This is the first meta-analysis evaluating the effect of chest compression first versus defibrillation first in patients having out-of-hospital cardiac arrest. We included four randomized, controlled clinical trials with 1503 subjects. Overall, our findings suggest that there was no significant difference between the two groups in general. However, our subgroup analyses of patients with a response interval >5 min found point estimates that pointed toward superiority of a chest compression-first approach and vice versa for the subgroup with response interval ≤5 min. The point estimate for the 1-year survival results pointed toward a lower 1-year mortality for chest compression-first patients, which was mainly driven by studies with longer EMS response times [15, 18]. However, the 95% confidence intervals of these subgroup and long-term analyses crossed 1.0, indicating insufficient precision of the effect estimates and resulting in statistical nonsignificance. These analyses were based on smaller patient numbers.
Rational for Chest Compressions Prior to Defibrillation
Chest compressions serve to empty the right ventricle (RV) and to avoid RV distension during VF, which helps to reduce the risk of occurrence of "nonperfusing" postdefibrillation rhythms (e.g., pulseless electrical activity or asystole) [30, 31]. Two experimental animal studies on ventricular defibrillation have demonstrated that chest compression first may improve defibrillation success in comparison to the standard defibrillation first approach. A randomized study in swine conducted by Berg et al. and a study by Niemann et al. in dogs both showed higher efficiency for chest compression prior to defibrillation [32, 33]. Data from a study conducted on humans showed that even short preshock pauses were found to strongly correlate with lower defibrillation success . Accordingly, a large observational study by Cobb et al. demonstrated improved survival for patients treated for out-of-hospital cardiac arrest after implementation of chest compression-first protocol compared to the preceding 42 months with the standard defibrillation-first approach . Similarly, a study including 886 patients of Bobrow et al. performed in Arizona implementing a protocol of 200 uninterrupted chest compressions before defibrillation (single shock) showed a remarkable increase in survival-to-hospital discharge, from 1.8% to 5.4% after protocol implementation [35, 36]. Yet, despite all of the above data from experimental and observational studies, our meta-analysis based on randomized clinical trials in humans shows that both treatments appear to be equivocal, with point estimates that favor chest compression first regarding long-term outcomes.
Several aspects could explain this controversy. First, findings from experimental animal studies may not apply to humans, especially since most models use electrical induction of ventricular fibrillation, which may not appropriately reflect the majority of cardiac arrests in humans . In a more recent study in swine using an acute myocardial ischemia model, 24-hr survival with a favorable neurological outcome was less likely when chest compressions were performed prior to defibrillation . Second, observational studies [27, 35] are more prone to confounding than randomized trials. Because we decided a priori to include only randomized, controlled trials in our meta-analysis, our results may differ from these large observational studies. Finally, it may be that the treatment effect of chest compression first may be dependent on the response interval from the time of call to EMS response. Further research, with patient-level data, will need to be conducted to assess whether this finding is consistent.
Short- versus longer-duration cardiac arrest
The possible difference in treatment effect for longer-lasting (response interval >5 min) makes plausible sense from a pathophysiological standpoint. Cardiac arrest (due to ventricular tachycardia/fibrillation (VT/VF)) is definitively not a static event. Rather, it is a dynamic process with sometimes continuous transitions starting with VT, transforming into coarse and then into fine amplitude VF and finally into asystole; these different electrocardiogram morphologies are obviously associated with different degrees of defibrillation success . During the course of VF high-energy phosphates are progressively depleted, which also decreases the chances for successful defibrillation .
Niemann et al. demonstrated the superiority chest compression first in a dog model , but found better outcomes for defibrillation first in a subsequent study . In this second study, VF duration was relevantly shorter (5 min versus 7.5 min in the first study). Another study conducted in dogs specifically evaluated different VF durations, showing differential results based on the duration of VF. For short-lasting VF arrests (< 3 min), defibrillation first was superior to chest compression first . It has to be considered, however, that most experimental animal studies used electrical induction of VF, which may not be identical to ischemia-induced VF . The study by Cobb et al. included in our analysis showed the most prominent benefit for chest compression first if response time was >4 min .
