Objective  To evaluate performance of GM (1, 1) model in predicting blood glucose, and the predictive ability of this model for prediction of different blood glucose ranges.

  Methods  Seventy-two-hour blood glucose sequence of fifty patients with T2DM was selected as study objects. GM (1, 1) model was established based on the metabolic algorithm to predict blood glucose levels after 5, 15 and 30 min. Fifty patients were randomly divided into control group(n=20) and experimental group(n=30). The blood glucose range(ranging from a to b) when the prediction error was low in the control group was obtained. According to the blood glucose range of the control group, the data of the experimental group was divided into group A(< a), group B(a to b), and group C(>b). Analysis was performed by Pearson correlation analysis and nonparametric test.

  Results  The blood glucose ranges for prediction duration of 5 min, 15 min and 30 min in the control group were 3.4 to 11.5 mmol/L, 3.3 to 11.4 mmol/L and 3.2 to 11.4 mmol/L, respectively. Correlation analysis showed that group B at differed prediction durations had the best predictive fit (P < 0.01). Nonparametric tests indicated that there was a statistical difference in prediction error between groups at the same prediction duration (P < 0.01), and the group A had the minimum error, followed by group B, and group C had the maximum error.

  Conclusion  GM (1, 1) model has better predictive efficacy at blood glucose range of 3.4 to 11.4 mmol/L.