Early Prediction of Malignant Brain Edema After Ischemic Stroke

Introduction

Malignant brain edema is a leading cause of early death after ischemic stroke,¹ which occurs in 10% to 78% of patients with ischemic stroke.² It is characterized by a malignant course of rapid neurological deterioration associated with massive cerebral swelling, leading to transtentorial herniation and death or poor functional outcome.³ Despite its devastating consequences, little evidence is available to inform prevention and management of malignant edema, apart from the decompressive hemicraniectomy.⁴

Given the lack of effective treatment, better understanding of the underlying pathophysiology and predictive factors for rapid deterioration could inform more accurate and faster prediction and, therefore, inform clinical practice for high-risk patients by the targeted implementation of more intensive monitoring and evidence-based interventions. A systematic review of predictors of life-threatening brain edema in ischemic stroke in 2008 identified 23 studies with 27 potential predictors and concluded that single predictors lacked sufficient predictive value.² Since then several studies of new predictive factors and models have been published, and thus, we performed a systematic review and critical appraisal of the evidence to determine the clinical utility of the available single predictors and predictive models. In addition, we focused on the time when each potential predictor had been assessed because early identification of patients at risk is important to inform timely clinical decisions.

Methods

The data that support the findings of this study are available from the corresponding author on reasonable request. We followed the guidelines for MOOSE (Meta-analyses Of Observational Studies in Epidemiology) to complete and report this study.⁵ Study protocol is available from http://www.crd.york.ac.uk/PROSPERO/display_record.php?ID=CRD42017075701.

Search Strategies and Study Selection

We searched Ovid Medline (from 1946) and Embase (from 1974) in September 2017 and updated in March 2018, with search terms ‘ischemic stroke’ AND ‘severe OR malignant OR edema’ AND ‘predict*’ (search strategies in Methods in the online-only Data Supplement). One researcher (Dr S. Wu) reviewed all abstracts, and 2 researchers (Dr S. Wu, R. Yuan) independently reviewed all full texts of potentially eligible studies and scrutinized reference lists of relevant articles. Discrepancies were resolved through joint-reassessment and discussing with another researcher (C. Wei).

Eligibility Criteria

We included observational studies of adult patients with acute ischemic stroke, defined malignant edema as a syndrome of clinical worsening (or death or requirement for decompressive hemicraniectomy) with imaging evidence of brain swelling, measured at least 1 potential predictor or developed a predictive model for malignant edema, and published a full-text report in a peer-reviewed journal. Predictive models were defined as a score or scale with ≥2 variables to assess the risk of developing malignant edema, with the overall discrimination performance being reported. There were no language restrictions. Studies were excluded if outcomes were defined as imaging-based edema or as clinical worsening or death not attributable to brain edema, or if the study assessed cross-sectional associations.

Quality Assessment

For each included study, 2 researchers (Dr S. Wu, Y. Wang) independently assessed 6 domains (study participation, study attrition, potential predictor measurement, outcome measurement, flow and timing, and data analysis) with 17 items of potential biases (Table I in the online-only Data Supplement).

Data Analysis

Odds ratios (ORs) and 95% CIs were extracted or calculated as the effect size for dichotomous variables and mean differences or standardized mean differences for continuous variables. Summary estimates of the strength of association were pooled by random-effects modeling. We had planned to estimate summary performance statistics of predictive models, but this was not performed because none of the included models had been externally validated. Heterogeneity was quantified by I² and was stratified by inclusion criteria on stroke severity and on whether all patients had received reperfusion therapies; for imaging factors, heterogeneity was stratified by imaging modalities and assessment time (subgroup analysis in Methods in the online-only Data Supplement). Publication bias was explored by funnel plots, where the estimate of effect size was plotted on the horizontal axis and the SE of the log-transformed estimate of effect size on a reversed scale of the vertical axis. For predictive models, we descriptively analyzed data according to the CHARMS (Checklist for Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies).⁶ For each model, we extracted or calculated C statistic for discrimination performance and sensitivity and specificity of each index score for classification performance. Statistical analyses were conducted in RevMan (Computer program), version 5.3 and Microsoft Office Excel 2007.

