Impact of transient return of spontaneous circulation and downtime on neurological outcomes following extracorporeal cardiopulmonary resuscitation for refractory out-of-hospital cardiac arrest: a single-center retrospective study

Abstract

Background

Despite recent advancements in cardiopulmonary resuscitation (CPR), the prognosis of out-of-hospital cardiac arrest (OHCA) remains poor. Extracorporeal cardiopulmonary resuscitation (ECPR) offers potential benefits to improve outcomes of patients with refractory OHCA. Identifying the factors associated with neurological outcomes at 180 days and their interactions remains critical for optimizing patient selection.

Methods

This single-center retrospective study analyzed consecutive patients with OHCA treated with ECPR at Kurume University Hospital between 2016 and 2023. Factors associated with the neurological outcomes at 180 days were evaluated using logistic regression and decision tree analyses.

Results

A total of 65 patients were included. Favorable neurological outcomes at 180 days were recorded in 18 (27.7%) patients. Transient return of spontaneous circulation (ROSC) (adjusted odds ratio (aOR): 8.37, 95% confidence interval (CI): 2.20–38.79) and shorter downtime (aOR: 0.95, 95% CI: 0.90–0.99) were independently associated with favorable neurological outcomes. Decision tree analysis revealed that the presence of transient ROSC and bystander CPR are upstream factors associated with favorable neurological outcomes, and that the absence of transient ROSC with prolonged downtime was associated with unfavorable outcomes.

Conclusion

Transient ROSC and shorter downtime were independently associated with favorable neurological outcomes at 180 days in patients treated with ECPR. Using a decision-tree model, we visualized how these key factors interact to influence long-term neurological recovery, highlighting their potential importance in refining the patient selection process for ECPR.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12872-026-06044-9.

Background

Out-of-hospital cardiac arrest (OHCA) remains a leading cause of mortality, despite significant advancements in cardiopulmonary resuscitation (CPR) over the past decade [1–3]. Survival with functional recovery to hospital discharge remains low, ranging from 5% to 10% [4].

In refractory cardiac arrest, extracorporeal CPR (ECPR) provides mechanical circulatory support (MCS) via veno-arterial extracorporeal membrane oxygenation (VA-ECMO), allowing the identification and treatment of reversible causes [5–9]. Several randomized trials have investigated ECPR versus conventional CPR with mixed results; some trials demonstrated improved survival with ECPR, while others showed no significant differences [5, 6, 10]. These studies indicated that ECPR may be effective under certain conditions. One critical factor is patient eligibility for ECPR; however, there is no consensus on the inclusion and exclusion criteria. The inclusion criteria in previous studies have varied, including age, initial rhythm, witnessed arrest, bystander CPR, and time from cardiac arrest to hospital arrival [5, 9, 11, 12]. Furthermore, most studies have focused on short-term endpoints such as 30-day survival or outcomes at hospital discharge. Data on long-term neurological recovery remain limited, and how various clinical factors interact to influence prognosis remains poorly understood [6, 9].

Therefore, the objective of this study was to identify factors associated with favorable neurological outcomes at 180 days and to perform an exploratory analysis of interactions between these factors.

Methods

Study population

This single-center retrospective study was conducted at the Advanced Emergency and Critical Care Center of Kurume University Hospital. We enrolled consecutive patients with OHCA who underwent ECPR at our institution between January 2016 and December 2023. Individuals aged < 18 years were excluded. The ECPR inclusion criteria were as follows: shockable rhythm on the patient’s initial ECG monitor, age < 75 years, independence in activities of daily living (ADL), and arrival at the hospital within 45 min of onset. For patients with non-shockable rhythm, the inclusion criteria were age < 75 years, independence in ADL, cardiac arrest witnessed by emergency medical service (EMS) teams, and suspicion of cardiogenic cardiac arrest based on symptoms, medical history, and physical examination. Individuals with a “do not attempt resuscitation” order, those with a family expressing a refusal willingness, and those with an existing terminal illness were excluded. Moreover, patients were excluded if the cause of cardiac arrest was actively suspected to be non-cardiogenic or if the physician assessed the patient as unlikely to survive owing to the absence of witnesses and signs of life, such as agonal respiration, pupillary light reflex, or any physical movement. These criteria were developed with reference to the inclusion criteria used in the SAVE-J study and other prior ECPR studies in Japan, and were adapted to fit our institutional workflow and real-world practice [9, 13].

