| 旷小羿,徐巧玲,徐伟,等.老年2型糖尿病患者血糖趋势预测模型构建初探[J].中国临床保健杂志,2022,25(2):208-212. |
| 老年2型糖尿病患者血糖趋势预测模型构建初探 |
| A preliminary study on the construction of a model for predicting blood glucose trends in elderly patients with type 2 diabetes mellitus |
| 投稿时间：2021-12-24 |
| DOI：10.3969/J.issn.1672-6790.2022.02.016 |
| 中文关键词: 糖尿病,2型 模型,统计学 预测 老年人 |
| 英文关键词: Diabetes mellitus,type 2 Models,statistical Forecasting Aged 〖FL |
| 基金项目:中央保健科研课题(2020YB05) |
|
| 摘要点击次数: 2416 |
| 全文下载次数: 2933 |
| 中文摘要: |
| 目的 血糖趋势预测模型的构建能有效协助医护人员管理2型糖尿病(T2DM)患者血糖,并促进其维持健康的生活方式。对比研究2种构模方式的优缺点,并探讨模型使用意义。方法 采集9例老年T2DM患者连续14 d的持续血糖监测数据(CGM),并结合在此期间的饮食(D)、运动(E)、用药、睡眠4大生活数据,运用支持向量回归(SVR)与基于长短期记忆单元的递归神经网络(LSTM-RNN)构建血糖趋势预测模型,模型的数据输入分为CGM、CGM+D、CGM+D+E这3种方式。结果 SVR模型表现随着预测范围的延长而下降(P<0.05),在预测范围(PH)=30 min时,饮食及运动数据的加入提升模型表现(P<0.05)。LSTM-RNN模型表现较为复杂,预测范围的延长对模型结果差异无统计学意义,在PH=45 min时,饮食数据的加入提升模型表现,且优于SVR模型45/60 min的预测结果。SVR与LSTM-RNN构建的预测模型均具有高度的个体化匹配度,EGA值在AB区域的占比达到了93.82%～99.77%。结论 SVR短时预测(15 min)优于LSTM-RNN,LSTM-RNN较长预测(45 min)优于SVR(45/60 min)；血糖趋势预测模型可对患者进行有效的热量相关指导,对该研究人群而言,饮食数据是影响血糖的主要因素,运动数据对血糖的影响不足。 |
| 英文摘要: |
| Objective The construction of a blood glucose trend prediction model can effectively assist healthcare professionals in the management of blood glucose in patients with type 2 diabetes mellitus (T2DM) and promote the maintenance of a healthy lifestyle.In this paper,we compare the advantages and disadvantages of two model construction approaches and discuss the implications of model use.Methods Continuous glucose monitoring (CGM) data were collected from nine elderly patients with T2DM for 14 consecutive days combining the four major life data of diet (D),exercise (E),medication and sleep during this period.A model of blood glucose trend prediction was constructed using support vector regression (SVR) and recurrent neural network based on long and short-term memory units (LSTM-RNN),and the data inputs of the model were categorized into CGM,CGM+D,and CGM+D+E.Results The SVR model performance decreased as the prediction horizons was extended (P<0.05),and at the prediction horizons (PH)=30 min,the addition of diet and exercise data improved the model performance (P<0.05).The LSTM-RNN model performance was more complex,and the difference in model results was not statistically significant for the extended prediction horizons.At PH=45 min,the addition of diet data improved the model performance and outperformed the prediction results of the SVR model at 45/60 min.Both SVR and LSTM-RNN constructed prediction models had a high degree of individualized matching,with the Clarke-EGA values reaching 93.82% to 99.77% in the AB region.Conclusions SVR short prediction (15 min) was better than LSTM-RNN and LSTM-RNN longer prediction (45 min) was better than SVR(45/60 min);the blood glucose trend prediction model provided effective calorie-related guidance to patients,and for this study population,dietary data were the main factor influencing blood glucose,with exercise data having an insufficient effect on blood glucose. |
|
查看全文
|
| 关闭 |
|
|
|