Derivation and validation of a risk score for predicting mortality among inpatients following rapid response team activation.

Abstract

Despite mature rapid response systems (RRS) for clinical deterioration, individuals activating RRS have poor outcomes, with up to one in four dying in hospital. We aimed to derive and validate a risk prediction tool for estimating risk of 28-day mortality among hospitalised patients following rapid response team (RRT) activation. Analysis of prospectively collected data on 1151 consecutive RRT activations involving 800 inpatients at a tertiary adult hospital. Patient characteristics, RRT triggers and actions, and mortality were ascertained from medical records and death registries. A multivariable risk prediction regression model, derived from 600 randomly selected patients, was validated in the remaining 200 patients. Main outcome was accuracy of weighted risk score (measured by area under receiver operator curve (AUC)) and performance characteristics for various cut-off scores. At 28 days, 150 (18.8%) patients had died. Increasing age, emergency admission, chronic liver disease, chronic kidney disease, malignancy, after-hours RRT activation, increasing National Early Warning Score, major/intense RRT intervention and multiple RRT activations were predictors of mortality. The risk score (0-105) in derivation and validation cohorts had AUCs 0.86 (95% CI 0.82 to 0.89) and 0.82 (95% CI 0.75 to 0.90), respectively. In the validation cohort, cut-off score of 32.5 or higher maximised sensitivity: 81.6% (95% CI 68.4% to 92.1%), specificity: 56.2% (95% CI 49.4% to 63.6%), positive likelihood ratio (LR): 1.9 (95% CI 1.5 to 2.3) and negative LR: 0.3 (95% CI 0.2 to 0.6). A validated risk score predicted risk of post-RRT death with more than 80% accuracy, helping to identify patients for whom targeted rescue care may improve survival.