Extracorporeal Membrane Oxygenation (ECMO) system and ECPR management

All patients included in the study were transferred directly to a fluoroscopy room within the Advanced Emergency and Critical Care Center. ECPR was performed by a dedicated multidisciplinary team. During daytime hours, the team generally consisted of 3–5 cardiovascular intensive care specialists, 1–2 emergency physicians, 1 surgeon, 1 neurosurgeon, and several junior residents. Emergency physicians managed airway stabilization and pharmacological interventions, while cardiovascular specialists performed the VA-ECMO cannulation. The nursing staff included at least two primary nurses from the emergency room supported by additional staff. Three clinical engineers (CEs) were exclusively assigned to the center, with two CEs managing the preparation and initiation of the ECMO system. During night shifts, the in-house team comprised 5 physicians (1 emergency physician, 1 cardiovascular specialist, 1 neurosurgeon, 1 surgeon, and 1 resident), 1–2 nurses, 1–2 radiologic technologists, and 1 CE capable of ECMO operation, supported by 1–2 cardiovascular specialists and 1 CE on call. Daytime was defined as the period from 8:00 to 18:00, and off-hours were defined as 18:00 to 8:00, regardless of the day of the week. Our institution employed the Terumo VA-ECMO system (CAPIOX™, Terumo, Tokyo, Japan) and the MERA Unified VA-ECMO system (UNIMO™, Senko Medical Trading Co., Tokyo, Japan) as the MCS devices. Additional MCS, including intraaortic balloon pumping (IABP) (Datascope CS100™/CS300™ or Cardiosave IABP Hybrid™, Getinge Japan, Tokyo, Japan) or a transcatheter heart pump (Impella^®, Abiomed Inc., Danvers, MA, USA), was performed based on the primary physician’s decision at the initiation of VA-ECMO and coronary artery angiography. Since its introduction in April 2019, Impella became the primary choice for adjunctive MCS, except in patients with bleeding complications or vascular access problems.

Femoral artery and vein cannulation were performed using 17–19 Fr and 21–25 Fr catheters, respectively, based on body size, using the percutaneous Seldinger technique while continuing conventional CPR (CCPR). In cases where multiple punctures were necessary, echo-guided punctures were performed, and wire manipulation was performed under fluoroscopic guidance for both procedures. Mechanical chest compression, using devices such as LUCAS^®3 (Physio-Control/Jolife AB, Lund, Sweden) or CLOVER3000™ (KOHKEN Medical CO. Ltd., Tokyo, Japan), was performed in some instances.

Post-ECPR management

The patients were transferred to the Cardiac Care Unit (CCU) within the Advanced Emergency and Critical Care Center, and their hemodynamic status was continuously monitored using a Swan-Ganz catheter, an oxygen saturation monitor attached to the right hand, and a regional oxygen saturation monitor attached to the forehead and front shins. VA-ECMO flow was maintained at 60 mL/kg/min to achieve a mean arterial pressure of ≥ 65 mmHg and a mixed venous oxygen saturation of > 60%. This flow rate was maintained until the serum lactate level decreased to < 2 mmol/L. Vasopressors, such as noradrenaline, were administered to maintain the mean arterial pressure, and red blood cell transfusions were performed to maintain hemoglobin levels above 10 g/dL while managing VA-ECMO.

