谵妄预测模型建立和评价的临床研究

To Establish and Evaluate a Prediction Model for Delirium

目的:通过数据分析建立谵妄的预测模型,并对模型进行评价和应用实践。方法:回顾性整理1521例自贡第四人民医院重症医学科住院患者的临床资料,按照是否发生谵妄,分为谵妄组(527例)和非谵妄组(994例),进行Logistic回归分析后建立模型。结果:谵妄预测模型建立为:L=APACHE Ⅱ评分+0.0558×总胆红素+1.191×尿素氮+4.098×饮酒史+2.009×呼吸机,L>51.39,即可预测谵妄发生。预测模型进行评价,模型系数的全局性Omnibus检验,χ~2=1025.74,P<0.001,有统计学意义。模型的灵敏度78.0%,特异度93.5%,漏诊率22.0%,误诊率6.5%。预测概率ROC曲线下面积为0.926,95%置信区间为(0.91,0.94)。结论:谵妄预测模型可运用于临床,对预防患者谵妄带来的不良结局有重要意义。

Objective： Our study was designed to establish and evaluate a predictive model for delirium based on data analysis. Methods： We retrospectively collected the clinical data of 1521 hospitalized patients in the de- partment of critical care medicine, Zigong Fourth People＇s Hospital. Patients were divided into delirium group （n= 527） and non-delirium group （n=994） according to whether or not the patient had delirium during hospitalization. We established the model after logistic statistical analysis of these data. Results： We established a prediction model for delirium by data analysis. L= APACHE II score ＋0.0558×total bilirubin value＋1.191 xurea nitrogen＋ 4.098xhistory of drinking＋2.009xmechanical ventilation. If the value of L is higher than 51.39, it is likely that delirium would occur. We evaluated the prediction model. The global Omnibus test of model coefficients, X^2= 1025.74, P〈0.001, which was statistically significant. The sensitivity of the model was 78%, specificity was 93.5%, missed diagnosis rate was 22%, and misdiagnosis rate was 6.5%. The area under the ROC curve was 0.926, 95% confidence interval （0.91,0.94）. Conclusion： The prediction model for delirium could be useful in clinical diagnosis and treatment to prevent the occurs of adverse outcomes.