Bayman Emine Ozgur, Dexter Franklin, Todd Michael M
From the Departments of Anesthesia and Biostatistics, University of Iowa, Iowa City, Iowa (E.O.B.); Division of Management Consulting, Department of Anesthesia, University of Iowa, Iowa City, Iowa (F.D.); and Department of Anesthesia, University of Iowa, Iowa City, Iowa (M.M.T.).
Anesthesiology. 2016 Feb;124(2):322-38. doi: 10.1097/ALN.0000000000000920.
One anesthesiologist performance metric is the incidence of "prolonged" (15 min or longer after dressing complete) times to extubation. The authors used several methods to identify the performance outliers and assess whether targeting these outliers for reduction could improve operating room workflow.
Time to extubation data were retrieved for 27,757 anesthetics and 81 faculty anesthesiologists. Provider-specific incidences of prolonged extubation were assessed by using unadjusted frequentist statistics and a Bayesian model adjusted for prone positioning, American Society of Anesthesiologist's base units, and case duration.
20.31% of extubations were "prolonged," and 40% of anesthesiologists were identified as outliers using a frequentist approach, that is, incidence greater than upper 95% CI (20.71%). With an adjusted Bayesian model, only one anesthesiologist was deemed an outlier. If an average anesthesiologist performed all extubations, the incidence of prolonged extubations would change negligibly (to 20.67%). If the anesthesiologist with the highest incidence of prolonged extubations was replaced with an average anesthesiologist, the change was also negligible (20.01%). Variability among anesthesiologists in the incidence of prolonged extubations was significantly less than among other providers.
Bayesian methodology with covariate adjustment is better suited to performance monitoring than an unadjusted, nonhierarchical frequentist approach because it is less likely to identify individuals spuriously as outliers. Targeting outliers in an effort to alter operating room activities is unlikely to have an operational impact (although monitoring may serve other purposes). If change is deemed necessary, it must be made by improving the average behavior of everyone and by focusing on anesthesia providers rather than on faculty.
麻醉医生的一项绩效指标是“延长”（包扎完成后15分钟或更长时间）拔管时间的发生率。作者使用了几种方法来识别绩效异常值，并评估针对这些异常值进行降低是否可以改善手术室工作流程。
检索了27757例麻醉和81名麻醉科教员的拔管时间数据。通过使用未调整的频率统计方法和针对俯卧位、美国麻醉医师协会基本单位和病例持续时间进行调整的贝叶斯模型，评估了特定提供者的延长拔管发生率。
20.31%的拔管是“延长的”，使用频率方法，40%的麻醉医生被确定为异常值，即发生率高于95%置信区间上限（20.71%）。通过调整后的贝叶斯模型，只有一名麻醉医生被视为异常值。如果由一名普通麻醉医生进行所有拔管操作，延长拔管的发生率变化可忽略不计（降至20.67%）。如果将延长拔管发生率最高的麻醉医生换成一名普通麻醉医生，变化同样可忽略不计（20.01%）。麻醉医生之间延长拔管发生率的变异性明显小于其他提供者。
与未调整的、非分层的频率方法相比，具有协变量调整的贝叶斯方法更适合于绩效监测，因为它不太可能将个体错误地识别为异常值。针对异常值以改变手术室活动的做法不太可能产生实际影响（尽管监测可能有其他用途）。如果认为有必要进行改变，必须通过改善每个人的平均行为并关注麻醉提供者而非教员来实现。
