Spot Sign Number Is the Most Important Spot Sign Characteristic for Predicting Hematoma Expansion Using First-Pass Computed Tomography Angiography
Analysis From the PREDICT Study
Background and Purpose—The spot sign score (SSS) provides risk stratification for hematoma expansion in acute intracerebral hemorrhage; however, external validation is needed. We sought to validate the SSS and assess prognostic performance of individual spot characteristics associated with hematoma expansion from a prospective multicenter intracerebral hemorrhage study.
Methods—Two hundred twenty-eight intracerebral hemorrhage patients within 6 hours after ictus were enrolled in the Predicting Hematoma Growth and Outcome in Intracerebral Hemorrhage Using Contrast Bolus CT (PREDICT) study, a multicenter prospective intracerebral hemorrhage cohort study. Patients were evaluated with baseline noncontrast computerized tomography, computerized tomography angiography, and 24-hour follow-up computerized tomography. Primary outcome was significant hematoma expansion (>6 mL or >33%) and secondary outcome was absolute and relative expansion. Blinded computerized tomography angiography spot sign characterization and SSS calculation were independently performed by 2 neuroradiologists and a radiology resident. Diagnostic performance of the SSS and individual spot characteristics were examined with multivariable regression, receiver operating characteristic analysis, and tests for trend.
Results—SSS and spot number independently predicted significant, absolute, and relative hematoma expansion (P<0.05 each) and demonstrated near perfect interobserver agreement (κ=0.82 and κ=0.85, respectively). Incremental risk of hematoma expansion among spot-positive patients was not identified for SSS (P trend=0.720) but was demonstrated for spot number (P trend=0.050). Spot number and SSS demonstrated similar area under the curve (0.69 versus 0.68; P=0.306) for hematoma expansion.
Conclusions—Multicenter external validation of the SSS demonstrates that the spot number alone provides similar prediction but improved risk stratification of hematoma expansion compared with the SSS.
Hematoma expansion occurs in approximately one third of patients with acute primary intracerebral hemorrhage (ICH) and is independently associated with early neurological deterioration, death, and disability.1,2 An accurate and reliable method to predict hematoma expansion in patients with acute ICH is needed to improve clinical prognostication and to assess efficacy of therapies targeted to selected patient cohorts.3 Intrahematoma contrast extravasation on computerized tomography (CT) angiography (CTA), coined the spot sign, is one of the most promising predictors of hematoma expansion.4 The spot sign was recently prospectively validated as an independent predictor of hematoma expansion in the Predicting Hematoma Growth and Outcome in Intracerebral Hemorrhage Using Contrast Bolus CT (PREDICT) multicentre cohort study.5 On the basis of radiological characteristics of the spot sign, a scoring system, the spot sign score (SSS), was devised reporting accurate risk stratification of hematoma expansion based on spot number, maximum spot size, and density.6 The SSS represents an important development toward an accurate predictive model for hematoma expansion; however, the single-center retrospectively developed scoring system is not externally validated. The purpose of this study was to validate the SSS for prediction of hematoma expansion and to perform an independent evaluation of spot characteristics most predictive of hematoma expansion.
Patient Cohort and Study Protocol
Patient data were obtained from the PREDICT study,5 a prospective multicenter cohort study (12 centers in 6 countries) of acute primary or anticoagulant-associated ICH patients presenting within 6 hours of onset. Inclusion criteria included age ≥18 years and ICH volume <100 mL.Exclusion criteria were known renal impairment, premorbid dependence defined as modified Rankin scale score >3, secondary cause of ICH (eg, tumor, arteriovenous malformation), deep coma (Glasgow Coma Scale ≤5), or major comorbid or terminal illness.
Baseline variables recorded included patient demographics, medical history, Glasgow Coma Scale and National Institutes of Health Stroke Scale, coagulation profile, and blood glucose. All data were collected and submitted to the coordinating center at the University of Calgary (Calgary, Canada). Study protocol was approved by the research ethics board at each participating center. Surrogate consent or written informed consent was obtained for each patient according to requirements established by each ethics board site.
