Outcomes of Extracorporeal Cardiopulmonary Resuscitation for In-Hospital Cardiac Arrest Among Children With Noncardiac Illness Categories

Abstract

OBJECTIVES:

The objective of this study was to determine the association of the use of extracorporeal cardiopulmonary resuscitation (ECPR) with survival to hospital discharge in pediatric patients with a noncardiac illness category. A secondary objective was to report on trends in ECPR usage in this population for 20 years.

DESIGN:

Retrospective multicenter cohort study.

SETTING:

Hospitals contributing data to the American Heart Association’s Get With The Guidelines-Resuscitation registry between 2000 and 2021.

PATIENTS:

Children (<18 yr) with noncardiac illness category who received greater than or equal to 30 minutes of cardiopulmonary resuscitation (CPR) for in-hospital cardiac arrest.

INTERVENTIONS:

None.

MEASUREMENTS AND MAIN RESULTS:

Propensity score weighting balanced ECPR and conventional CPR (CCPR) groups on hospital and patient characteristics. Multivariable logistic regression incorporating these scores tested the association of ECPR with survival to discharge. A Bayesian logistic regression model estimated the probability of a positive effect from ECPR. A secondary analysis explored temporal trends in ECPR utilization. Of 875 patients, 159 received ECPR and 716 received CCPR. The median age was 1.0 [interquartile range: 0.2–7.0] year. Most patients (597/875; 68%) had a primary diagnosis of respiratory insufficiency. Median CPR duration was 45 [35–63] minutes. ECPR use increased over time (p < 0.001). We did not identify differences in survival to discharge between the ECPR group (21.4%) and the CCPR group (16.2%) in univariable analysis (p = 0.13) or propensity-weighted multivariable logistic regression (adjusted odds ratio 1.42 [95% CI, 0.84–2.40; p = 0.19]). The Bayesian model estimated an 85.1% posterior probability of a positive effect of ECPR on survival to discharge.

CONCLUSIONS:

ECPR usage increased substantially for the last 20 years. We failed to identify a significant association between ECPR and survival to hospital discharge, although a post hoc Bayesian analysis suggested a survival benefit (85% posterior probability).

Approximately 15,200 children require cardiopulmonary resuscitation (CPR) for in-hospital cardiac arrests (IHCAs) annually in the United States (1). Pediatric IHCA survival rates have improved over time from 18% in 2000 to 38% in 2018, but a substantial proportion of patients do not achieve return of spontaneous circulation (ROSC) (2, 3). Thus, extracorporeal CPR (ECPR), the implementation of venoarterial extracorporeal membrane oxygenation (VA-ECMO) during CPR, is used as a rescue therapy for refractory IHCA (2–4).

The association of ECPR with pediatric IHCA survival outcomes has been reported in several studies, but the majority have largely focused on children with underlying cardiac disease (5–12). In this patient population, ECPR is associated with higher rates of survival to hospital discharge when compared with conventional CPR (CCPR) (6). Thus, based on the currently available evidence, international CPR guidelines and International Liaison Committee on Resuscitation consensus statements advise consideration of ECPR for children with existing cardiac disease who have IHCA in settings with existing ECMO protocols (13, 14). Data regarding the association of ECPR with IHCA outcomes in children without primary cardiac diagnoses are lacking. Guidelines specifically acknowledge these lack of data and are unable to provide recommendations in the noncardiac population (10, 13–15). Thus, ECPR deployment in noncardiac patients is determined based on hospital-specific practice guidelines (16, 17).

To address this critical knowledge gap, the primary objective of this study was to determine the association between the use of ECPR and survival to hospital discharge in children with noncardiac illness category who received greater than or equal to 30 minutes of CPR for IHCA reported to the American Heart Association’s Get With The Guidelines-Resuscitation registry (GWTG-R). We hypothesized that rates of survival to hospital discharge would be higher in children who received ECPR than children who received CCPR. A secondary objective was to determine temporal trends in ECPR usage in the noncardiac population.