The core body temperature was regulated between 34 ℃ and 36 ℃ during the initial 24 h and then gradually increased by 0.25 ℃ per hour until a core body temperature of 37 ℃ was achieved. Once serum lactate levels dropped below 2 mmol/L, the weaning process from VA-ECMO was initiated. Following successful weaning from VA-ECMO, removal of the IABP or Impella was performed upon recovery of the patient’s cardiac function. Unfractionated heparin was used as the anticoagulant, with activated thromboplastin time monitored every 6 h (target range: 50–60 s) or activated clotting time assessed multiple times daily (target range: 160–180 s).

Endpoints and definitions

The primary outcome was the identification of factors associated with favorable neurological outcomes at 180 days in patients who underwent ECPR for refractory OHCA. The secondary outcome was the interactions between the factors associated with favorable neurological outcomes.

Neurological outcomes were assessed using the Glasgow- Pittsburgh Cerebral Performance Category (CPC). CPCs 1 and 2 were considered favorable neurological outcomes, while CPCs 3, 4, and 5 were considered poor outcomes. Initial shockable rhythm was defined as ventricular fibrillation or pulseless ventricular tachycardia. Transient ROSC was defined as any palpable pulse or measurable blood pressure ≥ 1 min before ECMO initiation, either before or after hospital arrival [14]. Downtime was defined as the duration from the witnessed cardiopulmonary arrest or the time the patient was last seen to be normal (in unwitnessed cases) to the initiation of VA-ECMO, including no-flow time (the interval from collapse to CPR initiation) and low-flow time (the interval from CPR initiation to VA-ECMO initiation) [15]. The aetiology of cardiac arrest was determined by attending cardiologists based on comprehensive clinical evidence, including findings from electrocardiography, coronary angiography, echocardiography, or computed tomography. Arterial blood gas analysis and biochemical tests were performed immediately after ECPR.

Statistical analysis

All statistical analyses were performed using the JMP Student Edition version 18.2.1 software (SAS Institute Japan, Tokyo, Japan). Statistical significance was defined as a two-sided p < 0.05. Continuous variables were expressed as medians (interquartile range [IQR]). Continuous and categorical variables were compared using the Wilcoxon rank-sum and the chi-squared tests, respectively. To determine the factors associated with a favorable neurological outcome at 180 days, univariate logistic regression analysis was conducted. Age, sex, witnessed, bystander CPR, shockable rhythm, transient ROSC, downtime, arterial pH, and serum lactate concentration were evaluated using univariate analysis. To identify the factors requiring adjustment, we constructed a directed acyclic graph (DAG) using DAGitty version 3.1 (https://www.dagitty.net). To account for the effects of confounding factors, we performed a logistic regression analysis using the adjustment sets identified by the DAG as independent variables to calculate adjusted odds ratios (aORs). A decision tree model was then constructed using the Partition platform in JMP, with the neurological outcome (CPCs 1–2 vs. CPCs 3–5) as the response variable and the same candidate predictors used in the univariate analysis as explanatory variables to explore interactions and stratification patterns among predictors. The split points were derived algorithmically using the partition platform in a data-driven manner without a subjective cutoff specification, ensuring the reproducibility of the splitting rule. To evaluate the stability and validity of the tree structure, variable importance was calculated based on 2,000 bootstrap resamples. This approach was employed to assess the robustness of the identified predictors and to minimize the risk of overfitting.

