Association of Neurocritical Care Services With Mortality and Functional Outcomes for Adults With Brain Injury

This systematic review and meta-analysis assesses data for adults with acute brain injury who were admitted to intensive care units to evaluate the association of neurocritical care with patient-centered outcomes.

Key Points

Question

What is the association of subspecialized neurocritical care with outcomes for adults with brain injury admitted to intensive care units?

Findings

In this systematic review and meta-analysis, there was a statistically significant 17% relative risk reduction in mortality and unfavorable functional outcomes for patients receiving neurocritical care compared with general care in critical care settings.

Meaning

Neurocritical care has the potential to improve survival in adults with acute brain injury.

Abstract

Importance

Neurocritical care (NCC) aims to improve the outcomes of critically ill patients with brain injury, although the benefits of such subspecialized care are yet to be determined.

Objective

To evaluate the association of NCC with patient-centered outcomes in adults with acute brain injury who were admitted to intensive care units (ICUs). The protocol was preregistered on PROSPERO (CRD42020177190).

Data Sources

Three electronic databases were searched (Ovid MEDLINE, Embase, Cochrane Central Register of Controlled Trials) from inception through December 15, 2021, and by citation chaining.

Study Selection

Studies were included for interventions of neurocritical care units (NCCUs), neurointensivists, or NCC consulting services compared with general care in populations of neurologically ill adults or adults with acute brain injury in ICUs.

Data Extraction and Synthesis

Data extraction was performed in keeping with PRISMA guidelines and risk of bias assessed through the ROBINS-I Cochrane tool by 2 independent reviewers. Data were pooled using a random-effects model.

Main Outcomes and Measures

The primary outcome was all-cause mortality at longest follow-up until 6 months. Secondary outcomes were ICU length of stay (LOS), hospital LOS, and functional outcomes. Data were measured as risk ratio (RR) if dichotomous or standardized mean difference if continuous. Subgroup analyses were performed for disease and models of NCC delivery.

Results

After 5659 nonduplicated published records were screened, 26 nonrandomized observational studies fulfilled eligibility criteria. A meta-analysis of mortality outcomes for 55 792 patients demonstrated a 17% relative risk reduction (RR, 0.83; 95% CI, 0.75-0.92; P = .001) in those receiving subspecialized care (n = 27 061) compared with general care (n = 27 694). Subgroup analyses did not identify subgroup differences. Eight studies including 4667 patients demonstrated a 17% relative risk reduction (RR, 0.83; 95% CI, 0.70-0.97; P = .03) for an unfavorable functional outcome with subspecialized care compared with general care. There were no differences in LOS outcomes. Heterogeneity was substantial in all analyses.

Conclusions and Relevance

Subspecialized NCC is associated with improved survival and functional outcomes for critically ill adults with brain injury. However, confidence in the evidence is limited by substantial heterogeneity. Further investigations are necessary to determine the specific aspects of NCC that contribute to these improved outcomes and its cost-effectiveness.

Introduction

Neurocritical care (NCC) is an emerging subspecialty that is dedicated to the care of critically ill patients with neurological disease, including devastating brain injuries. This has led to the development of subspecialty training pathways and NCC centers in some countries. However, the introduction of specific NCC services, particularly specialized intensive care units (ICUs), is yet to be adopted globally despite prior work suggesting this may improve outcomes for patients with brain injuries.^1,2,3

Cohorting patients under a specialized multidisciplinary team is not without precedent.^4,5 The associations of neurocritical care units (NCCUs) and services with outcomes have also been previously explored in systematic literature reviews by Kramer and Zygun² in 2011 and 2014³ and updated in an editorial by Kramer and Couillard¹ in 2020. However, several questions remain about NCC as a concept, particularly as the dominant mechanisms underpinning improved patient outcomes are not clear. This may reflect the clinical expertise of the medical and nursing team under the leadership of a neurointensivist, as well as a specialty multidisciplinary allied health team enhancing patient recovery. The introduction of an NCCU or service also typically involves the addition of advanced multimodal neuromonitoring and/or institution of standardized NCC protocols.

