基于机器学习的住院患者压力性损伤分析与预测

Pressure Injury Analysis and Prediction Based on Machine Learning Methods

摘要:

压力性损伤是护理工作的重点，也是评价护理质量的重要指标，设计合理的评估量表和科学预测是预防的关键措施。基于传统的12个指标，再新增3个风险指标，设计更全面的风险评估量表；据此收集一段时间内住院患者的信息，采用卡方检验分析对损伤有显著影响的指标，将患者分为入院时和院内获得性压力性损伤两类，分析其特征、产生部位和分布科室。基于支持向量机、概率神经网络和广义回归神经网络3种方法建立预测模型，在支持向量机中，采用高斯核函数构建模型，并使用遗传算法优化核函数参数。比较4种场景下3种方法的预测精度，支持向量机的预测准确率最高，达到84.68%，另外2种方法的准确率较低，均为82.78%。

Abstract:

Pressure injury is the focus of nursing， and an important index to evaluate the quality of nursing. Designing reasonable assessment scale and scientific prediction is the key measure to prevent it. Based on the 12 factors， three new risk factors are added in this paper， thus a more comprehensive scale is designed and patients are surveyed. The chi-square test is used to find the factors that have significant impact on pressure injury. Patients are divided into two categories， PIOA （pressure injury on admission） and HAPI （hospital acquired pressure injury）. Then， the characteristics of patients， the locations， and departments are analyzed. Three machine learning methods， support vector machine， probabilistic neural network， and general regression neural network are applied to construct the prediction model. The Gaussian kernel function is used in the SVM model， and the genetic algorithm is adopted to optimize the parameters. The prediction accuracy of the three models are compared in four scenarios. The SVM model， which has optimized parameters， has the highest accuracy of 84.68% while the accuracy of PNN and GRNN are equal to 82.78% and lower than SVM.