METHODS

Overview

This was a retrospective cohort study using data from the GWTG-R registry of in-hospital resuscitation events using Utstein-style data reporting (18). The design and reporting of GWTG-R have been described previously (www.heart.org/resuscitation). The Children’s Hospital of Philadelphia Institutional Review Board determined this study exempt as it did not meet criteria for human subject research (IRB#21018505).

Inclusion and Exclusion Criteria

Pediatric IHCA index events (first IHCA in a given hospitalization) with a total CPR duration of greater than or equal to 30 minutes between January 1, 2000, and May 4, 2021, were eligible for inclusion. Subjects were included if they had an illness category classified as “medical noncardiac” or “surgical noncardiac” and excluded if they had a cardiac illness category. Patients with a “trauma” illness category were also excluded due to the anticipation that ECPR would be rarely used in this patient population and that the overall low number of patients in this category would lead to difficulty with achieving balance in statistical models. This definition does not fully exclude patients with cardiac diagnoses, as the definition of illness category in the GWTG-R registry is based on the subject’s primary diagnosis at the time of the event. For example, a subject with a history of a structural cardiac defect admitted with pneumonia would be classified as “medical noncardiac” at the time of event, while a patient admitted with myocarditis would be classified as “medical cardiac.” A CPR duration threshold of at least 30 minutes was selected to avoid inclusion of: 1) patients who required CPR during cannulation for ECMO that preceded cardiac arrest and 2) CCPR patients for whom teams were not committed to similar resuscitation efforts as what would be required for ECPR (e.g., minimize inclusion of CCPR patients who received CPR without reasonable expectation of survival). The ECPR group was defined by patients who received “cardiopulmonary bypass/extracorporeal CPR” under the GWTG-R registry category for non-drug interventions and had sustained return of circulation (ROC) for at least 20 minutes. Thus, only events with successful cannulation were included as we theorized there could be interinstitutional discrepancies in recording ECPR as an intervention if the cannulation was not completed.

To be eligible, patients were required to be treated in a hospital with at least 3 years of registry data during the study period and at least three total GWTG-R entries in the calendar year of the event. Patients were only included if they were treated in a hospital with at least one ECPR case meeting our inclusion criteria in the same calendar year or any earlier year to compare patients with ECPR vs. CCPR, rather than patients treated at ECPR vs. non-ECPR centers. GWTG-R does not report ECMO volume per center, so this served as a surrogate for determining a center’s available ECPR system and resources, which has shown association with survival (19). However, the analysis of ECPR usage temporal trends did not exclude patients based on ECPR utilization of the center. Events were excluded if they occurred outside predefined inpatient locations (ICUs, diagnostic/intervention areas, and general inpatient wards) or the emergency department. Patients were excluded if data on non-drug interventions, survival to hospital discharge, sex, or pre-existing conditions were missing. All acute stroke patients were excluded based on the relative contraindication for ECMO. Only years with complete data were included in the analysis of ECPR usage trends; this was limited to events between January 1, 2000, and December 31, 2020.

Outcomes

The primary outcome was survival to hospital discharge. An exploratory outcome was survival with a favorable outcome (pediatric cerebral performance category [PCPC] score of 1, 2, or 3 or no change from admission PCPC score) in the subset of patients with available admission and discharge PCPC data.

Statistical Analyses

Categorical variables were summarized as counts and percentages and continuous variables as median, first, and third quartiles. Fisher exact test (with a Monte Carlo approximation when response had more than two categories) compared categorical variables and Wilcoxon rank-sum test compared continuous variables between ECPR and CCPR patients.