Two hundred sixty-eight patients were enrolled in PREDICT from June 24, 2006, to September 6, 2010. Forty patients were excluded from the primary analysis of hematoma expansion (14 treated with recombinant activated factor VII before follow-up CT; 15 treated with surgical evacuation before follow-up CT; 7 died before follow-up CT, and 4 did not have a follow-up CT for unknown reasons). Two hundred twenty-eight (85%) patients had a 24-hour follow-up CT and were included in the study analysis.
Imaging performed included baseline noncontrast CT (NCCT), CTA, and 24-hour follow-up NCCT. Time from ictus to NCCT, CTA, and 24-hour follow-up scan was recorded. Delayed angiographic or postcontrast imaging was not included in the study protocol. CT protocols were performed according to standard institutional technique reflecting a pragmatic observational study approach. Site scanners and specific CT protocols are listed in Table I in the online-only Data Supplement.
CTA source images were independently examined for spot sign presence by 2 experienced neuroradiologists and a radiology resident blinded to 24-hour follow-up CT and hematoma expansion. The spot sign definition is as previously reported.5,7 Images were viewed under optimal CT windows for spot detection (width 200, level 100). For each spot sign, maximum attenuation and axial size, location (peripheral, defined as within ≤5 mm of hematoma margin, and central), relative attenuation compared with the ipsilateral distal supraclinoid internal carotid artery density (or basilar if infratentorial ICH), and relative attenuation compared with the superior sagittal sinus were recorded. Regions of high density within the hematoma not meeting criteria of at least double the density of background hematoma used for spot definition in PREDICT were recorded as subthreshold spots and excluded from spot sign number and SSS analyses. Differences in reader interpretation were resolved by consensus. If >1 spot was identified, characterization was performed on the largest spot. A SSS was calculated for each patient as previously published6: 1 to 2 spot signs, 1 point; ≥3 spot signs, 2 points; maximum axial dimension ≥5 mm, 1 point; maximum density ≥180 HU, 1 point.
Baseline and 24-hour NCCT hematoma volumes were analyzed with Quantomo, a computerized-planimetry software (Cybertrial Inc, Calgary, Canada),8,9 and performed by an neurologist blinded to CTA images. ICH location and presence of intraventricular hemorrhage were recorded from baseline NCCT images by the neuroradiologists.
Primary outcome was clinically significant hematoma expansion defined as ICH volume increase >6 mL or >33%.4–6,10,11 Secondary outcome measures were hematoma expansion as defined by absolute and relative volume increase expressed in milliliter and percentage, respectively.
Independent association of the SSS with primary outcome was assessed using multivariable logistic regression adjusting for potential confounders of hematoma expansion, including baseline international normalized ratio (INR), mean arterial pressure, time to baseline NCCT, ICH volume, blood glucose, intraventricular hemorrhage presence, and age.12,13 Association with secondary continuous outcomes was assessed with log-transformed multivariable linear regression controlling for the same variables.12 Diagnostic performance for prediction of the primary outcome was assessed for each SSS category, and overall area under the curve (AUC) was assessed using receiver operating characteristic analysis. Exploratory analysis was performed for patients not meeting spot criteria, proposed in the original SSS derivation study,6 and for nonanticoagulated (INR ≤1.2) patients.
Individual spot sign characteristics were examined with receiver operating characteristic analysis to determine optimal AUC for primary outcome prediction.6 Multivariable logistic regression, using generalized estimating equations to account for potential clustering of spot characteristics by site, was used to determine spot characteristics independently associated with primary outcome.
An exploratory analysis of subthreshold spots was performed to determine whether sensitivity of the spot sign could be improved. McNemar test was used to compare differences in sensitivity and specificity.
To determine whether the SSS improved prediction of primary outcome compared with individual spot characteristics, we performed a parallel analysis between SSS and spot characteristics independently associated with primary outcome. Specifically, we compared receiver operating characteristic curves between SSS and individual significant spot characteristics to determine whether there was an overall difference in primary outcome discrimination. Subsequently, multivariable logistic regression models were created, adjusting for the same potential confounders of hematoma expansion used in the SSS model; however, SSS was replaced with individual significant spot characteristics within the model. Comparison of logistic regression model discrimination was performed using the c-statistic. To further compare SSS and individual significant spot characteristics, tests for trends in proportions and continuous outcomes were assessed using the Cochrane–Armitage test and Cuzick nonparametric test, respectively.
