Delays in Cardiopulmonary Resuscitation, Defibrillation, and Epinephrine Administration All Decrease Survival in In-Hospital Cardiac Arrest

Nicholas G. Bircher, M.D.; Paul S. Chan, M.D., M.Sc.; Yan Xu, Ph.D.


Anesthesiology. 2019;130(3):414-422. 

In This Article

Materials and Methods

Patient Population

The Get With The Guidelines–Resuscitation (formerly known as the National Registry of Cardiopulmonary Resuscitation) is an American Heart Association–sponsored prospective multicenter observational registry of in-hospital cardiac arrest. The design of the Get With The Guidelines–Resuscitation database has been previously described.[2] Briefly, all patients with cardiac arrest (defined as the absence of a palpable central pulse, apnea, and unconsciousness) and without do-not-resuscitate orders are enrolled by hospital quality improvement personnel who have received specialized training. Patients eligible for enrollment are identified from multiple sources, including but not limited to cardiac arrest flow sheets, hospital paging system logs, and routine checks of code carts. Standardized reporting using Utstein-style definitions[5] are used for patient variables and outcomes. This study was approved by the Institutional Review Board of the University of Pittsburgh, Pittsburgh, Pennsylvania. Our statistical analysis plan was approved the National Registry of Cardiopulmonary Resuscitation Adult Research Task Force on April 9, 2009, before accessing the data. This analysis and manuscript were approved by the Executive Database Steering Committee in accordance with the Get With The Guidelines Publication Policy.[6]

Between 2000 and 2008, we identified 132,950 patients with an in-hospital cardiac arrest within Get With The Guidelines–Resuscitation with complete comorbidity data for our model. We excluded 21,212 episodes of recurrent arrest to focus on index in-hospital cardiac arrest events (Figure 1). As we evaluated the effect of time to initiation of CPR on outcomes, we excluded 21,334 patients with an unwitnessed in-hospital cardiac arrest and 6,998 patients without information on time to initiation of CPR. We also excluded 5,549 patients with implausible time to initiation of CPR (i.e., negative times [n = 5,036] and time to initiation of CPR of 7 min or more [n = 513]). Our study population comprised 77,857 patients with a witnessed in-hospital cardiac arrest with time to initiation of CPR of 0 to 6 min. We excluded 8,954 patients for missing values for times from CPR to defibrillation or epinephrine treatment. We also excluded patients missing survival data and for negative or outlier times from CPR to defibrillation or epinephrine treatment (11,528 excluded patients, outliers as defined by Tukey box plot of more than 11 min in the defibrillation group or more than 9 min in the epinephrine group). The final sample sizes were 11,002 in the defibrillation group and 46,310 in the epinephrine group.

Figure 1.

Study cohort selection. TTC, time to initiation of cardiopulmonary resuscitation.

Independent Variables and Study Outcomes

Our two main independent variables were established a priori and were (1) time to initiation of CPR and (2) time from the initiation of CPR to treatment, defined as either defibrillation or epinephrine. Time to initiation of CPR was defined as the difference between the recorded clock time for the determination of pulselessness and the recorded clock time for the beginning of chest compressions. Similarly, because we aimed to study the influence of delay (as opposed to the influence of shockable vs. nonshockable rhythms), time to treatment was defined as the difference between the recorded clock time for either the first defibrillation attempt or the administration of epinephrine and time for initiation of CPR. For patients who received both defibrillation and epinephrine treatments, this interval was defined by whichever intervention was recorded as being given first. Survival to discharge was established a priori as our primary outcome.

Statistical Analysis

Baseline differences between the defibrillation- and epinephrine-treated groups were examined. Continuous variables were compared with the median and Kruskal–Wallis tests, and categorical variables were compared using the chi-square test. We then constructed multivariable logistic regression models to examine the associations between survival as an outcome and time to initiation of CPR and time from CPR to defibrillation or epinephrine treatment as ordinal categorical predictors. To maximize statistical power and to allow comparison of the two groups, we developed models in which the groups were combined, as well as separate models for each group. To further enhance statistical power, in post hoc exploratory analysis, we evaluated various binning strategies (supplemental table 1,, describing the partitioning of categories for each binning strategy) in addition to univariate and pointwise analysis for time to initiation of CPR and time to treatment. We dichotomized time to initiation of CPR into ranges of 0 to 2 and 3 to 6 min. We also categorized time to defibrillation and time to epinephrine treatment into ranges 0 to 2, 3 to 5, 6 to 8, and 9 to 11 min.

After screening study variables for collinearity, we included the following covariates in our model: age, sex, race, whether the patient was monitored, location of cardiac arrest, initial rhythm, illness category (medical cardiac, medical noncardiac, surgical cardiac, surgical noncardiac), and comorbidities present within 24 h of cardiac arrest (table 1; supplemental table 2,, depicting a complete list of group baseline characteristics). Moreover, we included in the model interventions in place at the time of cardiac arrest, including mechanical ventilation, various vasopressors, and other invasive procedures (supplemental table 2,, depicting a complete list of group baseline characteristics). In post hoc testing, we assessed possible clustering effects at the hospital level (i.e., between hospitals) in three separate analyses: (1) adding facility as a covariate to our model, (2) our model using the generalized estimating equation (details in supplemental table 3,, detailing model evaluation), and (3) a two-stage hierarchical analysis using facility and location (within the hospital), as well as the other covariates in our model. We also did post hoc sensitivity analysis by excluding patients who received defibrillation for a nonshockable rhythm or epinephrine for a shockable rhythm. In addition, we performed fractional polynomial analysis and logistic regression diagnostics using Stata/SE 15.1 (StataCorp LLC, USA). All other analyses were conducted using SPSS 22 to 25 (IBM SPSS, USA) and were assessed at a two-sided significance level of 0.05.