Gabriel Rodney A, Waterman Ruth S, Kim Jihoon, Ohno-Machado Lucila
From the Departments of *Biomedical Informatics and †Anesthesiology, University of California, San Diego, San Diego, California.
Anesth Analg. 2017 May;124(5):1529-1536. doi: 10.1213/ANE.0000000000001827.
A predictive model that can identify patients who are at an increased risk for prolonged postanesthesia care unit (PACU) stay could help optimize resource utilization and case sequencing. Although previous studies identified some predictors, there is not a model that only utilizes various patients demographic and comorbidities, that are already known preoperatively, and that may affect PACU length of stay for outpatient procedures requiring the care of an anesthesiologist.
We collected data from 4151 patients at a single institution from 2014 to 2015. The data set was split into a training set (cases before 2015) and a test set (cases during 2015). Bootstrap samples were chosen (R = 1000 replicates) and a logistic regression model was built on the samples using a combined method of forward selection and backward elimination based on the Akaike Information Criterion. The trained model was applied to the test set. Model performance was evaluated with the area under the receiver operating characteristic (ROC) Curve (AUC) for discrimination and the Hosmer-Lemeshow (HL) test for goodness-of-fit.
The final model had 5 predictor variables for prolonged PACU length of stay, which included the following: morbid obesity, hypertension, surgical specialty, primary anesthesia type, and scheduled case duration. The model had an AUC value of 0.754 (95% confidence interval 0.733-0.774) on the training set and 0.722 (95% confidence interval 0.698-0.747) on the test set, with no difference between the 2 ROC curves (P = .06). The model had good calibration for the data in both the training and test data set indicated by nonsignificant P values from the HL test (P = .211 and .719 for the training and test set, respectively).
We developed a predictive model with excellent discrimination and goodness-of-fit that can help identify those at higher odds for extended PACU length of stay. This information may help optimize case-sequencing methodologies.
一个能够识别出麻醉后监护病房（PACU）停留时间延长风险增加的患者的预测模型，有助于优化资源利用和病例排序。尽管先前的研究确定了一些预测因素，但尚无一个仅利用术前已知的各种患者人口统计学和合并症的模型，这些因素可能会影响需要麻醉医生护理的门诊手术的PACU停留时间。
我们收集了2014年至2015年期间一家机构4151例患者的数据。数据集被分为训练集（2015年之前的病例）和测试集（2015年期间的病例）。选择了自助抽样（R = 1000次重复），并基于赤池信息准则，采用向前选择和向后消除的组合方法，在样本上建立了逻辑回归模型。将训练好的模型应用于测试集。使用受试者工作特征（ROC）曲线下面积（AUC）进行区分评估模型性能，并使用Hosmer-Lemeshow（HL）检验进行拟合优度评估。
最终模型有5个预测PACU停留时间延长的变量，包括：病态肥胖、高血压、手术专科、主要麻醉类型和预定病例持续时间。该模型在训练集上的AUC值为0.754（95%置信区间0.733 - 0.774），在测试集上为0.722（95%置信区间0.698 - 0.747），两条ROC曲线之间无差异（P = 0.06）。HL检验的P值不显著，表明该模型在训练集和测试数据集中对数据均有良好的校准（训练集和测试集的P值分别为0.211和0.719）。
我们开发了一个具有出色区分度和拟合优度的预测模型，该模型有助于识别PACU停留时间延长几率较高的患者。这些信息可能有助于优化病例排序方法。