Interobserver agreement for spot sign presence, number, density, and overall SSS was calculated using the multirater κ statistic and intraclass correlation coefficient for categorical and continuous variables, respectively. Values of κ 0.21 to 0.4, 0.41 to 0.6, 0.61 to 0.8, and 0.81 to 1 were considered fair, moderate, substantial, and nearly perfect, respectively. Intraclass correlation coefficient values of <0.4, 0.4 to 0.75, and >0.75 were considered to demonstrate poor, fair to good, and excellent agreement. Statistical significance was defined as P≤0.05 for all tests. Analysis was performed using SAS 9.2 (SAS Institute Inc, Cary, NC) and R, version 2.13.2.
Two hundred twenty-eight patients (mean±SD, 68.7±14.0 years; 57% men) met the study inclusion criteria. Baseline demographic characteristics are described in detail elsewhere.5 In brief, median (interquartile range, IQR) time from onset to baseline NCCT and CTA was 2.3 (1.5–3.8) and 2.7 (1.7–4.2) hours, respectively. Median (IQR) time from baseline to follow-up NCCT was 23.6 (19.7–25.7) hours. Median (IQR) Glasgow Coma Scale, mean arterial blood pressure, and INR were 15 (12–15), 121 (103–136) mm Hg, and 1.04 (1.00–1.12), respectively. Baseline ICH location was deep in 161 (71%), lobar in 62 (27%), and infratentorial in 5 (2%) patients. Intraventricular hemorrhage was present in 84 (37%) patients. Median (IQR) baseline and 24-hour ICH volume were 12.4 (5.8–24.5) and 15.7 (7.1–36.1) mL, respectively. In total, 73 (32%) patients had significant hematoma expansion. Median absolute hematoma expansion and relative hematoma expansion were 1.0 (0.0–5.8) mL and 9.8 (–0.1 to 35.0) %, respectively.
Spot Sign Score Validation
Multivariable logistic regression for primary outcome prediction, controlling for confounders, demonstrated that SSS was independently associated with hematoma expansion criteria (OR of 1.93 per 1-point increase of SSS, 95% CI, 1.32–2.86; P=0.001). Similarly, multivariable linear regression for secondary outcome also demonstrated independent associations with absolute and relative hematoma expansion (P<0.001 and P=0.009, respectively). Diagnostic performance of the SSS for prediction of hematoma expansion is listed in Table 1. Risk of significant, absolute, and relative hematoma expansion by SSS is listed in Table 2. Interobserver agreement for SSS was near perfect (κ=0.82; 95% CI, 0.74–0.89).
Exclusion of 3 spots not meeting the 120 HU spot sign cutoff resulted in sensitivity reduction of spot presence to 49% (95% CI, 37%–61%) with increased specificity to 86% (95% CI, 80%–91%). Neither sensitivity nor specificity was significantly changed (McNemar test, P=0.317 and P=0.083, respectively) and no significant difference in overall AUC occurred (0.68 versus 0.68; P=0.978). Exclusion of anticoagulated patients (INR >1.2; n=60) marginally increased the overall AUC for SSS prediction of hematoma expansion to 0.70 (95% CI, 0.62–0.79).
Spot Sign Characteristics
Spot sign characteristics by primary outcome are listed in Table 3. Interobserver agreement for spot presence and spot number were near perfect (κ=0.90; 95% CI, 0.91–0.99; κ=0.85; 95% CI, 0.77–0.92, respectively). Multirater agreement for spot density and size measurement was excellent (intraclass correlation coefficient, 0.92; 95% CI, 0.81–0.97; intraclass correlation coefficient, 0.83; 95% CI, 0.58–0.92, respectively).
