Neurocrit Care DOI 10.1007/s12028-015-0154-5
ORIGINAL ARTICLE
TURN Score Predicts 90-day Outcome in Acute Ischemic Stroke Patients After IV Thrombolysis David Asuzu1 • Karin Nystro¨m2 • Joseph Schindler1,2 • Charles Wira1,2 David Greer1,2 • Janet Halliday2 • Kevin N. Sheth1,2
•
Ó Springer Science+Business Media New York 2015
Abstract Background and Purpose We developed the TURN score for predicting symptomatic intracerebral hemorrhage (sICH) after IV thrombolysis. Our purpose was to evaluate its ability to predict 90-day outcome. Methods We retrospectively analyzed data from 303 patients who received IV rt-PA during the NINDS rt-PA trial. Severe outcome was defined as 90-day modified Rankin scale (mRS) scores C5, 90-day Barthel index (BI) scores 2. Excellent outcome was defined as 90-day mRS scores B1, 90-day BI scores C95 and 90-day GOS scores = 1. Agreement between TURN and 90-day outcome was assessed by univariate logistic regression reporting odds ratios (OR) and by areas under the receiver operating characteristic curves (AUROC). TURN was also compared with 6 other scores for predicting sICH or severe outcome. Results TURN predicted 90-day mRS C5 with OR 5.73, 95 % confidence interval (3.60, 9.10), P < 0.001 and AUROC 0.83, 95 % confidence interval (0.77, 0.89).
Electronic supplementary material The online version of this article (doi:10.1007/s12028-015-0154-5) contains supplementary material, which is available to authorized users. & Kevin N. Sheth [email protected] David Asuzu [email protected] 1
Yale School of Medicine, 333 Cedar Street, New Haven, CT 06510, USA
2
Yale Department of Emergency Medicine, Division of Vascular Neurology, Division of Neurocritical Care and Emergency Neurology, Yale Department of Neurology, Yale-New Haven Hospital, New Haven, CT, USA
TURN also predicted 90-day mRS B1 with OR 5.24, 95 % confidence interval (3.43, 7.99), P < 0.001 and AUROC 0.80, 95 % confidence interval (0.74, 0.85). TURN predicted 90-day mRS C5 with OR significantly higher than DRAGON (2.30, P = 0.01), ASTRAL (1.18, P < 0.001), HAT (2.89, P = 0.05) and SEDAN (2.16, P = 0.01), and with AUROC significantly higher than SPAN-100 (0.64, P < 0.001) and SEDAN (0.71, P = 0.01). Likewise, TURN predicted 90-day mRS B1 with OR significantly higher than Stroke-TPI (2.89, P = 0.05), DRAGON (2.29, P = 0.01), ASTRAL (1.15, P < 0.001), HAT (2.71, P = 0.04) and SEDAN (2.15, P = 0.01), and with AUROC significantly higher than SPAN-100 (0.58, P < 0.001) and SEDAN (0.70, P = 0.01). Similar results were obtained using 90-day BI and 90-day GOS scores. Conclusions TURN predicted 90-day outcome with comparable or better accuracy compared to several existing clinical scores. Keywords IV thrombolysis Severe outcomes Predictive scores Intracerebral hemorrhage
Introduction Stroke remains a leading cause of mortality and morbidity in the United States. The National Institute of Neurological Disorders and Stroke (NINDS) trial demonstrated improved 90-day outcomes in ischemic stroke patients treated with IV recombinant tissue plasminogen activator (rt-PA) [1]. Based on the results of this trial, the United States Food and Drug Administration (FDA) approved IV rt-PA for acute ischemic stroke treatment, and both the American Heart Association and the American Academy of Neurology subsequently endorsed IV rt-PA use in selected
123
Neurocrit Care
patients [2, 3]. Other trials have confirmed improved 90-day outcomes for acute ischemic stroke patients treated with IV rt-PA, including the European Cooperative Acute Stroke Study (ECASS) trials I–III [4–6] and the Safe Implementation of Thrombolysis in Stroke-Monitoring Study (SITS-MOST) [7]. IV rt-PA use is limited by an increased risk for symptomatic intracerebral hemorrhage (sICH) [8] and by other factors such as delayed presentation after stroke. Clinical tools for predicting sICH are needed to selectively identify patients who may benefit from IV rt-PA therapy without a substantially increased risk for sICH. We have described the TURN score (thrombolysis risk using mRS and NIHSS), a simple clinical tool for predicting sICH risk after IV rt-PA therapy based on two clinical variables: prestroke modified Rankin scale (mRS) score and admission National Institutes of Health Stroke Scale (NIHSS) score [9]. sICH likely contributes directly to severe 90-day outcomes, as the prevalence of both are concordantly increased by the presence of the same risk factors such as hyperglycemia [10]. We therefore hypothesized that our simple clinical tool for predicting sICH also predicts severe 90-day outcomes, and further tested its ability to predict excellent 90-day outcome in an independent patient dataset.
