GM(1,1)模型对T1DM和T2DM患者的低血糖预警价值

Value of GM (1,1) model for early warning of hypoglycemia in patients with T1DM and T2DM

目的探讨GM(1,1)模型在T1DM和T2DM患者中对低血糖事件的预警价值。方法收集2017年10月至2017年12月在河南省人民医院住院的DM患者50例,其中T1DM患者7例,T2DM患者43例。以连续72h的血糖数据为研究对象,通过Matlab2018运行GM(1,1)模型,得到5 min、15 min、30 min后的预测血糖值,以平均绝对误差(MAE)、平均相对误差(MRE)、均方根误差(RMSE)及Clarke误差网格评价该模型的预测性能。以灵敏度(Sens)、虚警率(FA)、漏警率(MA)等指标评价低血糖预警效果,并进一步分析在T1DM、T2DM患者中的差异。结果 5 min预测时,GM(1,1)模型的MAE、MRE、RMSE分别是0.48、6.6%、0.68;15 min预测时,MAE、MRE、RMSE分别是0.71、9.6%、1.00;30 min预测时,MAE、MRE、RMSE分别是1.03、13.9%、1.48。Clarke误差网格表明该模型对血糖的预测精度符合要求。5 min预测时,T1DM患者中,低血糖预警的Sens、MA、FA分别为79.1%、20.8%、2.4%;T2DM患者中,Sens、MA、FA分别为78.9%、21.1%、1.7%。15 min预测时,T1DM患者中,Sens、MA、FA分别为71.7%、28.3%、3.8%;T2DM患者中,Sens、MA、FA分别为70.6%、29.4%、2.9%。30 min预测时,T1DM患者中,Sens、MA、FA分别为63.3%、36.7%、5.9%;T2DM患者中,Sens、MA、FA分别为61.7%、38.3%、4.4%。两独立样本t检验表明无论在哪个预测时长,该模型在T1DM与T2DM患者中的低血糖预警效果无统计学差异(P>0.05)。结论 GM(1,1)模型能有效预警低血糖,且在T1DM、T2DM患者中预警效果一致。

Objective To explore the early warning value of the GM(1,1) model for hypoglycemic events in patients withT1 DM and T2 DM.Methods Fifty consecutive glycemic data of DM patients, 7 with T1 DM and 43 with T2 DM, were included. The predicted blood glucose values for the next 5, 15, and 30 min were obtained by running the GM(1,1) model through Matlab2018, and the model was evaluated by the mean absolute error(MAE), mean relative error(MRE), root mean square error(RMSE), and Clarke’s error grid. The effects of hypoglycemia early warning were evaluated by sensitivity(Sens), miss alarm rate(MA), and false alarm rate(FA), and the differences were further analyzed among patients with T1 DM or T2 DM.Results MAE, MRE, and RMSE for the GM(1,1) model were 0.48, 6.6%, 0.68 for 5 min prediction, 0.71, 9.6%, 1.00 for 15 min prediction, and 1.03, 13.9%, 1.48 for 30 min prediction, respectively. The Clarke error mesh analysis indicated that the model’s prediction accuracy for blood glucose met the requirements. When predicted at 5 min, the Sens, MA, FA of hypoglycemia warning were 79.1%, 20.8%, 2.4% in T1 DM patients and 78.9%, 21.1%, 1.7% in T2 DM patients. At 15 min prediction, Sens, MA and FA were 71.7%, 28.3%, 3.8% in T1 DM and 70.6%, 29.4%, 2.9% in T2 DM patients. At 30 min prediction, Sens, MA, FA were 63.3%, 36.7%, 5.9% in T1 DM and 61.7%, 38.3%, 4.4% in T2 DM patients. Two independent sample T-tests indicated that the model had no statistically significant difference in the effect of early warning of hypoglycemia in patients with T1 DM versus T2 DM, regardless of which prediction time was long(P>0.05).Conclusion GM(1,1) model was effective in alerting hypoglycemia, and the effect of alerting was consistent among patients with T1 DM or T2 DM.