Analysis of “Simple Post-Processing of Continuous Glucose Monitoring Measurements Improves Endpoints in Clinical Trials”

Abstract

Jensen et al used continuous glucose monitoring (CGM) data of the Dexcom G4 Platinum (DG4P) sensor obtained in a clinical efficacy and safety study of Novo Nordisk’s new insulin Fiasp® to calculate the CGM time delay versus plasma glucose (PG) and self-measured blood glucose (SMBG) measurements (9-10 min). Shifting the CGM signal by 9 min backward in time versus PG and SMBG data improved the analytical accuracy of the DG4P sensor and the reliability of clinical research endpoint (hypoglycemia, postprandial glucose increments) detection. Since this method takes advantage of post-processing of CGM data, it is particularly suited for the optimization of data processing in clinical studies. In contrast, real-time corrections of time delays need predictive algorithms.

Continuous glucose monitoring (CGM) systems determine glucose concentration in interstitial fluid. Since glucose enters the interstitial volume by diffusion from the capillary bed and diffusion needs time an inherent time delay between blood and interstitial glucose concentration exists. This leads to differences in glucose concentration of these compartments at the same point in time. The faster glucose changes take place in capillary blood compared with diffusion time the larger these differences tend to be.

Jensen et al ¹ describe in the current issue of Journal of Diabetes Science and Technology the identification of the average time delay in adult subjects with type 1 diabetes of the Dexcom® G4 Platinum (DG4P; Dexcom, San Diego, CA) in an efficacy and safety clinical trial of the fast acting insulin Fiasp® (Novo Nordisk A/S, Bagsvaerd, Denmark) using CGM as well as rather frequently measured plasma glucose (PG) and self-measured blood glucose (SMBG) data. The average time delay (9 to 10 min) is used to improve the analytical accuracy of the CGM device by shifting the CGM data back in time with respect to PG and SMBG measurements. Mean absolute relative difference (MARD) was significantly improved by 1.0 percentage points with PG data and 1.8 percentage points with SMBG data as reference. In addition, the authors demonstrate a better reliability of hypoglycemia detection: Shifting the CGM data by −9 min versus PG data improved the sensitivity of Level 1 hypoglycemia (PG ⩽ 70 mg/dL) detection by 5.9 percentage points and of Level 2 hypoglycemia (PG < 54 mg/dL) detection by even 11.4 percentage points. In addition, postprandial differences between CGM and PG measurements were significantly reduced by using the time shifted CGM data.

Time delay was defined as the time shift of CGM data that leads to the minimum mean absolute difference (MAD) of CGM versus PG and SMBG data, respectively. The described methods for data acquisition and time shift determination are reliable and take advantage of the huge amount of CGM and blood glucose data which were acquired during the study. This takes into account that time delay determination is prone for errors. ² Best condition for time delay determination is to have as many PG or SMBG measurements as possible during fast and large blood glucose variations. Under these conditions measurement errors of PG/SMBG and CGM data are not too large compared with the differences in glucose concentration of the two compartments.

The reported average delay of 9 to 10 min is reasonable and in general agreement with former studies on CGM time delays. One remark: In the discussion of the reliability of their time delay results the authors explain divergent results in a study by Sinha et al ³ by the use of healthy subjects in that study. In fact, the study was conducted with—otherwise healthy—subjects with type 1 diabetes, 12 adolescents and 12 adults. Time delays of the DG4P for adolescents was 5.6 ± 0.9 min and for adults 8.1 ± 0.7 min. Taking the adult results and the usual measurement errors into account there is no meaningful difference between the time delay results of the two studies.

The authors mention that performance improvements might be even better when using an individual time delay for shifting the CGM data in time. Nevertheless, they stayed with a general time shift approach, which is meaningful: Though there are serious indications that individual differences of time delays exist (contributing to the wide range of individual time delays measured in addition to measurement errors) and that these might be rather constant over time in a person ² its existence is not yet fully confirmed. ³ Moreover, since optimal conditions for time delay measurements are not fully met in the study of Jensen et al the reliability of individual time delays would have been limited.

The authors rightly remark that the optimal time shift for DG4P sensors must not be used for other CGM devices. Beyond the physiological time delay there exists a physical delay (caused, eg, by the diffusion barriers inside the CGM sensor) as well as an algorithmic time delay (caused, eg, by filtering algorithms which reduce noise from the raw sensor signal or by real-time time delay correction algorithms) which might be different for other CGM systems. ² The transferability of results might also be limited because the optimal time shift might depend on the actual conditions of a clinical study: systematically different distribution of the rate-of-change of glucose variations might affect time delays as well as the age of the participants. ³ The authors also emphasize that extremely nonphysiologic situations as in clamp studies can affect time delays as well. In general it is evident that much more work is necessary to get a better understanding of the time delay phenomenon and generally of the processes controlling the exchange of glucose between capillary blood and interstitial fluid.

The authors concentrate in their article on the potential of time shifted CGM measurements for improving the reliability of the detection of clinical research endpoints by post-processing of CGM/PG/SMBG data obtained in a clinical study. This is an important yet special application for the use of time delay measurements. In such a clinical setting it is possible to optimize the data processing within a given data set and to apply optimized parameters to the data set itself (in this case the optimal time shift of the CGM data). This is different if CGM data are to be corrected for a time delay in a real-time setting, for example, when people with diabetes use a CGM system in daily life in order to control their glycemic state. In this case a real-time correction algorithm using a predetermined time shift is necessary and the representativeness of the data used to find the predetermined time shift is of importance and needs particular attention.

For the sake of completeness it should be noticed that blood samples obtained by Alternate Site Testing (AST) from hairy skin (eg, the forearm) are not suited as reference for CGM time delay measurements. AST blood itself has a time delay to fingertip blood depending on the actual skin blood perfusion. ⁴

To summarize: the study by Jensen et al demonstrates the usefulness of a simple time delay correction of CGM data to improve the reliability of the detection of clinical research endpoints based on CGM measurements.