摘要
目的分析颅脑损伤患者继发精神障碍的影响因素,同时考虑病例和非病例间数量不平衡的特点,构建基于SMOTE算法的logistic预警模型。方法根据2008年山东省18家医院的颅脑损伤患者继发精神障碍的数据,利用logistic回归分析筛选影响因素并建立基于原始数据的预警模型;在此基础上,采用SMOTE过抽样算法改进数据集,并构建基于改进数据集的精神障碍预警模型。结果额叶脑挫伤、弥漫性轴索损伤、并发颅内感染、颞叶硬膜下血肿、颅盖骨线性骨折、颅内积气、患者性别和颅脑损伤严重程度(GCS评分)均为颅脑损伤患者发生精神障碍的危险因素;而基于SMOTE过抽样算法所构建预警模型的预测效果明显优于利用原始数据所建模型的效果。结论基于SMOTE过抽样算法所构建的预警模型能更准确预测颅脑损伤患者继发的精神障碍。
Objective To analysis the risk factors of psychiat- ric illness following traumatic brain injury and to construct an early warning system based on SMOTE algorithm. Methods According to traumatic brain injury patients＇ data from 18 Shandong＇s hospitals in 2008, the risk factors of psychiatric illness following traumatic brain injury were analys- ized and selected by logistic regression analysis to establish an early warn- ing system. Basing on the system, the risk factors were improved and select- ed by SMOTE algorithm to establish an better early warning system of psy- chiatric illness. Results The risk factors include frontal cerebral contu- sion, diffuse axonal injury,intracranial infection, temporal lobe subdural he- matoma,calvaria bone linear fracture, intracranial pneumatosis, gender and craniocerebral injury（GCS）. The early warning system basing on SMOTE algorithm was obviously superior to the one basing on logistic regression a- nalysis. Conclusion The early warning model basing on SMOTE algo- rithm could be used to predict psychiatric illness following traumatic brain injury better.
出处
《中国卫生统计》
CSCD
北大核心
2013年第6期790-793,共4页
Chinese Journal of Health Statistics
基金
国家科技支撑计划资助项目(项目编号2008BAI52B03)
关键词
SMOTE算法
过抽样
颅脑损伤精神障碍
SMOTE algorithm
Over sampling
Cranio- cerebral injury
Psychiatric illness
