Yuping Zeng1, He He1, Jun Zhou1, Mei Zhang1, Hengjian Huang2, Zhenmei An3. 1. Department of Laboratory Medicine, West China Hospital, Sichuan University, Chengdu, China. 2. Department of Laboratory Medicine, West China Hospital, Sichuan University, Chengdu, China. Electronic address: huanghenjian@sina.com. 3. Department of Endocrine and Metabolism, West China Hospital, Sichuan University, Chengdu, China.
Abstract
BACKGROUND: Due to a high prevalence of thalassemia in southwest China, the diagnostic value of glycated hemoglobin A1c (HbA1c) is limited in the local population. Glycated albumin (GA) must also be measured for glucose monitoring. We sought to explore the relationships between HbA1c and GA. METHODS: We analyzed 3,414 participants and allocated to four groups: GA > 14% and HbA1c > 5.7% (group 1), GA > 14% and HbA1c ≤ 5.7% (group 2), GA ≤ 14% and HbA1c > 5.7% (group 3), and GA ≤ 14% and HbA1c ≤ 5.7% (group 4). We used stepwise multivariable logistic regression analysis to study the inconsistency of HbA1c and GA. Furthermore, we explored their association using multiple linear regression (MLR), random forest regression (RFR), and 3 blended models. Finally, we performed sensitivity analyses by changing the thresholds of HbA1c (6.5%) and GA (12% or 16%). RESULTS: There were 934 participants in group 1, 86 in group 2, 964 in group 3, and 1,430 in group 4. Age, high-density lipoprotein-cholesterol concentration, and red blood cell count were associated with the discordance in HbA1c and GA values. We constructed an RFR model that included MLR predictions as independent variables and could explain 97.80% of the variance in HbA1c in the training set, and 91.65% in the cross-validation set. Our results remained robust in 3 sensitivity analyses. CONCLUSIONS: HbA1c and GA values are inconsistent in the population we studied. A model that blends MLR and RFR can be used to correct HbA1c values when conflicting HbA1c and GA values are encountered in patients.
BACKGROUND: Due to a high prevalence of thalassemia in southwest China, the diagnostic value of glycated hemoglobin A1c (HbA1c) is limited in the local population. Glycated albumin (GA) must also be measured for glucose monitoring. We sought to explore the relationships between HbA1c and GA. METHODS: We analyzed 3,414 participants and allocated to four groups: GA > 14% and HbA1c > 5.7% (group 1), GA > 14% and HbA1c ≤ 5.7% (group 2), GA ≤ 14% and HbA1c > 5.7% (group 3), and GA ≤ 14% and HbA1c ≤ 5.7% (group 4). We used stepwise multivariable logistic regression analysis to study the inconsistency of HbA1c and GA. Furthermore, we explored their association using multiple linear regression (MLR), random forest regression (RFR), and 3 blended models. Finally, we performed sensitivity analyses by changing the thresholds of HbA1c (6.5%) and GA (12% or 16%). RESULTS: There were 934 participants in group 1, 86 in group 2, 964 in group 3, and 1,430 in group 4. Age, high-density lipoprotein-cholesterol concentration, and red blood cell count were associated with the discordance in HbA1c and GA values. We constructed an RFR model that included MLR predictions as independent variables and could explain 97.80% of the variance in HbA1c in the training set, and 91.65% in the cross-validation set. Our results remained robust in 3 sensitivity analyses. CONCLUSIONS: HbA1c and GA values are inconsistent in the population we studied. A model that blends MLR and RFR can be used to correct HbA1c values when conflicting HbA1c and GA values are encountered in patients.