Results

Study population and baseline characteristics

Between January 2016 and December 2023, 66 consecutive patients with OHCA underwent ECPR. Among them, one patient aged < 18 years was excluded, leaving 65 eligible patients. Missing data were present only for arterial pH and serum lactate levels (n = 2 per group). Because the proportion of missing data was small and unlikely to influence the results, these variables were handled using a complete-case analysis without imputation. Of these patients, 18 (27.7%) had favorable neurological outcomes at 180 days. The median patient age was 61 (IQR, 53–69) years, and 89.2% of patients were men (Table 1). In 86.2% of patients, the initial rhythm was shockable, and bystander CPR was performed in 72.3% of the patients. The most common cause of cardiac arrest was ischemic heart disease (80.0%), including acute coronary syndrome, which was observed in 53.8% of patients. The median time from call to hospital arrival was 33 (IQR: 28–42) min, the response time was 9 (IQR: 8–11), the time on scene was 10 (IQR: 8–13) min, the time from departure from the field to hospital arrival was 13 (IQR: 7–21) min, the time from hospital arrival to ECMO initiation was 17 (IQR: 14–23) min, and the downtime was 49 (IQR: 38–60) min (Table 2). Arterial blood gas analysis immediately after ECPR revealed that the median pH value was 6.98 (IQR: 6.85–7.11), and the median serum lactate level was 13.6 (IQR: 11.2–16.0) mmol/L. Transient ROSC was recorded in 38.5% of the patients. For post-resuscitation therapy, target temperature management was achieved in 90.8%, and the percutaneous coronary intervention was performed in 49.2% of patients. As additional MCS to VA-ECMO, IABP and Impella were used in 35.4% and 46.2% of the patients, respectively.

Table 1.

Baseline characteristics stratified by neurological outcomes at 180 days

	All (n = 65)	CPC 1–2 (n = 18)	CPC 3–5 (n = 47)	P values**
Age, years	61 [ 53–69 ]	63 [49–71]	61 [54–68]	0.907
Males, n (%)	58 (89.2)	15 (83.3)	43 (91.5)	0.343
Body mass index, kg/m2	24.5 [22.6–26.6]	24.7 [21.5–26.6 ]	24.5 [23.1–27.0 ]	0.698
Past history, n (%)
Hypertension	33 (55.0)	10 (55.6)	23 (54.8)	0.955
Hyperlipidemia	24 (39.3)	5 (27.8)	19 (44.2)	0.232
Diabetes mellitus	23 (37.7)	4 (22.2)	19 (44.2)	0.107
Smoking	36 (63.2)	9 (52.9)	27 (67.5)	0.297
Chronic heart failure	10 (17.0)	5 (27.8)	5 (12.2)	0.142
Prior myocardial infarction	9 (15.0)	4 (22.2)	5 (11.9)	0.305
Prior stroke	9 (14.8)	2 (11.1)	7 (16.3)	0.604
Initial cardiac rhythm
Shockable	56 (86.2)	16 (88.9)	40 (85.1)	0.693
Pulseless electrical activity	9 (13.8)	2 (11.1)	7 (14.9)	0.693
Witnessed arrest	52 (80.0)	16 (88.9)	36 (76.6)	0.268
Bystander CPR	47 (72.3)	16 (88.9)	31 (66.0)	0.065
Use of epinephrine in the field	35 (57.4)	10 (62.5)	25 (55.6)	0.630
Aetiology of cardiac arrest
Ischemic heart disease	52 (80.0)	11 (61.1)	41 (87.2)	0.019
Acute coronary syndrome	35 (53.8)	7 (38.9)	28 (59.6)	0.134
Non-ischemic heart disease	3 (4.6)	1 (5.6)	2 (4.3)	0.823
Idiopathic arrhythmia	6 (9.2)	4 (22.2)	2 (4.3)	0.025
Pulmonary embolism	1 (1.5)	1 (5.6)	0 (0.0)	0.103
Others	3 (4.6)	1 (5.6)	2 (4.3)	0.823
off-hours admission	26 (40.0)	7 (38.9)	19 (40.4)	0.910

N Number of patients, CPC Cerebral performance category, CPR Cardiopulmonary resuscitation

* Numbers show median value [interquartile range] or n/N (%) in brackets

** P values were calculated using Pearson’s chi-square test or the Wilcoxon rank-sum test for comparisons between CPC 1–2 and CPC 3–5 groups

Table 2.

