Long-Term Neuropsychiatric, Neurocognitive, and Functional Outcomes of Patients Receiving ECMO

Abstract

Background and Objectives

Despite the common occurrence of neurologic complications during extracorporeal membrane oxygenation (ECMO) support, data on long-term neuropsychiatric, neurocognitive, and functional outcomes are sparse. We aimed to determine the prevalence of long-term neuropsychiatric symptoms, neurocognitive and functional impairment, and favorable neurologic outcomes in adult patients who receive ECMO.

Methods

PubMed, Embase, Cochrane, Web of Science, and Scopus were searched for text related to ECMO and neuropsychiatric, neurocognitive, and functional outcomes from inception to May 3, 2023. Our primary outcome was the prevalence of neuropsychiatric symptoms (pain/discomfort, anxiety, depression, posttraumatic stress disorder [PTSD], and sleep disturbance) at long-term (≥6 months) follow-up. Our secondary outcomes were the prevalence of neurocognitive impairment (memory, attention, and reasoning), functional impairment (daily activities, physical activity/mobility, and personal/self-care), and favorable neurologic outcomes (Cerebral Performance Category ≤2, modified Rankin scale ≤3, or Glasgow Outcome Scale ≥4). This study was registered in PROSPERO (CRD42023420565).

Results

We included 59 studies with 3,280 patients (median age 54 years, 69% male). The cohort consisted of 86% venoarterial (VA)–ECMO (n = 2,819) and 14% venovenous (VV)–ECMO (n = 461) patients. More than 10 tools were used to assess neuropsychiatric and neurocognitive outcomes, indicating a lack of standardization in assessment methodologies. The overall prevalence of neuropsychiatric symptoms was 41% (95% CI 33%–49%): pain/discomfort (52%, 95% CI 42%–63%), sleep disturbance (37%, 95% CI 0%–98%), anxiety (36%, 95% CI 27%–46%), depression (31%, 95% CI 22%–40%), and PTSD (18%, 95% CI 9%–29%). The prevalence of neurocognitive impairment was 38% (95% CI 13%–65%). The prevalence of functional impairment was 52% (95% CI 40%–64%): daily activities (54%, 95% CI 41%–66%), mobility (41%, 95% CI 28%–54%), and self-care (21%, 95% CI 13%–31%). The prevalence of neuropsychiatric symptoms in VV-ECMO patients was higher than that in VA-ECMO patients (55% [95% CI 34%–75%] vs 32% [95% CI 23%–41%], p = 0.01), though the prevalence of neurocognitive and functional impairment was not different between the groups. The prevalence of favorable neurologic outcomes was not different at various follow-ups: 3 months (23%, 95% CI 12%–36%), 6 months (25%, 95% CI 16%–35%), and ≥1 year (28%, 95% CI 21%–36%, p = 0.68).

Discussion

A substantial proportion of ECMO patients seemed to experience neuropsychiatric symptoms and neurocognitive and functional impairments at long-term follow-up. Considerable heterogeneity in methodology for gauging these outcomes exists, warranting the need for standardization. Multicenter prospective observational studies are indicated to further investigate risk factors for these outcomes in ECMO-supported patients.

Introduction

Although extracorporeal membrane oxygenation (ECMO) is increasingly used for cardiac and respiratory failure,¹ it is associated with significant complications, most notably acute brain injury (ABI) including ischemic stroke, intracranial hemorrhage, and hypoxic-ischemic brain injury, which are associated with 2–3 times increased mortality.^2-4 Despite the common occurrence of ABI while on ECMO (up to 33% when neuromonitoring is implemented),⁵ its impact on long-term outcomes remains poorly understood.

Prior meta-analyses have studied neurologic outcomes, defined as Cerebral Performance Category (CPC) scores at discharge in ECMO patients.^6-8 In a meta-analysis of 1,057 extracorporeal cardiopulmonary resuscitation (ECPR) patients, the prevalence of favorable neurologic outcomes at 30 days or longer postdischarge was 21%.⁶ In a meta-analysis of 13 studies of venoarterial (VA)–ECMO (non-ECPR) patients, the prevalence of favorable neurologic outcomes at any time was higher at 41%.⁸ However, there is a large knowledge gap in understanding the long-term outcomes at 6 months or greater in ECMO survivors with limited observational cohort studies that are confounded by significant heterogeneity in the (1) methodologies used to evaluate these outcomes and (2) time points for follow-up.

A recent retrospective cohort study found that, compared with non-ECMO intensive care unit (ICU) survivors, ECMO survivors were associated with a moderately increased risk of mental health diagnoses or social problems postdischarge.⁹ Furthermore, a meta-analysis of patients with acute respiratory distress syndrome (ARDS) treated with venovenous (VV)–ECMO had a lower Short-Form 36 (SF-36) score, indicating a worse quality of life, compared with non-ECMO patients with ARDS.¹⁰ However, this study included only a very limited number of articles and patients (3 studies, n = 245), focused on a specific ECMO subgroup, and did not comprehensively quantify the prevalence of neuropsychiatric symptoms and neurocognitive and functional impairment.