In 2002, Weisfeldt et al. proposed a three-phase time-sensitive model for treatment of sudden cardiac arrest: the electrical phase (early phase during the first around 0-4 min where immediate defibrillation may be optimal, the circulatory phase (4-10 min) where predefibrillation chest compressions could be meaningful, and the metabolic phase (> 10 min), where survival rates are poor in general . The authors stated in their editorial that "phase-specific research is needed to extend knowledge of the importance of time on resuscitation, such as testing early defibrillation and public access defibrillation programs during the electrical phase and testing chest compression and vasoconstrictors first during the circulatory phase." . Our findings support the view of Weisfeld et al. as illustrated in Figure 4 and as shown in the subgroup analyses of patients with longer versus those with shorter response intervals.
Limitations of this study
It has to be considered that nonstratified overall results showed odds ratios very close to 1.0; that is, no treatment effect with fairly narrow confidence (precision) intervals and with very little heterogeneity. In contrast, OR point estimates pointed toward superiority of predefibrillation chest compressions for those cardiac arrests with prolonged EMS response, while in patients with shorter EMS intervals these OR estimates pointed toward superiority of a defibrillation-first approach (Figures 5 and 6). Owing to the smaller sample sizes in these subgroups, confidence intervals were wider due to reduced precision of these estimates. The confidence intervals for these subgroup analyses crossed 1.0; i.e., the result was statistically not significant. It is possible that there is in fact a difference that was not detected by our analysis due to limited statistical power. An interaction between optimal treatment and response time is further supported by the observation that the odds ratios were influenced by the average response intervals of the individual studies (Figure 3 and Additional file 1). However, the meta-regression analysis (Additional file 1), even though in line with the findings of the subgroup analyses, has to be interpreted with care because it is based on summary measure (mean response intervals of each study) and not on individual response intervals. Meta-analyses are useful for synthesizing the literature and to explore areas for further exploration rather than to provide a definitive conclusion. Future research based on this meta-analysis could be conducted with patient-level data to assess whether the overall pooled results are consistent with the individual-level data.
RCT data are considered the "golden standard" and superior to observational studies. Clearly, the latter are more prone to be biased by confounding, and, accordingly, we considered RCT exclusively in this meta-analysis. Nevertheless, there are caveats for RCT also ; this is especially true in the context of human emergency medicine research. The vast majority of patients assessed for inclusion in these trials were finally not eligible because of predefined exclusion criteria or owing to logistical reasons. Thus, the patient selection associated with RCT potentially complicates generalizability of findings into routine clinical practice. For example, bystander CPR rate ranged from 54-64% in three of the included trials, while the AHA estimates the average bystander CPR rate in the United States to be 31.4% . Future research will need to be conducted on communities that may be more generalizable than the study populations in this analysis.
A further limitation of this study is the heterogeneity of the study protocols. Three of the four included trials use the 2000 guidelines with a "three-shock protocol" [16–18],
while one study utilized a single shock application (as advocated in the current 2005 guidelines) in the chest compression first group . All four studies did not control for the quality of chest compressions. The quality of chest compressions has a key impact on outcome and is often insufficient, even for in-hospital cardiac arrests  and even in some experimental studies . We cannot exclude that the quality of compressions in the included studies was insufficient, and as a consequence, the studies were unable to show a benefit. Because of the differences in study protocols, we chose to use a random effects model rather than a fixed-effect model for data analysis.
Finally, we did not have the complete set of individual patient data, and our analyses are thus based on study-level data. Therefore, we could not adjust the analysis for covariables. For example, the 1-year survival data for the study by Jost et al.  are based on Kaplan-Meier survival estimates, which showed a survival probability of 10.6% in the intervention group and 7.6% in the control group (P = 0.45).