[Average blood glucose construction: a multi-level Bayes model approach and application].

ABSTRACT

OBJECTIVE: To explore the average blood glucose construction method based on the multi-level Bayes model and evaluate the example application. METHODS: We generate simulated data with multi-level Bayes model. Three methods were utilized to construct the average blood glucose at the same time, then we compared the result with each other. A cohort study method was used to select 12321 participants aged over 45 y who without stroke in a community in Suzhou and was followed up from 2011 to 2018, of which 53. 7% were male. Mean blood glucose calculated by the most accurate complete Bayesian method was divided into six groups. The Cox regression model was used to analyze the effect of mean blood glucose on the incidence of fatal stroke. RESULTS: 1000 times of simulation result showed that the average mean blood glucose estimation calculated by the complete Bayesian method was 0. 278, the average of blood glucose estimation was 0. 527 mmol/L, and the average correlation coefficient with the actual blood glucose was r=0. 898. During the follow-up period, 153 fatal strokes occurred. Association was found between the mean blood glucose and the risk of fatal stroke(P<0. 05). The average risk of blood glucose over 140 mg/dL was 2. 304 times that of 90-99 mg/dL(HR=2. 304, 95%CI 1. 151-4. 613) after the adjustment of effects. CONCLUSION: The complete Bayesian multi-level latent variable model can accurately estimate the average blood glucose.