The Role of Imaging in Clinical Stroke Scales That Predict Functional Outcome: A Systematic Review

Introduction

Acute ischemic stroke is the second leading cause of death worldwide and a leading cause of long-term disability. ^1,2 A risk score that predicts functional outcomes after stroke has several potential uses. It could help guide realistic prognostic expectations for patients, families, and physicians and aid in the planning of long-term care. It could be used in nonrandomized studies to control for case-mix variation and in controlled clinical trials to define individual clinical end points, to select suitable patients, and to reduce required sample size. ^{3 –6} It could be used for risk adjustment when evaluating stroke providers. In order to be useful and applicable to clinical practice and management, a prognostic model must be validated, easy to implement, and contain variables that are readily available for all patients.

Many scores exist that aim to predict poststroke functional outcome. Some of these scores are simple, with the simplest including solely a scale of stroke severity (National Institutes of Health Stroke Scale [NIHSS]), while others involve more inputs that require additional time and/or resources, such as neuroimaging. The added value of neuroimaging in predicting functional outcomes is important to critically evaluate, given the significant potential costs of imaging. These costs are both monetary and as potential harm to patients including possible treatment delays secondary to additional diagnostic imaging, radiation risks, and risk of contrast nephropathy. The goal of our study was to systematically review stroke scores that predict functional outcomes, with a focus on comparing clinical scores that incorporate imaging with those that do not incorporate imaging.

Methods

A systematic search was performed to comprehensively identify studies that use a clinical score/scale to predict functional outcome after ischemic stroke. We defined a score as a tool that could be administered during a patient’s hospital admission in which various variables (including but not limited to clinical, physiologic, and/or imaging) were evaluated and given a numeric value, with these values being totaled to produce a final score that corresponded to the probability of a good or poor outcome. Risk discrimination performance was assessed using area under the receiver operating characteristics curve (AUC) based on the favorable or unfavorable functional outcome used to evaluate each score. This study was exempt from institutional review board approval because it only involved analysis of publicly available, published, aggregate data.

Results

Study Selection

We screened a total of 3330 abstracts from which 14 scores were ultimately deemed to meet all inclusion criteria for the systematic review. Study selection steps are summarized in Figure 1. Table 1 shows the risk scores included in this review and includes information regarding the study populations, components, and performance of each score. Most of the study populations were from clinical trials for acute stroke or regional stroke registries. Neuroimaging was included as a predictor variable in half (7 of the 14) of the scores included. The mRS score at 3 months was the most common functional outcome (predicted by 13 [93%] of these scores); the BI and GOS were also used as outcomes. The AUCs from the derivation studies ranged from 0.64 to 0.84 for scores that included imaging and 0.64 to 0.94 for scores without imaging predictors. We found that 7 scores (50%) were validated externally for functional outcome, while 2 additional scores were validated using the same population as the derivation cohort. Among these 7 scores, 4 scores (57%) that contained imaging variables were externally validated and 3 nonimaging scores (43%) were externally validated. Table 2 shows the scores grouped by ischemic stroke patient population, specifically referring to their treatment status with intravenous (IV) thrombolysis or endovascular therapy; we assumed the National Institutes of Health Stroke Scale (NIHSS) and Shortened versions of the National Institutes of Health Stroke Scale (sNIHSS) scores were derived in patient populations without either treatment, because these scores were derived before widespread use of these acute ischemic stroke treatments.

OPEN IN VIEWER

Figure 1.

The flow diagram of studies included in the review.

OPEN IN VIEWER

Table 1.

Summary of Clinical Stroke Scores That Predict Functional Outcomes Included in This Review.

