Predicting hemorrhagic transformation in patients not submitted to reperfusion therapies

Research Article Predicting hemorrhagic transformation in patients not submitted to reperfusion therapies , Joao Brainer Clares de Andrade, MD,* † Jay P. Mohr, MD,†,1 Fabricio Oliveira Lima, MD, PhD,‡,§,2 Joao Jos
e de Freitas Carvalho, MD, MSc,§,3 Victor Aguiar Evangelista de Farias, MSc,{,4 Jamary Oliveira-Filho, MD, PhD,#,5 Octavio Marques Pontes-Neto, MD, PhD,k,6 Rodrigo Bazan, MD, PhD,**,7 Kristel Larisa Back Merida, MD,††,8 Luisa Franciscato, MD,k,9 Matheus Mendes Pires, MD,#,10 Gabriel Pinheiro Modolo, MD,**,11 , Mayara Silva Marques, MD,††,8 Renata Carolina Acri Nunes Miranda,* 12 and , Gisele Sampaio Silva, MD, MPH, PhD* 13 Background: Well studied in patients with ischemic stroke after reperfusion therapies (RT), hemorrhagic transformation (HT) is also common in patients not treated with RT and can lead to disability even in initially asymptomatic cases. The best predictors of HT in patients not treated with RT are not well established. Therefore, we aimed to identify predictors of HT in patients not submitted to RT and create a user-friendly predictive score (PROpHET). Material and Methods: Patients admitted to a Comprehensive Stroke Center from 2015 to 2017 were prospectively evaluated and randomly selected to the derivation cohort. A multivariable logistic regression modeling was built to produce a predictive grading score for HT. The external From the *Universidade Federal de S~ao Paulo, Sao Paulo, SP, Brazil; †Columbia University, Doris and Stanley Tananbaum Stroke Center, 710 W 168th St. Neurological Institute of New York. 6TH Floor. NI 614. ZIP 10032. New York City, NY, USA; ‡Universidade de Fortaleza, Fortaleza, Cear
a, Brazil; §Hospital Geral de Fortaleza, Ceara, Brazil; {Universidade Federal do Ceara, Brazil; #Universidade Federal da Bahia, Brazil; kUniversidade de S~ ao Paulo, Ribeir~ao Preto, Brazil; **Universidade Estadual Paulista, Brazil; and ††Hospital Instituto de Neurologia de Curitiba, Brazil. Received January 14, 2020; revision received April 22, 2020; accepted May 5, 2020. Corresponding author. E-mail: joao.brainer@unifesp.br. 1 Columbia University, Doris and Stanley Tanabaum Stroke Center, 710W 168th St. Neurological Institute of New York. 6th Floor. Phone +12123058033. 2 Hospital Geral de Fortaleza, Ceara, Brazil, Rua Avila Goulart, 900, Fortaleza, CE, Brazil. Phone +558531013209. 3 Hospital Geral de Fortaleza, Ceara, Brazil, Rua Avila Goulart, 900, Fortaleza, CE, Brazil, Phone +558531013209. 4 Universidade Federal do Cear
a, Campus de Quixada. Quixada, CE, Brazil, Phone: +5585981559918. 5 Praça XV de Novembro, SN. Salvador, BA, Brazil. Phone: +55713283-5577. 6 Hospital das Clínicas, Floor 4th. Ribeir~ao Preto, SP, Brazil. Phone: +551636021000. 7 Av. Professor M
ario Rubens Guimar~aes Montenegro. Botucatu, SP. Brazil, Phone: +551438801220. 8 R. Jeremias Maciel Perretto, 300. Curitiba, PR, Brazil. Phone: +554130288545. 9 Hospital das Clínicas, Floor 4th. Ribeir~ao Preto, SP, Brazil. Phone: +551636021000. 10 Praça XV de novembro, SN. Salvador, BA, Brazil. Phone: +55713283-5577. 11 Av. Professor M
ario Rubens Guimar~aes Montenegro. Botucatu, SP. Brazil. Phone: +551436021000. 12 Hospital Israelita Brasileiro Albert Einstein. Rua Albert Einstein, 627. S~ ao Paulo, SP, Brazil . 13 Universidade Federal de Sao Paulo, Sao Paulo, Brazil. Rua Napoleao de Barros, 715. Sao Paulo, SP, Brazil. Phone +551150899200. 1052-3057/$ - see front matter © 2020 Elsevier Inc. All rights reserved. https://doi.org/10.1016/j.jstrokecerebrovasdis.2020.104940 Journal of Stroke and Cerebrovascular Diseases, Vol. 29, No. 00 (), 2020: 104940 1 J.B.C. DE ANDRADE ET AL. 2 validation was assessed using datasets from 7 Comprehensive Stroke Centers using the area under the receiver operating characteristic curve (AUROC). Results: In the derivation group, 448 patients were included in the final analysis. The validation group included 2,683 patients. The score derived from significant predictors of HT in the multivariate logistic regression analysis was male sex (1 point), ASPECTS  7 (2 points), presence of leukoaraiosis (1 point), hyperdense cerebral middle artery sign (1 point), glycemia at admission 180 mg/dL (1 point), cardioembolism (1 point) and lacunar syndrome (-3 points) as a protective factor. The grading score ranges from -3 to 7. A Score 3 had 78.2% sensitivity and 75% specificity, and AUROC of 0.82 for all cases of HT. In the validation cohort, our score had an AUROC of 0.83. Conclusions: The PROpHET is a simple, quick, cost-free, and easyto-perform tool that allows risk stratification of HT in patients not submitted to RT. A cost-free computerized version of our score is available online with a userfriendly interface. Keywords: Ischemic stroke—Hemorrhagic transformation—Complication— Vascular