Prostaglandin Administration and Outcomes in Children on Extracorporeal Membrane Oxygenation

Introduction

Extracorporeal membrane oxygenation (ECMO) provides life-saving support for children with severe cardiac or respiratory failure, with more than 1500 newborns and 2500 pediatric patients supported worldwide each year. ¹ Although essential for survival in these critically ill populations, ECMO profoundly disrupts normal hemostatic balance and places patients at substantial risk for both major bleeding and thrombosis.^2,3 Mitigating these complications remains one of the central challenges in pediatric ECMO management. Contemporary anticoagulation practices rely largely on unfractionated heparin or bivalirudin, ⁴ agents that primarily target the plasma coagulation cascade and do not directly address platelet dysfunction, a key contributor to clot formation in the ECMO circuit.^5,6 This therapeutic gap highlights the need to explore alternative or adjunctive approaches that more comprehensively modulate hemostasis and may ultimately improve patient outcomes.

Both PGE₁ and PGI₂ inhibit platelet function by elevating intracellular cAMP levels through G-protein coupled receptor activation, which triggers protein kinase A (PKA)-mediated phosphorylation of multiple substrate proteins that suppress platelet activation, adhesion, and aggregation. ⁷ However, they differ significantly in their receptor selectivity, potency, and duration of action. PGI₂ acts primarily through the IP receptor with high affinity and is more potent than PGE₁ as an inhibitor of platelet aggregation against most agonists (ADP, adrenaline, collagen). In contrast, PGE₁ exhibits dual receptor activity, binding both stimulatory IP receptors and inhibitory EP3 receptors, with higher affinity for EP3 than for IP receptors, creating a concentration-dependent balance between inhibitory and potentially pro-aggregatory effects. Both prostaglandins are widely used in children with pulmonary hypertension and have an established safety record in this population. ⁸ Unlike anticoagulants that impair coagulation factors, PGE₁ and PGI₂ modulate platelet function, making them promising adjuncts in ECMO anticoagulation.

Existing evidence suggests potential benefits of prostaglandin-based anticoagulation. A meta-analysis of 17 studies in critically ill patients receiving continuous renal replacement therapy demonstrated that PGI₂-based strategies significantly reduced bleeding compared with traditional anticoagulation regimens. ⁹ Moreover, a randomized, double-blind, placebo-controlled phase II trial in 114 adult ECMO patients showed that adding PGE₁ to heparin was safe and was associated with fewer bleeding and thrombotic events. ¹⁰ Despite these promising results, the therapeutic implications of prostaglandins in pediatric ECMO remain largely unexamined. And in practice, prostaglandins may be introduced to mitigate circuit thrombosis despite therapeutic anticoagulation rather than primarily to reduce bleeding risk.

To address this knowledge gap, the goal of this study is to evaluate whether prostaglandin-modulated anticoagulation is associated with improved bleeding and thrombotic outcomes in critically ill children supported with ECMO. By defining the potential clinical benefits of prostaglandins, this study aims to inform whether PGI₂ or PGE₁ should be considered as adjuncts to standard anticoagulation strategies in pediatric ECMO.

Material and Methods

Outcomes

The primary outcomes of the study were bleeding and clotting events, identified through diagnostic codes in PHIS. Intracranial hemorrhage was defined using ICD-9 codes 430, 431, and 432.9, as well as ICD-10 codes I60xx, I61x, and I62xx. Stroke was defined using ICD-9 codes 434.xx and ICD-10 codes I634xx (see supplemental material; Table 1). Overall bleeding complications were defined as intracranial, pulmonary, and/or gastrointestinal hemorrhages. Overall clotting complications were defined as limb ischemia, pulmonary embolism, and/or stroke. Secondary outcomes included hospital length of stay, measured in days from admission to discharge; ECMO duration, defined as the number of days with an ECMO flag; and in-hospital mortality, defined as death occurring during the same hospitalization. The tertiary outcome was total hospital charges adjusted by the Center for Medicare & Medicaid Services (CMS) price index for the hospital's location, as reported in PHIS, which served as a proxy for financial impact.

OPEN IN VIEWER

Table 1.

Demographics.