Furthermore, acute brain injuries (ABIs) are markedly diverse in their etiology and management, and the outcomes of specific neurological conditions require review. For example, subarachnoid hemorrhage (SAH) and traumatic brain injury (TBI) manifest substantial mortality and morbidity with limited therapeutic options. Determining whether NCC improves the outcomes of these patients is therefore paramount. We performed a systematic review and meta-analysis to explore the associations of NCC delivery with outcomes in critically ill adults with ABIs.

Methods

Study Design

This systematic review and meta-analysis was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guideline⁶ (eFigure 1 in the Supplement). The protocol was preregistered in the International Prospective Register of Systematic Reviews (PROSPERO) database (CRD42020177190) and published as a peer-reviewed article in March 2021.⁷

Eligibility Criteria

We included all peer-reviewed, English-language observational and interventional studies published as full articles. Studies analyzing subgroups derived from previously published data were excluded.

The study population included adult patients (aged ≥18 years) who were admitted to an ICU with an ABI. Studies that included a minority of patients younger than 18 years or that defined their population as aged 16 years and older were also included. Acute brain injuries included acute ischemic stroke (AIS), intracerebral hemorrhage (ICH), SAH, and TBI. Studies that examined all patients with neurological illness not limited by diagnosis were included. Studies that limited the admitting diagnosis to specific neurological conditions not involving ABI (eg, spinal cord injury or myasthenia gravis) or clinical syndromes (eg, status epilepticus) were excluded.

Interventions

Interventions of interest were different models of NCC delivery, including the introduction of an NCCU, neurointensivist, or neurointensivist-led multidisciplinary team. An NCCU was defined as an ICU dedicated to the care of patients with neurological illness who were exclusively managed by this specialized unit excepting logistical circumstances. A neurointensivist was a physician accredited in NCC who was ideally board-certified or nationally recognized.

These interventions were compared with care provided by general ICUs or another non-NCCU. We excluded studies examining the introduction of standardized management protocols for NCC conditions.

Outcomes

The primary outcome of interest was all-cause mortality at the longest follow-up within 6 months. Secondary outcomes included ICU length of stay (LOS), hospital LOS, and functional outcomes at the longest follow-up within 6 months, at or after hospital discharge. To assess LOS, we extracted the means and SDs for each participant group. If medians and IQR were reported, these were converted to means and SDs through the quantile estimation method.⁸ Functional outcomes were categorized dichotomously as favorable or unfavorable. Unfavorable outcomes were defined as a Glasgow Outcome Scale score less than 4, an Extended Glasgow Outcome Scale score less than 5, or a modified Rankin Scale score greater than 3. If necessary, we approximated dichotomous data from reported percentages or figures in the included studies.

Data Sources

We searched the following electronic databases from inception to December 15, 2021: Ovid MEDLINE and Epub Ahead of Print, In-Process and Other Non-Indexed Citations, Daily, and Versions; Embase Classic and Embase; and Cochrane Central Register of Controlled Trials. Previous reviews and included studies were screened for relevant citations. The search strategies are presented in eTable 1 in the Supplement.

Data Collection and Analysis

Studies identified through the search strategy and citation chaining were uploaded to Covidence and duplicates removed.⁹ Two authors (X.P. and J.R.) independently screened abstracts and categorized as “exclude,” “include,” and “maybe,” with the latter 2 progressing to the next stage of screening. Full texts of the remaining studies were uploaded to Covidence for the second stage of screening by the 2 reviewers. Discrepancies were resolved through discussion with a third reviewer (A.A.U.). The final included studies were collated for data extraction and analysis.

Two reviewers (X.P. and J.R.) extracted data independently from the eligible studies using a data extraction form. We abstracted the following information: study characteristics, participant characteristics, sample size, interventions, comparators, and outcomes.