neurology © 2020 Elsevier Inc. All rights reserved. Background Hemorrhagic transformation (HT) a complication of ischemic stroke strongly associated with neurological deterioration, short and long-term disability, as well as death1. Much better studied in patients treated with reperfusion therapies (RT), HT is no less important in patients not treated with RT with an incidence that reaches 43%, with approximately 20% being associated with neurological worsening1-3. Furthermore, these patients represent the majority of stroke patients since approximately 80% of ischemic stroke patients in developed countries and up to 98% in low and middle-income countries do not receive RT4. Therefore, HT is a frequent and potentially dangerous complication5 that deserves better characterization in patients not submitted to RT6. Predicting HT in patients not submitted to RT may be helpful for establishing the timing for initiation of acutestage therapies like anticoagulation, as well as for monitoring and prognostication7. All previously published scales of HT prediction in patients with acute ischemic stroke were developed focusing on patients treated with RT6,8. Molecular effects of thrombolytics and the importance of the reperfusion itself probably have a determinant influence upon the risk of HT, making such scales inappropriate for patients not treated with RT3,9. In addition, in eligible patients for RT we found some differences such as time from the symptoms onset, size of the ischemia, influence of blood pressure at admission and the ratio of recanalization in comparison to patients not eligible for these therapies  some authors have pointed out that as a limitation for an accurate prediction of HT based on data from only these treated patients2,3,6,8,9. The purpose of this study was therefore to identify predictors of HT in patients not submitted to RT and develop a user-friendly predictive score of hemorrhagic transformation in patients not treated with RT (PROpHET), evaluating clinical, laboratory and neuroimaging findings. Methods Patients All patients with a diagnosis of ischemic stroke not treated with RT admitted to a Comprehensive Stroke Center from February 2015 to October 2017 were eligible. We prospectively generated a random sample of these patients to be followed over time. At each admission, a random number generator was used to determine whether the patient would be included in the derivation cohort. The validation cohort was comprised of all other patients not included in the derivation cohort during the same time period in our institution and of patients admitted to other six Comprehensive Stroke Centers (CSC). For these patients, data were collected retrospectively from medical records from 2014 to 2018, and the Institutional Review Board (IRB) waived the need for informed consent. We adopted the Oxfordshire Community Stroke Project classification system for subtypes of ischemic stroke. Included in that classification, lacunar Stroke has been widely reported in the literature as a protective factor against HT3,8,9. Lacunar syndrome was defined as the sudden or gradual onset of a focal neurological deficit lasting > 24 h including pure motor hemiparesis, pure sensory stroke, sensorimotor stroke, ataxic hemiparesis or dysarthria-clumsy hand.. We excluded patients who did not have both a head CT scan at admission and a followup neuroimaging within 2 to 7 days. We assessed the etiology of the events based on the Trial of Org 10172 in Acute Stroke Treatment - Stop Stroke Study (TOAST SSS). Two board-certified vascular neurologists completed and supervised this assessment. Patients with incomplete evaluation were classified as ‘Undetermined Cause’. At hospital transfer/discharge, patients from our derivation cohort had their functionality assessed by our local trained nursing staff using the modified Rankin score. PREDICTIVE SCORE OF HEMORRHAGIC TRANSFORMATION Hemorrhagic transformation HT was diagnosed using international imaging criteria10 (European Cooperative Acute Stroke Study, ECASS II trial) when evidence of blood or hemoglobin products within the recent ischemic area was present on neuroimaging within seven days of admission in both derivation and validation groups. In our derivation cohort, a radiological subgroup classification of HT was also assessed based on the ECASS II trial10. Hemorrhagic infarction 1 (HI1) was defined as small petechiae along the margins of the infarct; hemorrhagic infarction 2 (HI2) as confluent petechiae within the infarcted area but no space-occupying effect; parenchymal hemorrhage (PH1) as blood clots in 30% or less of the infarcted area with some slight spaceoccupying effect; and parenchymal hemorrhage (PH2) as blood clots in more than 30% of the infarcted area with substantial space-occupying effect. Symptomatic HT was defined