Gasping as a predictor of short- and long-term outcomes in patients with cardiac arrest: a systematic review and meta-analysis

2.1 Search strategy

The literature of cardiac arrest and gasping from PubMed, Embase, Medline, Cochrane, and Web of Science up to September 2020 were searched, categorized, and analyzed with the search terms “cardiac arrest”, “CPR”, “heart arrest”, “out-of-hospital”, and “gasping”, “agonal breathing”, “agonal respiration”, “abnormal breathing”. In this analysis, only articles published in English were included.

2.2 Selection criteria

The primary analysis of randomized trials or secondary analysis of observational cohort studies were eligible for inclusion by comparing clinical outcomes between patients with and without gasping during cardiac arrest. All the studies were selected by two independent reviewers based on the following criteria: (a) the subjects were human adults; (b) eligible clinical outcomes include ROSC, survival to discharge, survival to one-month, or Cerebral Performance Category (CPC) score; (c) sufficient published data for estimating the risk ratio (RR) with 95% CI. Studies were excluded if there was no sufficient data for pooling.

2.3 Quality assessment

The quality of a selected study was assessed by using a semi-quantification principle star system, consists of eight items in three forums including crowd selection, comparability, exposure assessment, and outcome evaluation utilizing the Newcastle-Ottawa scale [16].

2.4 Data extraction

Data was extracted by two independent reviewers from each eligible study. ROSC was the primary outcome variable, while survival to hospital discharge, survival to one-month, and one-year survival with Cerebral Performance Category (CPC) score $\le$ 2 were the secondary outcome variables.

The study characteristics included the first author’s names, publication years, countries, type of recordings, study designs, and the total number of patients who suffered CA.

2.5 Statistical analysis

All analyses were conducted with Review Manager (version 5.3). Risk ratios (RR) and 95% CIs were calculated to determine the association of gasping and CA. The pooled RR and 95% CI were calculated by fixed-effect or random-effect model. If there was significant heterogeneity (P $<$ 0.05, I${}^2$$>$ 50%), a random-effects model was adopted; otherwise, fixed-effects models were applied if there was no significant heterogeneity (P $\ge$ 0.05, I${}^2$$\le$ 50%). Heterogeneity was tested by the Q-test based on the Chi-square or I${}^2$ statistic test. Whenever heterogeneity was present, sensitivity analyses were performed by the absence of one trial in each turn to test the influence of a single study on the overall pooled estimate. Publication bias was evaluated by visually inspecting funnel plots.

3.1 Study selection

Initially, the keywords search returned 407 records, of which 225 duplicated studies and 64 studies that referred to animals were excluded. After scanning the abstracts of the remaining potential eligible studies, 12 reviews, 5 case reports, and 83 irrelevant subjects of gasping and outcomes in patients with CA were further excluded. After full-text articles were assessed for eligibility, 11 studies were excluded, 5 of which failed to provide enough data and 6 reported no outcomes of interest. Finally, 8 studies from 7 articles [5, 6, 17, 18, 19, 20, 21], which involved patients who suffered a cardiac arrest, qualified for the inclusion criteria in the meta-analysis to estimate pooled data. Marterns et al. [18] reported two different databases (1983-1987 (1) and 1991-1993 (2)). The flow of studies, including the reasons for inclusion and exclusion of studies, is shown in Fig. 1.

Fig. 1.Selection of studies for the meta-analysis.

3.2 Study characteristic and validity assessment

The characteristics of the studies are shown in Table 1. All seven studies were observational, which included prospective [19, 20, 21] and retrospective cohort studies. Only one study [20] involved an in-hospital cardiac arrest, the others were out-of-hospital cardiac arrest. Among the seven cohorts, four studies from three articles reported the ROSC rate of gasping compared with no gasping patients, while six studies from five articles reported survival to hospital discharge, three studies reported favorable neurological outcomes, and four studies reported the relationship between gasping and cardiac rhythm during cardiac arrest. The quality assessments of the seven studies are shown in Table 2.

