基于递归神经网络的糖尿病药物疗效预测模型

Predictive Model of Diabetes Drug Efficacy Based on Recurrent Neural Network

目的利用已有的医疗记录信息,建立糖尿病药物疗效预测模型,增强Ⅱ型糖尿病患者长期使用抗糖尿病药物控制血糖的效果。方法提出一种基于递归神经网络的糖尿病药物疗效预测模型(SPM)。模型利用之前,把所有医疗记录的顺序序列作为输入,输出当前处方的预测治疗效果。结果将SPM与基准模型在某医院的真实医疗数据集上进行对比。实验表明,使用SPM后药物疗效预测精度得到提高,对于多记录患者,预测精度提升效果十分明显。结论 SPM在临床工作中具有实用价值。

Objective To establish a predictive model for efficacy of diabetes drugs with existing medical record and enhance the efficacy of blood glucose control for long-term users of antidiabetic drug. Methods A predictive model(SPM) based on recurrent neural network was proposed, which used the sequence of all previous medical records as input to output the predicted therapeutic effect of current prescription. Results The SPM model was compared with the benchmark model on the real medical data set of a hospital. The experiment showed that the accuracy of prediction of drug efficacy was improved after using SPM. For patients with multiple records, the accuracy of prediction was significantly improved. Conclusion SPM has practical value in clinical work.