Post Hoc Analysis

We performed a post hoc literature search in May 2018 to identify randomized controlled trials (RCTs) to provide more robust evidence for the effect of reperfusion therapies and successful revascularization on the development of malignant edema (Methods in the online-only Data Supplement). Reperfusion therapies included either intravenous thrombolysis (ie, intravenous administration of alteplase or other thrombolytic agents) or endovascular interventions (ie, intraarterial thrombolysis, mechanical clot disruption or retrieval, or both). Revascularization was defined as either recanalization of the arterial occlusion or reperfusion of the downstream territory.⁷

Results

Figure 1 shows literature search and study selection. We included 38 studies (31 cohort studies^8–38 and 7 case-control studies,^39–45 in 49 articles) with 3278 patients (mean age 49 to 78 years; male proportion 35%–74%). Study characteristics are summarized in Table II in the online-only Data Supplement. Twenty-five studies reported the assessment time for malignant edema, all within 30 days after stroke (median 6 days; interquartile range, 4–7).^{9–14,16,17,19,21–23,25,28,29,31–33,35,37,39–43} In 31 cohort studies, 31% (796/2546) of patients developed malignant edema (for individual studies: median 31%; range 10%–78%). We identified 24 clinical factors, 7 domains of imaging markers, 13 serum biomarkers, other physiological biomarkers, and 4 predictive models.

• Download figure
• Download PowerPoint

Of 17 items for the risk of bias, the most common bias lied in the inadequate sample size (high risk in all 38 studies), followed by nonprespecified (24 studies) or nonreported assessment time for edema (13 studies), and data presented in inappropriate forms (16 studies; Figure I in the online-only Data Supplement). We found asymmetry of funnel plots for potential predictors, with a trend towards symmetrical as the number of studies increases. For most plots, the individual effect sizes distributed symmetrically aside the pooled effect size, with most studies showed an SE of log(OR) around 0.5 or log(SEM) around 0.25 (Figure II in the online-only Data Supplement).

We identified 11 factors related to demographics or medical history, which could be collected from patients immediately after their admission (Table III in the online-only Data Supplement). Of these factors, only younger age was associated with malignant edema (n=2075; pooled mean difference, −4.42; 95% CI, −6.63 to −2.22 years),^{8,10,11,13,15,16,20,21,24,26–29,31,32,34–36,40–43} with heterogeneity (df=21, I²=88%, P<0.0001) that could not be explained by stroke severity (Figure 2A) or whether receiving reperfusion therapies (Figure 2B). Four studies each reported a cutoff age ranging from 60 to 75 years.^9,18,42,46

• Download figure
• Download PowerPoint

We identified 11 stroke characteristics, which could be determined within 24 hours after admission (Table III in the online-only Data Supplement). Patients presenting with higher National Institutes of Health Stroke Scale (NIHSS) scores,^{8,10,12,15,20,24,26–28,30–33,35,37,43,44,46,47} depressed consciousness,^{9,21,25,27,36,45} gaze palsy,^9,19,25,45 nausea or vomiting,^21,41 or receiving ventilation^16,19,48 were more likely to develop malignant edema, whereas there was no association with hemispheric lateralization of infarct,^{9,11–13,16,18,19,24–27,30,31,35–37,40,41,44,45,47} large arterial atherosclerosis,^{10,12,16,26–28,33,35–37,40,42,43} cardioembolism,^{10,12,16,19,26–28,33,35–37,40,42,43} headache,^21,41 fever,^21,41 or osmotic therapies.^30,48 Fourteen studies reported median values of admission NIHSS (Table IV in the online-only Data Supplement): in unselected patients regarding stroke severity (n=568)^{8,12,28,43,44} and all but one study of patients with major artery occlusion (n=239),^15,20,37 malignant group (median NIHSS all ≥17: 17–20) had higher NIHSS scores than nonmalignant group (median NIHSS all ≤15: 5.5–15); in all patients with clinically severe stroke (n=526; median NIHSS ≥17 in both groups in all studies)^{10,24,30,33,35} and 1 study of patients with major artery occlusion (n=52; median NIHSS ≤15 in both groups),²⁷ higher NIHSS scores were not associated with further clinical deterioration. Nine studies reported cutoff values of NIHSS (11–20) with different measures of their predictive values for malignant edema.^{12,26,27,31,32,37,43,46,47}

