SMBG glucose values, it would be desirable to standardize reporting of parameters for glycemic variability. More than two dozen criteria for variability are available. ^{19,20,26 –28,35,36} The community should promote use of a consistent terminology for identical concepts.

Several computer programs are currently available to compute the parameters describing glycemic variability ^{20,30,37 –41} (see supplementary material for Rodbard ²⁰ at http://online.liebertpub.com/doi/suppl/10.1089/dia.2009.0015/suppl_file/Appendix-2.xls). It would be desirable to compare these programs to verify that they generate similar results when applied to a wide variety of datasets.

As noted by Bergenstal et al., ¹ nearly all measures of glycemic variability are highly correlated with the overall SD. The IQR, mean average glucose excursion, Mean of Daily Differences (MODD), and Continuous Overlapping Net Glycemic Action (CONGA _n and CONGA_1–24) are highly correlated with SD _T ^{26 –28} (see supplementary material for Rodbard et al. ²⁶ at http://online.liebertpub.com/doi/suppl/10.1089/dia.2009.0077/suppl_file/Supp_Data.pdf). Several subtypes of the SD (e.g., the within-day SD _w and the between-day SD _b ) are also closely correlated with the overall SD _T . ^{11,12,19,20,26,28}

Artificial Intelligence and Clinical Decision Support

Barriers to Adoption of a Standardized Report

Consider the myriad of devices used to display the report and analysis of glucose data, which include, in addition to desktop workstations, glucose meters, mobile devices, insulin pumps, CGM devices, and electronic medical records. The variety of size, shape, resolution, color capabilities, and type of the display impacts the choice of what can be displayed and how it can be displayed. The type of report desired also depends on the clinical applications ranging from ambulatory patients to hospitals and intensive care units.

Many device-manufacturing companies have made considerable investments in the development of proprietary software, some with patented features. Many users are likely to be reluctant to change operating procedures, software, and the content and form of the report.

The Future

A shared standard that reduces the cognitive burden on the users should promote increased and more effective use of CGM and SMBG and thereby help to improve the quality of patient care. ^{1 –3} Device and software developers should strive to evolve toward a common core report, with similar content, method of calculation, terminology, symbols, format, color coding, instructions, and educational materials for interpretation of results. The standardized 1-page report ^{1 –3} can be accompanied by more detailed analyses customized to individual patients, clinicians, clinical situations and applications, and devices.

Conclusions

We must continually strive toward achieving standardization, to provide simplicity, brevity, clarity, efficiency, and effectiveness of communication. The AGP with its “dashboard” described by Bergenstal et al. ¹ is an important model for the development of a consensus standardized report. There are many valuable approaches to data analysis, display, and interpretation that were necessarily excluded from the Recommendations ¹ for the sake of brevity. We must retain sufficient flexibility to permit customization to diverse clinical situations and environments, and for research applications.