as blood at any site in the brain on the CT scan or MRI and documentation by the clinical team of neurological deterioration causing a decrease in the NIHSS score of 4 or more points5,10. Neuroimaging All patients underwent computed tomography (CT) at admission and a follow-up neuroimaging within 2 to 7 days as per the institution protocol. Follow-up images were performed using magnetic resonance imaging (MRI) in 213 (43.3%) patients in the derivation cohort. All neuroimages were evaluated by radiologists not involved in patient care and not aware of the clinical syndrome or functional status of the patients. The Alberta Stroke Program Early CT score (ASPECTS), the presence of a hyperdense MCA sign and of leukoaraiosis were a part of the standard radiology report in our institution. The hyperdense MSA sign was assessed by measurements of the absolute attenuation of the affected and normal vessels (absolute density of the affected MCA of >43 Hounsfield units and the MCA ratio of >1.2 on a non-contrast CT scan were considered as a positive hyperdense MCA sign)6. The ASPECTS was performed by trained radiologists. In the validation group, both CT and MRI were also adopted to determine the presence of HT within 7 days from hospital admission. Patients with a visible hypodensity on not eligible areas (posterior circulation) for ASPECTS at admission were excluded from the final analysis. Patients without any visible hypodensity at admission were classified as having ASPECTS 10. Statistical analysis Descriptive statistics were used to report patients’ characteristics. Appropriateness of parametric testing was assessed with the Kolmogorov-Smirnov test. The independent samples t test was used to compare means between patients with and without HT. Nonparametric 3 data were compared using the Mann-Whitney test. Categorical variables were compared with the Chi-square test. Variables that had an association with HT with a p-value <0.1 were selected for multivariable analysis. A multivariable model was built using the likelihood ratio test for comparison between models. The calibration of the model was assessed with the Hosmer-Lemeshow test. Independent predictive factors identified in logistic regression analyses were then used to produce a predictive grading score for HT. The score of each variable was defined based on coefficients of the multivariable logistic equation by rounding to the nearest integer. Bootstrapping was used to reduce bias in performance estimates. We assessed the discrimination of the score using the area under the receiver operating characteristic (AUROC) curve. The optimal cut-point of the PROpHET was defined using the Youden Index. The AUROC of derivation and validation cohorts were compared11. We assessed the AUROC of PROpHET and published predictive scores of HT in the derivation and validation cohorts from our center in Brazil (N = 986) due to the large availability of data to complete the validation. These scores were then compared11 in terms of their accuracy. Statistical analysis was performed with SPSS 25 software (Chicago, Ill., USA) and MedCalc 19.0.4. Standard protocol approvals, registrations, patient consents and data availability The hospital IRB approved this study. All the national ethical requirements were observed to support the research. In the derivation group, we obtained informed consent from the patient or next of kin. The IRB waived the need for informed consent for the validation group as the data collection was retrospective. All data used in the analyses are presented in the tables and figures. Results From 2,432 patients with acute ischemic stroke admitted to our institution from February 2015 to October 2017, 577 underwent reperfusion therapy, 734 were excluded due to inadequate imaging or refusal of consent, and 448 were prospectively randomly selected for the derivation cohort. A total of 2,683 patients were retrospectively selected for the validation group, 538 patients from our institution and 2,145 from the other six CSC. For the 448 patients included in the derivation cohort, the incidence of HT was 21.2% (N = 95, 95% CI 17.625.2%), and 9.8% (N = 44/448) were symptomatic HT. Demographic and baseline characteristics of patients with and without HT are shown in Table 1. The incidence of HT among patients included in our validation cohort (N = 2,683) was 13.15% (N = 353/2,683). All patients were evaluated with a follow-up neuroimaging within 2 to 7 days (3 [2,4] days in patients with HT versus 3 [2, 5] in patients without HT, p = 0.6). CT scan J.B.C. DE ANDRADE ET AL. 4 Table 1. Baseline clinical characteristics of patients in the derivation group Variable Overall N = 448 HT group N = 95 No HT N = 353 p-value Age, years, mean (SD) Male sex, % Clinical presentation Baseline NIHSS score, points, median [IQR] LACS, % TACS, % Cortical