Of all spot characteristics, spot sign number demonstrated the greatest AUC (0.68; 95% CI, 0.62–0.75; P<0.001) and was maximized when patients were categorized according to 0, 1, 2 to 3, and ≥4 spots (AUC, 0.69; 95% CI, 0.62–0.75; P<0.001). Univariate analysis did not demonstrate any associations for spot size, density, relative density measurements, or location with primary outcome. Multivariable logistic regression for primary outcome prediction with individual SSS components as covariates (spot number, size, and density), using generalized estimating equations to account for potential clustering of spot characteristics by site, demonstrated significance only for spot number (P=0.037). Effects of increasing spot number on hematoma expansion are listed in Tables 4 and 5. For patients with 0 (n=167), 1 (n=36), 2 (n=15), 3 (n=5), and ≥4 (n=5) spots, 21.5%, 52.8%, 73.3%, 40.0%, and 100% had significant hematoma expansion, respectively.
Multivariable logistic regression for primary outcome prediction, controlling for the same covariates used in the multivariable regression used to validate the SSS, demonstrated an independent association between spot number and primary outcome (OR, 2.42; 95% CI, 1.57–3.93; P<0.001). Multivariable linear regression also demonstrated significant associations with absolute and relative expansion (each P<0.001).
Exploratory analysis of subthreshold spots (n=14; median [IQR] density of 111 [102–115] HU) demonstrated that inclusion of spots with >102 HU density (n=10) increased sensitivity for prediction of hematoma expansion to 59% (95% CI, 47–70) from 51% (McNemar test, P=0.014), whereas specificity was decreased to 82% (95% CI, 75–88) from 85% (McNemar test, P=0.046).
Parallel Analysis: Spot Sign Score Compared With Spot Sign Number
Receiver operating characteristic analysis comparing SSS with spot sign number category demonstrated no significant difference in AUC for primary outcome prediction (0.69 versus 0.68; P=0.306). Multivariable logistic regression comparison of SSS and spot number models, each adjusted for the same confounders, demonstrated no significant difference in primary outcome discrimination with spot sign number compared with SSS (c-statistic: 0.796 versus 0.785, respectively; P=0.304).
Tests for trend in primary outcome risk with increasing SSS were not demonstrated among spot-positive patients (P trend=0.720). Similarly, increasing SSS was not associated with absolute and relative hematoma expansion (P trend=0.182 and 0.470, respectively). Exclusion of spots<120 HU and patients with INR >1.2 did not impact tests for trend in primary outcome (P trend=0.172). Spot number category showed significant trends for hematoma expansion risk (P=0.050), and absolute (P<0.001) and relative hematoma expansion (P<0.001).
In agreement with the original SSS derivation cohort6 and a separate prospective single-center study,14 the present multicenter study confirms that the SSS is an independent predictor of hematoma expansion. Analysis of individual spot characteristics demonstrated that only spot number was associated with hematoma expansion prediction, whereas spot density, size, and relative attenuation were not. A parallel analysis demonstrated that the overall predictive value of spot sign number was similar to the SSS. However, increasing SSS was not associated with increased risk of hematoma expansion among spot-positive patients, whereas increasing spot number alone was.
The inability to observe an association between spot density and size with hematoma expansion despite potential biological plausibility may be multifactorial. Importantly, by design, participating sites in PREDICT used local CTA protocols reflecting the current state of CTA-use worldwide. CT attenuation measurements without contrast medium alone have demonstrated large variations in absolute HU measurements in phantoms between CT vendors and acquisition parameters.15 Additional differences relating to varying contrast medium concentrations, volume, injection speed, acquisition triggering, and timing between sites potentially further contribute to variations in spot density and size. Accounting for potential clustering of spot characteristics among sites and examining relative attenuation measurements, we were unable to observe an association between spot density and size with increasing risk of expansion. Therefore, the use of spot density and size to risk stratify hematoma expansion in a multicenter setting may be impractical in the absence of protocol harmonization. Further, precise measurement of density and size may be practically challenging in the acute setting. Despite protocol heterogeneity, we were able to demonstrate a significant association between spot number and increasing hematoma expansion risk, suggesting that spot number alone provides sufficient risk stratification for expansion risk as applied to current real-world clinical practice. An advantage of spot number compared with SSS is that number should be easily assessable by clinicians in the emergency department without the need to calculate a score.