	No ProstaglandinN = 19020	PGE1N = 2208	PGI2N = 725	Both PGE1 and PGI2N = 137	P Value
Age (days)	101 (5, 1237.5)	2(1, 7)	9 (1, 322)	2 (1, 5)	<.001
Neonates	39.8% (n = 7570) (39.1, 40.5)	94.2% (n = 2080) (93.2, 95.1)	57.2% (n = 415) (53.6, 60.8)	90.5% (n = 124) (85.6, 95.4)	<.001
Sex (male)	54.2% (n = 10,309) (53.5, 54.9)	56.4% (n = 1245) (54.3, 58.5)	51.7% (n = 375) (48.1, 55.4)	54.0% (n = 74) (45.7, 62.4)	.118
VA ECMO	81.6% (n = 15,520) (80.8, 82.4)	89.9% (n = 1985) (88.1, 91.2)	86.2% (n = 625) (82.2, 90.2)	79.7% (n = 109) (69.8, 89.5)	<.001
Central Cannulation	18.1% (n = 3,443) (17.3, 19.0)	18.4% (n = 406) (16.2, 20.8)	15.2% (n = 110) (11.5, 19.8)	10.9% (n = 15) (5.4, 20.8)	<.001
Pulmonary hypertension	12.3% (n = 2,339) (11.9, 12.8)	7.0% (n = 1546) (6.0, 8.1)	29.8% (n = 216) (26.3, 33.1)	19.7% (n = 27) (13.1, 26.3)	<.001
Sepsis	27.6% (n = 5250) (27.0, 28.2)	12.3% (n = 272) (11.0, 13.8)	21.8% (n = 158) (18.8, 24.8)	10.2% (n = 14) (5.2, 15.3)	<.001
Cardiac failure	54.0% (n = 10,271) (53.3, 54.7)	58.0% (n = 1281) (56.0, 60.1)	54.5% (n = 395) (50.9, 58.1)	52.6% (n = 72) (44.2, 60.1)	.004
Respiratory failure	48.0% (n = 9130) (47.3, 48.7)	14.0% (n = 309) (12.6, 15.4)	40.0% (n = 290) (36.4, 43.6)	25.6% (n = 35) (18.2, 32.9)	<.001
Coagulopathy	2.8% (n = 5,326) (2.6, 3.1)	2.2% (n = 486) (0.6, 3.5)	2.5% (n = 181) (1.4, 3.6)	0.7% (n = 10) (0, 2.2)	<.001
Congenital heart disease	56.7% (n = 10,784) (56.0, 57.4)	84.1% (n = 1857) (82.6, 85.6)	60.4% (n = 438) (56.9, 64.0)	80.3% (n = 110) (73.6, 86.6)	<.001

Results are presented as medians and interquartile ranges or as percentages and 95% confidence intervals.

Abbreviations: ECMO, extracorporeal membrane oxygenation; PGE₁, prostaglandin E₁; PGI₂, prostaglandin I₂; VA, veno-arterial.

Results

Population

A total of 22,061 pediatric patients who received ECMO were included in the study. The median age at initiation was 50 days (IQR 3, 842). Most patients (82.0% [18,090/22,061], 95% CI 81.9, 83.4) were supported with veno-arterial ECMO (the remaining 18.0% [3971/22,061] of patients were supported with veno-venous ECMO), and 18.0% (3971/22,061) (95% CI 17.3, 18.8) underwent central cannulation. Pulmonary hypertension was present in 12.4% (2736/22,061) (95% CI 12.0, 12.9). The most common comorbidities and clinical conditions were congenital heart disease (59.7% [13,170/22,061], 95% CI 59.0, 60.3), cardiac failure (54.4% [12,001/22,061], 95% CI 53.7, 55.1), and respiratory failure (44.2% [9751/22,061], 95% CI 43.6, 44.9), sepsis (25.8% [5692/22,061], 95% CI 25.2, 26.4), and coagulopathies (2.6% [5,736/22,061], 95% CI 2.4, 2.8). Most patients were treated exclusively with heparin (86.8% [19,149/22,061], 95% CI 86.3, 87.3). A very small proportion received bivalirudin as their sole anticoagulant (0.16% [353/22,061], 95% CI 0.11, 0.22), and 11.2% (2471/22,061] of patients (95% CI 10.8, 11.7) received both heparin and bivalirudin during their course of care. Only a small minority of patients (381/22,061, 1.7%) did not receive unfractionated heparin or bivalirudin during ECMO support; among these, 7/381 (1.8%) received PGE₁ and 1/381 (0.3%) received PGI₂. PGE₁ was administered in 10.6% (2338/22,061) of patients (95% CI 10.2, 11.0), and PGI₂ in 3.8% (838/22,061 (95% CI 3.6, 4.1). Table 1 describes the population according to the use of prostaglandins, and Figure 1 shows the temporal trend of PGE₁ and PGI₂ over time.

OPEN IN VIEWER

Figure 1.