To assess the association of ECPR vs. CCPR with outcomes, propensity-weighted regression was used to reduce the effect of variables that confound the relationship between CPR type and outcomes (age, sex, race, illness category, pre-existing medical conditions, event location, first documented rhythm, time of event: night/weekend vs. weekday, hospital size, calendar year of event, and CPR duration) (20). This robust approach involved three main steps. First, a logistic regression model was created to estimate each patient’s probability (propensity) to receive ECPR and a butterfly plot of propensities was created by ECPR status (Supplemental Fig. 1, http://links.lww.com/CCM/H468). For patients missing race, “missing” was coded as a separate category that was used to achieve balance in the propensity-weighted regression. Second, patients were weighted using stabilized inverse probability of treatment weights. Using this method rather than propensity matching allowed a greater number of patients to be included in the final cohort. This method applied a weight to each patient based on the patient’s probability to receive treatment to create balance across the potential confounders between the two groups (21). Differences in proportions were compared between patients with ECPR and CCPR and summarized as counts and percentages in the original and weighted cohorts. Groups were considered well-balanced when the absolute standardized difference was <0.10 after propensity weighting (20). Third, a logistic regression model was created using the weighted cohort to assess the association of ECPR with survival to hospital discharge. This model was then adjusted for any covariate that did not meet the criteria for sufficient balance after propensity weighting. Because of our hypothesis that CPR duration would be strongly associated with both receiving ECPR as well as survival to hospital discharge, CPR duration was included as an a priori covariate in the logistic regression model despite achieving our goal balance through propensity weighting (22, 23). Site was included as a random effect to account for within-hospital correlation. Neurologic outcomes were evaluated for those patients included in the propensity-weighted regression, but due to the small number of patients with available PCPC scores, an adjusted odds ratio (OR) was not calculated.

Specific attention was taken to control for CPR duration in the final models. The odds of survival decrease sharply for each minute of CPR initially, but the slope quickly levels out. In order to appropriately capture the relationship, a piecewise linear spline of CPR duration was used as the covariate for CPR duration. This allowed the slope to change (become less steep) at 45 and 60 minutes after the start of CPR. CPR duration was also truncated after 120 minutes to address the lack of change in odds of survival after that time.

A post hoc Bayesian estimation of the logistic regression model was created to determine the probability of a positive effect of ECPR on survival to hospital discharge. For the purposes of this analysis, minimally informative priors (scaled Student t distributions with 3 degrees of freedom, centered at 0, with scale parameter resulting in priors with sd = 175) were chosen for all regression parameters without incorporation of subjective prior beliefs for estimation to be driven by the data while applying the same propensity weighting described earlier. The aim was to estimate the posterior probabilities that the treatment effect exceeded a range of potential values for the minimal clinically important treatment effect (OR > 0.9, OR > 0.95, OR > 1, OR > 1.05, OR > 1.10, OR > 1.15, OR > 1.20, OR > 1.25). This range was selected to include the potential for harm (OR < 1) in addition to the probability of treatment benefit. This method was included to provide a clinically relevant evaluation of benefit of a rare therapy that has been accepted by many into practice for rescue of refractory IHCA.

The usage of ECPR over time was assessed using the trends test. All analyses were performed using SAS 9.4 (SAS Institute, Cary, NC) or R 4.0.1 (R Foundation for Statistical Computing, Vienna, Austria) with reported p values based on a two-sided alternative and deemed significant if less than 0.05.

RESULTS

Unweighted Cohort Characteristics

After sequential exclusions (Fig. 1), 875 patients were included in the primary analysis; 159 received ECPR and 716 received CCPR. The unweighted prearrest patient and event characteristics are described in Table 1, arrest characteristics in Table 2, and hospital characteristics of the patients in the final cohort are described in Supplemental Table 1 (http://links.lww.com/CCM/H468). The ECPR cohort patients were older, had lower frequency of baseline CNS depression, and higher frequency of pre-existing diagnosis of pneumonia. Patients receiving ECPR had a longer median duration of CPR 56 [interquartile range: 45–82] vs. 43 [34–58] minutes for CCPR (p < 0.001). Patients receiving ECPR were more likely to have an arterial catheter in place prior to arrest, received more epinephrine doses, had a shorter epinephrine dosing interval, and had higher frequency of receiving alternative pharmacologic interventions during arrest (calcium, bicarbonate, and alternative vasopressors). Information on the demographics of patients excluded for treatment in non-ECPR performing centers is provided in Supplemental Table 2 (http://links.lww.com/CCM/H468).