Clinical findings and resuscitation process stratified by neurological outcomes at 180 days

	All (n = 65)	CPC 1–2 (n = 18)	CPC 3–5 (n = 47)	P values**
Time from call to hospital arrival, min	33 [28–42]	33 [25–38]	33 [28–43]	0.393
Response time, min	9 [8–11]	10 [7–11]	9 [8–11]	0.809
Time on scene, min	10 [8–13]	10 [8–12]	10 [7–15]	0.637
Departure from the field to hospital arrival, min	13 [7–21]	11 [7–15]	13 [8–21]	0.590
Hospital arrival to ECMO time, min	17 [14–23]	17 [13–22]	17 [14–25]	0.730
Downtime, min	49 [38–60]	41 [32–50]	51 [42–65]	0.009
Serum lactate concentration, mmol/L	13.6 [11.2–16.0]	13.7 [9.7–15.3]	13.6 [11.4–16.8]	0.215
Arterial pH	6.98 [6.85–7.11]	7.01 [6.94–7.20]	6.97 [6.83–7.08]	0.096
Transient ROSC	25 (38.5)	12 (66.7)	13 (27.7)	0.004
Target temperature management	59 (90.8)	16 (88.9)	43 (91.5)	0.746
PCI procedure	32 (49.2)	7 (38.9)	25 (53.2)	0.302
Additional MCS on ECMO
IABP	23 (35.4)	3 (16.7)	20 (42.6)	0.051
Impella	30 (46.2)	11 (61.1)	19 (40.4)	0.134
Duration of ECMO, hours	90.2 [51.9–125.2]	91.8 [68.3–111.1]	90.2 [51.3–133.2]	0.889
Successful ECMO weaning	41 (63.1)	18 (100)	23 (48.9)	< 0.001
Survival at 30 days	30 (46.2)	18 (100)	12 (25.5)	< 0.001
Survival at 180 days	23 (35.4)	18 (100)	5 (10.6)	< 0.001

N Number of patients, CPC Cerebral performance category, ECMO Extracorporeal membrane oxygenation, IABP Intra-aortic balloon pumping, MCS Mechanical circulatory support, PCI Percutaneous coronary intervention, PVAD Percutaneous Ventricular Assist Device, ROSC Return of spontaneous circulation, TTM Targeted temperature management

* Numbers show median value [interquartile range] or n/N (%) in brackets

** P values were calculated using Pearson’s chi-square test or the Wilcoxon rank-sum test for comparisons between CPC 1–2 and CPC 3–5 groups

Comparison between favorable and unfavorable outcomes

Between the favorable and unfavorable neurological outcome groups, there were no significant differences in baseline characteristics such as age and sex (Table 1). However, regarding the cause of arrest, the incidences of ischemic heart disease (61.1 vs. 87.2%) and idiopathic arrhythmia (22.2% vs. 4.3%) were higher in the favorable group. Regarding the resuscitation process, the downtime was shorter in the favorable outcome group (median 41 vs. 51 min, p = 0.009) and transient ROSC was more frequently observed (66.7% vs. 27.7%) (Table 2). There were no significant differences in serum lactate concentration between the two groups (median 13.7 vs. 13.6 mmol/L, p = 0.215).

Factors associated with favorable neurological outcomes at 180 days

Univariate analysis revealed bystander CPR (odds ratio [OR], 4.13; 95% confidence interval [CI], 1.00–28.18), transient ROSC (OR, 5.23; 95% CI, 1.68–17.91), and downtime (OR, 0.95; 95% CI, 0.90–0.98) were identified as significant prediction of favorable neurological outcomes. The adjusted analysis using the DAG-identified adjustment set demonstrated that transient ROSC (aOR, 8.37; 95% CI, 2.20–38.79) and downtime (aOR, 0.95; 95% CI, 0.90–0.99) remained significantly and independently associated with favorable neurological outcomes (Table 3). For age, sex, and witnessed arrest, the DAG identified no confounders requiring adjustment; therefore, only unadjusted estimates are reported. The specific adjustment sets used for each variable in the DAG-based analysis are detailed in the Supplementary Table. The DAG was constructed to visualize the hypothesized causal relationships among pre-hospital factors and outcomes (Fig. 1).