In this study, we performed a systematic review and meta-analysis in adult VA-ECMO and VV-ECMO patients. We hypothesized that (1) long-term neuropsychiatric symptoms and neurocognitive and functional impairment occur more often than not; and (2) there is a significant heterogeneity in timing and methods in reporting these outcomes.

Methods

This systematic review was reported in accordance with the 2020 Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines.¹¹ Institutional review board approval was not required because this study represents a secondary analysis of aggregated public datasets and did not directly involve human subject research.

Search Strategy

We searched MEDLINE (PubMed), Embase, Cochrane, Web of Science, and Scopus for subject headings and text related to ECMO and neuropsychiatric, neurocognitive, and functional outcomes from inception to May 3, 2023. Controlled vocabulary, such as Medical Subject Headings and Emtree terms, when appropriate, were used in combination with keywords for the concepts of ECMO and neuropsychiatric, neurocognitive, and functional outcomes. The search was not limited by language. Two investigators (A.K. and J.K.K.) independently reviewed and evaluated each study for eligibility using Covidence. Articles that met the inclusion criteria were obtained and reviewed. References of the included articles were also screened for additional studies (i.e., “citation searching”). The full search strategy can be found in the eAppendix 1 (links.lww.com/WNL/D337).

Inclusion and Exclusion Criteria

Studies were eligible if (1) neuropsychiatric, neurocognitive, and functional outcomes were reported in VA-ECMO or VV-ECMO patients and (2) if these outcomes were evaluated at a postdischarge status. All observational studies and randomized controlled trials that reported usage of ECMO for adults and reported neuropsychiatric, neurocognitive, and functional outcomes were marked eligible. Articles that did not contain information regarding neuropsychiatric, neurocognitive, or functional outcomes in adult ECMO patients postdischarge were excluded. These articles were discussed among authors before being excluded. Abstracts without full texts such as conference proceedings, case reports, editorials, commentaries, systematic reviews and meta-analyses, narrative reviews, animal studies, and studies with the pediatric/neonatal population (younger than 18 years) were excluded.

Study Selection and Data Extraction

Two investigators (A.K. and J.K.K.) individually screened and extracted each article based on titles and abstracts. Disagreements were resolved by a third reviewer (S.K.). Full texts of potential studies were then reviewed. Data were then extracted into an Excel spreadsheet (Microsoft Corp., Redmond, WA). Study characteristics (design, country, and sample size), patients' demographics and characteristics (age, sex, body mass index, and comorbidities), time of follow-up, ECMO cannulation information (ECMO mode, cannulation strategy, indications, and duration), hospital information (ICU and hospital length of stay), and proportions of neuropsychiatric, neurocognitive, and functional outcomes were collected. Extractions were independently conducted and discussed if discrepancies arose.

Definitions and Outcomes

A summary of all methodologies used to assess each outcome is included in eTable 1 (links.lww.com/WNL/D337). Studies assessed neuropsychiatric, neurocognitive, and functional outcomes by questionnaire. Specifically, our primary outcome was neuropsychiatric symptoms (pain/discomfort, anxiety, depression, posttraumatic stress disorder [PTSD], and sleep disturbance). Our secondary outcomes were neurocognitive impairment (memory, attention, and reasoning issues), functional impairment in outcomes such as daily activities, physical activity/mobility, and personal/self-care, typically defined by the European Quality of Life 5 Dimensions Questionnaire (EuroQol-5D), and favorable neurologic outcome, defined as a CPC score ≤2, modified Rankin scale (mRS) ≤3, or Glasgow Outcome Scale (GOS) ≥4. Favorable neurologic outcome proportions included patients who did not survive during their index hospitalization. Long-term follow-up was defined as ≥6 months. In our PROSPERO registration of this study, our primary outcome was favorable neurologic outcome at discharge or follow-up while we secondarily aimed to assess neuropsychiatric outcomes. Although we still analyzed both outcomes, we elected to switch the primary and secondary outcomes because we realized that there was a greater gap in the literature for neuropsychiatric outcomes while there was already some literature assessing favorable neurologic outcomes in ECMO patients. We also noted in PROSPERO that we would compare short (<6-month follow-up) vs long-term (≥6-month follow-up) durations for each outcome. On pooling data in our meta-analysis, the only short-term outcome we were able to assess was favorable neurologic outcome at ≥3-month follow-up. There was insufficient data for appropriately assessing neuropsychiatric symptoms and neurocognitive and functional impairment at this follow-up time point.

Risk of Bias Assessment

Two investigators (A.K. and J.K.K.) independently reviewed and evaluated the risk of bias of each included study. The Newcastle-Ottawa scale was used to assess the quality of case-control and cohort studies.¹² Patient selection, comparability, and assessment of outcome or exposure were the 3 domains of the Newcastle-Ottawa scale. The Cochrane risk-of-bias tool 2 (RoB 2) was used to assess the quality of randomized controlled trials.¹³ Scores ranging from 0 to 9 points were assigned to each included study. High-quality studies had 6 points or more. Any disagreements were resolved by a third investigator (S.K.).

