Prognostic Factors Associated With Intracranial Hemorrhage and Ischemic Stroke During Venovenous Extracorporeal Membrane Oxygenation: A Systematic Review

Abstract

OBJECTIVES:

Venovenous extracorporeal membrane oxygenation (ECMO) is a life-preserving intervention for patients with respiratory failure refractory to conventional mechanical ventilation. Intracranial hemorrhage (ICH) and ischemic stroke are life-threatening complications associated with venovenous ECMO. Despite this, little is known regarding the prognostic factors associated with these adverse neurologic events. We conducted a systematic review that characterizes these predictors of ICH and ischemic stroke during venovenous ECMO.

DATA SOURCES:

We conducted a comprehensive search of MEDLINE and Embase via the Ovid interface.

STUDY SELECTION:

We developed and performed a literature search to identify articles that evaluated ICH and ischemic stroke in adult patients undergoing venovenous ECMO. We excluded studies based on design, target population, and outcomes.

DATA EXTRACTION:

Data were extracted manually by one reviewer. Risk of bias assessment was completed using the Quality in Prognostic Studies approach for each included study. Prognostic factors associated with ICH and ischemic stroke that were identified in two or more included studies were evaluated through the Grading of Recommendations, Assessment, Development, and Evaluation approach.

DATA SYNTHESIS:

Three hundred thirty-three studies met criteria for screening. Seventeen studies met final inclusion criteria. Seventeen studies addressed predictors of ICH. Five studies demonstrated an increased risk of ICH with lower pH before venovenous ECMO (moderate certainty). Five studies demonstrated an increased risk of ICH with greater decreases in Paco₂ pre- to post-venovenous ECMO cannulation (moderate certainty). Four studies addressed predictors of ischemic stroke; however, there were no predictors of ischemic stroke identified in two or more of the included studies.

CONCLUSIONS:

This systematic review demonstrates that abnormalities and changes in blood gas parameters from pre- to post-venovenous ECMO cannulation are probably associated with increased risk of ICH. Additional high-quality studies dedicated to probable predictors of these adverse neurologic events are crucial to understanding the pathophysiology of ICH and ischemic stroke in this population and informing clinical practice to mitigate the risk of these life-threatening events.

KEY POINTS.

Question: This review aims to characterize prognostic factors associated with intracranial hemorrhage (ICH) and ischemic stroke in adult patients undergoing venovenous extracorporeal membrane oxygenation (ECMO).

Findings: Lower pH before venovenous ECMO cannulation and a greater decrease in Paco₂ pre- to post-venovenous ECMO cannulation were probably associated with an increased risk of ICH. Elevated Paco₂ before ECMO cannulation, greater increase in Pao₂ from pre- to post-ECMO cannulation and use of vasoactive medications may have been associated with increased risk of ICH. No consistent predictors of ischemic stroke in this population were identified for assessment.

Meaning: Abnormalities and changes in blood gas parameters from pre- to post-venovenous ECMO cannulation are probably associated with increased risk of ICH.

Venovenous extracorporeal membrane oxygenation (ECMO) is a life-preserving intervention for patients with hypoxia and respiratory failure refractory to conventional mechanical ventilation (1, 2). Neurologic injury, including intracranial hemorrhage (ICH) and ischemic stroke are known complications associated with venovenous ECMO, both of which are associated with increased mortality (3, 4). Despite this, little is known regarding the pathophysiology and predictors for ICH and ischemic stroke in patients undergoing venovenous ECMO.

