Determinants of regression from impaired fasting glucose to normoglycemia: Development and validation of nomograms

Abstract

Aim: To explore the determinants of regression to normoglycemia among non-diabetes subjects with IFG using nonlaboratory- based and laboratory-based models, and to develop and validate nomograms for predicting the regression to normoglycemia. Method: This study was based on cross-sectional data of 1197 individuals who had IFG (FPG between 5.6-6.9 mol/l) diagnosed at recruitment. Within 18 months, these subjects were invited for a detailed assessment including a standardized questionnaire to collect demographic information, lifestyle factors and medication use, anthropometric assessment including blood pressure, body weight and waist circumference, and laboratory tests including repeated FPG, 2h-OGTT, HbA1c, estimated glomerular filtration rate (eGFR) and full lipid profile. Two stepwise logistic regression models were established to determine the non-laboratory-based and laboratory-based risk factors that were associated with regression from IFG to normoglycemia on a randomly selected training dataset (n=810). The models were validated on the remaining testing dataset (n=387). Area under the receiver-operating characteristic curve (AUC) and Hosmer-Lemeshow test were used to evaluate discrimination and calibration of the models. Two nomograms were further draw based on the two models. Results: Of the eligible study subjects, 180 had normoglycemia based on the repeated FPG and 2h-OGTT results. The non-laboratory-based model, with score ranging from 0 to 54, showed that subjects without central obesity, hypertension, statin medication, and with lower FPG at recruitment and enough physical activity were more likely to regress to normoglycemia. While the model with additional inclusion of laboratory test results revealed that IFG subjects without central obesity, hypertension, and with lower FPG at recruitment, lower triglycerides level and HbA1c level had higher likelihood to regress to normoglycemia. This model had a score ranging from 0 to 63. Both models demonstrated acceptable discrimination [AUC: 0.682 (0.614-0.749) and 0.747 (0.686-0.809), respectively) and calibration (Hosmer-Lemeshow test: p=0.486 and 0.795, respectively). Discussion: These data suggested that controlling central obesity, hypertension, glucose level, abnormal lipid profile and increasing physical activity could improve the likelihood of regression to normoglycemia among subjects with IFG. Two nomograms were created with acceptable discrimination and calibration in predicting normoglycemia for subjects with IFG.