Statistical Analysis

Weighted mean and SD calculations were performed for continuous variables related to patient's demographics and clinical characteristics. The prevalence of neuropsychiatric symptoms and neurocognitive and functional impairment was reported in each study based on the number of patients with the outcome divided by the number of ECMO survivors at each follow-up time-point. The prevalence of favorable neurologic outcome was reported based on the number of patients with favorable neurologic outcome divided by the total number of ECMO patients, including patients who died during the index hospitalization. Separately, we also reported the prevalence of favorable neurologic outcome using only ECMO survivors (excluding patients who died).

Random-effects models with the inverse variance method were used for meta-analyses of the prevalence of each outcome due to substantial heterogeneity between studies because random-effects meta-analyses permit for heterogeneity through the assumption that underlying effects correspond to a normal distribution; the Freeman-Tukey double-arcsine transformation was used for all meta-analyses. For the between-study variance tau², the Sidik-Jonkman estimator was used,^14,15 and the Hartung-Knapp method was used for adjustment of confidence intervals.¹⁶ The Cochrane Q statistic (χ² test) was used to test for the presence of heterogeneity, and the magnitude of the heterogeneity was assessed with the I² statistic, ranging from 0% to 100%.¹⁷

Sensitivity analyses for the prevalence of neuropsychiatric symptoms and neurocognitive and functional impairment were performed with (1) retrospective vs prospective studies and (2) VA-ECMO vs VV-ECMO due to high heterogeneity among included studies. Prespecified variables were determined due to limited availability and being theorized to have an associated effect with the response variable. Meta-regression with prespecified variables (age, sex, ECMO modality—VA-ECMO vs VV-ECMO, and ECMO duration) was performed with the prevalence of any neuropsychiatric symptom as the response variable. All statistical analyses were performed using R Studio (R 4.1.2).

Standard Protocol Approvals, Registrations, and Patient Consents

The study protocol was registered on PROSPERO (CRD42023420565). No institutional review board approval was required for this study because it uses deidentified data.

Data Availability

All data used in this manuscript are deidentified and publicly available.

Results

Our search identified 9,477 studies, yielding 181 full-text articles that were assessed for eligibility. Of these, 59 studies (10 prospective cohort, 3 retrospective case-control, 45 retrospective cohort, and 1 randomized controlled trial) were included (Figure 1).

Figure 1. PRISMA Flowchart for the Creation of Our Study Cohort.

N/A = not available; PRISMA = Preferred Reporting Items for Systematic Reviews and Meta-Analyses.

Baseline Characteristics

Of 3,280 patients analyzed, 2,253 (69%) were male and 1,027 (31%) were female individuals, and the weighted mean of all median ages and median body mass indices were 54 years (SD = 7.2) and 27 kg/m² (SD = 1.8), respectively (Table 1). VV-ECMO patients were younger than VA-ECMO patients (median age 42 years, SD = 5.6 vs median age 56 years, SD = 5.5, p < 0.001). VA-ECMO patients comprised most of the study population (n = 2,819, 86%) when compared with VV-ECMO patients (n = 461, 14%). The most common indication for VA-ECMO was ECPR (n = 1,720, 61%), followed by cardiogenic shock (n = 506, 23%, Table 1). The most common indication for VV-ECMO was acute respiratory failure (n = 325, 71%) while the remaining patients' indications were unknown or classified as other indications. Most of the VA-ECMO patients were peripherally cannulated (n = 2,245, 80%); the remaining patients were centrally cannulated (n = 574, 20%). VV-ECMO patients were cannulated through single-lumen cannula (2 sites, n = 200, 41%) vs double-lumen cannula (1 site, n = 6, 1%), while the remaining 58% of the VV-ECMO patients were not properly identified in their respective studies. VV-ECMO patients were supported longer than VA-ECMO patients (median 12.5, SD = 3.6 days vs median 3.7, SD = 1.6 days, p < 0.001).

Table 1.

Baseline and Clinical Characteristics of Patients Who Underwent Extracorporeal Membrane Oxygenation, Were Discharged, and Screened for Neuropsychiatric Symptoms, Neurocognitive and Functional Impairment, and Favorable Neurological Outcomes