Subgroup Analysis

We conducted a prespecified subgroup analysis for the primary outcome, stratified by the following disease processes: SAH, ICH, AIS, and TBI. We conducted a post hoc subgroup analysis by intervention type, comparing models of NCC delivery at a unit or personnel level, ie, the introduction of a neurointensivist or neurointensivist-led team.

Risk-of-Bias Assessment

The Risk of Bias in Nonrandomized Studies of Intervention (ROBINS-I) tool was used for quality assessment of each study, which consists of 7 domains.¹⁰ Two reviewers (X.P. and J.R.) independently assessed the risk of bias of studies for the primary outcome, classifying each domain and overall risk of bias as low, moderate, serious, or critical risk, with discrepancies resolved by discussion between the 2 authors.

Statistical Analysis

We anticipated substantial between-study heterogeneity and as such used a random-effects model to pool effect sizes. For dichotomous data, ie, mortality and functional outcomes, we calculated a pooled estimate of risk ratio (RR) with a 95% CI using a random-effects model according to the Mantel-Haenszel method.¹¹ For continuous data, ie, LOS outcomes, we calculated a pooled estimate of standardized mean difference (SMD) with 95% CI using a random-effects model according to the Hedges g formula.¹² We used SMD because LOS was calculated across all publications with the same unit, ie, days. The CIs around the pooled effects were calculated using Knapp-Hartung adjustments.¹³

The Paule and Mandel method was used to estimate the between-study variance (τ²), and the confidence intervals were calculated using the Q profile method.^14,15 The homogeneity assumption was measured by the I², which described the percentage of total variation across the studies due to heterogeneity rather than chance. I² was calculated from the basic results obtained from a typical meta-analysis as I² = 100% × (Q − df) / Q, where Q is the Cochran heterogeneity statistic. A value of 0% indicates no observed heterogeneity, and larger values indicate increasing heterogeneity. Heterogeneity was further evaluated using prediction intervals.¹⁶ Subgroup analyses were performed with a fixed-effects model, that is, assuming an independent τ². Publication bias was evaluated by visual inspections of funnel plots and the Egger test.¹⁷

All analyses were conducted in R version 4.1.2 (R Foundation for Statistical Computing) and RStudio version 2021.9.1.372.^18,19 A P value less than .05 was considered significant. We assessed confidence in the evidence for each assessed outcome using the Grading of Recommendations Assessment, Development and Evaluation framework.²⁰

Sensitivity Analyses

We performed the following prespecified sensitivity analyses concerning the primary outcome to evaluate the robustness of our results.

1. Analyzing studies with low or moderate risk of bias.

2. Analyzing studies that did not restrict their study population by disease process.

3. Analyzing studies with the longest reported mortality beyond ICU mortality.

4. Analyzing studies by geographical locations, ie, North America, Europe, and rest of the world.

We performed a post hoc sensitivity analysis by analyzing studies by last year of recruitment and identifying and excluding influential studies because of the degree of between-study heterogeneity.

Results

After 5659 nonduplicated published records were screened, 26 studies fulfilled the inclusion criteria (eFigure 1 in the Supplement).^{21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46} Eight studies were excluded following full-text assessment for eligibility (eTable 2 in the Supplement). Eight studies were excluded that were previously included in past systematic reviews (eTable 3 in the Supplement).

All studies were nonrandomized observational cohort studies, and all but 3 were retrospective. The last year of recruitment ranged from 1988 to 2018. Total sample size ranged from 74 to 13 543 participants. The intervention for 15 studies was the introduction of an NCCU, with 6 comparing the same unit before and after the unit change. Reported components of an NCCU after the unit change included the introduction of a neurointensivist; trained medical, nursing, and allied health staff; advanced neuromonitoring; and/or the introduction of protocols (eTable 4 in the Supplement). Eleven studies examined the introduction of a neurointensivist or neurointensivist-led team (eTable 5 in the Supplement). Further detailed study characteristics are summarized in the Table. Reported outcomes of each study are summarized in eTable 6 in the Supplement. Only 4 studies compared withdrawal of care as a cause of mortality, with 3 reporting no statistical differences^42,43,45 and 1 reporting increased withdrawal-of-care deaths in the non-NCCU cohort.³⁸

Table. Characteristics of Included Studies From the Systematic Review.

