Blood glucose profile evaluation with model-based approach using continuous glucose monitoring data
Hanna Kunina (1), Jenny Y Chien (2), Parag Garhyan (2), Jeanne S Geiser (2), Maria C. Kjellsson (1)
(1) Department of Pharmacy, Uppsala University, Uppsala, Sweden (2) Eli Lilly and Company, Indianapolis, USA
Introduction:
Continuous glucose monitoring (CGM) has revolutionized type 1 (T1D) and type 2 (T2D) diabetes management, as it provides real-time, dynamic blood glucose (BG) measurements and the possibility to directly improve glycemic control. However, the usage of CGM in drug development is challenging due to unreliable information about meals, both time and size. Several algorithms have been developed for the analysis of CGM data in patients without endogenous insulin production (i.e., T1D), where the time and size of meals are assessed, relying on information about short-acting insulin injections, which are administered adjacent to meals [1,2]. For patients with endogenous insulin production (i.e., T2D), the application of these algorithms is limited as these patients commonly receive once-daily administrations of long-acting insulin, and lack information on the level of insulin deficiency and insulin sensitivity, in addition to the missing information on meals. While previously published algorithms can predict unannounced meal time and size for T1D patients based on CGM data, none of them have provided a solution that enables prospective simulations of unperformed experiments in patients with T2D.
Objectives:
The overall goal of this project is to develop a pharmacometric approach to assess drug effect in patients with residual insulin secretion using CGM data. As a first objective, we aimed to develop a model that assessed meal occurrence and size using CGM data and to use the model output to quantify meal variability for prospective simulations.
Methods:
The CGM data from 68 insulin-naïve adult patients with T2D from the IMAGINE 2 RCT[3] were analyzed using the integrated glucose-insulin (IGI) model [4]. The baseline CGM data, i.e., prior to insulin treatment, were collected for up to 7 days and consisted of glucose measurements at 5 minutes intervals. In the initial analysis, each day for every patient was treated as a unique individual resulting in 449 individuals (i.e., Npatients*Ndays). The underlying dynamics of glucose and insulin, including bidirectional feedback, were set by fixing the physiological parameters of the IGI model to the published values for a population with T2D. It was assumed that a meal could be distinguished hourly, and the occurrence and maximum bioavailable meal size (in g of glucose) was estimated using a logit transformation to scale the fraction of maximum bioavailable meal to dose hourly. Both between-day (BDV) and between-hour (BHV) variability of the meal intake was estimated in the 24-hour time frame, with BHV implemented as inter-occasion variability. The analysis was performed in NONMEM v.7.5.0 [5] using the method SAEM with PsN v.5.2.0 [6]. Data management and graphical evaluations of the results were performed in R v.4.0.4 [7]. The computations were enabled by resources at Uppsala Multidisciplinary Center for Advanced Computational Science (UPPMAX).
Results:
The IGI model was successfully adapted to estimate the meal occurrence and size using CGM data from patients with poorly controlled T2D and parameters were estimated with high certainty (maximum relative standard error 4.3%). The model identified 3 meal peaks in the population, corresponding to breakfast, lunch, and dinner at 8:00 am, 1:00 pm, and 7:00 pm, of 44 g, 40 g, and 43 g bioavailable glucose, respectively. A clear nadir was also identified at 2:00 am. However, variability between individuals and days was considerable. The BDV, representing between-patient and between-day variability, and BHV were estimated to be 249% (RSE=4.3%) and 300% (RSE=0.75%), respectively. Analysis of empirical Bayes estimates showed that patients with small, frequent meals and nighttime meals were more difficult to predict than patients with large, infrequent meals and meals during the day. Predictions of dynamic insulin, although not sampled, were available in this approach.
Conclusions:
We have herein illustrated how to determine the time and size of meals in patients with poorly controlled T2D using CGM data using a pharmacometric, model-based approach. The developed framework allows for the evaluation of comprehensive CGM data and may assist in the advancement of antidiabetic drug development. The future perspective includes the application of the developed approach to investigate dose-response after the initiation of long-acting insulin therapy.
References:
[1] Xie J, Wang Q. A Variable State Dimension Approach to Meal Detection and Meal Size Estimation: In Silico Evaluation Through Basal-Bolus Insulin Therapy for Type 1 Diabetes. IEEE Trans Biomed Eng. 2017;64(6):1249-1260.
[2] Zheng M, Ni B, Kleinberg S. Automated meal detection from continuous glucose monitor data through simulation and explanation. J Am Med Inform Assoc. 2019;26(12):1592-1599.
[3] Davies MJ, Russell-Jones D, Selam J-L, Bailey TS, Kerényi Z, Luo J, Bue-Valleskey J, Iványi T, Hartman ML, Jacobson JG, Jacober SJ for the IMAGINE 2 Study Investigators. Basal insulin peglispro versus insulin glargine in insulin-naïve type 2 diabetes: IMAGINE2 randomized trial, Diabetes Obes Metab 2016, 18, 1055–1064.
[4] Silber HE, Jauslin PM, Frey N, Gieschke R, Simonsson US, Karlsson MO. An integrated model for glucose and insulin regulation in healthy volunteers and type 2 diabetic patients following intravenous glucose provocations. J Clin Pharmacol. 2007;47(9):1159-1171.
[5] Beal SL, Sheiner LB, Broeckmann A, Bauer RJ. NONMEM 7.5 User's Guides (1989-2020). ICON plc, Gaithersburg, MD.
[6] Lindbom L, Ribbing J, Jonsson EN. Perl-speaks-NONMEM (PsN) – a Perl module for NONMEM related programming. Comput Methods Programs Biomed. 2004; 75: 85- 94.
[7] R Core Team. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing; 2022. http://www.R-project.org/