	All ECMO (n = 3,280)a	VA-ECMO (n = 2,819, 86%)	VV-ECMO (n = 461, 14%)
Demographics
Age, y, median (SD)b	53.91 (7.15)	55.76 (5.49)	41.95 (5.61)
Male, n (%)c	2,253/3,280 (69)	1,763/2,683 (66)	231/400 (58)
Female, n (%)c	1,027/3,280 (31)	627/2,683 (23)	126/400 (32)
Body mass index, kg/m2, mean (SD)b	26.95 (1.82)	26.40 (1.47)	29.09 (1.25)
Comorbidities, n (%)
Coronary artery disease	304/1,639 (19)	304/1,639 (19)	0 (0)
Chronic kidney disease	49/986 (5)	49/986 (5)	0 (0)
Diabetes	211/1,426 (15)	207/1,366 (15)	4/60 (7)
Hypertension	296/761 (39)	296/761 (39)	0 (0)
Hyperlipidemia	60/299 (20)	60/299 (20)	0 (0)
Charlson comorbidity score	1.30 (0.46)	1.46 (0.38)	2 (0.15)
ECMO indications, n/N (%)
ECPR (VA-ECMO)	N/A	1,720/2,819 (61)	N/A
Cardiogenic shock (VA-ECMO)	N/A	506/2,819 (23)	N/A
Postcardiotomy shock (VA-ECMO)	N/A	33/2,819 (1)	N/A
Acute respiratory failure (VV-ECMO)	N/A	N/A	325/461 (70)
Unknown/others	N/A	560/2,819 (20)	136/461 (30)
Cannulation strategy, n/N (%)
Peripheral (VA-ECMO)	N/A	2,245/2,819 (80)	N/A
Central (VA-ECMO)	N/A	29/2,819 (1)	N/A
Single-lumen (VV-ECMO)	N/A	N/A	200/461 (41)
Double-lumen (VV-ECMO)	N/A	N/A	6/461 (1)
Unknown	N/A	545/2,819 (19)	278/461 (58)
ECMO duration, d, mean (SD)b	5.4 (4.0)	3.7 (1.6)	12.5 (3.6)
ICU LOS, d, mean (SD)b	11.5 (12.9)	5.9 (4.8)	27.8 (14.9)
Hospital length of stay, d, mean (SD)b	27.7 (20.2)	24.7 (20.7)	30.8 (21.1)

Abbreviations: ARDS = acute respiratory distress syndrome; BMI = body mass index; ECMO = extracorporeal membrane oxygenation; ECPR = extracorporeal cardiopulmonary resuscitation; ICU = intensive care unit; LOS = length of stay; N/A = not available; VA = venoarterial; VV = venovenous.

^aThere are some discrepancies between VA-ECMO and VV-ECMO vs total ECMO numbers because certain patients were not able to be distinguished as VA-ECMO vs VV-ECMO at certain time points of follow-up. The following continuous variables had missing data: ECMO duration: 2,142/3,280 patients; ICU LOS: 1,454/3,280 patients; Hospital LOS: 1,081/3,280 patients; And Charlson comorbidity score: 562/3,280 patients.

^bAge, BMI, ECMO duration, ICU length of stay, and hospital length of stay are reported as the weighted mean with standard deviation of mean.

^cSex: male and female sex did not always sum to all patients because the number of male or female individuals was not always reported for the subgroups of VA-ECMO and VV-ECMO, though they may have been reported for all (total) ECMO patients.

Neuropsychiatric, Neurocognitive, and Functional Outcomes (Raw Numbers)

The weighted mean of the median follow-up times for screening of all neuropsychiatric, neurocognitive, and functional outcomes was 15.8 months (SD = 17.3, Table 2). The follow-up time was longer in VV-ECMO patients than in VA-ECMO patients (median 27.1, SD = 35 months vs 13.4, SD = 14.9 months, p < 0.001). The raw numbers for the prevalence of neuropsychiatric symptoms, neurocognitive and functional impairment, and favorable neurologic outcome are listed in Table 2.

Table 2.

Neuropsychiatric Symptoms, Neurocognitive and Functional Impairment, and Favorable Neurologic Outcomes of Patients Who Underwent Extracorporeal Membrane Oxygenation and Were Discharged

	Total (n = 3,280)a	VA-ECMO (n = 2,819)	VV-ECMO (n = 461)
Length of time since follow-up, mo, mean (SD)	15.83 (19.24)	13.4 (14.9)	27.1 (35)
Neuropsychiatric symptoms, n/N (%)
Pain/discomfort	171/346 (49)	29/52 (56)	102/219 (47)
Anxiety	203/574 (35)	78/239 (33)	110/256 (43)
Depression	176/574 (31)	65/239 (27)	93/256 (36)
Posttraumatic stress disorder	48/265 (18)	36/202 (18)	9/35 (26)
Sleep disturbance	27/83 (33)	27/83 (33)	N/A
Neurocognitive outcomes, n/N (%)
Memory, attention, or reasoning impairment	61/173 (35)	16/91 (18)	31/54 (57)
Functional outcomes, n/N (%)
Daily activity impairment	134/253 (53)	31/69 (45)	71/126 (56)
Mobility/physical activity impairment	112/261 (43)	25/52 (48)	60/134 (45)
Personal/self-care impairment	48/216 (22)	14/52 (27)	24/113 (21)
Favorable neurologic outcome
CPC ≤2, mRS ≤3, GOS ≤4 at 3-mo follow-up	215/931 (23)	215/931 (23)	N/A
CPC ≤2, mRS ≤3, GOS ≤4 at 6-mo follow-up	203/869 (23)	172/749 (23)	21/55 (38)
CPC ≤2, mRS ≤3, GOS ≤4 at 1 y or longer follow-up	228/883 (26)	221/870 (25)	7/13 (54)

Abbreviations: CPC = Cerebral Performance Category; GOS = Glasgow Outcome Scale; mRS = modified Rankin scale; VA = venoarterial; VV = venovenous.

