医院感染危险度的预测预报

Prediction and Forecast of Nosocomial Infection Risk Factor

目的 研究影响医院感染发生的主要因素 ,建立对住院患者医院感染危险度进行预测预报的统计模型。方法 采用医院信息系统记录的 2 73 5 2例住院患者的临床信息 ,经过数据清理和标准编码后 ,建立预测医院感染的logsitic回归模型并将医院感染危险度分为四级。结果 影响医院感染的主要危险因素有患者年龄、最高体温、诊断个数、住院天数、病重天数、输血次数、抗生素使用、是否放疗、是否翻身、X常规送检次数、介入操作是否可能造成感染、有无糖尿病、病种、介入操作次数、手术麻醉类型、入院科室。将logistic回归模型的预测概率大于 0 3 5作为医院感染的诊断截点 ,诊断试验的特异度为 0 995 ,误诊率为 0 0 0 5 ,ROC曲线下面积AUC为 0 986。通过决策树分析 ,可将医院感染危险度分为四个级别 :Pr <0 0 2 82 7为低度 ,0 0 2 82 7≤Pr <0 13 40 0为中度 ,0 13 40 0≤Pr <0 5 0 692为高度 ,Pr≥ 0 5 0 692为极度危险。结论 在运行医院信息系统的医院中 ,将住院患者的在线信息经标准化编码后代入模型 ,可对该患者医院感染危险度进行监控和预测预报 。

Objective To study the main influence factors of nosocomial infection(NI) and establish a model to predict and forecast the risk of NI on patients in hospitals. Methods Clinical data of 27352 inpatients extracted from hospital information system were sorted out and coded, and a logistic regression model about the probability of NI was established. The risk of NI was divided into four scales. Results With multiple factor analysis,16 risk factors of NI were identified, which were age, high body temperature, numbers of diagnosis, days of staying in hospital and seriousness, numbers of routine test for urine, times of blood transfusion, use or without use of antibiotic and radiotherapy, turning over the bodies or not, relationships between infection and interventional operations, with or without diabetes, categories of diseases based on ICD-9, numbers of interventional operations, type of anesthesia and department of admission. If NI was judged when predicted probability(Pr)of logistic regression model exceeded 0 35, the specificity and false diagnostic rate of diagnostic test were 0 995 and 0 005 respectively, and the area under ROC curve was 0 986. According to decision tree method, the risk of NI was classified into four degrees: low (Pr<0 02827), moderate (0 02827≤Pr<0 13400), high (0 13400≤Pr<0 50692) and extreme(Pr≥0 50692). Conclusions Logistic regression model can be used as a tool for NI forecasting. When online clinical data of inpatients are standardized and put into the model, hospital information system will supervise, predict and forecast the risk of NI on each patient, which can provide supplementary information about diagnosis and control of NI.