Methods Patient Data We retrospectively analyzed data from the NINDA rt-PA Stroke Study, a multicenter, prospective, double-blind, placebo-controlled randomized trial from January 1991 to October 1994 [1]. 9 patients were excluded due to incomplete data. Data from the NINDS trial were purchased from the National Technical Information Service (http:// www.ntis.gov/) using internal funds from the Yale Department of Neurology. Clinical data were converted to Microsoft Excel format using Statistical Analysis System software (SAS Institute Inc, Cary, NC) and decoded using instructions included in the CD-ROM and in accordance with published guidelines [11]. This study was approved by the Yale Human Investigation Committee and the Yale Human Research Protection Program. Written informed consent was not required for reviewing retrospective deidentified patient data. Outcome Measures Severe 90-day outcome was defined according to previously published studies as a 90-day mRS score C5, a Barthel index (BI) score 2, and excellent 90-day outcome was defined as a 90-day mRS score B1, a BI score C95 or a GOS score = 1 [12–14]. The mRS score measures overall independence after a stroke with moderate to good interobserver agreement [15, 16]. A score of 0 indicates no symptoms, a score of 5 indicates severe disability, and a score of 6 indicates death. The BI measures ability to perform activities of daily living after a stroke [17]. Patients able to perform all activities of daily living such as eating, bathing, walking and using the toilet receive a score of 100. The GOS is a global assessment of function [18], and ranges from 1 to 5 with a score of 1 indicating mild disability and a score of 5 death. Clinical Scores We calculated TURN for each patient as previously described: TURN = -4.65 + (mRS * 0.27) + (NIHSS * 0.10) [9]. We used the orthogonal function (-TURN) to predict excellent outcome. We previously compared eight scores predicting symptomatic intracranial hemorrhage in ischemic stroke patients treated with IV rt-PA [19]. We calculated six of those scores for each patient using the NINDS study dataset: Stroke-Thrombolytic Predictive Instrument (Stroke-TPI), DRAGON, Stroke Prognostication using Age and NIH Stroke Scale-100 (SPAN-100), Acute Stroke Registry and Analysis of Lausanne (ASTRAL), hemorrhage after thrombolysis (HAT) and SEDAN. Detailed derivations of each score have been published elsewhere [20–24] and summarized in Supplemental Table S7 [19]. We excluded Post-thrombolysis Risk Score (PRS) and Safe Implementation of Treatments in Stroke Symptomatic Intracerebral Hemorrhage (SITS-ICH) due to unavailability of required data. Statistical Analyses Association between sICH and severe 90-day outcome was assessed by univariate logistic regression reporting odds ratios, and by linear weighted Kappa [25]. Association between measures of 90-day outcome and clinical scores was assessed using univariate logistic regression reporting odds ratios. Goodness-of-fit was assessed using Hosmer– Lemeshow statistics, with P > 0.05 considered a statistically significant indicator of goodness-of-fit [26]. Overall accuracy was assessed using nonparametric areas under the Receiver Operating Characteristic curve (AUROC). Standard errors were calculated by the DeLong method [27]. Clinical scores were compared using two-sample t-tests with unequal variance using Welch’s approximation for degrees of freedom [28]. P values less than 0.05 (twotailed) were considered statistically significant. GraphPad
Neurocrit Care
Prism 6.0 software was used for statistical figures (GraphPad Software, La Jolla, California USA). Statistical analyses were performed using STATA 13 software package (StataCorp LP, College Station, Texas).