Annual proportion of ECMO patients receiving prostaglandins. Stacked bars show the percentage of patients treated with PGE₁ only, PGE₁ and PGI₂, PGI₂ only, and no prostaglandin for each ECMO start year. Abbreviation: ECMO, extracorporeal membrane oxygenation.

As shown in Figure 2, among patients treated with prostaglandins, most did not have a documented diagnosis of pulmonary hypertension. Specifically, 92.2% (2036/2208) of patients who received PGE₁ had no pulmonary hypertension (95% CI 91.1, 93.3), and 71.8% (521/725) of those who received PGI₂ were not diagnosed with pulmonary hypertension (95% CI 68.8, 74.8). Conversely, among patients with pulmonary hypertension, the large majority (85.6%, 16,281/19,020) did not receive either PGE₁ or PGI₂ (95% CI 84.3, 86.9).

OPEN IN VIEWER

Figure 2.

Venn diagram illustrating the overlap between pulmonary hypertension, PGE₁ administration, and PGI₂ administration among pediatric patients supported with ECMO. Proportional areas represent the number of patients in each group and overlap category. Abbreviation: ECMO, extracorporeal membrane oxygenation.

PGE₁

Adjusting for site effect, and patient and ECMO characteristics, PGE₁ was not independently associated with intracranial hemorrhage (adjusted OR 1.01, 95% CI 0.79, 1.29, P = .92; Figure 3A) or gastrointestinal hemorrhage (adjusted OR 0.97, 95% CI 0.52, 1.81, P = .93; Table 2). In contrast, PGE₁ infusion was independently associated with an increased risk of pulmonary hemorrhage (adjusted OR 1.24, 95% CI 1.08, 1.43, P = .003; Table 2). PGE₁ was not associated with ischemic stroke (adjusted OR 1.02, 95% CI 0.82, 1.28, P = .84; Figure 3B), pulmonary embolism (adjusted OR 0.57, 95% CI 0.27, 1.19, P = .14; Table 2), or arterial thromboembolic events (adjusted OR 0.91, 95% CI 0.74, 1.11, P = .34; Table 2).

OPEN IN VIEWER

Figure 3.

Forest plots of GLMM of intracranial hemorrhage (A) and ischemic stroke (B).

OPEN IN VIEWER

Table 2.

Outcomes Associated with Prostaglandin Infusion on ECMO.

Outcomes Associated with PGE1	Adjusted Odds Ratio	95% CI	P Value
Intracranial hemorrhage	1.01	0.79, 1.29	.92
Ischemic stroke	1.02	0.82, 1.28	.84
Pulmonary hemorrhage	1.24	1.08, 1.43	.003
Gastrointestinal hemorrhage	0.97	0.52, 1.81	.93
Pulmonary embolism	0.57	0.27, 1.19	.14
Arterial clot	0.91	0.74, 1.11	.34
ECMO duration	2.31	1.29, 3.32	<.001
Hospital length of stay	2.00	−1.06, 5.06	.20
Mortality	1.61	1.47, 1.77	<.001
Hospital Charges	1.14	1.10, 1.18	<.001

Outcomes associated with PGI2	Adjusted Odds Ratio	95% CI	P value
Intracranial hemorrhage	0.96	0.72, 1.29	.79
Stroke	0.88	0.65, 1.18	.38
Pulmonary hemorrhage	1.21	1.00, 1.47	.047
Gastrointestinal hemorrhage	1.13	0.66, 1.95	.66
Pulmonary embolism	0.96	0.46, 1.99	.91
Arterial clot	1.08	0.80, 1.46	.60
ECMO duration	7.72	6.15, 9.28	<.001
Hospital length of stay	14.54	9.81, 19.26	<.001
Mortality	1.71	1.47, 1.97	<.001
Hospital charges	1.51	1.43, 1.60	<.001

The generalized linear mixed model adjusted for pulmonary hypertension, age category (<28 days vs ≥28 days), ECMO characteristics (veno-arterial vs veno-venous vs unknown, central cannulation vs peripheral vs unknown), anticoagulation (heparin vs bivalirudin), concomitant infusion of prostaglandin I₂, patient diagnoses (respiratory failure, cardiac failure, congenital heart disease, sepsis, and coagulopathies), and site (using a random effect).

Abbreviations: PGE₁, prostaglandin E₁; PGI₂, prostaglandin I₂; ECMO, extracorporeal membrane oxygenation.