Table 3.

Univariate and adjusted analysis using DAG for associated factor of favorable neurological outcomes at 180 days

Variables	Univariate			adjusted analysis using DAG
	OR	95%CI	P value	aOR	95%CI	P value
Age, year	0.98	0.94–1.03	0.467	Not applicable
Male	0.47	0.09–2.58	0.361	Not applicable
Witnessed	2.44	0.57–16.98	0.246	Not applicable
Bystander CPR	4.13	1.00–28.18	0.049	3.50	0.82–24.39	0.095
Shockable rhythm	1.40	0.30–10.07	0.687	1.15	0.21–8.99	0.871
Transient ROSC	5.23	1.68–17.91	0.004	8.37	2.20-38.79	0.002
Downtime, min	0.95	0.90–0.98	0.004	0.95	0.90–0.99	0.006
Arterial pH	12.79	0.59–380.14	0.105	8.32	0.10-986.37	0.352
Serum lactate concentration, mmol/L	0.91	0.78–1.04	0.156	0.980	0.79–1.20	0.850

aOR Adjusted odds ratio, CI Confidence interval, CPR Cardiopulmonary resuscitation, DAG Directed acyclic graph, OR Odds ratio, ROSC Return of spontaneous circulation

Fig. 1.

DAG illustrating the causal structure for neurological outcomes. Directed acyclic graph (DAG) illustrating the hypothesized causal structure among pre-hospital factors and neurological outcomes at 180 days. Arrows indicate hypothesized causal directions. CPR, Cardiopulmonary resuscitation; DAG, Directed acyclic graph; Lac, Lactate; ROSC, Return of spontaneous circulation

Data analysis using the decision-tree model

Decision-tree model analysis revealed that the first split was based on whether transient ROSC was observed (Fig. 2). In the group with transient ROSC, the next split was based on whether bystander CPR was performed. In the group without transient ROSC, the subsequent split was determined based on whether the downtime was < 49 min. Figure 2 demonstrates that cases without transient ROSC had fewer patients with favorable neurological outcomes at 180 days (15.0%). In the group with transient ROSC and bystander CPR, 11 out of 16 patients (68.8%) had favorable neurological outcomes at 180 days. In the group without transient ROSC, but with a downtime < 49 min, six out of 18 patients (33.3%) had favorable neurological outcomes at 180 days. In contrast, in the group without transient ROSC but with a downtime ≥ 49 min, no patients had favorable neurological outcomes at 180 days. To evaluate the stability of these findings, variable importance was calculated using 2,000 bootstrap resamples, which consistently identified transient ROSC and downtime as the most significant predictors (Supplementary Figure).

Fig. 2.

Decision-tree model analysis of data of patients treated with ECPR using a tree model. Decision-tree model for neurological outcomes at 180 days in patients treated with ECPR. Each node displays the number of patients and the proportion with favorable (CPC 1–2) and unfavorable (CPC 3–5) outcomes. Stacked bars represent the distribution of CPC 1–2 (cross-hatched) and CPC 3–5 (horizontal lines). CPC, Cerebral Performance Category; CPR, Cardiopulmonary resuscitation; ECPR, Extracorporeal cardiopulmonary resuscitation; ROSC, Return of spontaneous circulation

Discussion

The major finding of this study was the independent association of the occurrence of transient ROSC and shorter downtime with favorable neurological outcomes at 180 days in patients treated with ECPR. The decision-tree model further demonstrated that transient ROSC with bystander CPR was associated with favorable neurological outcomes. In contrast, none of the patients in the group without transient ROSC and with a model-derived downtime threshold of approximately 49 min, as identified by the decision tree, exhibited favorable neurological outcomes.