Figure 1.

Patient selection flow diagram of sequential exclusions of patients included in the final cohort separated by extracorporeal cardiopulmonary resuscitation (ECPR) vs. conventional CPR (CCPR) status. PCPC = pediatric cerebral performance category, ROC = return of circulation.

TABLE 1.

Prearrest Patient and Event Characteristics by Extracorporeal Cardiopulmonary Resuscitation Status

	ECPR Status
Characteristics	Conventional Cardiopulmonary Resuscitation (n = 716)	ECPR (n = 159)	Overall (n = 875)	p
Patient and event characteristics
Age at CPA event (yr)	1.0 [0.1, 7.0]	2.0 [0.6, 8.0]	1.0 [0.2, 7.0]	0.015a
Age at CPA event				0.021a
<1 mo	180 (25.1%)	21 (13.2%)	201 (23.0%)
1 mo–< 1 yr	155 (21.6%)	37 (23.3%)	192 (21.9%)
1 yr–<8 yr	208 (29.1%)	61 (38.4%)	269 (30.7%)
8–≤18 yr	173 (24.2%)	40 (25.2%)	213 (24.3%)
Sex				0.218b
Male	404 (56.4%)	81 (50.9%)	485 (55.4%)
Female	312 (43.6%)	78 (49.1%)	390 (44.6%)
Race
White	358 (50.0%)	74 (46.5%)	432 (49.4%)
Black/others	229 (32.0%)	48 (30.2%)	277 (31.7%)
Unknown	129 (18.0%)	37 (23.3%)	166 (19.0%)
Location of event				0.099b
ICU	558 (77.9%)	126 (79.2%)	684 (78.2%)
Emergency department	72 (10.1%)	11 (6.9%)	83 (9.5%)
General inpatient	57 (8.0%)	9 (5.7%)	66 (7.5%)
OR	29 (4.1%)	13 (8.2%)	42 (4.8%)
Night, weekend, or holiday event	384 (53.6%)	76 (47.8%)	460 (52.6%)	0.189b
Pre-existing conditions
Acute CNS nonstroke event	59 (8.2%)	8 (5.0%)	67 (7.7%)	0.190b
Baseline depression in CNS function	149 (20.8%)	16 (10.1%)	165 (18.9%)	0.002b
Cardiac malformation-acyanotic	47 (6.6%)	8 (5.0%)	55 (6.3%)	0.589b
Cardiac malformation-cyanotic	28 (3.9%)	7 (4.4%)	35 (4.0%)	0.823b
Congenital malformation (noncardiac)	146 (20.4%)	30 (18.9%)	176 (20.1%)	0.743b
Hepatic insufficiency	45 (6.3%)	10 (6.3%)	55 (6.3%)	1.000b
Hypotension/hypoperfusion	195 (27.2%)	51 (32.1%)	246 (28.1%)	0.242b
Major trauma	25 (3.5%)	2 (1.3%)	27 (3.1%)	0.203b
Metastatic or hematologic malignancy	69 (9.6%)	18 (11.3%)	87 (9.9%)	0.558b
Metabolic/electrolyte abnormality	148 (20.7%)	34 (21.4%)	182 (20.8%)	0.830b
Pneumonia	68 (9.5%)	25 (15.7%)	93 (10.6%)	0.032b
Renal insufficiency	90 (12.6%)	22 (13.8%)	112 (12.8%)	0.694b
Respiratory insufficiency	493 (68.9%)	104 (65.4%)	597 (68.2%)	0.398b
Sepsisc	206 (28.8%)	53 (33.3%)	259 (29.6%)	0.251b

CPA = cardiopulmonary arrest, ECPR = extracorporeal cardiopulmonary resuscitation.