Results First, we assessed for correlation between sICH and 90-day severe outcome defined as mRS C5, BI 2 using univariate logistic regression with sICH as the independent variable. sICH predicted 90-day mRS C5 with odds ratio 10.1, 95 % confidence interval (4.0, 25.6), P < 0.001. sICH also predicted 90-day BI 2 with odds ratio 14.5, 95 % confidence interval (4.2, 49.7), P < 0.001 (Table 1). We verified agreement between sICH and 90-day severe outcome using linear weighted Kappa [25]. We found 80 % agreement between sICH and 90-day mRS C5, j = 0.27, P < 0.0001, 72 % agreement with 90-day BI 2, j = 0.21, P < 0.0001 (Table 1). Therefore, we established a direct correlation between sICH and 90-day severe outcome, with the strongest correlation between sICH and 90-day mRS C5. We next assessed the ability of the sICH predictive score TURN to predict 90-day severe outcome using univariate logistic regression. TURN predicted mRS C 5 with odds ratio 5.73, 95 % confidence interval (3.60, 9.10), P < 0.001. Goodness-of-fit was assessed as Hosmer– Lemeshow v2 of 3.63 using 10 groups, P = 0.89, demonstrating good agreement. We found an AUROC of 0.83, 95 % confidence interval (0.77, 0.89), confirming good overall accuracy. We verified the ability of TURN to predict 90-day severe outcome using the BI and the Glasgow Outcome Score (GOS). TURN predicted 90-day BI 2 with odds ratio 5.17, 95 % confidence interval (3.42, 7.80), P < 0.001. Goodness-of-fit analysis yielded a Hosmer–
Lemeshow v2 of 7.93 for 90-day BI < 60 using 10 groups, P = 0.44, and Hosmer–Lemeshow v2 of 9.0 for 90-day GOS >2 using 10 groups, P = 0.34, indicating good agreement. TURN also predicted 90-day BI 2 with AUROC of 0.81, 95 % confidence interval (0.76, 0.86) verifying good overall accuracy. Next, we compared TURN to six other clinical scores including Stroke-TPI, DRAGON, SPAN-100, ASTRAL, HAT, and SEDAN using two-sample t-tests with unequal variance and Welch’s approximation for degrees of freedom. TURN predicted 90-day mRS C5 with odds ratio significantly higher than for DRAGON, ASTRAL, HAT, and SEDAN (P < 0.05, Supplemental Table S1; Fig. 1a). Similar results were obtained for 90-day BI 2 (Supplemental Table S3; Fig. 1e). None of the 6 scores tested predicted 90-day severe outcomes with higher odds ratios than TURN. We also compared prediction accuracy for 90-day outcomes between TURN and existing scores. AUROC for TURN using 90-day mRS C5 was significantly higher than SPAN-100 and SEDAN (P < 0.05, Supplemental Table S1; Fig. 1b). Similar results were obtained using 90-day BI 2 (Supplemental Tables S2, S3; Fig. 1d, f). None of the other scores yielded an AUROC significantly higher than TURN, demonstrating its strength of association and predictive accuracy compared to existing scores. Next, we examined the distribution of 90-day outcomes in rt-PA-treated patients in the NINDS trial dataset and found an approximate bimodal distribution (Fig. 2). 42.5 % of patients had excellent outcomes (90-day mRS B1) and another 23.4 % had severe adverse outcomes (90-day mRS C5). Clinical scores predicting adverse outcomes may thus predict excellent outcomes using their orthogonal functions. To test this, we evaluated the ability of the orthogonal TURN score (-TURN) to predict 90-day excellent outcome using univariate logistic regression with -TURN as the independent variable and 90-day mRS B1 as the dependent variable. -TURN predicted 90-day mRS B 1 with odds ratio 5.24, 95 % confidence interval (3.43, 7.99), P < 0.001. Likewise, -TURN predicted 90-day BI C95
Table 1 Agreement between symptomatic intracerebral hemorrhage (sICH) and severe 90-day outcome Odds ratio
95 % confidence interval
z
P > |z|
Agreement (%)
j
Z
P>Z
90-day mRS C5 90-day Barthel index
ORIGINAL ARTICLE
TURN Score Predicts 90-day Outcome in Acute Ischemic Stroke Patients After IV Thrombolysis David Asuzu1 • Karin Nystro¨m2 • Joseph Schindler1,2 • Charles Wira1,2 David Greer1,2 • Janet Halliday2 • Kevin N. Sheth1,2
•
Ó Springer Science+Business Media New York 2015