Our understanding of the development of ICH and ischemic stroke in venovenous ECMO patients is limited, largely due to the lack of studies that address these complications. While several studies evaluate ICH and ischemic stroke in patients undergoing venovenous ECMO, these studies often do not investigate potential prognostic factors related to these neurologic outcomes (5, 6). Additionally, there are studies that evaluate all adverse neurologic events (e.g., ICH, ischemic stroke, hypoxia-induced brain injury, seizure) as a composite outcome in this population, despite these various complications likely occurring due to different pathophysiologic mechanisms (7–9). This is supported by evidence suggesting that the prevalence of these adverse events differs between one another (3, 4). Finally, studies report on ICH and ischemic stroke collectively across a population receiving different ECMO modalities, including venovenous ECMO, venoarterial ECMO, and hybrid modalities (10–14). This is despite the prevalence of these adverse outcomes varying across different ECMO modalities, leading to gaps in the current understanding of the pathophysiology and predictors of ICH and ischemic stroke across different modalities (15). As such, the aim of this study was to conduct a systematic review that identifies and summarizes potential predictors of ICH and ischemic stroke separately, in adult patients undergoing venovenous ECMO.

MATERIALS AND METHODS

Study Design

This systematic review followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (16). Furthermore, the Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies checklist was used to structure the study (17). The review protocol was registered through PROSPERO (CRD42023399089).

Data Sources and Search Strategy

With the assistance of a health sciences librarian, we developed and performed a literature search through MEDLINE (Ovid) and Embase (Ovid) using controlled vocabulary and keywords (Supplemental Fig. 1, http://links.lww.com/CCM/H622) from time of inception to November 16, 2023.

Study Selection

We included all English language full text articles of randomized controlled trials, retrospective and prospective cohort studies, and cross-sectional studies. We excluded case series and case reports, as these would not present a point estimate of our outcomes of interest. To meet inclusion criteria, studies must have included patients 16 years old or older undergoing venovenous ECMO, and must have reported on prognostic factors, or predictors associated with ICH or ischemic stroke. Articles were only included if ICH or ischemic stroke events met clinical criteria as defined by the Extracorporeal Life Support Organization registry database definitions (18). We excluded studies that did not report on predictors of ICH or ischemic stroke in venovenous ECMO and studies that only included composite neurologic outcomes. We excluded those studies that reported on a mixed population of ECMO patients (e.g., those that included data on venoarterial and venoarterial-venous ECMO) without reporting on data specific to the venovenous ECMO subgroup.

Data Extraction

We used the Covidence software (Melbourne, Australia) to standardize the review strategy and methods between two reviewers (S.T., A.J.W.). Articles were compiled using the initial search strategy. Each reviewer individually conducted an abstract review of studies identified in the literature search. The reviewers then met to discuss and resolve conflicts. Thereafter, each reviewer individually conducted a full-text review of the remaining studies. The reviewers then met to discuss and resolve conflicts. Data on both predictors and negative findings related to ICH and/or ischemic stroke were extracted for each included article.

Risk of Bias Assessment

We assessed each included study for risk of bias using the Quality in Prognostic Studies (QUIPS) approach (19). This is an evidence-based tool to assess the risk of bias for studies of prognostic factors, and grades each study as having either a low, moderate, or high risk of bias. Consistent with QUIPS guidelines, observational studies start as having a high risk of bias based on study design and can be downgraded to moderate or low based on factors related to study participation, study attrition, prognostic factor measurement, outcome measurement, and study confounders. Two reviewers (A.J.W., S.T.) participated in QUIPS risk of bias assessment.

Prognostic Factors Assessment

Overall certainty of prognostic factors of ICH or ischemic stroke that were identified in two or more studies were evaluated through the Grading of Recommendations, Assessment, Development, and Evaluation (GRADE) approach by two reviewers (S.T., A.J.W.) (20). The importance of each predictor to patient outcome (ICH or ischemic stroke) was assessed by a critical care trained and practicing physician (S.T., D.G.). The reviewers used the GradePro Guideline Development Tool online software (Hamilton, ON, Canada) to complete GRADE assessments. Each prognostic factor is categorized as having a high, moderate, low, or very low level of certainty based on study methodology, as well as assessments of prognostic factor precision, consistency, risk of bias, directness, and publication bias. Consistent with GRADE guidelines, observational data start as high certainty evidence, but can be downgraded based on other characteristics. When defining the association between a prognostic factor and an outcome through certainty, GRADE guidelines describe high certainty as associated, moderate certainty as probably associated, low certainty as may have been associated, and very low certainty as uncertain. Importance, or how clinically relevant each prognostic factor is to the patient outcome, is then determined as being critical, important, or of low importance by experts in the field—in this case, critical care specialists who actively engages in venovenous ECMO care (S.T., D.G.).

