非计划再入院风险预测研究

为改善患者身体健康，降低非计划再入院率，减轻患者负担和社会资源浪费，本研究基于我国某区域卫生信息平台的医疗数据，利用机器学习方法，构建了非计划再入院风险预测模型.不同于已有仅预测了再入院概率的研究，本研究通过将风险预测建模为多分类问题，实现了在时间和可能性两个维度对再入院风险进行预测.通过调整机器学习算法参数设置，构建了基于神经网络、随机森林和支持向量机算法的3大类共10个再入院风险备选预测模型.基于真实数据集的实验结果表明，在备选风险预测模型中，使用多项式核函数的支持向量机模型预测效果最好，预测准确率达到96.65%.本研究成果可以使医疗机构基于患者历史医疗数据，从时间和可能性两个维度更全面、精准地评估患者再入院风险，进而采取必要的干预措施，降低非计划再入院率. 

Research on Risk Prediction of Unplanned Readmission

To improve patients’ health, decrease unplanned hospital readmission rate, alleviate patients’ burden and prevent social resources waste, an unplanned hospital readmission risk prediction model was built, utilizing machine learning method and based on a dataset collected from a regional health care information platform of China. Different from existing works which only predict readmission risk, this research tried to model the problem from a multi-class classification view and predict readmission time and probability simultaneously. 10 classifiers were built by adjusting the parameters of neural network, random forest and support vector machine. Experiments on real dataset showed that the support vector machine classifier using polynomial kernel function performed best in terms of prediction accuracy, which was about 96.96%. The research result can assess readmission risk more precisely in time and probability based on patients’ historical health care data. With the help of the result, medical agencies can adopt proper interventions and reduce unplanned hospital readmission rate.