symptoms, % Neuroimaging ASPECTS on admission, median [IQR] Hyperdense cerebral artery sign, % Leukoaraiosis, % MRI performed, % Petechial (types I and II)10, % Parenchymatous hematoma type I10, % Parenchymatous hematoma type II10, % BP and glucose level Glycemia (mg/dL) median [IQR] Glycemia 180 mg/dL, % SBP (mmHg), mean (SD) DBP (mmHg), mean (SD) Laboratory parameters Hemoglobin (g/dL), median [IQR] WBC (/mm3), median [IQR] 66.7 (14.4) 56 63 (13.4) 71.5 65 (15.6) 51.8 0.17 0.001 14 [7,20] 17.2 48.4 62.2 18 [14,21] 2.1 66.3 76.4 12 [6,19] 21.4 43.6 59.5 < 0.001 < 0.001 < 0.001 0.01 8 [6,9] 12.7 45.5 42.8 - 9 [7,10] 9 45 46.2 - < 0.001 < 0.001 0.72 0.007 - 126 [106,170] 21.6 155 (30.5) 85 (19.1) 7 [6,8] 26.3 47.4 30.5 58 42 27.4 14.7 148 [110,221] 32.6 147 (26.5) 85 (16) 119 [106,176] 18.7 156 (29.7) 85 (18.2) 0.008 0.005 0.01 0.97 13.4 [12.3,14.4] 10220 [8160, 12385] 221350 [184800, 278300] 1.08 [1.04,1.14] 1.07 [0.96,1.18] 1 [0.8,1.3] 35 [27,52] 181.6 (48.8) 115.8 (42.1) 125 (69.6) 13.6 [12.9,14.8] 10570 [8131, 13840] 205300 [192800, 258850] 1.1 [1.03,1.15] 1.13 [1.06,1.36] 1 [0.8,1.4] 39 [25,68] 177.6 (55.1) 113 (44.2) 124 (64.4) 13.6 [12.4,14.6] 9649 [7736, 12000] 212100 [184650, 259050] 1.08 [1.03,1.14] 1.03 [0.97,1.18] 0.9 [0.8,1.27] 33 [25,45] 183.2 (51.1) 117.8 (46) 135.2 (73.9) 0.26 0.02 0.14 0.44 0.78 0.27 0.39 0.42 0.23 32 71 15.6 28.1 3.6 1.6 41 66.3 25.2 28.4 3.1 1.1 29.4 72.2 13 28 3.7 1.7 0.03 0.25 0.006 0.99 0.99 0.99 22.5 38.9 17.8 < 0.001 12 [5,20] 23.4 7.6 16 [14,40] 5.2 14.7 12 [4,20] 28.6 5.6 < 0.001 < 0.001 0.007 Platelets, median [IQR] PTT (s), median [IQR] aPTT (s), median [IQR] Creatinine (mg/dL), median [IQR] Ureia (mg/dL) median [IQR] Total cholesterol (mg/dL), mean (SD) LDL cholesterol (mg/dL), mean (SD) Triglycerides (mg/dL), mean (SD) Medical history, % Diabetes Mellitus Arterial Blood Hypertension Heart failure Prior use of ASA only Prior use of oral anticoagulant In-hospital anticoagulation within 7 days from admission TOAST Cardioembolism (%) Outcomes at discharge, % NIHSS at discharge mRS 0-2 at discharge In-hospital mortality 0.93 SD = Standard deviation. NIHSS = NIH Stroke Scale. IQR = Interquartile range. LACS = Lacunar syndrome. TACS = Total anterior circulation syndrome. Mrs = Modified Rankin Scale. ASPECTS = Alberta Stroke Program Early CT Score. MRI = Magnetic Resonance Imaging. BP = Blood pressure. SBP = Systolic Blood Pressure. DBP = Diastolic Blood Pressure. WBC = White Blood Cells. PTT = Partial Thromboplastin Time aPTT = Activated Partial Thromboplastin Time. ASA = Acetylsalicylic acid. TOAST = Trial of Org 10172 in Acute Stroke Treatment Classification. PREDICTIVE SCORE OF HEMORRHAGIC TRANSFORMATION 5 Table 2. Adjusted regression logistic model used to derive variables of PROpHET Variable Coefficients Odd ratio (OR) 95% confidence interval p-value Sex male Glycemia  180 mg/dL on admission ASPECTS score  7 on admission Hyperdense cerebral middle artery sign Leukoaraiosis Lacunar syndrome TOAST  Cardioembolism 1.255 1.146 1.735 0.834 0.667 -2.768 1.256 3.50 3.14 5.67 2.30 1.94 0.12 3.51 1.856.64 1.566.33 3.0010.68 1.104.80 1.043.61 0.0070.54 1.86-6.61 < 0.001 0.001 < 0.001 0.02 0.03 0.01 < 0.001 ASPECTS = Alberta Stroke Program Early CT Score. TOAST = Trial of Org 10172 in Acute Stroke Treatment Classification and MRI (33.7%, N = 151/448) were both used as followup neuroimaging. Patients without HT underwent MRI more often than patients with HT (46.2 vs. 30.5%, p = 0.007). The most frequent radiological groups10 of HT were hemorrhagic infarction types I and II (58%, N=40/ 95). Parenchymal hemorrhage type I (N = 26/95) was associated with symptomatic HT (ꭓ2 = 0.03). The PROpHET grading score was derived from the adjusted logistic regression model (Table 2) and is presented in Table 3. The accuracy of the derivation model based on 1,000 bootstrap replicates was 83% (8087.6%). The calibration was estimated by the Brier score (0.122) and the Hosmer-Lemeshow goodness-of-fit test (chisquare 8.7, df 8, p = 0.36). The absolute incidence of HT in patients increased per each group point (Fig. 2 Fig. 1, parts A and B). The total incidence of symptomatic cases of HT is shown in Fig. 2 Fig. 1 (part C). The optimal cutpoint determined by the Youden’s Test was 3. This cutpoint showed a sensitivity of 78.2% (68-86), specificity of 75% (70-79) and a predictive negative value of 92.8% (9095.1) for HT in general, and sensitivity of 78% (62.489.4), specificity of 75% (7080) and a predictive negative value of 96.5% (93.698) for symptomatic cases of HT. The calibration of the score (Table 3) was assessed using the Hosmer-Lemeshow goodness-of-fit test (chi square 2.58, df 6, p = 0.86). Table 3. The PROpHET (-3 to 7 points) score for prediction of the hemorrhagic transformation in ischemic stroke patients not submitted to reperfusion therapies Variable Points Sex