主动脉夹层患者院内死亡的Bayes判别分析研究

Bayes Discriminant Analysis for In-hospital Death of Aortic Dissection

背景 主动脉夹层(AD)是心血管系统的一种危急重症,具有发病急、病死率高的特点,既往缺少对AD患者院内结局预测的有效模型,随着近几年发病人数的不断增加,临床上迫切需要有效可靠的患者院内结局预测模型.目的 建立Bayes判别方程,预测AD患者院内死亡可能性,为临床制定有效的治疗方案提供参考.方法 采用统一标准收集2010年1月—2015年12月在余杭区第一人民医院(228例)、新疆医科大学第一附属医院(325例)就诊的553例AD患者的临床资料(一般情况、基础体征、检查指标).根据患者院内是否存活分为存活组(n=470)和死亡组(n=83),通过单因素分析筛选差异变量.再由计算机软件随机挑选患者作为样本集,通过Bayes判别分析建立Bayes判别方程.最后,利用剩余患者作为检验集,将其资料代入已建立的Bayes判别方程验证其准确性.结果 AD患者平均年龄(51.5±12.3)岁;男女比例为3.89:1;院内病死率为15.0%(83/553).最终纳入Bayes判别方程的变量为:发病时间(a1)、DeBakey分型(a2)、尿素氮(a3)、随机血糖(a4)、糖化血清蛋白(a5)、间接胆红素(a6)、国际标准化比率(INR)(a7)、纤维蛋白原(a8),得到Bayes判别方程:Q1=1.174×a1+6.813×a2+0.323×a3+0.213×a4+10.522×a5+0.171×a6+25.656×a7+1.014×a8-39.843;Q2=-13.336×a1+27.131×a2-1.928×a3-5.030×a4+35.574×a5-0.658×a6+287.333×a7-3.509×a8-1707.601.即将患者上述参数(a1~a8)分别带入Bayes判别方程中,当Q1>Q2时,认为该患者应归为存活组;当Q1<Q2时,认为该患者应归为死亡组.经检验集数据检验,符合率为98.85%,误判率为1.15%(P=0.003);经全部患者数据检验,符合率为98.73%,误判率为1.27%(P<0.001).结论 Bayes判别方程可从统计学角度对AD患者的院内结局进行初步预测.

Background Aortic dissection(AD) is an acute critical disease of the cardiovascular system characterized by acute onset and high mortality.There is lack of studies in which effective models for predicting the in-hospital outcome in patients with AD are reported.However,there is an urgent clinical need for an effective and reliable model for predicting in-hospital outcomes of AD as AD incidence has recent increases.Objective To develop Bayes formulas to predict in-hospital death of AD patients,providing a reference for the formulation of effective treatments of this disease.MethodsAccording to the unified inclusion and exclusion criteria of this study,553 cases of AD were enrolled during January 2010 to December 2015,including 228 from First People’s Hospital of Yuhang District,and 325 from the First Teaching Hospital of Xinjiang Medical University.Clinical data such as general personal characteristics,basic signs,and laboratory indices were collected.By the in-hospital outcome,participants were divided into the survival group(n=470) and death group(n=83).Differential variables were identified based on the results of univariate analysis.A sample set was selected using a computerbased random number generator,and their data were included in the Bayes formulas derived from Bayes discriminant analysis.The remaining participants were defined as a test set,and their data were included in the Bayes formulas to verify the accuracy of the formulas.Results The participants had a mean age of(51.5±12.3) years,a ratio of male to female of 3.89∶1 and an in-hospital mortality rate of 15.0%(83/553).The variables finally substituted in the Bayes formulas were onset time(a1),DeBakey types(a2),urea nitrogen(a3),random plasma glucose(a4),glycated serum protein(a5),indirect bilirubin(a6),international normalized ratio(a7),and fibrinogen(a8),and the derived formulas were Q1=-1.174×a1+6.813×a2+0.323×a3+0.213×a4+10.522×a5+0.171×a6+25.656×a7+1.014×a8-39.843 and Q2=-13.336×a1+27.131×a2-1.928×a3-5.030×a4+35.574×a5-0.658×a6+287.333×a7-3.509×a8-1 707.601.And the significance of the formulas could be explained as:after putting the aforementioned variables(a1-a8) of an AD patient into the formulas,if the value of Q1 is greater than that of Q2,then the patient is determined as a survivor,otherwise he is a deceased.The formulas were found with an accuracy rate of 98.85% and an error rate of 1.15%(P=0.003)after being tested with the data of the participants in the test set,and were identified with an accuracy rate of 98.73%,and an error rate of 1.27%(P<0.001)after being tested with the data of all the participants.Conclusion Our Bayes formulas can be used to predict the in-hospital outcome of AD patients from a statistical perspective.