ECPR is a resuscitation method for patients with refractory cardiogenic cardiac arrest who cannot be treated with CCPR [5–9]. However, it is not effective for all patients and should be reserved for those with reversible causes and the potential for neurological recovery. As ECPR requires substantial medical resources and costs [16], identifying patients most likely to benefit from it is essential. Many institutions have therefore developed their own inclusion criteria [12], and numerous studies have examined individual prognostic factors [17–23]. However, neurological outcomes are likely influenced by complex interactions among multiple factors. Our decision-tree analysis complements these previous reports by visualizing such interactions and stratifying patients according to clinically meaningful combinations of variables. Decision-tree models have been widely utilized in resuscitation medicine to visualize complex interactions between prognostic factors [24, 25]. To enhance the methodological rigor, we employed a DAG to identify the minimal adjustment set for each variable, thereby minimizing the risk of over-adjustment bias in our logistic regression analysis. Furthermore, given the limited sample size and the inherent risk of overfitting in tree-based models, we evaluated the stability of our decision-tree structure using 2,000 bootstrap resamples. The resulting variable importance rankings consistently identified downtime and transient ROSC as the most robust predictors (Supplementary Figure), reinforcing the internal validity of our patient stratification. In the present study, the fact that the decision-tree splits were closely aligned with the significant variables identified in the DAG-based adjusted logistic regression supports the medical plausibility of our model. While the small sample size in certain nodes generally requires cautious interpretation, the high variable importance of transient ROSC and downtime confirmed via bootstrap resampling suggests that these findings provide robust and valuable exploratory insights into patient stratification for ECPR.

In this study, transient ROSC and downtime were associated with neurological outcomes after 180 days. Previous reports have shown that transient ROSC is independently associated with favorable outcomes in patients treated with ECPR, regardless of the timing of ROSC [23]. Regarding the association between downtime and neurological outcomes, current guidelines recommend establishing ECPR within 60 min of cardiac arrest [26]. Additionally, results from a multicenter registry in Japan reported favorable outcomes with a low-flow time of < 40 min [27]. In contrast, our study found no association between blood lactate concentration, pH, and neurological outcomes. This discrepancy may be attributed to the timing of blood sampling immediately after ECMO initiation, which may reflect both initial injury severity and reperfusion effects rather than purely the pre-ECMO physiological status [28].

An analysis of the interaction between factors revealed that favorable neurological outcomes were achieved in the group in which transient ROSC was observed and bystander CPR was performed. This suggests that bystander CPR may act as an upstream facilitator of transient ROSC rather than an independent downstream predictor, consistent with previous studies showing that patients who received bystander CPR were more likely to achieve ROSC than those who did not [29, 30]. Conversely, patients who previously achieved ROSC but experienced re-arrest during resuscitation had worse outcomes [31]. Based on these findings, patients who underwent bystander CPR and achieved transient ROSC, but experienced subsequent cardiac arrest may be suitable candidates for ECPR. By contrast, none of the patients in the group without transient ROSC and with a downtime duration identified by the decision-tree analysis achieved a favorable neurological outcome. This finding suggests that a prolonged ischemic burden may outweigh the potential benefits of ECPR when immediate restoration of cerebral perfusion cannot be achieved. Considering that downtime includes the duration from hospital arrival to ECMO initiation, patients without transient ROSC may be suitable candidates for ECPR only if they are located within a limited geographical area near the hospital. To overcome the practical constraints of time and distance, prehospital ECPR has already been implemented in some regions and has demonstrated logistical feasibility, particularly through mobile ECMO teams or Helicopter Emergency Medical Services [32–34]. Such strategies may be theoretically relevant for patients without transient ROSC, but with potentially reversible pathology, in whom the window for neurological salvage is highly time dependent. Further studies are warranted to determine whether earlier circulatory support in the prehospital phase can mitigate the detrimental effects of prolonged low-flow time in this subgroup.