Abstract Background and Purpose We developed the TURN score for predicting symptomatic intracerebral hemorrhage (sICH) after IV thrombolysis. Our purpose was to evaluate its ability to predict 90-day outcome. Methods We retrospectively analyzed data from 303 patients who received IV rt-PA during the NINDS rt-PA trial. Severe outcome was defined as 90-day modified Rankin scale (mRS) scores C5, 90-day Barthel index (BI) scores 2. Excellent outcome was defined as 90-day mRS scores B1, 90-day BI scores C95 and 90-day GOS scores = 1. Agreement between TURN and 90-day outcome was assessed by univariate logistic regression reporting odds ratios (OR) and by areas under the receiver operating characteristic curves (AUROC). TURN was also compared with 6 other scores for predicting sICH or severe outcome. Results TURN predicted 90-day mRS C5 with OR 5.73, 95 % confidence interval (3.60, 9.10), P < 0.001 and AUROC 0.83, 95 % confidence interval (0.77, 0.89).
Electronic supplementary material The online version of this article (doi:10.1007/s12028-015-0154-5) contains supplementary material, which is available to authorized users. & Kevin N. Sheth [email protected] David Asuzu [email protected] 1
Yale School of Medicine, 333 Cedar Street, New Haven, CT 06510, USA
2
Yale Department of Emergency Medicine, Division of Vascular Neurology, Division of Neurocritical Care and Emergency Neurology, Yale Department of Neurology, Yale-New Haven Hospital, New Haven, CT, USA
TURN also predicted 90-day mRS B1 with OR 5.24, 95 % confidence interval (3.43, 7.99), P < 0.001 and AUROC 0.80, 95 % confidence interval (0.74, 0.85). TURN predicted 90-day mRS C5 with OR significantly higher than DRAGON (2.30, P = 0.01), ASTRAL (1.18, P < 0.001), HAT (2.89, P = 0.05) and SEDAN (2.16, P = 0.01), and with AUROC significantly higher than SPAN-100 (0.64, P < 0.001) and SEDAN (0.71, P = 0.01). Likewise, TURN predicted 90-day mRS B1 with OR significantly higher than Stroke-TPI (2.89, P = 0.05), DRAGON (2.29, P = 0.01), ASTRAL (1.15, P < 0.001), HAT (2.71, P = 0.04) and SEDAN (2.15, P = 0.01), and with AUROC significantly higher than SPAN-100 (0.58, P < 0.001) and SEDAN (0.70, P = 0.01). Similar results were obtained using 90-day BI and 90-day GOS scores. Conclusions TURN predicted 90-day outcome with comparable or better accuracy compared to several existing clinical scores. Keywords IV thrombolysis Severe outcomes Predictive scores Intracerebral hemorrhage
Introduction Stroke remains a leading cause of mortality and morbidity in the United States. The National Institute of Neurological Disorders and Stroke (NINDS) trial demonstrated improved 90-day outcomes in ischemic stroke patients treated with IV recombinant tissue plasminogen activator (rt-PA) [1]. Based on the results of this trial, the United States Food and Drug Administration (FDA) approved IV rt-PA for acute ischemic stroke treatment, and both the American Heart Association and the American Academy of Neurology subsequently endorsed IV rt-PA use in selected
123
Neurocrit Care
patients [2, 3]. Other trials have confirmed improved 90-day outcomes for acute ischemic stroke patients treated with IV rt-PA, including the European Cooperative Acute Stroke Study (ECASS) trials I–III [4–6] and the Safe Implementation of Thrombolysis in Stroke-Monitoring Study (SITS-MOST) [7]. IV rt-PA use is limited by an increased risk for symptomatic intracerebral hemorrhage (sICH) [8] and by other factors such as delayed presentation after stroke. Clinical tools for predicting sICH are needed to selectively identify patients who may benefit from IV rt-PA therapy without a substantially increased risk for sICH. We have described the TURN score (thrombolysis risk using mRS and NIHSS), a simple clinical tool for predicting sICH risk after IV rt-PA therapy based on two clinical variables: prestroke modified Rankin scale (mRS) score and admission National Institutes of Health Stroke Scale (NIHSS) score [9]. sICH likely contributes directly to severe 90-day outcomes, as the prevalence of both are concordantly increased by the presence of the same risk factors such as hyperglycemia [10]. We therefore hypothesized that our simple clinical tool for predicting sICH also predicts severe 90-day outcomes, and further tested its ability to predict excellent 90-day outcome in an independent patient dataset.