male Glycemia  180 mg/dL on admission ASPECTS score  7 on admission Hyperdense cerebral middle artery sign Leukoaraiosis Lacunar syndrome TOASTCardioembolism 1 1 2 1 1 -3 1 ASPECTS = Alberta Stroke Program Early CT Score. TOAST = Trial of Org 10172 in Acute Stroke Treatment Classification The median score was 3 [3, 4] and 1 [0, 3] points in patients with and without HT, respectively (p < 0.001), Fig. 1 Fig. 2, part A. Scores were similar between asymptomatic and symptomatic HT cases (p = 0.35), Fig. 1 Fig. 2, part B. The corresponding odds ratios (OR) for PROpHET score 3 for HT in general and symptomatic cases were 10.6 (95% CI 6.018.68, p < 0.001) and 10.6 (95% CI 4.923.3, p < 0.001), respectively. The ROC analysis in the derivation group showed an AUROC of 0.82, 95% CI 0.780.86, p < 0.001 (Fig. 3, part A). The AUROC for symptomatic HT cases was 0.83 (95% CI 0.790.87, p < 0.001), Fig. 3, part B. A computerized tool based on a logistic regression model with continuous variables was built. This model had an AUC of 0.83 (95% CI 0.790.88), Hosmer-Lemeshow test of 0.45 and Brier score of 0.1 in the derivation group. There was no difference in accuracy between the model with continuous variables and our grading-score (Table 3). This tool allows estimating the probability of HT within 7 days from the hospital admission and it is publicly available at www.prophet-score.com. Validation A total of 2,683 patients from seven different CSC were included in the validation cohort. In this group, the median of score was 3 [2,4] and 0 [-2,2] points in patients with and without HT, respectively (p < 0.001), Fig. 1Fig 2, part C. The AUROC of the PROpHET was 0.83 (95% CI 0.810.86, p < 0.001), Fig. 3, part B. There was no statistical difference in the AUROCs between the derivation and validation cohorts. The optimal cutpoint determined by the Youden’s test was also 3. The sensitivity and specificity of the cutpoint were 66% (6070.4) and 87.4% (8688.7), respectively in the validation cohort. The positive predictive value was 44.1 (4147.3), and the negative predictive value was 94.3 (93.595.1). The Brier score was 0.08. The computerized tool performance assessed with AUROC in the validation group (N = 2,683) had an AUC 0.82 (95% CI 0.790.88, p < 0.001), Fig. 3, part D. The Brier score was 0.09. The overall odds ratio for each point in PROpHET for HT in the validation cohort was 2.07 J.B.C. DE ANDRADE ET AL. 6 Figure 1. Distribution of patients (%) according to PROpHET points in the derivation and validation cohorts (X2 < 0.001). A. Derivation group (N=448), X2 < 0.001. B. Validation group (N=2,683), X2 < 0.001. C. Derivation group (N=417), X2 < 0.001. (95% CI 1.912.24, p < 0.001). The performance of the model using continuous variable (computerized tool) in the validation group is showed in Fig. 3, part D. Discussion The PROpHET is an accurate and easy-to-use predictive score of a dreaded neurological complication in patients not treated with RT. In low and middle-income countries, this group of patients represents up to 98% of all cases of AIS4. Despite the relatively high incidence of HT in patients not treated with RT, there are few predictive scores focusing in this patient population and all of them included patients submitted to RT6,12,13. The pathophysiology of HT in patients not treated with RT might be very different from the phenomenon in patients treated with either thrombolysis or thrombectomy because reperfusion can have effects on the blood-brain-barrier (BBB). For instance, molecular effects of the tissue plasminogen activator (tPA) lead to direct damage on the BBB, also promoting dilation of the cerebral vasculature6,14,15. These effects could mask many clinical and laboratory characteristics, making the prediction of HT in patients not treated with tPA a challenge. Therefore, specific scores for this patient population are warranted6,14,15. PROpHET comprises seven known conditions associated to HT (hyperdense MCA sign, male sex, hyperglycemia at admission, cardio-aortic embolism, leukoaraiosis and extensive lesions). All of these elements are part of routine clinical evaluation of patients with AIS and can be easily determined at the bedside. In general, patients with a high risk of HT based on the PROpHET are males with elevated baseline blood glucose, extensive lesions, presence of leukoaraiosis and cardio-aortic embolism, including nonlacunar syndromes. The literature reports at least 11 predictive scores of HT  but all of those included patients treated with IV thrombolysis6,7,12,13,16,17,18,19,20,21,22. Therefore, specific scores for the patient population not managed with RT are warranted. Overall these scores included 19 variables. All the variables included in PROpHET were described by at least one published score, except for leukoaraiosis. However, leukoaraiosis has been