^aWilcoxon rank-sum test.

^bFisher exact test (Monte Carlo approximate for tables larger than 2 × 2).

^cIn Get With The Guidelines-Resuscitation registry data until the year 2018 used the term septicemia.

TABLE 2.

Arrest Characteristics by Extracorporeal Cardiopulmonary Resuscitation Status

	ECPR Status
Characteristics	Conventional Cardiopulmonary Resuscitation (n = 716)	ECPR (n = 159)	Overall (n = 875)	p
Interventions in place at time of arrest
Arterial cathetera	161 (22.5%)	57 (35.8%)	218 (24.9%)	<0.001b
Dialysisa	33 (4.6%)	6 (3.8%)	39 (4.5%)	0.832b
Type of assisted ventilation				0.823b
None	151 (21.1%)	35 (22.0%)	186 (21.3%)
Invasive assisted ventilation	485 (677%)	109 (68.6%)	594 (67.9%)
Noninvasive assisted ventilation	80 (11.2%)	15 (9.4%)	95 (10.9%)
Vasoactive agenta	260 (36.4%)	71 (44.7%)	331 (37.9%)	0.058b
Arrest characteristics
First documented rhythm				0.005b
Bradycardia with poor perfusion	335 (50.4%)	55 (36.7%)	390 (47.9%)
Asystole	122 (18.3%)	25 (16.7%)	147 (18.0%)
Pulseless electrical activity	168 (25.3%)	62 (41.3%)	230 (28.2%)
Ventricular tachycardia	15 (2.3%)	3 (2.0%)	18 (2.2%)
Ventricular fibrillation	6 (0.9%)	1 (0.7%)	7 (0.9%)
Other	19 (2.9%)	4 (2.7%)	23 (2.8%)
Unknown	51 (7.1%)	9 (5.7%)	60 (6.9%)
Cardiopulmonary resuscitation duration (min)c	43.0 [34.0, 58.5]	56.0 [45.0, 82.0]	45.0 [35.0, 63.0]	<0.001d
Total number of epinephrine dosese	7.0 [4.0, 10.0]	11.0 [6.5, 15.0]	7.0 [4.0, 11.0]	<0.001d
Time to first epinephrine (min)f	1.0 [0.0, 3.0]	1.0 [0.0, 3.0]	1.0 [0.0, 3.0]	0.895d
Epinephrine dosing interval (min)g	7.2 [4.6, 12.5]	5.1 [4.0, 8.5]	6.6 [4.5, 11.7]	<0.001d
Vasopressor(s) other than epinephrine bolus	284 (39.7%)	94 (59.1%)	378 (43.2%)	<0.001b
Epinephrine bolus	680 (95.0%)	155 (97.5%)	835 (95.4%)	0.210b
Sodium bicarbonate	519 (72.5%)	133 (83.6%)	652 (74.5%)	0.003b
Calcium chloride or gluconate	450 (62.8%)	131 (82.4%)	581 (66.4%)	<0.001b
Atropine	184 (25.7%)	42 (26.4%)	226 (25.8%)	0.842b
Lidocaine	52 (7.3%)	12 (7.5%)	64 (7.3%)	0.867b
Amiodarone	31 (4.3%)	9 (5.7%)	40 (4.6%)	0.527b

ECPR = extracorporeal cardiopulmonary resuscitation.

^aMissing data on one patient.

^bFisher exact test (Monte Carlo approximate for tables larger than 2 × 2).

^cTruncated after 120 min.

^dWilcoxon rank-sum test.

^eMissing data on 70 patients.

^fTruncated after 30 min. Missing data on 77 patients who received at least one dose of epinephrine.