The post hoc search identified 10 additional studies for thrombolysis, endovascular interventions, and revascularization (Figure III and Table V in the online-only Data Supplement).^49–58 Patients who had received intravenous thrombolysis showed a higher risk for developing malignant edema than those received control treatment in RCTs (n=4986; OR, 1.34; 95% CI, 1.01–1.77; df=4; I²=0%)^{49,52,53,56,58} but a lower risk in observational studies^{8,12,16,18,27,33–37} (Figure IV in the online-only Data Supplement). After excluding the single RCT reporting a significant association (n=3035; OR, 1.65; 95% CI, 1.12–2.45),⁵⁸ the pooled association of the remaining RCTs was nonsignificant (n=1951; OR, 1.06; 95% CI, 0.71–1.60).^49,52,53,56 RCTs identified no association between endovascular interventions and the development of malignant edema (n=938; OR, 0.94; 95% CI, 0.70–1.27; df=2; I²=0%),^51,54,57 consistent with the findings in observational studies (Figure V in the online-only Data Supplement).^{12,15,16,27,30,33,35–37,43,44,56} Revascularization within 24 hours after onset was associated with a lower risk of malignant edema (n=1600; OR, 0.37; 95% CI, 0.24–0.57; df=12; I²=53%),^{15,18–20,29,34,35,47,50,54,55,59} where the association was significant regardless whether revascularization was defined as recanalization or reperfusion (Figure 3A) or whether all patients had received reperfusion therapies or not (Figure 3B).

• Download figure
• Download PowerPoint

Patients with malignant edema had a larger infarct volume on admission than those without (n=929; standardized mean difference, 2.57; 95% CI, 1.82–3.31; df=12; I²=93%; P<0.0001), where heterogeneity was explained by difference in imaging modalities (computed tomography [CT] versus diffusion-weighted imaging [DWI]; Figure 4A) and the time of DWI assessment (Figure 4B) but not CT time (Figure 4C).^{8,11,15,18,26,27,29,31,32,36,37,42,44} Five studies (n=335) measured DWI infarct volume within 14 hours after onset in patients with severe stroke and defined a cutoff volume (82–145 cm³), where the volume >145 cm³ showed the best predictive values (Table VI in the online-only Data Supplement).^{18,31,32,36,37}

• Download figure
• Download PowerPoint

Six studies investigated parenchymal hypoattenuation on CT within 40 hours after stroke, where the extent >50% of the middle cerebral artery territory was associated with malignant edema (n=420; OR, 5.33; 95% CI, 2.93–9.68; df=3; I²=24%; P=0.27), with the larger extent indicating the stronger association (Figure 5)^{9,21,22,24,39,41} and the ischemia involving other territory increased the risk than middle cerebral artery alone (n=833; OR, 7.10; 95% CI, 4.53–11.14).^{16,21,24,29–31,33,41} The association between hypoattenuation >50% of the middle cerebral artery territory and malignant edema was significant both in patients with severe stroke (n=358; OR, 4.27; 95% CI, 2.63–6.94; df=2; I²=0%; P=0.72)^21,24,41 and unselected stroke patients (n=62; OR, 20.08; 95% CI, 4.05–99.51)³⁹ and regardless of whether CT was performed within 6 hours (n=157; OR, 5.54; 95% CI, 2.36–13.03),^24,41 18 hours (n=62; OR, 20.08; 95% CI, 4.05–99.51),³⁹ or 40 hours (n=201; OR, 3.78; 95% CI, 2.10–6.80)²¹ after stroke.

• Download figure
• Download PowerPoint

Six studies (n=737) reported controversial findings for the ASPECTS (Alberta Stroke Program Early CT Score).^{15,20,30,33,42,47} Other early infarct signs associated with malignant edema included moderate mass effect (ventricle compression)^{21,30,39,41,42,45} and major mass effect (midline shift^39,42,45 or compression of basal cisterns^30,39; 4 studies reported different cutoff values of midline shift^17,32,46,48), but not hyperdense artery signs,^{21,22,30,39,41} minor mass effect (sulcal effacement),^21,22,41,42 or basal ganglia involvement^{21–23,38,41} (Table III in the online-only Data Supplement). Brain atrophy was associated with a lower risk of malignant edema when defined by objective measures.^23,32,60 Seven studies investigated hemorrhagic transformation^{21,28–30,34,35,44} but no association in the study specifically assessing the presence of hemorrhagic transformation before edema (P=0.18).²¹