^aThere are some discrepancies between VA-ECMO and VV-ECMO vs total ECMO numbers because certain patients were not able to be distinguished as VA-ECMO vs VV-ECMO at certain time points of follow-up. The sample size for calculations pertaining to favorable neurologic outcomes accounts for all ECMO patients at baseline of the study, including those who died later. The sample size for calculations pertaining to neuropsychiatric, neurocognitive, and functional outcomes includes only ECMO patients who survived and were screened at the time point of interest (i.e., 6 months or longer).

Risk of Bias Assessment

The overall median risk of bias was approximately 8 (SD = 1), indicating no elevated risk of bias across all included observational studies (eAppendix 2, links.lww.com/WNL/D337). The RoB 2 tool showed a low overall risk of bias for the 1 randomized controlled trial¹⁸ included in our analysis, suggesting this study was of high quality (eAppendix 3).

Assessment Methods for Each Outcome

Studies used a wide variety of methods to assess outcomes, including different types of questionnaires. The SF-36 was the most frequently used tool (10/22 studies, 38%, eTable 1, links.lww.com/WNL/D337) followed by the Impact of Event Scale (9/22 studies, 35%) to assess neuropsychiatric symptoms. Other forms such as Hospital Anxiety and Depression scale (8/26, 31%), EuroQol-5D (7/26, 27%), World Health Organization Quality of Life questionnaire (1/26, 4%), Posttraumatic Stress Disorder Checklist 6 instrument (1/26, 4%), and unspecified questionnaire (1/26, 4%) were also used. Many studies used more than 1 form of assessment to assess outcomes, indicating heterogeneity in assessment methodology.

Neuropsychiatric Symptoms

Of 59 final studies, 26 studies (n = 790) reported proportions of neuropsychiatric symptoms.^19-44 The prevalence of any neuropsychiatric symptom among all ECMO patients at long-term follow-up was 41% (95% CI 33%–49%, Figure 2). The most prevalent neuropsychiatric symptom was pain/discomfort (52%, 95% CI 42%–63%), followed by sleep disturbance (37%, 95% CI 0%–98%), anxiety (36%, 95% CI 27%–46%), depression (31%, 95% CI 22%–40%), and PTSD (18%, 95% CI 10%–28%, eFigures 1–5, links.lww.com/WNL/D337).

Figure 2. Meta-Analysis of the Prevalence of Any Neuropsychiatric Symptom (Pain/Discomfort, Anxiety, Depression, Sleep Disturbance, and PTSD) in Extracorporeal Membrane Oxygenation Patients at 6 Months or Longer Follow-Up.

PTSD = posttraumatic stress disorder.

Neurocognitive Impairment

Seven studies (n = 194) reported proportions of neurocognitive impairment.^{19,21,27,29,45-47} The prevalence of neurocognitive impairment (memory, attention, or reasoning issues) at long-term follow-up was 38% (95% CI 13%–65%, Figure 3).

Figure 3. Meta-Analysis of the Prevalence of Any Neurocognitive Impairment (Memory, Attention, or Reasoning) in Extracorporeal Membrane Oxygenation Patients at 6 Months or Longer Follow-Up.

Prevalence of Long-Term Functional Impairment

Seven studies (n = 274) reported proportions of functional impairment.^{20,22,25,26,28,41,48} The prevalence of functional impairment was 52% (95% CI 40%–64%, Figure 4) at long-term follow-up. The most prevalent functional impairment was daily activities (54%, 95% CI 41%–66%), followed by physical activity/mobility (41%, 95% CI 28%–54%), and personal/self-care (21%, 95% CI 13%–31%, eFigures 6–8, links.lww.com/WNL/D337).

Figure 4. Meta-Analysis of the Prevalence of Any Impaired Functional Outcome (Daily Activities, Physical Activity/Mobility, and Personal/Self-Care) in Extracorporeal Membrane Oxygenation Patients at 6 Months or Longer Follow-Up.

Favorable Neurologic Outcome

Thirty studies reported proportions of favorable neurologic outcome.^{15,18,21,29,49,50,e1-e25} The prevalence of favorable neurologic outcome (including nonsurvivors) was not different at each follow-up time point: 3 months (23%, 95% CI 12%–36%, Figure 5A), 6 months (25%, 95% CI 16%–35%, Figure 5B), and ≥1 year (28%, 95% CI 21%–36%, p = 0.68, Figure 5C). The prevalence of favorable neurologic outcome (survivors only) was also not different at each follow-up: 3 months (59%, 95% CI 26%–88%, eFigure 9A, links.lww.com/WNL/D337), 6 months (84%, 95% CI 56%–100%, eFigure 9B), and ≥1 year (90%, 95% CI 75%–99%, p = 0.10, eFigure 9C).