3.3 Meta-analysis

3.3.1 ROSC

Four studies in three articles estimated the association of gasping with ROSC in patients suffering a cardiac arrest (Fig. 2). When low heterogeneity across the studies (Chi${}^2$ = 5.97, df = 3, P = 0.11, I${}^2$ = 50.0%) was obtained, a fixed-effect model was applied to estimate a pooled RR of 2.08 (95% CI: 1.93-2.25), indicating that patients who suffered a cardiac arrest with gasping were 2.08 times more likely to develop ROSC compared to those without gasping. A funnel plot was utilized to assess the publication bias of all the studies (Fig. 3). We excluding a single study each time for sensitivity analysis. The result did not change the statistical outcome when the studies were excluded.

Fig. 2.Forrest plot showing the pooled estimate of the effect between gasping and no gasping. OR, risk ratio; CI, confidence limit; M-H: Mantel Haenszel method.

Fig. 3.Funnel plot of pooling of gasping effect on different outcome. The hollow dots and dotted line indicate individual studies and 95% confidence intervals, respectively.

3.4 Survival to discharge

In six studies involving 8746 patients after cardiac arrest, 331 out of 1464 patients (22.6%) survived to hospital discharge in the gasping group, while 432 out of 7282 patients (5.9%) in no gasping group survived to discharge. In view of the very low heterogeneity among these studies (Chi${}^2$ = 3.23, df = 5, P = 0.66, I${}^2$ = 0%) the fixed-effects model was then adopted. A pooled analysis indicated that there was a significant difference of survival to discharge between the patients who gasped during cardiac arrest compared with those who did not (RR = 3.75, 95% CI = 3.27-4.29, P $<$ 0.001) (Fig. 2). As shown in Fig. 3, a funnel plot was used to assess the publication bias of all studies; the points representing the evaluated studies in the funnel plot were concentrated at the top, indicating that the studies had high precision and large sample sizes. A sensitivity analysis was performed by the same method above which excluding every single study each time and the results did not change the outcome statistically by excluding any studies, indicating that the outcome showed a high stability.

3.5 CPC $\le$ 2

Only three studies showed outcomes on the association of gasping with neurologically favorable one-year survival. These cohort studies were included to estimate the pooled RR of 8.02 (95% CI: 5.56, 11.57; P $<$ 0.001) (Fig. 2). Although the pooled result of the trials showed a significantly increased chance to achieve a good neurological outcome when patients gasped during cardiac arrest, the association of gasping and CPC $\le$ 2 was highly heterogeneous (Chi${}^2$ = 10.34; df = 2; P = 0.006; I${}^2$ = 81%). The funnel plots depicted in Fig. 3 were generally asymmetrical, indicating the presence of publication bias. Although only three observational studies were involved in this analysis, sensitivity analysis demonstrated that the absence of any study each time did not influence the outcome of the statistics, indicating that the outcome was stable.

3.6 Shockable rhythm

Meta-analysis of four studies showed a significant increase in a shockable cardiac rhythm during cardiac arrest in gasping patients compared with patients with no gasping. A fixed-effect model estimated the pooled RR as 2.82 (95% CI: 2.50, 3.18; P $<$ 0.001) with low heterogeneity (Chi${}^2$ = 5.76; df = 3; P = 0.12; I${}^2$ = 48%) (Fig. 2). The patients were much more likely to have a shockable cardiac rhythm during cardiac arrest who were gasped. A funnel plot was used to assess the publication bias of all studies. The funnel plot (Fig. 3) was almost symmetrical, indicating that publication bias may exist but was not apparent. A sensitivity analysis was performed as the same method above. The results showed a decreased heterogeneity (I${}^2$ = 95% to I${}^2$ = 48%) and higher RR (2.04 to 2.82) when excluding the study of Bobrow2018 [6]. Since the data from the Bobrow2018 study only referred to ventricular fibrillation (VF) but not a totally shockable rhythm, we deleted the data from this study. However, the absence of any study data did not change the outcome statistically since the affected quantity was small.