Limitations

Our study has some potential limitations. First, this was a single-center retrospective study with a relatively modest sample size, which may have limited the generalizability of the findings. Although we used a directed acyclic graph (DAG) to identify minimal adjustment sets and reduce the risk of over-adjustment, this approach does not address model instability caused by the limited number of outcome events. Therefore, the adjusted estimates may be imprecise and should be interpreted with caution. Furthermore, many of the participants were male, reflecting the known epidemiological characteristics of refractory OHCA. This sex imbalance may further limit the generalizability of our findings to female patients. Furthermore, due to the small number of cases in some terminal nodes of the decision tree, the risk of overfitting cannot be entirely excluded, and the results should be interpreted with caution. Second, the study spanned an 8-year period, during which significant changes occurred in our institution’s clinical practices regarding ECPR. Specifically, our institution did not adopt an aggressive ECPR strategy until 2016, and over the subsequent years, our experience may have increased, leading to a more refined patient selection process. In addition, the Impella was not introduced for left ventricular venting at our institution until April 2019. These temporal changes in clinical approaches and technological adoption inevitably introduce variability in outcomes. Moreover, the indications for ECPR were not based on a fully standardized protocol and, in some cases, depended on the discretion of the treatment team, especially for patients near the borderline of the predefined criteria. Therefore, patient selection was not entirely uniform, and patients perceived as more likely to have favorable outcomes may have been preferentially treated with ECPR. This potential selection bias may have influenced the observed associations. Therefore, the findings should be interpreted as associative rather than causal. Despite these limitations, our findings provide valuable preliminary insights into the factors associated with favorable neurological outcomes after ECPR. Therefore, future prospective studies with larger sample sizes are warranted to establish the appropriate patient selection criteria.

Conclusions

Transient ROSC and shorter downtime were associated with favorable neurological outcomes at 180 days in patients treated with ECPR. While these exploratory findings from a limited sample require validation, they suggest that both factors may be important for refining patient selection and prognostic assessment in clinical practice.

Abbreviations

ADL

Activities of daily living

CCU

Cardiac care unit

CEs

Clinical engineers

CI

Confidence interval

CPR

Cardiopulmonary resuscitation

CPC

Cerebral performance category

CCPR

Conventional cardiopulmonary resuscitation

DAG

Directed acyclic graph

EMS

Emergency medical service

ECMO

Extracorporeal membrane oxygenation

ECPR

Extracorporeal cardiopulmonary resuscitation

IQR

Interquartile range

IABP

Intraaortic balloon pumping

MCS

Mechanical circulatory support

OR

Odds ratio

OHCA

Out-of-hospital cardiac arrest

ROSC

Return of spontaneous circulation

VA-ECMO

Venoarterial extracorporeal membrane oxygenation

Authors’ contributions

TH contributed to the conceptualization, methodology, investigation, formal analysis, visualization, writing, review, and editing of the original draft. TS contributed to project administration, investigation, data curation, writing, review, and editing of the original draft. KM contributed to the formal analysis, validation, writing, review, and editing of the original draft. Y M, KS, KH, MO, OT, and Y F contributed to the writing, review, and editing of the original draft. All authors read and approved the final manuscript.

Funding

Not applicable.

Data availability

The datasets are not publicly available due to ethical and privacy restrictions, but are available from the corresponding author on reasonable request with approval from the Ethics Committee of Kurume University.

Declarations

Ethics approval and consent to participate

The study protocol was approved by the Kurume University Hospital Ethics Committee (Approval: No. 24291) and was conducted in accordance with the Declaration of Helsinki. Owing to the retrospective design of the study, patient consent was obtained using the opt-out method, as approved by the ethics committee.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.