Figure 5. Meta-Analysis of the Prevalence of Favorable Neurologic Outcome (CPC Score ≤2, mRS ≤3, or GOS ≥4) in Extracorporeal Membrane Oxygenation Patients at (A) 3 Months, (B) 6 Months, and (C) 12 Months or Longer Follow-Up.

CPC = Cerebral Performance Category; GOS = Glasgow Outcome Scale; mRS = modified Rankin scale.

Sensitivity Analysis

In a sensitivity analysis, the prevalence of neuropsychiatric symptoms in retrospective studies was comparable with that of prospective studies (39% [95% CI 30%–48%] vs 46% [95% CI 22%–71%], p = 0.52, eFigure 10, links.lww.com/WNL/D337). The prevalence of neuropsychiatric symptoms in VV-ECMO patients was higher than that in VA-ECMO patients (55% [95% CI 34%–75%] vs 32% [95% CI 23%–41%], p = 0.01, eFigure 11).

The prevalence of neurocognitive impairment in retrospective studies was comparable with that in prospective studies (32% [95% CI 1%–76%] vs 44% [95% CI 0%–100%], p = 0.65, eFigure 12, links.lww.com/WNL/D337). The prevalence of neurocognitive impairment was similar in VV-ECMO vs VA-ECMO patients (50% [95% CI 0%–100%] vs 24% [95% CI 0%–95%], p = 0.30, eFigure 13).

Last, the prevalence of functional impairment in retrospective studies was comparable with that in prospective studies (50% [95% CI 41%–59%] vs 55% [95% CI 18%–90%], p = 0.67, eFigure 14, links.lww.com/WNL/D337). The prevalence of functional impairment was similar in VV-ECMO vs VA-ECMO patients (55% [95% CI 28%–80%] vs 45% [95% CI 28%–63%], p = 0.33, eFigure 15).

Risk Factors

In meta-regression, ECMO configuration (VV vs VA-ECMO) was not independently associated with long-term neuropsychiatric symptoms when adjusted with age, sex, and ECMO duration (eTable 2, links.lww.com/WNL/D337). However, this analysis was limited by sample size, missing data, and significant bias/heterogeneity.

Discussion

Herein, we performed a large and comprehensive systematic review and meta-analysis that evaluated the long-term neuropsychiatric, neurocognitive, and functional outcomes in patients receiving ECMO. Overall, in studies reporting on each of these outcomes, a significant portion of patients experienced long-term neuropsychiatric symptoms (41%, Figure 2) and neurocognitive (38%, Figure 3) and functional impairment (52%, Figure 4). At 1-year follow-up or longer, only 28% (Figure 5C) of ECMO patients (nonsurvivors included) had a favorable neurologic outcome. However, when counting only survivors, 90% had a favorable neurologic outcome (eFigure 9, links.lww.com/WNL/D337), suggesting ECMO patients who are successfully discharged can achieve good neurologic recovery over time.

Neuropsychiatric, neurocognitive, and functional outcomes hold paramount importance across various cardiac surgery populations including coronary artery bypass graft surgery, aortic surgery, valve procedures, and left ventricular assist device implantation.^8,e26 For ECMO patients, these outcomes are especially critical because they are associated with long-term mortality, quality of life, and economic impact.^e27,e28 Accordingly, they are recognized as core outcome measures in ECMO research.^e29 Our meta-analysis reveals a concerning prevalence of neuropsychiatric, neurocognitive, and functional impairments in ECMO patients; yet no research has previously been undertaken to systematically evaluate these important outcomes. Although prior literature on these outcomes have suggested ECMO patients are at risk of neuropsychiatric symptoms and neurocognitive and functional impairment,^10,e30-e32 it is difficult to adequately compare our results with those of previous studies because prevalence calculations for neuropsychiatric, neurocognitive, and functional outcomes were not performed.

There are meta-analyses for the prevalence of these neurologic outcomes in other ICU populations. Studies indicate that approximately one-third of ICU survivors experienced depression disorders, anxiety, PTSD and disabilities in functional outcomes.^e33,e34 In our analysis, ECMO patients were at higher risk of neuropsychiatric symptoms and neurocognitive impairment: approximately 40% of ECMO patients manifested these outcomes, which can further increase the risk of developing new psychiatric disorders. They were also at higher risk of functional impairment (52%), compared with general ICU survivors (32%).^e33 A retrospective study using Canadian health administrative databases demonstrated that ECMO treatment was significantly associated with increased risk of new mental health diagnoses or social problems,⁹ although the methodology used in that study, relying on the International Statistical Classification of Diseases, Tenth Revision codes for diagnoses, carries inherent limitations, such as a lack of granular data and potential coding errors. Given these observations, our findings highlight the critical need for routine mental health evaluations and robust support mechanisms as a part of comprehensive long-term management in ECMO patients. Still, it should be duly noted that ECMO itself may not actually increase the prevalence of neuropsychiatric symptoms and neurocognitive impairment; rather, the severity of critical illness that is treated with ECMO may be the underlying reason. Accordingly, multicenter prospective observational studies in ECMO are necessary to investigate the causal inference and association between ECMO support and neurologic sequelae in ECMO patients.

