An estimation model for postprandial blood glucose levels in patients with type 2 diabetes mellitus

纪立农,张雪莲,冯晨,马骁,杨文英

DOI: https://doi.org/10.3760/cma.j.cn115791-20210224-00102

2021-01-01

Abstract:Objective:To develop and validate a model for postprandial blood glucose (PBG) estimation in Asian patients with type 2 diabetes mellitus (T2DM).Methods:The estimation model for PBG was developed and validated based on glycated hemoglobin (HbA 1c) and finger-based fasting blood glucose (FBG). The training cohort was drawn from a phase 4, multinational, randomized, controlled, 26-week trial comparing the efficacy and safety of insulin lispro low-mixture and insulin lispro mid-mixture as an insulin starter in Asian T2DM patients. The glycemic data at baseline (before insulin initiation) and 26 weeks were used for developing the estimation model for PBG. The validation cohort was drawn from a multicenter, randomized, and pragmatic trial comparing the efficacy and safety of basal insulin analog and insulin analog mid-mixture as an insulin starter in Chinese T2DM patients with inadequate glycemic control (HbA 1c ≥7.5%) following oral antihyperglycemic medication (OAM) treatment. The glycemic data at baseline and 24 weeks were used for validating the estimation model. Based on the data distribution of the training cohort, a linear regression model including two variables was used to develop the estimation model. The β regression coefficient of each selected model predictor and α intercept were estimated using the datasets which consisted of FBG, PBG, and HbA 1c. FBG and HbA 1c were selected as independent variables for the estimation for PBG. The goodness of fit was assessed using the decisive factor R 2. For model validation, the PBG value was estimated using this model and compared with the real PBG. The mean of individual differences between estimated PBG and real PBG and the 95% confidence interval (CI) were calculated to evaluate the accuracy of the estimation model. Additionally, the root mean square error (RMSE) was calculated. Results:The clinical trials for the training and validation cohorts enrolled, respectively, 403 and 814 patients with T2DM. Based on data distribution through scatter plot, a linear regression model including two variables, HbA 1c and FBG, was used for developing PBG estimation model: PBG (Unit: mmol/L) = 1.5×HbA 1c (Unit: %)+0.5×FBG (Unit: mmol/L)-4.1 (R 2=0.57). In the validation cohort, the estimated PBG was (13.8±3.7) mmol/L and the real PBG was (13.7±4.6) mmol/L with the individual differences between the estimated PBG and real PBG being (0.1±3.4) mmol/L, 95%CI being -0.1 to 0.3, and P being 0.525. The RMSE for the estimation model was 3.40 mmol/L. Conclusion:Patients with T2DM treated with OAM or insulin could estimate their PBG with HbA 1c and FBG using this estimation model with reasonable accuracy.