^gDetined as total resuscitation duration after the first dose of epinephrine divided by the total number of administered epinephrine doses

Model Development and Weighting

Balance was achieved for the majority of propensity model covariates (Supplemental Table 3, http://links.lww.com/CCM/H468), with the exception of race, sex, hypotension/hypoperfusion, sepsis/pneumonia, and baseline depression in CNS function, which did not meet the goal of less than 0.10 for sufficient balance in the propensity model. Based on this residual imbalance, these covariates were included in the multivariable logistic regression model. Supplemental Data (http://links.lww.com/CCM/H468) include the characteristics of the six patients not included in the final cohort due to their propensity scores (Supplemental Table 4, http://links.lww.com/CCM/H468) and the primary drivers of the propensity model (Supplemental Table 5, http://links.lww.com/CCM/H468). Later event year, hospital size, and the absence of baseline CNS depression were the covariates with the greatest association with receiving ECPR.

Outcomes

Before propensity weighting, survival to hospital discharge was similar between groups (ECPR: 34/159 [21.4%] vs. CCPR: 116/716 [16.2%]; p = 0.13; Table 3). Data regarding PCPC are also included in Table 3 but was limited due to a degree of missingness (58% missing data among survivors). In the multivariable logistic regression model incorporating propensity weights, the adjusted OR of survival to hospital discharge with ECPR was 1.42 (95% CI, 0.84–2.40; p = 0.19). The Bayesian analysis estimated an 85.1% posterior probability of a population-level OR greater than 1 for survival to hospital discharge with ECPR vs. CCPR (Table 4).

TABLE 3.

Outcomes by Extracorporeal Cardiopulmonary Resuscitation Status: Univariate Outcomes

	ECPR Status
Outcomes	Conventional Cardiopulmonary Resuscitation	ECPR	p
Sustained return of spontaneous circulation	306/716 (42.7%)	0/159 (0.0%)	<0.001a
24-hr survival	204/716 (28.5%)	117/159 (73.6%)	<0.001a
Survival to hospital discharge	116/716 (16.2%)	34/159 (21.4%)	0.130a
PCPC at hospital dischargec			0.211b
1	17/44 (38.6%)	11/19 (57.9%)
2	9/44 (20.5%)	3/19 (15.8%)
3	8/44 (18.2%)	2/19 (10.5%)
4	9/44 (20.5%)	2/19 (10.5%)
5	1/44 (2.3%)	1/19 (5.3%)
Change in PCPC from admit to hospital dischargec			0.143b
0	36/44 (81.8%)	12/19 (63.2%)
1	3/44 (6.8%)	4/19 (21.1%)
2	2/44 (4.5%)	1/19 (5.3%)
3	2/44 (4.5%)	1/19 (5.3%)
4	1/44 (2.3%)	1/19 (5.3%)
Hospital discharge with PCPC of 1, 2, or no worse than baselinec	39/44 (88.6%)	15/19 (78.9%)	0.434a
Hospital discharge with PCPC of 1, 2, 3, or no worse than baselinec	40/44 (90.9%)	16/19 (84.2%)	0.422a

ECPR = extracorporeal cardiopulmonary resuscitation; PCPC = pediatric cerebral performance category.

^aFisher exact test.

^bWilcoxon rank-sum test.

^cSummarized on those who survived to hospital discharge.

TABLE 4.

Probability of Treatment Effect With Extracorporeal Cardiopulmonary Resuscitation by Bayesian Analysis

	OR > 0.90	OR > 0.95	OR > 1	OR > 1.05	OR > 1.10	OR > 1.15	OR > 1.20	OR > 1.25
P(OR)	0.93	0.89	0.85	0.81	0.76	0.71	0.65	0.60

OR = odds ratio.

This model assumes minimally informative prior beliefs for all logistic regression parameters.

Supplemental Figure 2 (http://links.lww.com/CCM/H468) depicts frequency and survival outcomes of ECPR and CCPR relative to CPR duration. Duration was distributed relatively evenly across all 15-minute duration intervals for the ECPR group, while more than half of CCPR events had CPR durations of 30–45 minutes.