PGI₂

PGI₂ infusion was not independently associated with intracranial hemorrhage (adjusted OR 0.96, 95% CI 0.72, 1.29, P = .79; Figure 3A) or gastrointestinal hemorrhage (adjusted OR 1.13, 95% CI 0.66, 1.95, P = .66; Table 2). PGI₂ infusion was associated with a higher risk of pulmonary hemorrhage (adjusted OR 1.21, 95% CI 1.00, 1.47, P = .047; Table 2), but was not associated with pulmonary embolism (adjusted OR 0.96, 95% CI 0.46, 1.99, P = .91) or arterial thromboembolic events (adjusted OR 1.08, 95% CI 0.80, 1.46, P = .60; Table 2). PGI₂ was also not associated with ischemic stroke (adjusted OR 0.88, 95% CI 0.65, 1.18, P = .38; Figure 3B).

Mortality

The hospital mortality rate was 39.2% (7456/19,020) in patients without prostaglandins, 47.9% (1058/2208) in patients who received PGE1, 51.3% (372/725) in patients who received PGI₂, and 60.6% (83/137) in patients who received both. The median survival time was 120 days in patients without prostaglandins, 93 days in those who received PGE₁, 97 days in those who received PGI₂, and 47 days in patients who received both (log-rank P < .001, Figure 4).

OPEN IN VIEWER

Figure 4.

Kaplan–Meier curves for hospital survival stratified by prostaglandin exposure. Patients were grouped into four categories: no prostaglandin exposure (light gray), PGE₁ alone (red), PGI₂ alone (green), and combined PGE₁ plus PGI₂ therapy (orange). Time was measured in hospital days from admission to discharge or death. Survival distributions differed significantly across the four groups (log-rank P < .001).

Adjusting for site effect, and patient and ECMO characteristics, PGE₁ and PGI₂ were independently associated with increased risk of death (adjusted OR 1.61, 95% CI 1.47, 1.77, P < .001; and adjusted OR 1.71, 95% CI 1.47, 1.97, P < .001, respectively).

Discussion

In this large, multicenter cohort of more than 22,000 pediatric ECMO patients, neither PGE₁ nor PGI₂ was associated with reductions in major bleeding or thrombotic complications. Both prostaglandins showed no independent association with intracranial hemorrhage, gastrointestinal hemorrhage, ischemic stroke, pulmonary embolism, or arterial thromboembolism after adjustment for patient and ECMO characteristics. However, each was associated with a modest but statistically significant increase in pulmonary hemorrhage. Beyond bleeding and clotting outcomes, prostaglandin exposure was consistently associated with markers of greater illness severity, including longer ECMO runs, longer hospital stays (for PGI₂), higher total hospital charges, and higher adjusted mortality. These findings suggest that, in contemporary pediatric ECMO practice as captured in PHIS, prostaglandin-modulated anticoagulation does not confer measurable protection against the major hemorrhagic or thrombotic complications. However, in clinical practice, prostaglandins may be initiated in response to refractory circuit thrombosis or high clot burden despite therapeutic anticoagulation, which would introduce confounding by indication and bias associations toward worse outcomes.

The observed association between prostaglandin exposure and increased mortality warrants careful interpretation. In our cohort, the median time to death exceeded three months among patients receiving prostaglandins, whereas the median ECMO duration was less than 2 weeks. This temporal separation suggests that excess mortality may reflect the underlying disease severity prompting prostaglandin use rather than a direct harmful effect of prostaglandins themselves. This interpretation is reinforced by the absence of an association between prostaglandin exposure and major hemorrhagic or thrombotic complications, events that typically drive ECMO-related mortality, apart from a modest increase in pulmonary hemorrhage. An additional consideration is the potential underreporting of pulmonary hypertension in administrative datasets. As shown in our cohort, most patients who received prostaglandins did not have a documented diagnosis of pulmonary hypertension. It is plausible that clinicians initiated prostaglandins for suspected pulmonary hypertension that was not formally coded. If a substantial proportion of prostaglandin recipients had unrecognized or undocumented pulmonary hypertension, this would explain their markedly longer ECMO runs, prolonged hospitalization, and increased mortality, since pulmonary hypertension itself is strongly associated with more complex physiology, slower recovery, and greater risk of late deterioration. Under this interpretation, prostaglandin use may serve as a surrogate marker for an unmeasured severity phenotype rather than a contributor to excess mortality. Prior evidence from the only randomized controlled trial to date, conducted in adults on venovenous ECMO, similarly demonstrated a numerically higher mortality in the PGE₁ arm (42% vs 17% mortality at 90 days) that did not reach statistical significance (P = .11). ¹⁰ Taken together, these findings indicate that although a causal effect of prostaglandins on mortality cannot be fully excluded, the more plausible explanation is confounding by indication, with prostaglandins preferentially administered to patients with more complex physiology or more severe underlying illness. Similarly, the absence of an observed reduction in thrombotic complications should be interpreted with caution. In clinical practice, prostaglandins may be initiated in response to evolving or established circuit thrombosis despite therapeutic anticoagulation. As such, reverse causation and confounding by indication may obscure a potential protective effect, particularly if prostaglandins are introduced after thrombotic processes have already begun. This raises the possibility that earlier initiation of prostaglandins, prior to the development of thrombotic complications, may yield different results, as suggested by randomized data in adults on VV-ECMO. ¹⁰

