目的：通过决策树算法整合多个临床常用指标，形成决策树模型，探讨决策树模型预测综合ICU机械通气患者拔管结果的价值。方法选取2008年12月—2012年12月武警后勤学院附属医院综合ICU收治的气管插管机械通气时间超过48 h的患者571例，对患者进行自主呼吸试验（ SBT），由主治医生判断患者是否耐受SBT。将SBT成功的患者分别纳入拔管成功（ ES）组和拔管失败（ EF）组。记录两组患者一般资料及在SBT开始后1、30、60 min测量患者呼吸浅快指数（ RSBI ）、分钟通气量（ VE ）、口腔阻断压（ P0.1）、P0.1× RSBI 和 SBT 30 min 时 RSBI 变化率（ΔRSBI30）。结果 SBT成功者451例，其中ES组376例，EF组75例。两组患者年龄、平均动脉压、急性生理和慢性健康状况评分Ⅱ（APACHE Ⅱ）比较，差异均无统计学意义（P>0.05）。ES组心率、机械通气时间低于EF组（P<0.05）。组间比较显示，SBT开始后1、30、60 min ES组RSBI、P0.1、P0.1× RSBI均低于EF组（P<0.001）；SBT开始后1、30、60 min两组VE比较，差异无统计学意义（ P>0.05）。组内比较显示：ES组RSBI、P0.1× RSBI在SBT开始后30、60 min均低于SBT开始后1 min，VE、P0.1在SBT开始后30、60 min均高于SBT开始后1 min（P<0.05）；EF组RSBI、VE、P0.1、P0.1× RSBI在SBT开始后30、60 min均高于SBT开始后1 min（P<0.05）。ES组ΔRSBI30为（98±36）%，低于EF组的（130±63）%（t=-6.200，P<0.001）。各生理学参数预测ES准确性比较，差异有统计学意义（χ2=53.4，P<0.05）。以P0.1× RSBI30≤384 cm H2O·次· min-1· L-1预测ES的ROC曲线下面积（AUC）为（0.87±0.03），与AUC=0.5比较，差异有统计学意义（ z=13.8，P<0.001）。CRT分析选择P0.1× RSBI30、ΔRSBI30作为解释变量对患者进行分类，决策树的目标变量为ES。决策树模型预测ES的灵敏度为100.0%，特异度为81.3%，阳性预测值为96.4%，阴性预测值为100.0%，准确性为96.8%，AUC为（0.91±0.02），与AUC=0.5比较，差异有统计学意义〔95%CI（0.88，0.93），z=17.96，P<0.001〕。决策树模型预测ES的AUC和准确性均高于P0.1× RS-BI30，差异有统计学意义（z =2.168，P =0.047；χ2=29.2，P <0.001）。结论决策树模型包含 P0.1× RSBI30、ΔRSBI30两个变量，对于综合ICU已经通过SBT的机械通气患者，该决策树可以准确预测其拔管结果。%Objective To explore the values of decision-making tree( DMT) in prediction of the extubation out-comes of patients with comprehensive ICU mechanical ventilation. Methods From December 2008 to December 2012, in De-partment of Respiratory and Critical Care Medicine of Affiliated Hospital of Armed Police Logistics Institute, 571 patients who had intubation and mechanical ventilation for more than 48 h underwent spontaneous breathing trial( SBT). The attending doctors judged whether patients tolerated SBT. The patients who had successful SBT were divided into extubation success( ES) group, those who failed in SBT divided into extubation failing ( EF ) group. The general information was recorded, rapid shallow breathing index(RSBI), minute ventilation(VE), mouth occlusion pressure(P0.1), P0.1 × RSBI were measured at minutes 1, 30, 60 after SBI, RSBI change rate at SBT 30 min(ΔRSBI30 ). Results A total of 451 patients were enrolled, 376 in ES group, 75 in EF group. There was no significant difference in age, mean arterial pressure, APACHE Ⅱ score between the 2 groups(P>0. 05). The heart rate was lower, mechanical ventilation time shorter in ES group than in EF group, the difference was significant(P<0. 05). In group comparison, RSBI, P0.1, P0.1 × RSBI were lower in ES group than in EF group at mi-nutes 1, 30, 60 after SBT starting(P<0. 001), there was no difference in VE between the 2 groups(P>0. 05). In intra-group comparison in ES group, RSBI, P0. 1 × RSBI were lower, VE, P0. 1 higher at minutes 30, 60 than at minute 1 after SBT, the difference was significant(P<0. 05);in EF group, RSBI, VE, P0.1, P0.1 × RSBI were higher at minutes 30, 60 than at minute 1(P<0. 05). ΔRSBI30 of ES group was(98 ±36)%, lower than that of EF group(130 ±63)%(t= -6. 200, P<0. 001). There was difference in accuracy of physiological parameters predicting ES(χ2 =53. 4, P<0. 05). P0. 1 × RSBI30≤384 cm H2O·breaths·min-1· L-1 AUC was(0. 87 ±0. 03), different from AUC=0. 5(z=13. 8, P<0. 001). CRT anal-ysis choosing P0. 1 × RSBI30 , △RSBI30 as explaining variable to classify the patients, the object variable of DMT was ES. The sensitivity of DMT mode predicting ES was 100. 0%, specificity 81. 3%, positive predictive value 96. 4%, negative predictive value 100. 0%, accuracy 96. 8%, AUC(0. 91 ± 0. 02), different from AUC=0. 5〔95%CI(0. 88, 0. 93), z=17. 96, P<0. 001〕. The AUC and accuracy of DMT predicting ES were both higher than P0. 1 × RSBI30 , the difference was significant( z=2. 168, P=0. 047;χ2 =29. 2, P<0. 001). Conclusion DMT contains 2 variables P0. 1 × RSBI30 , △RSBI30 . For the pa-tients who have had SBT mechanical ventilation, DMT can predict extubation outcomes accurately.
展开▼