After compiling and extracting the data, we conducted a systematic review rather than a meta-analysis for several reasons. First, there is a small volume of literature available on this topic. Second, there is heterogeneity as to what data were collected, and how it was collected, across the various studies evaluating predictors of ICH and ischemic stroke in venovenous ECMO populations. Finally, statistical analysis and reporting of odds ratios was variable across different studies, which restricted the ability to perform a meta-analysis.

RESULTS

Study Selection and Characteristics

The PRISMA diagram can be found in Figure 1. We identified 333 articles in our initial electronic search. After abstract screening, 65 full-text articles underwent full-text review for inclusion. Of those, 17 full-text articles were used for data extraction, comprising 16 observational cohort studies and one nonrandomized experimental study. All studies were published from 2015 onward, with 15 of the studies having been published from 2021 onward. Study characteristics are outlined in Supplemental Table 1 (http://links.lww.com/CCM/H622).

Figure 1.

Preferred Reporting Items for Systematic Reviews and Meta-Analyses flow diagram. This diagram was made using Covidence software (Covidence systematic review software, Veritas Health Innovation, Melbourne, Australia; www.covidence.org). ECMO = extracorporeal membrane oxygenation, ICH = intracranial hemorrhage, VA = venoarterial.

Risk of Bias Assessment

QUIPS risk of bias assessment results are outlined in Supplemental Table 2 (http://links.lww.com/CCM/H622). Of the 17 included studies, all were assessed as having either moderate to high risk of bias in at least one of the assessed domains. Sixteen of the 17 included articles were observational cohort studies and thus were initially graded as high risk of bias. Generally, studies had a low risk of bias when considering QUIPS category of study attrition given retrospective designs. Most studies were considered to have a high risk of bias due to prognostic factors not identified a priori, outcome measurement (brain imaging) not being standardized in all patients, confounding factors being identified based on available data and not predetermined which resulted in having limited ability to adjust for confounding given observational designs.

Prognostic Factors Associated With ICH

Sixteen studies addressed predictors of ICH. All studies that were deemed to be very low, low, or moderate level of certainty were down-graded due to issues with inconsistency or imprecision on certainty assessment. A summary of the findings from the included studies, including statistical analysis of the covariates (p, odds ratio, and CIs when reported) are outlined in Supplemental Table 3 (http://links.lww.com/CCM/H622). The GRADE findings are outlined in Table 1.

TABLE 1.

Grade Assessment of Predictors of Intracranial Hemorrhage

Prognostic Factor	No. of Studies	Certainty Assessment					Certainty	Importance
		Study Design	Risk of Bias	Inconsistency	Indirectness	Imprecision
Lower pH before cannulation	6	Nonrandomized studies	Not serious	Not seriousa	Not serious	Seriousb	⨁⨁⨁◯ Moderate	IMPORTANT
Elevated Pco2 before cannulation	9	Nonrandomized studies	Not seriousc	Seriousd	Not serious	Seriousb	⨁⨁◯◯ Low	IMPORTANT
Greater decrease in Pco2 from pre- to post-cannulation	9	Nonrandomized studies	Not serious	Not seriousc,e	Not serious	Seriousc	⨁⨁⨁◯ Moderate	IMPORTANT
Greater increase in Po2 from pre- to post-cannulation	5	Nonrandomized studies	Not serious	Seriousc	Not serious	Seriousb	⨁⨁◯◯ Low	IMPORTANT
Lower platelet concentration throughout treatment	9	Nonrandomized studies	Not serious	Very seriousc	Not serious	Extremely seriousb	⨁◯◯◯ Very low	IMPORTANT
Higher activated partial thromboplastin time levels	6	Nonrandomized studies	Not serious	Very seriousc	Not serious	Extremely seriousb	⨁◯◯◯ Very low	IMPORTANT
Higher lactate dehydrogenase concentration throughout treatment	2	Nonrandomized studies	Seriousc	Seriousb	Not serious	Seriousc	⨁⨁◯◯ Low	NOT IMPORTANT
Acute kidney injury and/or higher creatinine concentration throughout treatment	6	Nonrandomized studies	Not serious	Seriousc	Not serious	Very seriousb	⨁◯◯◯ Very low	NOT IMPORTANT
Higher total bilirubin concentration throughout treatment	4	Nonrandomized studies	Not serious	Seriousc	Not serious	Extremely seriousb	⨁◯◯◯ Very low	NOT IMPORTANT
Vasoactive medication use	4	Nonrandomized studies	Not serious	Seriousf	Not serious	Seriousb	⨁⨁◯◯ Lowc	NOT IMPORTANT