Figure 2 displays the use of ECPR over time. Compared with reference year of 2000, ECPR usage increased over time (test of trends p < 0.001).

Figure 2.

Percentage of pediatric patients with a noncardiac illness category receiving greater than or equal to 30min of cardiopulmonary resuscitation (CPR) for in-hospital cardiac arrests who received treatment with extracorporeal CPR (ECPR) out of all the CPR events in this cohort by event year.

DISCUSSION

Using a large, multicenter national IHCA database and robust methods of controlling for potential confounders, we did not identify a statistically significant difference in survival to hospital discharge between ECPR compared with CCPR in children with noncardiac illness categories who received at least 30 minutes of CPR. Using a Bayesian probabilistic approach, we were able to estimate an 85% probability of a positive effect on survival to hospital discharge with the usage of ECPR. Across a 20-year span of cardiac arrest data reported to the GWTG-R registry, we demonstrated an increase in the percentage of children treated with ECPR for prolonged IHCA greater than or equal to 30 minutes.

To the best of our knowledge, this study represents the largest study of pediatric ECPR in the noncardiac population and enhances our understanding of the epidemiology and outcomes of patients treated with this therapy. Contrary to our hypothesis, we were unable to identify an association between ECPR and survival to hospital discharge in this noncardiac illness category cohort. Prior studies have shown better outcomes among cardiac patients who receive ECPR but have not focused on the outcomes of this unique population (5, 7, 9). Despite including two decades of GWTG-R registry data, the number of ECPR patients included in this study who underwent greater than or equal to 30 minutes of CPR remained relatively low (n = 159) and the width of the 95% CI for ECPR’s adjusted OR (95% CI, 0.84–2.40) suggests that inadequate statistical power may have hampered our ability to detect a positive finding. This small sample size also limited the ability to analyze survival based on event year, which may have shown improvement in ECPR outcomes over the last two decades.

To that end, the Bayesian analysis suggests a potential positive treatment effect of ECPR on survival to discharge. This statistical approach calculates the probability of treatment benefit—in this case, 85%—and this probability can be used as a more intuitive measure of evidence for a population-level benefit of ECPR. This calculation of probability provides an additional method for interpreting data compared with the frequentist calculation of p value and 95% CI, which are not synonymous with probability of treatment effect (24, 25). In the analysis of a rare but clinically used treatment such as ECPR, this probabilistic estimate of treatment benefit can add to our interpretation of these data and potentially aid in clinical decision making regarding the use of ECPR. In addition to the 85% probability of survival benefit with the use of ECPR, the low (7%) probability of an OR for survival less than 0.9 suggests that ECPR is very unlikely to expose patients to a considerable risk of harm. However, we can only assess harm defined as increased odds of death with the data available. These findings support the ongoing usage and development of ECPR systems.

Compared with children with cardiac illness, for whom ECPR is associated with survival benefit, ECPR outcomes in this cohort may differ for both systems and physiologic reasons. From a systems standpoint, ECPR is a resource intensive therapy with success relying on a multidisciplinary response team. As the majority of these systems in pediatrics have been focused on the cardiac population (26, 27), cardiac center teams may be more efficient in the deployment of ECPR. The cardiac cohort comprised largely of surgical cardiac patients (6), who often have an established relationship with a cardiothoracic surgeon and at times the ability to cannulate an open chest. Conversely, in a recent survey of pediatric noncardiac ECPR programs in North America, most noncardiac cannulations are performed by general surgeons, who are less likely to be involved in the patient’s care before IHCA (16). System factors may increase time to cannulation and CPR duration before ROC in the noncardiac population. In our study, the median time to ROC for those who received ECPR was 56 minutes, whereas in recent studies of ECPR for cardiac patients the median time to ROC was 25 minutes (28). In this cohort of prolonged cardiac arrest, we observed high rates of calcium and bicarbonate administration, despite current recommendations to avoid the universal use of these therapies (29–31). This observation may provide insight into the physiologic derangements observed during prolonged cardiac arrest or may indicate common use of these medications regardless of indication in children requiring prolonged CPR. From the physiologic standpoint, IHCA etiology differs between these populations. Patients with noncardiac diagnoses and refractory IHCA may be more likely to have heterogenous physiologic derangements and multisystem organ dysfunction leading up to the IHCA than their cardiac counterparts (2, 32–35).