Score Name	Derivation Population	Predictors	Use of Imaging	Outcomes	AUC/Performance	Validation
ASTRAL 8	Patients with acute ischemic stroke having initial mRS <2, Centre Hospitalier Universitaire Vaudois (CHUV), Switzerland, 2003 to 2010	Age, stroke severity based on NIHSS, time delay from onset to admission >3 hours, range of visual field defects, admission glucose, level of consciousness	None	3-Month mRS >2	AUC = 0.85 (derivation); AUC = 0.82 to 0.89 (validation range) 9,10	External validation in Greek and Chinese populations
BOAS 11	Patients with acute ischemic stroke having initial mRS <4, S. Orsola-Malpighi Hospital, Bologna, Italy, 2007 to 2008	Initial NIHSS score, age, need of urinary catheter, oxygen administration, persistence of paralysis in an upper limb	None	9-Month mRS >2	AUC = 0.89 (derivation); AUC = 0.84 (internal validation)	Internal validation on separate test population; no external validation
DRAGON 12	Patients with ischemic stroke treated with IV alteplase, Helsinki University Central Hospital, Helsinki, Finland, 1995 to 2010	Dense cerebral artery sign/early infarct signs on admission head CT, prestroke mRS score >1, age, blood glucose, onset to treatment time (OTT), NIHSS score	CT hypodensity and/or dense cerebral artery sign	3-Month mRS 3 to 6 (derivation and initial external validation); 3-month mRS 0 to 2 (external validation) or mRS 4 to 6 13	AUC = 0.84 (derivation); AUC = 0.84 (validation for mRS 0-6); AUC = 0.79 (validation for mRS 0-2) 13	Internal validation with bootstrap; external validation in Swiss and Danish populations
MRI-DRAGON 14	Patients with ischemic stroke treated with IV alteplase, Hôpital Sainte-Anne, Paris, France, 2003 to 2012	DWI ASPECTS, prestroke mRS score >1, age, blood glucose, OTT, NIHSS score, M1 (proximal middle cerebral artery) occlusion	MRI (DWI ASPECTS)	3-Month mRS 3 to 6	AUC = 0.83	Internal validation with bootstrap
HAT 15	tPA-treated group in the National Institute of Neurological Disorders and Stroke study, 2001 to 2007	NIHSS score, extent of hypodensity on CT scan, serum glucose at baseline, history of diabetes	CT hypodensity	3-Month mRS 0 to 2	AUC = 0.75	None for functional outcome
HIAT 16	Intra-arterial recanalization therapy (IAT)–treated patients with stroke at University of Texas Houston Stroke Center, Houston, Texas, 1998 to 2007	Age, baseline NIHSS score, admission glucose, diabetes, heart disease, previous stroke, absence of mismatch on the pretreatment MRI	Pre-IAT MRI (DWI lesion and perfusion deficit mismatch)	Hospital discharge mRS 4 to 6 (derivation); 3-month mRS 4 to 6 (validation)	None in derivation. AUC = 0.68 in external validation 17	External validation with Houston Stroke Center patients 17
HIAT2 17	IAT-treated patients with stroke at University of Texas Houston Stroke Center, Houston, Texas, 2003 to 2011	Age, baseline NIHSS score, admission glucose, diabetes, hypertension, atrial fibrillation, clot-burden score, ASPECTS	CT ASPECTS	3-Month mRS 4 to 6	AUC = 0.70 (derivation)	External validation with US patients (Atlanta, GA)
iScore 18	Patients with acute ischemic stroke in Ontario, Canada, 2003 to 2008	Age, sex, stroke severity, stroke subtype, atrial fibrillation, congestive heart failure, cancer, renal dialysis, preadmission disability, glucose	None	30-Day mRS 3 to 6 (derivation); 3-month mRS 3 to 6 (validation)	AUC = 0.79 (derivation); AUC = 0.81 to 0.82 (external validation range) 19,20	External validation in French and Korean populations
NAV 21	Patients with acute stroke treated with endovascular therapy, 2 US stroke centers, 2009 to 2011	Age, baseline NIHSS, decreased cerebral blood volume (estimated by CT perfusion)	CT perfusion	90-Day mRS 0 to 2	AUC = 0.77	None
NIHSS 22	Patients with ischemic stroke, multiple North American sites, 1993 to 1994	NIHSS score	None	3-Month GOS =3 or 4	AUC = 0.94	None
sNIHSS-4 23	Patients with ischemic stroke, multiple North American trial sites, 1997	sNIHSS-4 score (shortened version of NIHSS score)	None	3-Month mRS >3	AUC = 0.73 (left hemispheric stroke) and 0.75 (right hemispheric stroke)	None
SAD 24	Patients with acute stroke having NIHSS ≥ 5 treated with endovascular therapy, multicenter prospective cohort, 2008 to 2013	Age, lesion volume on DWI	MRI (DWI)	3-Month mRS 4 to 6	AUC = 0.82 (derivation); AUC = 0.69 (external validation)	External validation (Massachusetts General Hospital)
SPAN 25	Patients with acute ischemic stroke, multicenter IV alteplase trial (treatment and placebo patients), 1991 to 1994	Age, NIHSS score	None	3-Month “favorable outcome” (mRS 0-1, NIHSS ≤ 1, BI ≥ 95, GOS score = 1)	c-Statistic = 0.64	None
THRIVE 26	Patients with acute stroke treated with endovascular therapy, multiple US trial sites, 2001 to 2003	NIHSS score, age, hypertension, diabetes, atrial fibrillation	None	3-Month mRS 0 to 3	AUC = 0.71 (derivation); AUC = 0.71 to 0.78 (external validation range)	Externally validated in other US clinical trial site 27 –30