reported as a risk factor for HT by other authors23,24,25. PROpHET predicted both symptomatic and asymptomatic cases of HT. The decision to include asymptomatic HT as an outcome was also driven by the worse short and long-term outcomes when compared to patients without HT7,14,15. Interestingly in our series, only PH type I, and not PH type II, was associated with symptomatic HT, contrasting with the existing literature3,10,17,21. This PREDICTIVE SCORE OF HEMORRHAGIC TRANSFORMATION 7 Figure 2. Comparison of medians according to PROpHET points (p <0.001). A. Derivation cohort. B. In the derivation cohort, a comparison between symptomatic cases and patients without HT. C. Validation group. unexpected finding can be explained by the small number of patients in this subgroup (N=14/95). We measured the extension of the ischemic lesions using ASPECTS, which graduates ischemic lesions in cortical and subcortical regions in the MCA territory. The ASPECTS itself has a high sensitivity for the prediction of HT26  but when we compare the ASPECTS itself to PROpHET, we find a higher accuracy of our score. This finding is congruent with the literature6,16. In the PROpHET we categorized ASPECTS into two groups: greater and lower or equal to 8. This ASPECTS cut point reflects an estimate of the involvement of > 1/3 of MCA territory16 and was already adopted by some authors16,27. Male gender was a risk factor for HT in our series in agreement with other authors17,18. In animal models28, the male gender is a risk factor for larger ischemic lesions and BBB damage. It is hypothesized that estrogen has a protective action on endothelial cells and hence on the BBB integrity and this might in part explain the association between male gender and HT28. We also found hyperglycemia at admission as an independent risk factor to HT. Hyperglycemia has been reported in 6 out of 9 known predictive scores for HT16,17,18,19,20,21. Hyperglycemia can lead to an active inflammatory cascade, impairment of the BBB and expansion of the ischemic lesion29. Interestingly, we have failed to find an association between HT and a previous history of diabetes mellitus (DM), in contrast to some other authors12,16,23. It is however possible that patients with high glucose at admission might indeed be unknown diabetics with poorly controlled glucose levels. Cardioembolism is a known risk factor for HT and was a predictive factor in our series as well6,18,14. Patients with cardioembolic strokes usually present with proximal occlusion of large vessels, leading to extensive ischemic lesions. Cardiac emboli are generally rich in red blood cells and fibrin, therefore more likely to be destroyed, leading to reperfusion of large ischemic areas30. Prior use of oral anticoagulants (N = 13/448) or aspirin alone (N = 126/448) was not a risk factor for HT in our series. Furthermore, there was no difference in the rates of HT between patients who were treated with anticoagulation (N = 7/448) within 7 days from hospital admission versus those who did not receive such treatment. The correlation between prior use of oral antithrombotic and HT is still unclear in the literature6,7,21,23. In our series, the frequency of the in-hospital treatment with aspirin within the first 48 h from hospital admission was actually lower among patients who had HT within 7 days. This finding is probably explained by an indication bias, as patients with HT at admission or within the first 48 h were probably less likely to be treated with antithrombotic. We found that clinical presentation as a lacunar syndrome was protective against HT. In general, patients with lacunar syndromes have small and deep ischemic lesions, most frequently sparing the cortex. Lesions with smaller volumes tend to have less HT, therefore it was expected that patients with lacunar syndromes would 8 J.B.C. DE ANDRADE ET AL. Figure 3. ROC curve for estimated HT based on PROpHET versus observed HT in derivation and validation cohorts. A. ROC analysis and AUC (0.82, 95% CI 0.780.86) in the derivation group. B. ROC analysis and AUC (0.83, 95% CI 0.790.87) in the derivation group considering the symptomatic HT group. C. ROC analysis and AUC (0.83, 95% CI 0.820.85) in the validation group. D. ROC analysis and AUC (0.82, 95% CI 0.800.84) of the computerized tool in the validation group (N=2,683). There was no difference between these curves (p>0.05). present lower rates of bleeding3,5,6,9,18. The variable lacunar syndrome was collinear with the NIHSS at admission as patients with lacunar strokes usually present with low NIHSS scores. Therefore, although NIHSS was a univariate predictor of HT, in the adjusted analysis lacunar stroke remained as a predictor, and NIHSS did not. Other authors also found this initial relationship between HT and NIHSS that was not confirmed in the multivariable regression