^aSome studies found statistically significant association (n = 5) between predictor and intracranial hemorrhage (ICH) while others found no statistically significant association (n = 1).

^bStudies show opposite directions of effect on ICH.

^cMeasures of effect and CIs not consistently reported.

^dSome studies found statistically significant association (n = 5) between predictor and ICH while others found no statistically significant association (n = 4).

^eSome studies found statistically significant association (n = 6) between predictor and ICH while others found no statistically significant association (n = 3).

^fSome studies found statistically significant association (n = 2) between predictor and ICH while others found no statistically significant association (n = 2).

Lower pH pre-venovenous EMCO cannulation was probably associated with ICH (Moderate Certainty). Five of six studies evaluating this parameter found a significant association between pH pre-venovenous EMCO cannulation and ICH, with odds ratios ranging from 0.03 to 0.1 (4, 21–24). Greater decrease in Paco₂ from pre- to post-ECMO cannulation was probably associated with ICH (Moderate Certainty) (21, 25, 28–31). Four of six studies evaluating this parameter found a significant association decrease in Paco₂ from pre- to post-ECMO cannulation and ICH, with odds ratios ranging from 1.22 to 6.02 (21, 25, 28–31). Higher Paco₂ pre-ECMO cannulation may have been associated with ICH development (low certainty) (4, 21, 24–26, 28, 29). A greater increase in Pao₂ from pre- to post-ECMO cannulation may have been associated with ICH (low certainty) (21, 26, 30, 32).

The relationship between lower platelet concentration throughout treatment and ICH was uncertain (very low certainty). Several studies found an association with lower platelet concentrations and ICH development. This relationship was demonstrated with: 1) lower mean/median platelet concentrations pre-ECMO cannulation (26, 28), 2) greater decrease in platelet concentrations from pre- to post-ECMO cannulation (26), 3) lower mean/median platelet concentration post-ECMO cannulation (25), and 4) lowest platelet concentration post-ECMO cannulation (30). Furthermore, other studies did not find an association between ICH and lower mean/median platelet concentrations pre-ECMO cannulation (21, 27), lower mean/median platelet concentration during ECMO treatment (22, 29), or the lowest platelet concentration post-ECMO cannulation (22, 31). The relationship between elevated activated partial thromboplastin time (aPTT) and ICH was also uncertain (very low certainty) (21, 22, 28–30, 32).

Higher lactate dehydrogenase (LDH) concentration may have been associated with ICH (low certainty) (23, 24). The relationship between acute kidney injury (AKI), higher serum creatinine concentration, and/or renal replacement therapy and ICH was uncertain (very low certainty) (4, 21, 22, 30–32). The use of vasoactive medications, including a diagnosis of septic shock may have been associated with ICH (low certainty) (23, 24, 30). In one study, a dose-response effect was found, with increased median norepinephrine requirements during venovenous ECMO treatment being associated with the development of ICH (24).

Prognostic Factors Associated With Ischemic Stroke

Four articles addressed prognostic factors related to ischemic stroke in venovenous ECMO (4, 30-32). Based on our parameters of only completing GRADE assessment of predictors that were associated with ischemic stroke in two or more included studies, there were no eligible predictive factors of ischemic stroke that could be identified in our systematic review.