Abbreviations: AUC, area under the curve; ASPECTS, Alberta Stroke Program Early CT Score; ASTRAL, Acute Stroke Registry and Analysis of Lausanne; BI, Barthel Index; BOAS, Bologna Outcome Algorithm for Stroke; CT, computed tomography; DRAGON, (hyper)Dense cerebral artery sign/early infarct signs on admission CT scan, prestroke modified Rankin Scale (mRS) score, Age, Glucose level at baseline, Onset-to-treatment time, and baseline National Institutes of Health Stroke Scale score; DWI, diffusion-weighted imaging; GOS, Glasgow Outcome Scale; HAT, hemorrhage after thrombolysis; HIAT, Houston intra-arterial recanalization therapy; IV, intravenous; MRI, magnetic resonance imaging; mRS, modified Rankin Score; NIHSS, National Institute of Health Stroke Scale; NAV, NIHSS, Age, and Volume; SAD, Stanford Age and Diffusion-Weighted Imaging; SPAN, Stroke Prognostication Using Age and NIHSS; THRIVE, Totaled Health Risk in Vascular Events; tPA, tissue plasminogen activator.

OPEN IN VIEWER

Table 2.

Summary of Clinical Stroke Scores by Patient Population.

Ischemic Stroke Patient Population	Potentially Relevant Scores	Externally Validated Scores
Treated with intravenous thrombolysis	DRAGON,a MRI-DRAGON,a HAT,a SPAN	DRAGONa
Treated with endovascular therapy	HIAT,a HIAT2,a NAV,a SAD,a THRIVE	HIAT, HIAT2, SAD, THRIVE
Not treated with either intravenous thrombolysis or endovascular therapy	NIHSS, sNIHSS
Not specific to treatment status	ASTRAL, BOAS, iScore	ASTRAL, iScore

Abbreviations: ASTRAL, Acute Stroke Registry and Analysis of Lausanne; BOAS, Bologna Outcome Algorithm for Stroke; HAT, hemorrhage after thrombolysis; HIAT, Houston intra-arterial recanalization therapy; MRI, magnetic resonance imaging; NIHSS, National Institute of Health Stroke Scale; SAD, Stanford Age and Diffusion-Weighted Imaging; SPAN, Stroke Prognostication Using Age and NIHSS; THRIVE, Totaled Health Risk in Vascular Events.

^aImaging as predictor.

Discussion

There are several scores that aim to predict functional outcomes after ischemic stroke, but they have yet to be summarized in a systematic review. We found 14 scores that predicted functional outcomes (mRS, BI, or GOS) and found substantial variation across these scores in terms of patient population, complexity of predictors used, and risk discrimination performance. Half of the scores included neuroimaging as a predictor. We did not find an appreciable difference in terms of risk discrimination performance between these scores and scores without neuroimaging, although direct comparison is markedly limited due to heterogeneity in both patient population-specific factors (such as acute stroke severity and/or treatment status) and outcomes measured; for instance, ASTRAL discrimination performance was assessed using 3-month mRS > 2, while BOAS discrimination performance was assessed using 9-month mRS > 2. In those scales in which imaging was used, 3 used simple noncontrast head CT to evaluate core infarct extent while 3 used MRI DWI-defined core infarct for prognostication. Of note, only 1 score (HIAT) required the use of perfusion imaging to define MR DWI and MR perfusion mismatch; another score (NAV) requires CT perfusion. No other scores required IV contrast, perfusion-weighted techniques, or other methods besides widely available noncontrast CT and MR DWI. Additionally, while all scores looked at initial imaging, some scores also incorporated follow-up imaging parameters into the analysis leading to further variability and heterogeneity.

