Abstract：
Rapid response systems designed to detect and respond to clinical deterioration often incorporate a multitiered, escalation response. We sought to determine the \'predictive strength\' of commonly used triggers, and tiers of escalation, for predicting a rapid response team (RRT) call, unanticipated intensive care unit admission, or cardiac arrest (events).
This was a nested, matched case-control study.
The study setting involved a tertiary referral hospital.
Cases experienced an event, and controls were matched patients without an event.
Sensitivity and specificity and area under the receiver operating characteristic curve (AUC) were measured. Logistic regression determined the set of triggers with the highest AUC.
There were 321 cases and 321 controls. Nurse triggers occurred in 62%, medical review triggers in 34%, and RRT triggers 20%. Positive predictive value of nurse triggers was 59%, that of medical review triggers was 75%, and that of RRT triggers was 88%. These values were no different when modifications to triggers were considered. The AUC was 0.61 for nurses, 0.67 for medical review, and 0.65 for RRT triggers. With modelling, the AUC was 0.63 for the lowest tier, 0.71 for next highest, and 0.73 for the highest tier.
For a three-tiered system, at the lowest tier, specificity of triggers decreases, sensitivity increases, but the discriminatory power is poor. Thus, there is little to be gained by using a rapid response system with more than two tiers. Modifications to triggers reduced the potential number of escalations and did not affect tier discriminatory value.
This was a nested, matched case-control study.
The study setting involved a tertiary referral hospital.
Cases experienced an event, and controls were matched patients without an event.
Sensitivity and specificity and area under the receiver operating characteristic curve (AUC) were measured. Logistic regression determined the set of triggers with the highest AUC.
There were 321 cases and 321 controls. Nurse triggers occurred in 62%, medical review triggers in 34%, and RRT triggers 20%. Positive predictive value of nurse triggers was 59%, that of medical review triggers was 75%, and that of RRT triggers was 88%. These values were no different when modifications to triggers were considered. The AUC was 0.61 for nurses, 0.67 for medical review, and 0.65 for RRT triggers. With modelling, the AUC was 0.63 for the lowest tier, 0.71 for next highest, and 0.73 for the highest tier.
For a three-tiered system, at the lowest tier, specificity of triggers decreases, sensitivity increases, but the discriminatory power is poor. Thus, there is little to be gained by using a rapid response system with more than two tiers. Modifications to triggers reduced the potential number of escalations and did not affect tier discriminatory value.
摘要：
目的：设计用于检测和应对临床恶化的快速反应系统通常包含多层，升级响应。我们试图确定常用触发器的“预测强度”，和升级的层次，为了预测快速反应小组(RRT)的呼叫，意外的重症监护室入院，或心脏骤停（事件）。
方法：这是一个嵌套的，匹配的病例对照研究。
方法：研究涉及三级转诊医院。
方法：病例经历了一个事件，和对照组为无事件的匹配患者.
方法：测量灵敏度和特异性以及受试者工作特征曲线（AUC）下的面积。逻辑回归确定具有最高AUC的一组触发因素。
结果：共321例，对照组321例。护士触发事件发生在62%，医疗审查触发34%，和RRT触发20%。护士触发因素的阳性预测值为59%，医学审查触发因素是75%，RRT触发因素为88%。当考虑对触发器的修改时,这些值没有不同。护士的AUC是0.61，0.67用于医学审查，和0.65的RRT触发器。通过建模，最低层的AUC为0.63，下一个最高的0.71，最高层为0.73。
结论：对于三层系统，在最低层，触发因素的特异性降低，灵敏度增加，但是歧视能力很差。因此，使用两层以上的快速反应系统几乎没有什么好处。对触发器的修改减少了潜在的升级次数，并且不影响层的歧视性价值。
方法：这是一个嵌套的，匹配的病例对照研究。
方法：研究涉及三级转诊医院。
方法：病例经历了一个事件，和对照组为无事件的匹配患者.
方法：测量灵敏度和特异性以及受试者工作特征曲线（AUC）下的面积。逻辑回归确定具有最高AUC的一组触发因素。
结果：共321例，对照组321例。护士触发事件发生在62%，医疗审查触发34%，和RRT触发20%。护士触发因素的阳性预测值为59%，医学审查触发因素是75%，RRT触发因素为88%。当考虑对触发器的修改时,这些值没有不同。护士的AUC是0.61，0.67用于医学审查，和0.65的RRT触发器。通过建模，最低层的AUC为0.63，下一个最高的0.71，最高层为0.73。
结论：对于三层系统，在最低层，触发因素的特异性降低，灵敏度增加，但是歧视能力很差。因此，使用两层以上的快速反应系统几乎没有什么好处。对触发器的修改减少了潜在的升级次数，并且不影响层的歧视性价值。