analysis7. In comparison to eight previously published scales that included either patients submitted or not to RT, the PROpHET proved to be the most accurate (p < 0.01) in our local sample (n = 986, patients from the derivation cohort and patients from our center included in the validation dataset). In this population, the accuracy of the other scores ranged from an AUROC of 0.55 (0.510.60)12 to 0.76 (0.720.80)16. If we evaluated only the published scores which also included patients not treated with RT6,12,13, the AUROC in our patient population ranged from 0.55 (0.510.60)13 to 0.73 (0.680.79)6. The SPAN10013score includes neurological severity at admission and age; the HTI6 includes the extension of the infarct, presence of atrial fibrillation, neurological severity, laboratory findings, and the presence of hyperdense middle cerebral artery sign; and the THIRVE12 includes age, neurological severity and co-morbidities, but not neuroimaging or laboratory findings. The PROpHET, on the other hand, consists of multiple clinical, laboratory, neuroimaging and epidemiology characteristics, having a good accuracy in predicting HT. Currently, there are no guidelines instructing to avoid therapies based on predictive scores of HT8,31. However, the knowledge about risk factors for HT in non-reperfused patients might help in the decision-making process of therapies for secondary prevention such as initiation of anticoagulation and deep venous thrombosis prophylaxis16. The PROpHET is useful in identifying patients at high or low risk of HT, providing an individualized prediction of HT. Besides, our score may help researchers evaluating acute stage therapies in stratifying patients as a lower or higher risk groups in the context of epidemiological and population-based studies or even clinical trials. Of course, the PROpHET alone should not be used to withhold evidence-based treatments but could help with the decision of the timing to initiate such therapies. PREDICTIVE SCORE OF HEMORRHAGIC TRANSFORMATION Our study has some limitations. First, neuroimaging was evaluated clinically by a single specialist, either a radiologist or a board-certified neurologist. These specialists, however, underwent prior training to standardize the neuroimaging reports. Second, patients with posterior circulation lesions were not included in our series. Finally, the validation cohort included patients from seven CSC using retrospective chart data. Conclusions In conclusion, the PROpHET is a simple, quick, costfree, and easy-to-perform tool that allows feasible risk stratification of HT in patients not submitted to RT. A free computerized version was made publicly available. Declaration of Competing Interest The authors state not disclosures related to the main subject of this research. Funding: Dr. Andrade's visiting scholarship at Columbia University, USA, is sponsored by the CAPES Foundation, Ministry of Education, Brazil. References 1. Berger C, Fiorelli M, Steiner T, Schabitz W-R, Bozzao L, Bluhmki E, et al. Hemorrhagic Transformation of Ischemic Brain Tissue:Asymptomatic or Symptomatic? Stroke 2001;32(6):1330-1335. 2. Toni D, Fiorelli M, Bastianello S, Sacchetti ML, Sette G, Argentino C, et al. Hemorrhagic transformation of brain infarct: predictability in the first 5 hours from stroke onset and influence on clinical outcome. Neurology 1996;46 (2):341-345. 3. Zhang J, Yang Y, Sun H, Xing Y. Hemorrhagic transformation after cerebral infarction: current concepts and challenges. Ann Transl Med 2016;2(8):1-7. 4. Martins SCO, Pontes-Neto OM, Alves CV, Freitas GRD, Filho JO, Tosta ED, et al. Past, Present, and Future of Stroke in Middle-Income Countries: The Brazilian Experience. Int J Stroke 2013;S100:106-111. 5. Seet RC, Rabinstein AA. Symptomatic intracranial hemorrhage following intravenous thrombolysis for acute ischemic stroke: a critical review of case definitions. Cerebrovasc Dis 2012;34:106-114. 6. Kalinin MN, Khasanova DR, Ibatullin MM. The hemorrhagic transformation index score: a prediction tool in middle cerebral artery ischemic stroke. BMC Neurol 2017;17(1):2-16. 7. Marsh EB, Llinas RH, Hillis AE, Gottesman RF. Hemorrhagic transformation in patients with acute ischaemic stroke and an indication for anticoagulation. Eur J Neurol 2013;20(6):962-967. 8. Yaghi S, Willey JZ, Cucchiara B, Goldstein JN, Gonzales NR, Khatri P, et al. Treatment and outcome of hemorrhagic transformation after intravenous alteplase in acute ischemic stroke: a scientific statement for healthcare professionals from the american heart association/American stroke association. Stroke 2017. 9