In patients with clinically severe stroke, no association was found between the presence of proximal arterial occlusion and malignant edema (n=48; OR, 0.75; 95% CI, 0.10–5.44; df=1; I²=40%; P=0.20).^24,59 In patients with arterial occlusion, more proximal site (n=399; OR, 4.89; 95% CI, 2.77–8.64; df=6; I²=19%; P=0.29),^{15,18,31,33–35,37} longer extent (n=224; OR, 3.78; 95% CI, 1.96–7.28; df=2; I²=0%; P=0.55),^19,36,47 and more segments (n=294; OR, 4.05; 95% CI, 2.24–7.34; df=2; I²=0%; P=0.80)^20,22,36 of occlusion were each associated with a higher risk of malignant edema. The malignant group had more severe deficits in cerebral blood perfusion within 24 hours after stroke than nonmalignant group, regardless of imaging modalities or measures used (Table VII in the online-only Data Supplement).^{8,9,12–14,24,25,27,36,59} In addition, patients with malignant edema had increased permeability of blood-brain barrier,^8,34 poor collateral circulation,^15,19,47 and more severe clot burden.⁴⁷ Thirteen serum biomarkers were included in meta-analysis (Table III in the online-only Data Supplement),^{10,11,13,16,21,26,28,30,33,36,41–43,45,47} whereas available data for other physiological biomarkers were limited.^{11,13,26,28,33,35,42,44,61–64}

Of 11 multifactorial models identified,^{11,20,21,28,30,33,35,36,46,60,65} 4 models fulfilled our definition of predictive models,^20,21,30,33 all included patients with severe stroke and assessed potential predictors within 24 hours after stroke (Table VIII in the online-only Data Supplement). Kasner index score²¹ and DASH score (DWI ASPECTS Score, Anterior Cerebral Artery Territory Involvement, M1 Susceptibility Vessel Sign, and Hyperglycemia)³³ were development studies, and MBE score (Malignant Brain Edema)²⁰ and EDEMA score (Enhanced Detection of Edema in Malignant Anterior Circulation Stroke)³⁰ were studies of development with internal validation, but none was externally validated. Factors in Kasner score²¹ and EDEMA score³⁰ were all routinely assessed on admission for stroke patients, whereas DASH score³³ required DWI and MBE score²⁰ required CT angiography. All models showed a C statistic of ≥0.70, while with a risk of overfitting because none satisfied the sample size criterion of >10 events per variable (Table IX in the online-only Data Supplement). Within the range of possible scores of each model, cutoff scores of Karsner score ≥2,²¹ DASH score ≥2,³³ MBE score ≥4,²⁰ and EDEMA score ≥3³⁰ showed the best classification considering both sensitivity and specificity (Table VIII in the online-only Data Supplement).

Discussion

We provide a comprehensive review of predictors and predictive models for malignant brain edema after ischemic stroke, particularly focusing on the time when the data would be available for each predictor in practice. Despite that the included studies were generally small and some had methodological flaws, we found that younger age, higher admission NIHSS, and larger parenchymal hypoattenuation on initial CT were consistently associated with an increased risk of the development of malignant course and all routinely available within 6 hours after stroke. A reduced risk of malignant edema was associated with successful revascularization but not with the administration of intravenous thrombolysis or endovascular interventions.

Previous studies found that patients with more severe neurological deficits were more likely to develop malignant edema.² Our study confirmed this finding and further proposed a narrower range of cutoff score (15–17) to distinguish patients at a risk for malignant edema from general stroke patients. Another predictor is the parenchymal hypoattenuation >50% of the middle cerebral artery territory on initial CT.² We further clarified that this predictor was reliable when assessed as early as within 6 hours up to 40 hours after stroke and for both general stroke patients and those with severe stroke. Although an absolute volume of infarct might be more accurate than a relative extent, it is difficult to reach consensus on a definite cutoff volume in practice because of various imaging measures used, and some imaging techniques, such as DWI, are not available in many stroke services.

