Reliability of Hemorrhagic Transformation Diagnosis in Acute Ischemic Stroke
Background and Purpose Diagnosis of hemorrhagic transformation (HT) could influence the prognosis and the management of acute ischemic stroke. The interobserver reliability of CT-scan HT classification is evaluated in the present study.
Methods Fifty 5-day CT scans of patients enrolled in the Multicenter Acute Stroke Trial–Italy (MAST-I) were reviewed independently by two neuroradiologists and one neurologist with CT training. They evaluated the presence and type of intraparenchymal HT (hemorrhagic infarction types I, II, and III and intracerebral hemorrhage) (five-item scale), as well as the presence of intraventricular and/or subarachnoid bleeding according to standardized definitions.
Results Agreement for exclusion of HT and intraventricular/subarachnoid bleeding was good between the neuroradiologists (κ=0.70 and κ=0.72) and excellent between the neurologist and each neuroradiologist (κ=0.87 and κ=0.77, κ=0.83, and κ=0.81, respectively). The overall agreement for the five-item HT scale between the two neuroradiologists was good (κw=0.65) because of discordance over the last three items. Better overall agreement was obtained with a three-item scale: no hemorrhage, petechial type I hemorrhagic infarction, and other HT (type II and type III hemorrhagic infarction and intracerebral hemorrhage) together (κw=0.82).
Conclusions Exclusion of HT is a reliable CT diagnosis when made by neuroradiologists and also by a neurologist with CT training. Five- and three-item scales of HT types showed good to excellent reliability. The validity of the scale for predicting short- and long-term outcome should be evaluated in future studies.
Studies using CT have shown that hemorrhage occurs in 10% to 43% of patients with cerebral infarct.1 2 3 4 Two distinct types of HT associated with a different prognosis can be recognized: HI and intrainfarct hematoma.2 4 5 6 Some authors have suggested further subdivisions of hemorrhage based on morphological patterns or extension of bleeding on CT,6 7 8 9 but no classification is widely accepted and reliability has never been verified. Because the diagnosis of hemorrhage could influence the management of patients with acute ischemic stroke and its typing could give prognostic information, it is important to establish the reliability of diagnosis. In the present study, we measured interobserver agreement of the CT-scan HT scale used in MAST-I.
Materials and Methods
We analyzed 50 CT scans of patients with acute cerebral infarction studied in MAST-I.9 The sample of 50 was extracted from all available control 5-day CT scans, 10 randomly from each of four MAST-I HT categories and 10 randomly from scans without HT lesion. The selected CT scans were reviewed on the same day by two neuroradiologists (E.B. as neuroradiologist 1 and C. De G. as neuroradiologist 2) independently of each other. In addition, all CT scans were reviewed by a neurologist with CT training (E.A.).
Observers were asked to express a judgment about the presence and type of hemorrhage on the basis of standardized definitions. HT of cerebral infarct was defined on noncontrasted CT scans as an area of increased density within an area of low attenuation in a typical vascular distribution. It was classified in four categories: (1) HI type I (petechial; small petechial and linear high attenuation region[s] within an infarct area); (2) HI type II (medium; homogeneous or heterogeneous region[s] of high attenuation within an infarct area); (3) HI type III (large; homogeneous region[s] of high attenuation of total presumed cerebral infarct area); and (4) intracerebral hemorrhage (homogeneous region of high attenuation exceeding the vascular territory of the presumed infarction). The latter category also included hemorrhagic lesion outside the area of infarction. Moreover, observers evaluated the same CT scans for the presence of intraventricular and subarachnoid bleeding defined as presence of a high attenuation area in the ventricular cavities and subarachnoid spaces, respectively. Answers were recorded on a specific separate form listing the possible CT-scan diagnosis. The only information that the investigators had was the time of stroke onset and the presentation symptoms; they were blind to treatment allocation, results of basal CT scan, clinical course, and modality of sampling. Two neuroradiologists worked at the same time in separate rooms, each evaluating half of the sample (first 25 CT scans by neuroradiologist 1, second 25 by neuroradiologist 2). At the end of this session, the material was switched by a third person, and each neuroradiologist reviewed the other 25 scans. The next day, all 50 CT scans were reviewed by the neurologist. There was no discussion or consultation between observers before the evaluation, and they did not discuss their findings with each other until the review was completed.
The degree of interobserver agreement was measured using κ statistics10 as follows: κ=(Po−Pe)/(1−Pe), where Po is the agreement observed and Pe the agreement expected by chance. The standard error and 95% CI of the κ value were determined as suggested by Fleiss.10 A value of κ=0 indicates that agreement is no better than chance, and κ=1 represents perfect agreement. The definition of Landis and Koch11 of strength of agreement was used: values of κ>0.80 represent excellent agreement beyond chance, κ<0.40 fair to poor agreement, between 0.40 and 0.60 moderate agreement, and between 0.60 and 0.80 good agreement. The degrees of disagreement were also considered and a weighted κ (κw) was calculated.10 Kappa values are unweighted unless otherwise specified.
