Development and validation of risk models to predict outcomes following in-hospital cardiac arrest attended by a hospital-based resuscitation team

Abstract

Aim: The National Cardiac Arrest Audit (NCAA) is the UK national clinical audit for in-hospital cardiac arrest. To make fair comparisons among health care providers, clinical indicators require case mix adjustment using a validated risk model. The aim of this study was to develop and validate risk models to predict outcomes following in-hospital cardiac arrest attended by a hospital-based resuscitation team in UK hospitals.

Methods: Risk models for two outcomes-return of spontaneous circulation (ROSC) for greater than 20min and survival to hospital discharge-were developed and validated using data for in-hospital cardiac arrests between April 2011 and March 2013. For each outcome, a full model was fitted and then simplified by testing for non-linearity, combining categories and stepwise reduction. Finally, interactions between predictors were considered. Models were assessed for discrimination, calibration and accuracy.

Results: 22,479 in-hospital cardiac arrests in 143 hospitals were included (14,688 development, 7791 validation). The final risk model for ROSC>20min included: age (non-linear), sex, prior length of stay in hospital, reason for attendance, location of arrest, presenting rhythm, and interactions between presenting rhythm and location of arrest. The model for hospital survival included the same predictors, excluding sex. Both models had acceptable performance across the range of measures, although discrimination for hospital mortality exceeded that for ROSC>20min (c index 0.81 versus 0.72).

Conclusions: Validated risk models for ROSC>20min and hospital survival following in-hospital cardiac arrest have been developed. These models will strengthen comparative reporting in NCAA and support local quality improvement.

Keywords: Cardiopulmonary resuscitation; Heart arrest; Hospital mortality; Models; Risk adjustment; Statistical.

Fig. 1

Relationship between age and: (A) return of spontaneous circulation greater than 20 min; (B) hospital survival. CI, confidence interval; ROSC, return of spontaneous circulation. Odds ratios and confidence intervals have been calculated relative to age 70 years (and therefore converge to an odds ratio of 1 at this point).

Fig. 2

Calibration plots for return of spontaneous circulation greater than 20 min (left) and hospital survival (right) in the development, validation and external validation datasets. ROSC, return of spontaneous circulation. Observed survival (with 95% confidence interval) plotted against predicted survival in ten equal sized groups, based on models using coefficients fitted in the development dataset.