DISCUSSION

To our knowledge, we report the first systematic review on prognostic factors associated with ICH and ischemic stroke in patients undergoing venovenous ECMO. Our systematic review reveals that there are limited data describing the predictors of these adverse neurologic outcomes in venovenous ECMO despite these complications being well-established in this population and life threatening. Arterial blood gas parameters (e.g., pH, Paco₂, Pao₂) pre-ECMO cannulation, during initial 24 hours of venovenous ECMO treatment, and changes from pre- to post-ECMO cannulation are probably associated with the development of ICH in this population, although with moderate certainty based on GRADE criteria. Other parameters that were identified throughout the literature as possible predictors of ICH and/or ischemic stroke, included both severity of illness and evidence of end-organ damage (e.g., use of vasoactive medications, elevated LDH, hyperbilirubinemia, AKI and/or elevated creatinine, use of renal replacement therapy). While this may indicate that severe illness is associated with the occurrence of ICH and/or ischemic stroke in this population, it is less likely that discrete biomarker abnormalities are implicated in the pathophysiology of these adverse neurologic events. Coagulopathy (thrombocytopenia, elevated aPTT) and anticoagulation strategy were also identified as possible predictors of ICH; however, more prospective studies are required to elucidate optimal coagulation parameters that mitigate the risk of adverse neurologic outcomes, while also preventing ECMO circuit malfunction.

The findings of this review confirm the limited quality evidence regarding neurologic injury in venovenous ECMO. Prospective studies frequently combine ICH and ischemic stroke into a composite outcome to increase event rate despite these outcomes likely having different mechanisms of injury (7–9). This review highlights the paucity in the literature surrounding the predictors of ischemic stroke in venovenous ECMO patients. While ischemic stroke is less common than ICH, it remains an independent predictor of morbidity and mortality in this population (3, 4). We feel this is an important gap in the literature that must be addressed to better understand the pathophysiology, and strategies to prevent, ischemic stroke in venovenous ECMO. No study identified had a standardized imaging protocol to identify ICH or ischemic stroke and thus likely under report events. We found that pH and Paco₂ reduction from pre- to post-venovenous ECMO start were the only two factors with a moderate GRADE of evidence to predict ICH (4, 21, 22, 27–30). These likely represent a similar mechanism of action given the relationship of low pH and Paco₂, although this mechanism has not been confirmed in venovenous ECMO (33). Although Paco₂ changes are likely associated with ICH, the timing of these Paco₂ changes is unclear as pre-venovenous ECMO Paco₂, Paco₂ reduction on venovenous ECMO initiation, and Paco₂ reduction from venovenous ECMO start to 24 hours post-venovenous ECMO have all been identified as potential exposure thresholds. These thresholds have not been identified a priori and investigated in prospective studies and likely represent available data used in retrospective cohorts.

Anticoagulation is frequently administered to patients on ECMO to prevent circuit thrombosis. This predisposes patients to coagulopathies that may precipitate ICH. A recent retrospective cohort study by Hofmaenner et al (22) demonstrated that an anticoagulation strategy targeting lower anti-Xa levels in patients on venovenous ECMO may reduce the risk of ICH without predisposing patients to more thrombosis-related complications. Kalbhenn et al (23) conducted a nonrandomized experiment to study this as well. The primary outcome was a nonsignificant reduction in the prevalence and severity of ICH in an experimental group with lower aPTT targets during anticoagulation and targeted factor replacement based on biochemical deficiencies compared with controls. The heterogeneity in anticoagulation and factor administration practices across studies, as well as only one of the two studies demonstrating statistical significance in the prevalence of ICH, prevented the inclusion of anticoagulation and factor replacement strategies as a predictor of ICH in GRADE analysis. Our group recognizes this evolving area of research and the physiologic plausibility of more conservative anticoagulation strategies improving ICH outcomes. We are hopeful for further studies that demonstrate the efficacy of this practice change in the future.