Methods Patient Data We retrospectively analyzed data from the NINDA rt-PA Stroke Study, a multicenter, prospective, double-blind, placebo-controlled randomized trial from January 1991 to October 1994 [1]. 9 patients were excluded due to incomplete data. Data from the NINDS trial were purchased from the National Technical Information Service (http:// www.ntis.gov/) using internal funds from the Yale Department of Neurology. Clinical data were converted to Microsoft Excel format using Statistical Analysis System software (SAS Institute Inc, Cary, NC) and decoded using instructions included in the CD-ROM and in accordance with published guidelines [11]. This study was approved by the Yale Human Investigation Committee and the Yale Human Research Protection Program. Written informed consent was not required for reviewing retrospective deidentified patient data. Outcome Measures Severe 90-day outcome was defined according to previously published studies as a 90-day mRS score C5, a Barthel index (BI) score 2, and excellent 90-day outcome was defined as a 90-day mRS score B1, a BI score C95 or a GOS score = 1 [12–14]. The mRS score measures overall independence after a stroke with moderate to good interobserver agreement [15, 16]. A score of 0 indicates no symptoms, a score of 5 indicates severe disability, and a score of 6 indicates death. The BI measures ability to perform activities of daily living after a stroke [17]. Patients able to perform all activities of daily living such as eating, bathing, walking and using the toilet receive a score of 100. The GOS is a global assessment of function [18], and ranges from 1 to 5 with a score of 1 indicating mild disability and a score of 5 death. Clinical Scores We calculated TURN for each patient as previously described: TURN = -4.65 + (mRS * 0.27) + (NIHSS * 0.10) [9]. We used the orthogonal function (-TURN) to predict excellent outcome. We previously compared eight scores predicting symptomatic intracranial hemorrhage in ischemic stroke patients treated with IV rt-PA [19]. We calculated six of those scores for each patient using the NINDS study dataset: Stroke-Thrombolytic Predictive Instrument (Stroke-TPI), DRAGON, Stroke Prognostication using Age and NIH Stroke Scale-100 (SPAN-100), Acute Stroke Registry and Analysis of Lausanne (ASTRAL), hemorrhage after thrombolysis (HAT) and SEDAN. Detailed derivations of each score have been published elsewhere [20–24] and summarized in Supplemental Table S7 [19]. We excluded Post-thrombolysis Risk Score (PRS) and Safe Implementation of Treatments in Stroke Symptomatic Intracerebral Hemorrhage (SITS-ICH) due to unavailability of required data. Statistical Analyses Association between sICH and severe 90-day outcome was assessed by univariate logistic regression reporting odds ratios, and by linear weighted Kappa [25]. Association between measures of 90-day outcome and clinical scores was assessed using univariate logistic regression reporting odds ratios. Goodness-of-fit was assessed using Hosmer– Lemeshow statistics, with P > 0.05 considered a statistically significant indicator of goodness-of-fit [26]. Overall accuracy was assessed using nonparametric areas under the Receiver Operating Characteristic curve (AUROC). Standard errors were calculated by the DeLong method [27]. Clinical scores were compared using two-sample t-tests with unequal variance using Welch’s approximation for degrees of freedom [28]. P values less than 0.05 (twotailed) were considered statistically significant. GraphPad
Neurocrit Care
Prism 6.0 software was used for statistical figures (GraphPad Software, La Jolla, California USA). Statistical analyses were performed using STATA 13 software package (StataCorp LP, College Station, Texas).