Source	Study years	Design	Location	Population group	No. of participants	Intervention	Control	Measured outcomes
								Mortality	ICU LOS	Functional outcomes	Hospital LOS
Burns et al,21 2013	2007-2010	Cohort, retrospective	US	ICH	74	NI	Before NI	In-hospital	Y	In-hospital	Y
Damian et al,22 2013	1996-2009	Cohort, retrospective	UK	ICH	8505	NCCU	General ICUa	In-hospital	Y	NA	Y
Diringer and Edwards,23 2001	1996-1999	Cohort, retrospective	US	ICH	1038	NCCU	General ICU	In-hospital	Y	NA	Y
Egawa et al,24 2016	2001-2014	Cohort, retrospective	Japan	SAH	234	NI	Before NI	In-hospital	Y	Hospital discharge	Y
Grieve et al,25 2016	2009-2011	Cohort, prospective	UK	TBI	2665	NCCU	General ICU	6 mo	Y	6 mo	Y
Jeong et al,26 2019	2010-2014	Cohort, retrospective	South Korea	General	2487	NCCU	General ICU (before)	ICU	Y	NA	NA
Josephson et al,27 2010	2003-2007	Cohort, prospective	US	SAH	512	NI	Before NI	In-hospital	Y	NA	NA
Kim SH et al,28 2020	2015-2018	Cohort, retrospective	South Korea	General	1557	NI	Before NI	In-hospital	NA	NA	NA
Kim TJ et al,29 2020	2007-2014	Cohort, retrospective	South Korea	AIS	1405	NI	No NI	3 mo	NA	NA	NA
Knopf et al,30 2012	2003-2011	Cohort, retrospective	US	AIS	1242	NCCUb	General ICU (before)	3 mo	Y	NA	Y
				ICH	491
				SAH	363
Ko et al,31 2019	2010-2016	Cohort, retrospective	South Korea	General	2199	NI	Before NI	6 mo	Y	NA	Y
Lombardo et al,32 2017	2008-2009	Cohort, retrospective	US	TBI	1290	NCCU	General ICU	In-hospital	NA	NA	Y
Lott et al,33 2009	2002-2005	Cohort, retrospective	US	AIS	3800	NCCU	General ICU	In-hospital	NA	NA	NA
				ICH	9743
McCredie et al,34 2018	2011-2013	Cohort, retrospective	Canada	TBI	9773	NCCU	General ICU	In-hospital	Y	NA	Y
Mielke et al,35 2019	1990-2013	Cohort, retrospective	Germany	SAH	537	NCCU	General ICU	6 mo	NA	6 mo	NA
Mirski et al,36 2001	1995-1997	Cohort, retrospective	US	ICH	128	NCCU	General ICU (before)	In-hospital	NA	Hospital discharge	NA
Patel et al,37 2002	1991-1997	Cohort, retrospective	UK	TBI	247	NCCU	General ICU (before)	6 mo	NA	6 mo	NA
Roberts et al,38 2019	2015-2016	Cohort, prospective	US	TBI	548	NCCU	Trauma ICU	30 d	Y	Hospital discharge	NA
Ryu et al,39 2017	2013-2016	Cohort, retrospective	South Korea	General	571	NI	Before NI	In-hospital	Y	NA	Y
				TBI (subgroup)	152
Samuels et al,40 2011	1995-2002	Cohort, retrospective	US	SAH	703	NI	Before NI	In-hospital	NA	NA	NA
Sekhon et al,41 2017	2010-2016	Cohort, retrospective	Canada	TBI	113	NI	Before NI	6 mo	Y	6 mo	NA
Soliman et al,42 2018	2016-2017	Cohort, retrospective	Saudi Arabia	General	572	NCCU	General ICU (before)	In-hospital	Y	NA	NA
Suarez et al,43 2004	1997-2000	Cohort, retrospective	US	General	2381	NI	Before NI	In-hospital	NA	NA	NA
Tran et al,44 2020	2015-2017	Cohort, retrospective	US	AIS-LVOc	128	NCCU	General ICU	90 d	NA	NA	NA
Varelas et al,45 2004	1999-2002	Cohort, retrospective	US	General	2366	NI	Before NI	In-hospital	Y	NA	Y
Wärme et al,46 1991	1980-1988	Cohort, retrospective	Sweden	TBI	121	NCCU	General ICU (before)	6 mo	NA	6 mo	NA