The precise mechanisms driving neuropsychiatric symptoms and neurocognitive and functional impairment in ECMO patients remain poorly understood. Proposed theories in ICU patients include accelerated neurodegeneration in predisposed individuals and newly acquired ABIs secondary to hypoxemia, shock, metabolic imbalances, and multiorgan failures.^e35 Subclinical lesions such as cerebral microhemorrhage and small vessel disease may also play a role in developing aforementioned symptoms specifically in ECMO patients because they are highly prevalent with frequencies up to 60% in brain autopsy study and 50% on MRI.^e36,e37 Still, these are detection frequencies because they do not confirm that they occurred during present illness. Even if cerebral small vessel diseases do not cause clinically significant ABIs, they may induce microstructural damage of the subcortical brain structures, chronic low-grade inflammation, and amyloid accumulation, ultimately causing psychiatric disorders and cognitive decline.^e38-e40

We observed a large heterogeneity in the measurement of neuropsychiatric symptoms and neurocognitive impairment across included studies (I² = 81% and I² = 87%, respectively). We theorized this high heterogeneity is largely because studies used diverse assessment tools and scales to evaluate such outcomes. Notably, we observed that almost half of all studies used more than 1 assessment form. Such variation in evaluation methods likely contributes to the observed heterogeneity in measuring neuropsychiatric and neurocognitive outcomes. By contrast, heterogeneity in methods assessing functional impairment was much lower (I² = 21%). This may be attributed to most studies largely using the EuroQol-5D (5/7, 72%) to measure this outcome. The use of such a standardized assessment could enhance consistency in measuring neuropsychiatric, neurocognitive, and functional outcomes. We recommend future studies to consider standardizing the time point of follow-up to be at 3, 6, and 12 months to evaluate neuropsychiatric symptoms, neurocognitive impairment, and functional outcomes. However, the standardization of instruments used to assess each domain should be determined by societal guidelines and the experts' consensus.

Furthermore, most follow-up assessments were not conducted at an in-person clinic setting. This may yield unreliable evaluations of patients' neurocognitive and functional outcomes because self-reported questionnaires were used. Since the coronavirus disease 2019 pandemic, telemedicine has been validated as a reliable method for follow-up evaluations, and patients have a higher likelihood of follow-up visits compared with in-person visits.^e41 The integration of remote assessments and telemedicine along with questionnaires is crucial for a comprehensive follow-up evaluation especially in ECMO patients who may have difficulties with attending in-person clinic visits given the degree of impairment in daily activities. Of importance, included studies conducted follow-up evaluations in different periods. Development of routine standardized assessments tools used across studies will help to improve the observed heterogeneity.

In subgroup analysis, neuropsychiatric symptoms were more prevalent in VV-ECMO (55%) patients, compared with VA-ECMO patients including ECPR patients (31%). This finding may be explained by the longer cannulation time and length of stay in the ICU for VV-ECMO patients compared with VA-ECMO patients because the contribution of acute psychiatric stress is duration dependent.^e42 Specifically, longer lengths of stay in the ICU are associated with somatic symptoms including pain, fatigue, and weakness, which are significant risk factors of neuropsychiatric symptoms such as depression.^e43 In addition, the longer a patient is supported on ECMO, the more likely they are to be using sedatives and bedridden.^e44 High sedative use is reported as a risk factor of depression and increased bedridden time can lead to muscle loss and worse somatic symptoms, increasing risk of depression.^e45 Existing literature suggests a higher frequency of ABI in VA-ECMO patients than in VV-ECMO patients, although 1 study showed that VA-ECMO and VV-ECMO patients have a similar rate of favorable neurologic outcome based on CPC score.⁸ Furthermore, though limited by sample size, the prevalence of favorable neurologic outcome seemed to improve in VV-ECMO patients when compared with that in VA-ECMO patients from 6-month to 1-year follow-up in our study. This finding may be attributed to VA-ECMO patients having higher complication (including ABI) and readmission rates when compared with that of VV-ECMO patients.^1,2,e46 Nevertheless, it is important to distinguish between these ABIs and longer-term neuropsychiatric and neurocognitive outcomes. More research is indicated to elucidate the mechanisms leading to neuropsychiatric symptoms and neurocognitive and functional impairments based on ECMO modalities. Such understanding could guide the development of tailored managements to mitigate these impairments in ECMO patients.

The main limitation was the heterogeneity in instruments used for neuropsychiatric symptoms and neurocognitive impairment assessment, which makes determining the exact prevalence of each outcome challenging. In addition, the broad range of follow-up endpoints, as noted by the large standard deviation of the median follow-up time, and overall diverse outcomes is also apparent. Most studies assessed were retrospective in nature, implying the possibility of recall bias confounding results and warranting the need for prospective studies. Patient characteristics such as age, ECMO-specific information such as cannulation strategy, preexisting comorbidities, severity of illness, and history of psychiatric illnesses could affect neuropsychiatric and neurocognitive outcomes, and these were not uniformly controlled across studies. Finally, it should be noted that most of our cohort was VA-ECMO, specifically ECPR patients, which may not be the most common indication for ECMO but is the most studied indication when investigating neurologic outcomes in the literature.