Our findings show that clinicians have multiple options for assessing poststroke functioning, but we did not find 1 score that is an obvious choice for all clinical situations. Most of the scores performed well in terms of risk discrimination and have been validated in internal or external populations (although only half of the scores were externally validated). Clinicians might consider the ease of calculating risk when choosing risk scores; the simplest scores are NIHSS alone, sNIHSS-4, and SPAN. Although they involve more inputs than the NIHSS-based scores, ASTRAL, BOAS, iScore, and THRIVE only require clinical inputs and can be calculated quickly. Patient factors could guide choices among these scores. For instance, ASTRAL was derived from patients with stroke having mRS < 2, while BOAS was derived from more severe patients (mRS < 4). The availability of CT (DRAGON, HAT, HIAT2, NAV [CT perfusion]) versus MRI (MRI-DRAGON, HIAT [MR perfusion], SAD) can guide choices among scores that included imaging predictors.

Our systematic review has several limitations that should be noted. First, we found substantial heterogeneity in terms of patient demographics, stroke severity, and how the scores defined their inputs and outputs as well as how functional outcome was measured. Additionally, some patient populations received no treatment, while others received pharmaceutical treatment and other vascular intervention. Although such heterogeneity introduces bias and makes it difficult to directly compare studies, as mentioned above, this variation could allow clinicians to tailor their choice of risk score for their patient population and the most relevant outcome of interest. Additionally, we could not perform a meta-analysis of risk discrimination performance due to the heterogeneity in risk score inputs and functional outcomes reported. While the majority of studies assessed functional outcome using mRS, some scores used other measures such as BI and GOS, which are themselves inherently different. The BI has been widely used to measure outcomes and is useful for planning rehabilitative strategies as it is a scale that measures 10 basic aspects of activity related to self-care and mobility ^3,37,38 ; the mRS assesses disability after stroke by measuring functional independence, incorporating the World Health Organization components of body function, activity, and participation ^3,39,40 ; similar to mRS, GOS is another ordered scale to assess outcomes with a key difference compared to mRS being the lack of distinction among patients with good outcomes, since this group encompasses full recovery and mild disability. ^3,41 Furthermore, the mRS measure itself has been shown to have interobserver variability, ⁴² which could further compound our analysis. Due to our selection criteria, there are some stroke studies that assessed functional outcome (such as but not limited to NIHSS from TOAST trial, ⁴³ NIHSS-time score, ⁴⁴ NIHSS-age score, ⁴⁵ stroke thrombolytic predictive instrument, ⁴⁶ 3-item scale, ⁴⁷ six simple variables, ⁴⁸ PLAN, ⁴⁹ SEDAN ⁵⁰ ) that were excluded as they were either operationalized as a statistical model as opposed to a score/scale that could be calculated at the bedside or they were not evaluated using the AUC/c statistic, but rather with disease association measures (such as odds ratios), which do not capture the full tradeoffs between sensitivity and specificity.

We found many studies (36 full-text papers reviewed, >50 other abstracts screened) that evaluated the use of imaging and clinical prognostic factors for poststroke functioning that were not included in our study because they did not include a score that could be calculated in a clinical setting. This discrepancy (>50 studies but only 7 scores that included neuroimaging) suggests that stroke imaging researchers ought to give more consideration to incorporating their findings into single, composite risk scores that can be more easily integrated to guide clinical decision-making.

Conclusion

Numerous stroke scales have been developed to predict functional outcome following acute ischemic stroke. There is substantial variation and heterogeneity among these scores in terms of their development and adoptability, including the use of neuroimaging as a predictor of poststroke outcomes. It is unclear whether imaging-based scores should be favored over simpler approaches that do not include imaging for predicting poststroke function, although the appropriate choice of risk score could depend on the treatment status and other patient factors.