Abbreviations: AIS, acute ischemic stroke; ICH, intracerebral hemorrhage; ICU, intensive care unit; LOS, length of stay; LVO, large vessel occlusion; NA, not applicable; NCCU, neurocritical care unit; NI, neurointensivist; SAH, subarachnoid hemorrhage; TBI, traumatic brain injury; Y, yes.

^a Comparisons made with general ICU with full neurological supports and with low neurological supports, with latter comparison evaluated.

^b Examined both the introduction of an NCCU and the introduction of an NI, with the former comparison evaluated.

^c Population evaluated was patients with AIS who required thrombectomy for LVO.

Risk of bias was assessed to be moderate in 22 studies, with the remaining 4 studies classed as serious. No study was assessed to be equivalent to a well-performed randomized clinical trial through the ROBINS-I tool criteria (eTable 7 in the Supplement).

Primary Outcome

Twenty-six studies including 55 792 patients were included in the meta-analysis of mortality outcomes.^{21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46} Critically ill adults with brain injury receiving care in an NCCU or by NCC specialized staff (n = 27 061) had a 0.83-fold lower risk of mortality (17% relative risk reduction; 95% CI, 0.75-0.92; P = .001), compared with those receiving care in a general ICU (n = 27 694) (Figure 1). Between-study heterogeneity variance was estimated at τ² = 0.04 (95% CI, 0.01-0.13), with an I² value of 86% (95% CI, 81%-90%) indicating substantial heterogeneity. The prediction interval ranged from g = 0.55 to 1.26, and as such, a harmful intervention effect cannot be excluded.

Figure 1. Meta-analysis of Mortality Outcomes as Risk Ratio (RR) in Critically Ill Adults With Brain Injury Receiving Specialized Neurocritical Care (NCC) and General Intensive Care.

Subgroup Analysis

There were 4 studies of AIS (n = 6575), 6 studies of ICH (n = 19 978), 5 studies of SAH (n = 2349), and 8 studies of TBI (n = 14 909) (Figure 2). Subgroup analysis of ICH demonstrated a 0.79-fold lower risk of mortality (95% CI, 0.64-0.98) among patients in an NCCU compared with a general ICU. In the remaining subgroups, no statistically significant intervention effect was identified. No significant difference was observed between studies of differing disease groups (P = .79).

Figure 2. Subgroup Analysis of Mortality Outcomes by Disease Process.

AIS indicates acute ischemic stroke; ICH, intracerebral hemorrhage; NCC, neurocritical care; RR, risk ratio; SAH, subarachnoid hemorrhage; TBI, traumatic brain injury.

Seventeen studies examined care in an NCCU, demonstrating a relative risk reduction of 18% (RR, 0.82; 95% CI, 0.72-0.95) compared with general care (Figure 3). Nine studies compared the presence of a neurointensivist or NCC team, demonstrating a relative risk reduction of 16%. While the estimated between-study heterogeneity differed considerably, with an I² of 91% for unit-level interventions and 23% for personnel-level interventions, the test for subgroup differences indicated no statistically significant difference between the 2 groups (unit RR, 0.82; 95% CI, 0.72-0.95; personnel RR, 0.84; 95% CI, 0.72-0.99; P = .79).

Figure 3. Subgroup Analysis of Mortality Outcomes by Intervention Type.

NCC indicates neurocritical care; RR, risk ratio.