A substantial proportion of ECMO patients experience neuropsychiatric symptoms and neurocognitive and functional impairments at 6 months or longer postdischarge. Large heterogeneity in assessing these domains exists, highlighting the need for standardized effective tools assessing these critical outcomes. Multicenter prospective observational studies are needed to investigate the effect of ECMO on these outcomes and to refine managements to optimize their long-term recovery and quality of life.

Glossary

ABI

acute brain injury

ARDS

acute respiratory distress syndrome

ECMO

extracorporeal membrane oxygenation

ECPR

extracorporeal cardiopulmonary resuscitation

EuroQol-5D

European Quality of Life 5 Dimensions Questionnaire

CPC

Cerebral Performance Category

GOS

Glasgow Outcome Scale

ICU

intensive care unit

mRS

modified Rankin scale

PTSD

posttraumatic stress disorder

RoB 2

risk-of-bias tool 2

SF-36

Short-Form 36

VA

venoarterial

VV

venovenous

Appendix 1. Authors

Name	Location	Contribution
Andrew Kalra, BS	Division of Cardiac Surgery, Department of Surgery, Johns Hopkins Hospital, Baltimore, MD; Sidney Kimmel Medical College, Thomas Jefferson University, Philadelphia, PA	Drafting/revision of the article for content, including medical writing for content; major role in the acquisition of data; study concept or design; and analysis or interpretation of data
Jin Kook Kang, MD	Division of Cardiac Surgery, Department of Surgery, Johns Hopkins Hospital, Baltimore, MD	Drafting/revision of the article for content, including medical writing for content; major role in the acquisition of data; and study concept or design
Shivalika Khanduja, MBBS	Division of Cardiac Surgery, Department of Surgery, Johns Hopkins Hospital, Baltimore, MD	Major role in the acquisition of data
Arjun K. Menta, BS	Division of Cardiac Surgery, Department of Surgery, Johns Hopkins Hospital, Baltimore, MD	Major role in the acquisition of data
Syed A. Ahmad, BS	Division of Neurosciences Critical Care, Department of Neurology, Neurosurgery, Anesthesiology and Critical Care Medicine, Johns Hopkins Hospital, Baltimore, MD	Major role in the acquisition of data
Olivia Liu, BS	Division of Neurosciences Critical Care, Department of Neurology, Neurosurgery, Anesthesiology and Critical Care Medicine, Johns Hopkins Hospital, Baltimore, MD	Major role in the acquisition of data
Emily Rodriguez, BS	Division of Cardiac Surgery, Department of Surgery, Johns Hopkins Hospital, Baltimore, MD	Major role in the acquisition of data
Marcus Spann, BA	Informationist Services, Johns Hopkins School of Medicine, Baltimore, MD	Major role in the acquisition of data; study concept or design
Adrian V. Hernandez, MD, PhD	Department of Pharmacy Practice, University of Connecticut School of Pharmacy, Storrs; Unidad de Revisiones Sistemáticas y Meta-análisis (URSIGET), Vicerrectorado de Investigación, Universidad San Ignacio de Loyola (USIL), Lima, Peru	Drafting/revision of the article for content, including medical writing for content; analysis or interpretation of data
Daniel Brodie, MD	Division of Pulmonary and Critical Care Medicine, Department of Medicine, Johns Hopkins University School of Medicine, Baltimore, MD	Drafting/revision of the article for content, including medical writing for content
Glenn J.R. Whitman, MD	Division of Cardiac Surgery, Department of Surgery, Johns Hopkins Hospital, Baltimore, MD	Drafting/revision of the article for content, including medical writing for content; study concept or design
Sung-Min Cho, DO, MHS	Division of Neurosciences Critical Care, Department of Neurology, Neurosurgery, Anesthesiology and Critical Care Medicine, Johns Hopkins Hospital, Baltimore, MD	Drafting/revision of the article for content, including medical writing for content; study concept or design; and analysis or interpretation of data

Appendix 2. Coinvestigators

Coinvestigators are listed at links.lww.com/WNL/D336

Study Funding

Supported by the NHLBI (1K23HL157610).

Disclosure

A. Kalra, J.K. Kang, S. Khanduja, A.K. Menta, S.A. Ahmed, O. Liu, E. Rodriguez, M. Spann, and A.V. Hernandez report no disclosures relevant to this manuscript. D. Brodie receives research support from and consults for LivaNova, has been on the medical advisory boards for Abiomed, Xenios, Medtronic, Inspira, and Cellenkos, is the President elect of the Extracorporeal Life Support Organization (ELSO) and the Chair of the Executive Committee of the International ECMO Network (ECMONet), and he writes for UpToDate. G.J.R. Whitman reports no disclosures relevant to this manuscript. S.-M. Cho is supported by the NHLBI (1K23HL157610). Go to Neurology.org/N for full disclosures.