Exploratory analysis of subthreshold intrahematoma contrast densities (without density twice that of background hematoma) demonstrated significant improvement in sensitivity for hematoma expansion prediction when included within the spot definition. Prior definitions used arbitrary absolute and relative density cutoff thresholds, yet the optimal threshold remains unknown.6,7 Inclusion of any contrast density seems more intuitive because faintly appearing densities may represent early imaging of a spot before maximal opacification.4,16 Prior studies have demonstrated varying rates of spot appearance indicating variable rates of contrast leak.6,17,18 Addition of delayed imaging to routine CTA demonstrates ≈8% more ICH patients with extravasation and increases sensitivity for prediction of hematoma expansion and poor clinical outcome.6,17,19 Recently, dynamic spot imaging using CT perfusion demonstrated 21% improved spot number detection. The peak time to maximal contrast density was 50 seconds after contrast bolus injection.20 Inclusion of dynamic or delayed imaging will yield the greatest number of spots detected and improve diagnostic performance for hematoma expansion prediction. Prognostic characteristics of dynamic and delayed spots will require further characterization; however, the number of spots detected will likely remain a critical determinant for risk stratification.
Potential study limitations include different inclusion/exclusion criteria from the original SSS studies, specifically inclusion of anticoagulated patients and slight variation in spot definition compared with the original SSS derivation study. Analysis of nonanticoagulated patients and exclusion of spots not meeting the original SSS derivation definition, however, did not significantly alter results. Also, specific contrast bolus concentration and CTA timing were not mandated in the PREDICT protocol precluding a detailed analysis of potential confounders and their effect on spot characteristics. Each center was, however, experienced in the use of CTA and used protocols represented the best clinical techniques.
Multicenter external validation of the SSS demonstrates that spot number alone provides similar hematoma expansion prediction and improved hematoma expansion risk stratification compared with the SSS using first-pass CTA. Spot number should aid ICH prognostication and may be useful in patient selection for future ICH therapeutic studies.
Sources of Funding
Dr Huynh was supported by a Canadian Institutes of Health Research Masters Award. Dr Dowlatshahi was supported by a Canadian Institutes of Health Research Fellowship Award and a University of Ottawa Department of Medicine Research Salary Award. Research support funding also provided by a Sunnybrook Health Sciences Brain Sciences Program Award. The PREDICT study was funded by an unrestricted grant from NovoNordisk Canada and Canadian Stroke Consortium.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.111.000410/-/DC1.
- Received September 26, 2012.
- Accepted January 14, 2013.
- © 2013 American Heart Association, Inc.
- Brott T,
- Broderick J,
- Kothari R,
- Barsan W,
- Tomsick T,
- Sauerbeck L,
- et al
- Wada R,
- Aviv RI,
- Fox AJ,
- Sahlas DJ,
- Gladstone DJ,
- Tomlinson G,
- et al
- Demchuk AM,
- Dowlatshahi D,
- Rodriguez-Luna D,
- Molina CA,
- Blas YS,
- Dzialowski I,
- et al
- Delgado Almandoz JE,
- Yoo AJ,
- Stone MJ,
- Schaefer PW,
- Goldstein JN,
- Rosand J,
- et al
- Dowlatshahi D,
- Kosior JC,
- Idris S,
- Eesa M,
- Dickhoff P,
- Joshi M,
- et al
- Broderick JP,
- Diringer MN,
- Hill MD,
- Brun NC,
- Mayer SA,
- Steiner T,
- et al
- Li N,
- Wang Y,
- Wang W,
- Ma L,
- Xue J,
- Weissenborn K,
- et al
- Ederies A,
- Demchuk A,
- Chia T,
- Gladstone DJ,
- Dowlatshahi D,
- Bendavit G,
- et al
- d’Esterre CD,
- Chia TL,
- Jairath A,
- Lee TY,
- Symons SP,
- Aviv RI
- Chakraborty S,
- Blacquiere D,
- Lum C,
- Stotts G
- Koculym A,
- Huynh TJ,
- Jakubovic R,
- Zhang L,
- Aviv RI