The major limitation to this review is the lack of high-quality studies evaluating predictors of ICH or ischemic stroke in patients managed with venovenous ECMO. In particular, there is a paucity in the number of prospective studies, with 16 of 17 included articles reporting retrospective cohort studies. This limits the interpretation of the available data, as few studies intentionally, prospectively study possible predictors of ICH and ischemic stroke. This introduces a set of biases in how factors and outcomes are defined and measured. Another bias of this study design includes a lack of control for confounding variables. Although most studies attempted to account for this using multivariate statistical analysis, the covariates included in multivariate models were based on available data and not pre-identified risk factors for ICH or ischemic stroke, which has been shown to increase risk of bias in prognosis studies (34). The lack of available studies led to the identification of variables that may contribute to the development of these adverse neurologic events in only one study. Such variables include the number of lumens in the venovenous ECMO setup, the size of cannula used, and anticoagulation strategies during cannulation (22, 35, 36). We chose to only include predictors associated with the development of ischemic stroke or ICH in our GRADE analysis if they were consistently identified as possible predictors in two or more studies. As such, additional rigorous studies designed with the prospective aim of evaluating predictors of ICH and ischemic stroke in patients undergoing venovenous ECMO are required.

There is also heterogeneity in the available literature as to how variables are measured. This makes it difficult to determine what factors related to changes in laboratory findings from pre- to post-cannulation may predict the development of ICH. For instance, several studies determined a binary threshold at which changes in Paco₂ from pre- to post-ECMO cannulation were or were not related to increased prevalence of ICH development (31, 32), whereas several other studies measured change in Paco₂ as a continuous variable and determined that higher change predicted ICH (21, 22). Another example of this heterogeneity was how and when various coagulation variables were measured, such as platelet concentrations and aPTT. Some studies reported these variables as mean/median levels pre-ECMO cannulation (29, 31), change in levels from pre- to post-ECMO cannulation (29), mean/median level post-ECMO cannulation (21), or lowest level post-ECMO cannulation (32). This heterogeneity of reporting practices inhibits our ability to interpret the relationship between these variables and the prevalence of neurologic events. Despite this, we outlined how these included studies reported their findings and their outcomes in Supplemental Table 3 (http://links.lww.com/CCM/H622). These findings of our review highlight the lack of standard as to when these laboratory and blood gas measurements are drawn in proximity to cannulation, and thus it is difficult to determine what factors specifically may predict ICH development. Factors to consider include absolute change, rate of change, and range of values both before and after venovenous ECMO cannulation correlate with ICH development.

There is colinearity between the stronger predictors associated with ICH in our study, such as low pH and high Paco₂. This demonstrates that physiologic abnormalities and changes that are interdependent likely contribute to the pathogenesis of ICH in this population; however, with such colinearity it is difficult to determine which parameters are more contributory. Despite this, we feel that our findings highlight the importance of further high-quality studies evaluating the effect of normalizing, and avoiding rapid changes to, blood gas parameters before and during venovenous ECMO.

CONCLUSIONS

There is a lack of high-quality data on prognostic factors associated with ICH and ischemic stroke in adult patients undergoing venovenous ECMO. This study demonstrates that abnormalities and changes in blood gas parameters from pre- to post-venovenous ECMO cannulation are probably associated with increased risk of ICH with moderate certainty. The heterogeneity regarding strategies used to assess what factors may be related to these adverse neurologic outcomes in this population prohibits researchers understanding of how to conduct high-quality, prospective studies aimed at investigating predictors of ICH and ischemic stroke in venovenous ECMO. This systematic review identifies possible predictors of ICH in this population to inform future high-quality studies seeking to intentionally investigate predictive factors of ICH and ischemic stroke in the adult venovenous ECMO population. These studies are critical to bettering our understanding of the pathophysiology, as well as the clinical strategies to mitigate the risks, of these life-threatening neurologic events.

ACKNOWLEDGMENTS

We thank the library of the College of Physicians and Surgeons of British Columbia for their expert advice in the development of our initial search strategy.

Supplementary Material

ccm-53-e400-s001.pdf ^{(667KB, pdf)}