Despite the absence of formal recommendations for ECPR in noncardiac patients and the lack of survival benefit in limited case series and prior reports (8, 36), we observed a significant temporal increase in ECPR use among centers reporting to the GWTG-R registry. This increased utilization of ECPR may reflect a combination of more aggressive treatment of refractory IHCA and recognition that prolonged CPR is not inherently futile. Although prolonged IHCA is associated with lower rates of survival compared with briefer IHCA events, some children do survive even after more than an hour of CPR (23). Increasing ECPR usage may contribute to improved survival in cases of prolonged IHCA, providing a rescue therapy for IHCA when ROSC is not otherwise achieved (5, 34, 35). Such anecdotal observations may contribute to institutional decisions to pursue this therapy in the noncardiac population in the absence of robust epidemiologic support. During this study period there has likely been substantial evolution and refinement of ECMO and ECPR methods and technology; however, limited sample size precluded a temporal analysis of survival in this trends analysis. The ideal design and patient eligibility for ECPR remains unknown and likely involves patient, location, and team factors beyond the primary illness category (5, 37, 38). Future work should focus on determining subsets of patients with the greatest survival benefit as well as the development of robust ECPR systems for populations outside of surgical cardiac patients.

Our registry-based data and retrospective analysis present several limitations. First, this retrospective cohort study cannot account for all potential forms of selection bias. We elected to restrict patients to only those with prolonged IHCA treated at ECPR centers but recognize this may introduce unmeasured selection bias, despite the use of propensity weighting and multivariable regression to control for suspected confounders. Second, the analysis was limited by the relatively small sample size of ECPR subjects compared with prior studies of ECPR in cardiac and adult populations. Third, the use of a multicenter data registry introduces the possibility of recording errors despite rigorous efforts on the part of the GWTG-R to ensure data integrity. Fourth, limitations in data granularity may have resulted in misclassifications, such as cases of postarrest ECMO deployment used following ROSC being classified as ECPR. Fifth, neurologic outcome data were limited due to missing PCPC data for many survivors, creating a notable challenge when reporting on survival after prolonged cardiac arrest. We recommend future study to address this knowledge gap in neurologic outcomes following noncardiac ECPR. Sixth, ECMO complication data are not recorded in the GWTG-R database, limiting the details of our survival outcomes. Finally, given limitations of data availability, we could not incorporate CPR mechanics into our interpretation of outcomes (39, 40). Given these limitations introduced through the use of retrospective data, there continues to be a need for prospective studies of ECPR in this population; whether equipoise exists for randomized controlled trials of ECPR vs. CCPR should be explored.

CONCLUSIONS

ECPR use in children with noncardiac illness has significantly increased in the last 20 years. Compared with CCPR, ECPR was not associated with a statistically significant increase in survival to discharge among children who required greater than or equal to 30 minutes of CPR. However, this may have been limited by statistical power and a Bayesian model estimated a positive effect of ECPR usage as well as low odds of harm with ECPR.

KEY POINTS.

Question:

The purpose of this investigation was to determine the association between the use of extracorporeal cardiopulmonary resuscitation (ECPR) and survival to hospital discharge in pediatric patients with a noncardiac illness category.

Findings:

The primary results of this investigation showed no statistically significant survival benefit with the usage of ECPR in a frequentist statistical model but did show a high probability of treatment benefit (85%) with a Bayesian statistical model.

Meaning:

This study did not identify a clear survival advantage with ECPR for noncardiac pediatric patients but the findings support the ongoing development of ECPR systems and further study in this population.