Secondary Outcome

Fourteen studies reported ICU LOS data with a total of 32 652 patients.^{21,22,23,24,25,26,28,31,34,38,39,41,42,45} Ten studies reported hospital LOS data with a total of 21 717 patients.^{21,23,24,25,28,31,32,34,39,45} There were no statistically significant differences in the pooled SMD for either ICU (SMD, 0.02 days; 95% CI, –0.17 to 0.20) or hospital LOS (SMD, 0.01 days; 95% CI, –0.17 to 0.18) (eFigures 2 and 3 in the Supplement), when considering the associations with NCCU or NCC services. The estimated between-study heterogeneity was substantial in both analyses.

Functional outcomes were reported in 8 studies including a total of 4667 patients.^{21,24,25,35,36,37,38,41,46} Critically ill adults with brain injury who received care in an NCCU or by NCC specialized staff had a 0.83-fold lower risk of an unfavorable functional outcome (17% relative risk reduction; 95% CI, 0.70-0.97; P = .03), compared with those receiving general care (Figure 4). Between-study heterogeneity variance was estimated at τ² = 0.02 (95% CI, 0.00-0.15), with an I² value of 63% (95% CI, 24%-82%) indicating moderate heterogeneity. The prediction interval ranged from g = 0.55 to 1.24, and as such, a harmful intervention effect cannot be excluded.

Figure 4. Meta-analysis of Poor Functional Outcomes in Critically Ill Adults With Brain Injury.

NCC indicates neurocritical care; RR, risk ratio.

Sensitivity Analyses

Sensitivity analyses concerning the primary outcome are summarized in eTable 8 in the Supplement. No definitive explanation for heterogeneity was identified, and the analyses did not materially change the findings from the main analysis.

Publication Bias and Certainty of Evidence

A funnel plot of studies reporting mortality data is presented in eFigure 4 in the Supplement. An Egger regression test did not indicate the presence of funnel plot asymmetry (intercept, 0.546; 95% CI, –0.92 to 2.01, t = 0.731, P = .48). Given that only nonrandomized observational data were available, with moderate to serious risk of bias and significant heterogeneity evident, the final level of certainty in the evidence was considered very low.

Discussion

This systematic review and meta-analysis of 26 studies demonstrated an association between dedicated NCCUs and personnel and a decreased risk of mortality (RR, 0.83; 95% CI, 0.75-0.92) and poor functional outcomes (RR, 0.83; 95% CI, 0.70-0.97) in critically ill adults with brain injury. Significant heterogeneity between studies was also noted. Our findings corroborate a previous study by Kramer and Zygun,³ which examined mortality outcomes between specialty neurological and general ICUs. Eighteen studies with a total of 41 391 patients were reviewed, with broader inclusion criteria for pediatric studies, conference abstracts, NCC protocols as interventions, and a study population of status epilepticus. The authors reported a reduction in mortality (OR, 0.72; 95% CI, 0.59-0.89) and better neurological recovery (OR, 0.70; 95% CI, 0.61-0.81) in patients receiving care in a neurological ICU.

Heterogeneity was anticipated, given the nature of the intervention and the ethical and organizational limits to performing randomized clinical trials, as well as the lack of a clear and consistent definition of NCC. In addition, the follow-up of patient outcomes differed widely across studies. Prespecified and post hoc sensitivity analyses did not isolate potential sources of heterogeneity or identify confounders that significantly altered the mortality outcome of the main analysis. In particular, there were no marked differences between last years of recruitment to indicate the differential effect on mortality is linked to general improvements in outcome over time, due to changes in technology and techniques.