In addition to initial stroke severity assessed clinically or radiologically, younger age is another predictor for which age-related brain atrophy is a possible confounder by providing buffering space for brain swelling. However, the optimal cutoff age for predicting malignant edema is unknown. Among 4 studies reporting cutoff ages,^9,18,42,46 a case-control study recruited patients <70 years and matched for stroke severity reported that malignant group had more patients aged <60 years than nonmalignant group⁴²; another study found that whether age ≥70 years did not change the risk for malignant edema⁴⁶; the other 2 studies (not in meta-analysis because median age was presented) reported that age ≥75 years increased risk of malignant edema.^9,18 In addition, in patients with malignant edema, those aged >75 years had larger infarct volume than those <75 years of age, even the age-related brain atrophy was taken into account.¹⁸ Furthermore, a study reported that the protective effect of older age no longer existed after controlling for the effect of infarct volume.³⁶ These findings, although from small studies, imply the possibility that patients aged >75 years are prone to more severe infarct that predisposes malignant edema, which worth further investigation.

There is a discrepancy between RCTs and observational studies about the association between intravenous thrombolysis and malignant edema, which could be explained by differences in baseline characteristics. First, apart from the IST-3 study (Third International Stroke Trial) that had relatively broad inclusion,⁵⁸ all other 4 RCTs had excluded patients with severe stroke syndrome or those with large cerebral infarction; both are important predictors for malignant edema. This difference in inclusion criteria was reflected in the change of the pooled effect size in the sensitivity analysis after excluding the IST-3 study. Therefore, the pooled results of RCTs might have been biased towards less severe stroke patients, whereas most included observational studies had enrolled patients with severe stroke who were more prone to malignant edema. Second, the effect of intravenous thrombolysis on malignant edema might have been confounded by baseline characteristics that influenced the choice of the intervention in the observational studies; for example, patients who had milder stroke and those who had earlier access to medical care were more likely to receive thrombolysis, where these characteristics are often associated with better prognosis. Therefore, we were unable to draw a firm conclusion of this association from either RCTs or observational studies, and future RCTs with broader inclusion are expected.

By performing meta-analysis for individual predictors, we were unable to analyze the effect of confounders. Therefore, we reviewed multifactorial models. where the confounding effects had been adjusted. Existing predictive models for malignant edema showed good discrimination but at a risk of overfitting because of small-sample effect, and none was externally validated. In addition, DASH score³³ and MBE score²⁰ showed better discrimination than Kasner score²¹ and EDEMA score,³⁰ at the cost of feasibility as the former 2 required advanced imaging techniques that may not be available in all stroke services. Furthermore, none of these models included age, an important predictor identified in this review. Large cohort studies are needed to externally validate these models and to improve them by including other important predictors identified in this review and future studies.

This review had several limitations. First, the random-effects modeling is subjected to the loss of efficiency if the assumed effect size is far from the true effect size because of a broad CI; for example, when there are substantial differences in baseline characteristics or only a small number of studies included. However, for variables in the current review, we found consistent results between random-effects modeling and fixed-effect modeling (data not shown). We believe that random-effects modeling used here is appropriate, and we are more confident when the positive inferences are drawn. Second, although we used broad search strategies and inclusion criteria, the asymmetry of funnel plots for some factors indicated potential publication bias because of the lack of large sample studies to provide more precise and reliable results. The included studies were generally small, and the number of studies for individual predictor was limited. Therefore, although some predictors, such as hyperdense artery sign, showed no association with malignant edema in this comprehensive review, they are worth further investigation because the sample size might have been underpowered to detect clinically important differences. Similarly, although we found associations between malignant edema and depressed consciousness, gaze palsy, nausea or vomiting, ventilation, and early mass effect on CT, they were investigated in the limited number of small studies, and these signs often indicate the existence of brain swelling, thus require further investigation.

Conclusions

Younger age, higher admission NIHSS, and larger parenchymal hypoattenuation on initial CT are early predictors for malignant edema after stroke, which are routinely available within 6 hours after stroke to inform timely clinical decisions. Successful revascularization reduces the risk for malignant edema. No firm conclusion could be drawn for other factors because of the limited data and diversity in study characteristics. Future studies with robust design are needed to explore optimal cutoff age and NIHSS scores for predicting malignant edema, to provide more reliable evidence for other potential predictors and to externally validate and improve existing models.

Acknowledgments

We thank Professor Peter Sandercock for his helpful comments to improve this study.

Footnotes