Predictive value of a tiered escalation response system: A case control study

Abstract

Objective: 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).

Design: This was a nested, matched case-control study.

Setting: The study setting involved a tertiary referral hospital.

Participants: Cases experienced an event, and controls were matched patients without an event.

Outcome measures: 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.

Results: 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.

Conclusion: 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.

Keywords: Observation and response charts; Rapid response systems; Rapid response teams; Track and trigger; Vital signs.