9. Alvarez-Sabín J, Maisterra O, Santamarina E, Kase CS. Factors influencing haemorrhagic transformation in ischaemic stroke. The Lancet Neurol 2013;12(7):689705. 10. Hacke W, Kaste M, Fieschi C, von Kummer R, Davalos A, Meier D, et al. Randomised double-blind placebo-controlled trial of thrombolytic therapy with intravenous alteplase in acute ischaemic stroke (ECASS II). Second Eur-Australas Acute Stroke Study Investig Lancet 1998;352(9136):1245-1251. 11. Delong ER, Delong DM, Clarke-Pearson DL. Comparing the Areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics 1988;44(3):837. 12. Flint AC, Faigeles BS, Cullen SP, Kamel H, Rao VA, Gupta R, Smith WS, Bath PM. Donnan GA; on behalf of the VISTA collaboration.THRIVE score predicts ischemic stroke outcomes and thrombolytic hemorrhage risk in VISTA. Stroke 2013;44:3365-3369. 13. Saposnik G, Guzik AK, Reeves M, Ovbiagele B, Johnston SC. Stroke prognostication using age and NIH stroke scale: SPAN-100. Neurology 2013;80:21-28. 14. Park JH, Ko Y, Kim W- J, Jang MS, Yang MH, Han M- K, et al. Is asymptomatic hemorrhagic transformation really innocuous. Neurology 2012;78(6):421-426. 15. Lei C, Wu B, Liu M, Chen Y. Asymptomatic hemorrhagic transformation after acute ischemic stroke: is it clinically innocuous? J Stroke Cerebrovasc Dis 2014;23(10):27672772. 16. Lou M, Safdar A, Mehdiratta M, Kumar S, Schlaug G, Caplan L, Searls D, Selim M. The HAT Score: a simple grading scale for predicting hemorrhage after thrombolysis. Neurology 2008;71:1417-1423. 17. Menon BK, Saver JL, Prabhakaran S, Reeves M, Liang L, Olson DM, Peterson ED, Hernandez AF, Fonarow GC, Schwamm LH, Smith EE. Risk score for intracranial hemorrhage in patients with acute ischemic stroke treated with intravenous tissue-type plasminogen activator. Stroke 2012;43:2293-2299. 18. Saposnik G, Fang J, Kapral MK, Tu JV, Mamdani M, Austin P, et al. The iScore predicts effectiveness of thrombolytic therapy for acute ischemic stroke. Stroke 2012;43:1315-1322. 19. Cucchiara B, Tanne D, Levine SR, Demchuk AM, Kasner S. A risk score to predict intracranial hemorrhage after recombinant tissue plasminogen activator for acute ischemic stroke. J Stroke Cerebrovasc Dis 2008;17:331-333. 20. Strbian D, Engelter S, Michel P, Meretoja A, Sekoranja L, Ahlhelm FJ, Mustanoja S, Kuzmanovic I, Sairanen T, Forss N, Cordier M, Lyrer P, Kaste M, Tatlisumak T. Symptomatic intracranial hemorrhage after stroke thrombolysis: the SEDAN score. Ann Neurol 2012;71:634-641. 21. Mazya M, Egido JA, Ford GA, Lees KR, Mikulik R, Toni D, Wahlgren N. Ahmed N; for the SITS Investigators. Predicting the risk of symptomatic intracerebral hemorrhage in ischemic stroke treated with intravenous alteplase: safe Implementation of Treatments in Stroke (SITS) symptomatic intracerebral hemorrhage risk score [published correction appears in Stroke. 2012;43:e102]. Stroke 2012;43:1524-1531. 22. Hallevi H, Barreto AD, Liebeskind DS, Morales MM, Martin-Schild SB, Abraham AT, et al. Identifying patients at high risk for poor outcome after intra-arterial therapy for acute ischemic stroke. Stroke 2009;40(5):1780-1785. 23. Whiteley WN, Slot KB, Fernandes P, Sandercock P, Wardlaw J. Risk factors for intracranial hemorrhage in J.B.C. DE ANDRADE ET AL. 10 24. 25. 26. 27. acute ischemic stroke patients treated with recombinant tissue plasminogen activator: a systematic review and metaanalysis of 55 studies. Stroke 2012;43:2904-2909. Wei C-C, Zhang S-T, Wang Y-H, Liu J-F, Li J, Yuan R-Z, et al. Association between leukoaraiosis and hemorrhagic transformation after cardioembolic stroke due to atrial fibrillation and/or rheumatic heart disease. J Neurol Sci 2017;378:94-99. Neumann-Haefelin T, Hoelig S, Berkefeld J, Fiehler J, Gass A, Humpich M, et al. Leukoaraiosis is a risk factor for symptomatic intracerebral hemorrhage after thrombolysis for acute stroke. Stroke 2006;37(10):24632466. Demchuk AM, Coutts SB. Alberta stroke program early CT score in acute stroke triage. Neuroimaging Clin N Am 2005;15(2):409-419. Sarraj A, Albright K, Barreto AD, Boehme AK, Sitton CW, Choi J, et al. Optimizing prediction scores for poor outcome after intra-arterial therapy in anterior 28. 29. 30. 31. circulation acute ischemic stroke. Stroke 2013;44 (12):3324-3330. Reeves M, Khoury J, Alwell K, Moomaw C, Flaherty M, Woo D, et al. Distribution of national institutes of health stroke scale in the cincinnati/northern kentucky stroke study. Stroke 2013;44(11):3211-3213. Lees K. Walters M. acute stroke and diabetes. Cerebrovasc Dis 2005;20(1):9-14. Shin JW, Jeong HS, Kwon H-J, Song KS, Kim J. High red blood cell composition in clots is associated with successful recanalization during intra-arterial thrombectomy. Plos One 2018;13(5):1-13. Powers WJ, Rabinstein AA, Ackerson T, Adeoye OM, Bambakidis NC, Becker K, et al. Guidelines for the Early Management of Patients With Acute Ischemic Stroke: 2019 Update to the 2018 Guidelines for the Early Management of Acute Ischemic Stroke: A Guideline for Healthcare Professionals From the American Heart Association/American Stroke Association. Stroke 2019;50(12).