基于数据挖掘技术的神经外科术后感染风险预测模型构建

A Risk Prediction Model Establishment for Postoperative Infection in Neurosurgery Based on Data Mining Technology

目的通过对医疗数据的深度挖掘与分析,构建神经外科患者术后感染风险预测模型,有效预测术后感染。方法回顾性收集北京市某综合医院神经外科中心患者术后数据,共计21239条样本。采用互信息法初步筛选特征变量,通过SMOTE算法解决类别不平衡问题,最终使用性能表现最优的随机森林模型训练得到神经外科术后感染风险预测模型。结果该预测模型的准确率为0.941,灵敏度为0.940,特异度为0.941,曲线下面积为0.985(参数调优),并得出出血量、诊断编码、出院病房、手术名称、术后血糖、术后白细胞绝对值是神经外科患者术后感染的重要特征。结论神经外科患者术后感染风险预测模型有助于临床决策、早期干预及预防术后感染的发生。

Objective To establish a risk prediction model for postoperative infection in neurosurgical patients,a methodology involved deep mining and analysis of medical data,and accurately and effectively predict postoperative infection as early as possible.Methods The postoperative data of patients were gathered from the neurosurgery department center of a general hospital in Beijing,comprising 21,239 samples.The mutual information method was applied to screen characteristic variables,and the class imbalance problem was addressed using SMOTE.Subsequently,a random forest model with the best performance was utilized to train a risk prediction model for postoperative infection in neurosurgical patients.Results The accuracy of the prediction model was 0.941,the sensitivity was 0.940,the specificity was 0.941,and the AUC was 0.985(with optimized parameters).It was concluded that the amount of blood loss,diagnostic code,discharge ward,operation name,postoperative blood glucose,and absolute value of postoperative white blood cells were important characteristics of postoperative infection in neurosurgical patients.Conclusion The risk prediction model for postoperative infection in neurosurgical patients proposed in this research can aid in clinical decision-making and early intervention,ultimately contributing to the prevention of postoperative infection.