For clinicians interested in using prostaglandins specifically as adjuncts to anticoagulation on ECMO, safety remains a central concern. Prostaglandins inhibit platelet function, and some may worry that adding them to standard heparin or bivalirudin regimens could heighten bleeding risk. Although we did not observe an increase in major intracranial or gastrointestinal hemorrhage, the modest increase in pulmonary hemorrhage and the association with increased mortality underscore that definitive conclusions regarding safety or efficacy cannot be drawn from these data. This is consistent with the only randomized controlled trial to date, conducted in adults on venovenous ECMO, which found that adding PGE₁ to heparin was associated with fewer overall bleeding and thrombotic complications compared with placebo. ¹⁰ It is important to acknowledge that we cannot determine whether bedside clinicians adjusted anticoagulation doses when prostaglandins were initiated. These findings should therefore be interpreted as hypothesis-generating rather than confirmatory with respect to safety.

This study has several limitations inherent to retrospective analyses using administrative datasets. First, PHIS does not capture the specific indication for ECMO, as diagnostic codes list diagnoses without clarifying which condition led to cannulation. The dataset also lacks granular clinical information such as laboratory values (including platelet counts, coagulation profiles, and anti-Xa levels). It further does not include medication dosing, including prostaglandin E1 dose, whose pro- and anti-platelet effects are dose dependent. Misclassification is possible because billing data may not fully reflect actual medication administration or duration, and coding errors may affect identification of bleeding or thrombotic events. We were also unable to identify patients receiving prostaglandins specifically for maintenance of ductal arteriosus patency or right ventricular decompression, populations that may carry distinct risks for pulmonary hemorrhage and confound observed associations. In addition, missing data for some variables, such as comorbidities, may limit the ability to adjust for confounding. Although analyses were adjusted for pulmonary hypertension, this diagnosis was uncommon among patients receiving prostaglandins, potentially contributing to the observed association with mortality. Because pulmonary hemorrhage was defined using administrative ICD codes, milder or transient bleeding events may not have been captured, and misclassification remains possible. PHIS does not allow determination of the clinical indication or timing of prostaglandin initiation relative to ECMO cannulation, and diagnostic codes reflect conditions documented at any point during hospitalization rather than necessarily at the time prostaglandins were started. In addition, administrative diagnostic codes cannot distinguish the severity of pulmonary hemorrhage, although there is substantial clinical variation between minimal bloody secretions in an endotracheal tube and life-threatening pulmonary bleeding. PHIS also captures only major complications, which means less severe events (such as chest tube output or bleeding at the cannulation site) are not recorded, as PHIS does not contain specific diagnostic codes for cannulation site bleeding or quantitative measures such as chest tube output, limiting our ability to assess less severe but clinically relevant bleeding complications, and potentially leading to underestimation of outcome rates. PHIS also does not provide reliable temporal linkage between surgical procedures and ECMO cannulation, preventing adjustment for the timing of cardiac surgery relative to prostaglandin exposure. As with all retrospective observational studies, causal inference cannot be drawn, and unmeasured confounders and treatment-selection biases may influence observed associations. Nonetheless, the inclusion of more than 20,000 patients across multiple centers supports the generalizability of these findings.

In conclusion, the role of prostaglandins as adjuncts to anticoagulation during ECMO remains uncertain. In this large multicenter analysis, prostaglandin E₁ and prostaglandin I₂ were not associated with reductions in major bleeding or thrombotic events. Associations with longer ECMO runs, prolonged hospitalization, and higher mortality likely reflect confounding by indication rather than direct harm. Given the observational design, potential for confounding by indication, and absence of demonstrated efficacy in reducing thrombotic events, these findings should not be interpreted as evidence supporting routine prostaglandin use as adjunct anticoagulation in pediatric ECMO. Prospective studies with standardized anticoagulation protocols and granular physiologic data are needed to determine whether prostaglandins confer benefit in selected patient subgroups.

Supplemental Material