Determining the intervention within NCC that produces the greatest effect on patient outcomes is challenging to isolate not only because of the variability in descriptions of components of an NCCU, but also because of the interconnectedness of each element. The presence of a neurointensivist alone would affect processes of care, influence the education of surrounding staff, and increase the use of advanced neuromonitoring modalities. As such, we included comparisons of both NCCUs and neurointensivists vs general care and did not identify significant subgroup differences between unit- or personnel-level interventions. In addition, assessing mortality data alone does not differentiate attitudes around withdrawal of care between neurointensivists and general intensivists and is a limitation in determining contributors to patient outcomes. Another possible factor relating to improved functional outcomes may concern differences in neuroprognostication, with studies indicating that neurointensivists were better at predicting poor functional outcomes compared with other physicians.^47,48

Although ABIs encompass a diverse range of conditions, there were no significant subgroup differences between SAH, ICH, TBI, and AIS. Critically, in the largest of these subgroups (ie, ICH), a statistically significant decrease in mortality was noted. Of interest, NCC appeared to have a minimal association with mortality in patients with AIS, albeit ward-based stroke care units are associated with improved mortality and functional outcomes.⁴ As such, we speculate that the presence of a multidisciplinary stroke unit as a stepdown location from the ICU (providing ongoing subspecialized care regardless of the subspecialty of the ICU) may have biased this analysis toward the null hypothesis.

While NCCUs appear to improve mortality and functional outcomes, the cost benefit is yet to be fully evaluated. Assessing LOS in critical care may be an indirect reflection of costs. In our meta-analysis, we did not identify a statistically significant difference in ICU and hospital LOS. However, individual studies have reported differences in LOS with some attribution to differences in overall costs. In evaluating the effect of a full-time neurointensivist to manage an NCCU in South Korea, a reduction in the median cost per patient was noted (due to an increase in the number of patients managed in the NCCU) despite an increase in overall costs.³¹ Mirski et al³⁶ also reported a decrease in total costs of care, attributed to shorter LOS in the NCCU as well as reduced resource utilization. However, 2 multicenter studies reported higher absolute costs with NCCUs over the short-term,^25,32,49 albeit improved long-term functional outcomes imply an overall monetary benefit.

Limitations

A major limitation of this study is the previously described heterogeneity of the included studies and the absence of gold-standard randomized clinical trials to assess the objective. This resulted in a very low grade of certainty in the evidence identified. However, it is noted that sensitivity analyses did not distinctly change the findings of the main analysis and may be an indication that the observed mortality outcome in our study may be close to the actual outcome. While the use of crude data allows for comparability between studies, it does not account for confounders, which may result in differing conclusions when considering unadjusted and adjusted outcomes in the individual studies. Although a comprehensive search strategy was developed to capture all relevant manuscripts, because of the authors’ backgrounds and resource availability, non-English databases were not searched, and although no non–English-language publications were identified, it is plausible for publications to have not been included as a result. We also prioritized peer-reviewed full-text articles; however, exclusion of conference abstracts and non–peer-reviewed studies could potentially introduce bias.

Conclusions

This systematic review and meta-analysis found that subspecialized NCC delivery was associated with decreased mortality and increased favorable functional outcomes compared with general intensive care. However, significant heterogeneity limits the degree of certainty in the level of evidence, and further evaluation is necessary to assess the long-term cost-effectiveness of implementing such subspecialized services.

Supplement.

eFigure 1. PRISMA flow diagram for review of neurocritical care interventions for critically ill brain-injured adults

eFigure 2. Meta-analysis of ICU length of stay as standardised mean difference

eFigure 3. Meta-analysis of hospital length of stay as standardised mean difference

eFigure 4. Funnel plot of studies reporting mortality outcomes of neurocritical care interventions for critically ill brain-injured adults

eTable 1. Electronic database and search strategy for neurocritical care interventions for critically ill brain-injured adults

eTable 2. Articles excluded from systematic review at full-text screening stage with rationale

eTable 3. Articles excluded from systematic review but included in previous systematic reviews

eTable 4. Components of a neurocritical care unit introduced following a unit level change from a general intensive care unit

eTable 5. Details of neurocritical care personnel in studies comparing the introduction of neurointensivists or neurointensivist-led teams

eTable 6. Reported benefits of measured outcomes for studies included in systematic review and meta-analysis

eTable 7. Risk of bias assessment for studies included in neurocritical care interventions for critically ill brain-injured adults

eTable 8. Sensitivity analyses of mortality outcomes comparison between general and neurocritical care interventions

eReferences