Results First, we assessed for correlation between sICH and 90-day severe outcome defined as mRS C5, BI 2 using univariate logistic regression with sICH as the independent variable. sICH predicted 90-day mRS C5 with odds ratio 10.1, 95 % confidence interval (4.0, 25.6), P < 0.001. sICH also predicted 90-day BI 2 with odds ratio 14.5, 95 % confidence interval (4.2, 49.7), P < 0.001 (Table 1). We verified agreement between sICH and 90-day severe outcome using linear weighted Kappa [25]. We found 80 % agreement between sICH and 90-day mRS C5, j = 0.27, P < 0.0001, 72 % agreement with 90-day BI 2, j = 0.21, P < 0.0001 (Table 1). Therefore, we established a direct correlation between sICH and 90-day severe outcome, with the strongest correlation between sICH and 90-day mRS C5. We next assessed the ability of the sICH predictive score TURN to predict 90-day severe outcome using univariate logistic regression. TURN predicted mRS C 5 with odds ratio 5.73, 95 % confidence interval (3.60, 9.10), P < 0.001. Goodness-of-fit was assessed as Hosmer– Lemeshow v2 of 3.63 using 10 groups, P = 0.89, demonstrating good agreement. We found an AUROC of 0.83, 95 % confidence interval (0.77, 0.89), confirming good overall accuracy. We verified the ability of TURN to predict 90-day severe outcome using the BI and the Glasgow Outcome Score (GOS). TURN predicted 90-day BI 2 with odds ratio 5.17, 95 % confidence interval (3.42, 7.80), P < 0.001. Goodness-of-fit analysis yielded a Hosmer–
Lemeshow v2 of 7.93 for 90-day BI < 60 using 10 groups, P = 0.44, and Hosmer–Lemeshow v2 of 9.0 for 90-day GOS >2 using 10 groups, P = 0.34, indicating good agreement. TURN also predicted 90-day BI 2 with AUROC of 0.81, 95 % confidence interval (0.76, 0.86) verifying good overall accuracy. Next, we compared TURN to six other clinical scores including Stroke-TPI, DRAGON, SPAN-100, ASTRAL, HAT, and SEDAN using two-sample t-tests with unequal variance and Welch’s approximation for degrees of freedom. TURN predicted 90-day mRS C5 with odds ratio significantly higher than for DRAGON, ASTRAL, HAT, and SEDAN (P < 0.05, Supplemental Table S1; Fig. 1a). Similar results were obtained for 90-day BI 2 (Supplemental Table S3; Fig. 1e). None of the 6 scores tested predicted 90-day severe outcomes with higher odds ratios than TURN. We also compared prediction accuracy for 90-day outcomes between TURN and existing scores. AUROC for TURN using 90-day mRS C5 was significantly higher than SPAN-100 and SEDAN (P < 0.05, Supplemental Table S1; Fig. 1b). Similar results were obtained using 90-day BI 2 (Supplemental Tables S2, S3; Fig. 1d, f). None of the other scores yielded an AUROC significantly higher than TURN, demonstrating its strength of association and predictive accuracy compared to existing scores. Next, we examined the distribution of 90-day outcomes in rt-PA-treated patients in the NINDS trial dataset and found an approximate bimodal distribution (Fig. 2). 42.5 % of patients had excellent outcomes (90-day mRS B1) and another 23.4 % had severe adverse outcomes (90-day mRS C5). Clinical scores predicting adverse outcomes may thus predict excellent outcomes using their orthogonal functions. To test this, we evaluated the ability of the orthogonal TURN score (-TURN) to predict 90-day excellent outcome using univariate logistic regression with -TURN as the independent variable and 90-day mRS B1 as the dependent variable. -TURN predicted 90-day mRS B 1 with odds ratio 5.24, 95 % confidence interval (3.43, 7.99), P < 0.001. Likewise, -TURN predicted 90-day BI C95
Table 1 Agreement between symptomatic intracerebral hemorrhage (sICH) and severe 90-day outcome Odds ratio
95 % confidence interval
z
P > |z|
Agreement (%)
j
Z
P>Z
90-day mRS C5 90-day Barthel index