The distribution of the answers on the presence and type of HT recorded by the neuroradiologists is summarized in Table 1⇓. Agreement was obtained in 30 of 50 CT scans (60%); a difference of one level was observed in 15 cases (30%) and of two levels in 5 (10%). The overall level of agreement was moderate, with an unweighted value of κ=0.49 (95% CI, 0.33 to 0.66), and good, with a weighted κ value of κw=0.65 (95% CI, 0.52 to 0.78). Analysis of individual categories (Table 2⇓, column 1) indicated that chance-corrected interobserver agreement was good for the diagnosis of hemorrhage exclusion (κ=0.70); only in 5 cases did neuroradiologists disagree, and in all cases one observer identified HI type I, whereas the other reported the absence of hemorrhage. Agreement was moderate for the diagnosis of HI type I (κ=0.58), lower for type II (κ=0.46), and fair for HI type III and intracerebral hemorrhage (κ=0.35 and 0.24, respectively). Better overall agreement was obtained when the last three items of the five-item scale for which disagreement was highest were considered together on a three-item scale (no hemorrhage, petechial type I HI, and other HT) (κ=0.76; 95% CI, 0.60 to 0.91; κw=0.82; 95% CI, 0.71 to 0.94).
The measure of agreement for occurrence and type of HT between the trained neurologist and each of two neuroradiologists is also shown in Table 2⇑ (columns 2 and 3). Overall agreement was good, with a value of κ=0.64 (95% CI, 0.49 to 0.80) in both cases. It was higher than that between the neuroradiologists for individual categories also. Agreement was excellent for the diagnosis of hemorrhage exclusion (κ=0.87 and κ=0.83).
We also calculated the level of agreement between all three observers, obtaining an overall value of κ=0.59 (95% CI, 0.51 to 0.67). The measures of agreement for individual categories are reported in Table 2⇑ (column 4).
Separate analysis of concordance in the diagnosis of intraventricular and/or subarachnoid bleeding showed complete agreement between the neuroradiologists in 45 of 50 cases (90%), with a value of κ=0.72 (95% CI, 0.49 to 0.95) (Table 3⇓). The level of agreement for the same items between the neurologist and each neuroradiologist was excellent, with a value of κ=0.77 (95% CI, 0.56 to 0.98) and κ=0.81 (95% CI, 0.59 to 1.00), respectively. A value of κ=0.76 (95% CI, 0.60 to 0.92) was obtained when the agreement among all three observers was calculated. Table 3⇓ also shows the level of agreement for intraventricular and subarachnoid bleeding separately.
The present study demonstrated that the CT diagnosis of HT exclusion is reproducible by different observers and that it could be reliably used in the assessment of patients with ischemic stroke in clinical practice and in multicenter clinical trials. Other authors have reported an agreement for diagnosis of HT exclusion between different experienced radiologists ranging from κ=0.13 to κ=0.56.12 In our study, we found an excellent agreement between two neuroradiologists and between each neuroradiologist and a trained neurologist, not only for the diagnosis of HT (κ=0.70, 0.87, and 0.83, respectively) but also for the presence of intraventricular and/or subarachnoid bleeding (κ=0.72, 0.77, and 0.81, respectively). On the basis of these results, it seems reasonable to assume that a trained neurologist is able to exclude intracranial bleeding with at least a 5-day CT scan. This diagnosis, as well as one of early CT signs of brain ischemia, could be of great importance in the management of acute stroke.13 In fact, reliable recognition of early infarction signs and absence of brain bleeding is fundamental when a rapid decision has to be made regarding thrombolytic and antithrombotic treatments.
The diagnosis of type of hemorrhagic lesion has a different use. The first HT subdivision of Jorgensen and Torvik14 was anatomopathological. Subsequently, many CT-scan classifications have been proposed using different criteria based on morphological features and extent and location of hemorrhagic lesion (Table 4⇓). The morphological criteria may identify types of hemorrhages with similar underlying pathogenetic mechanisms. With this criterion, petechial HI (scattered punctate, speckled, or mottled appearance; sometimes called “small” or “mild” hemorrhage) and confluent hemorrhage (generally called more severe HI or parenchymal hematoma if it has a homogeneous appearance) have been identified.1 2 6 7 8 9 12 15 16 17 18 19 20 22 The extension criterion could be useful for predicting clinical outcome, since various authors have suggested that the largest lesions are associated with severe prognosis.2 6 7 15 16 Bleeding extent has been estimated both in absolute terms12 16 (eg, by measuring the maximum diameter of the lesion) and in relative terms with respect to the presumed infarcted area.6 7 9 20 A classification based on location of hemorrhagic lesions has been used in only one study.6 Despite the numerous classifications present in the literature, none has been tested to assess interobserver reliability. In the present study, in which we tested the reliability of a scale based on both morphological and extension criteria, we found good agreement overall between two neuroradiologists (κw=0.65). Investigating whether the level of agreement could be further improved, we found that the discordance over the last three items of the scale, mostly based on the extension criterion, could explain in large part the disagreement between the two neuroradiologists. In fact, it is difficult to differentiate a medium homogeneous or heterogeneous hemorrhagic lesion within an infarct area (type II HI) from a large homogeneous lesion covering the total presumed infarct area (type III HI), as well as from a homogeneous hemorrhage exceeding the vascular territory of the presumed infarction or located outside the area of infarction (intracerebral hemorrhage). On the other hand, the agreement for diagnosis of petechial hemorrhagic infarction type I, for which a morphological criterion was used, was higher. Separating petechial type I HI from the other types of HT, we obtained a simplified three-item scale (no hemorrhage, petechial type I HI, and other HT) that yielded an overall κ value higher than that of the original five-item scale (κw=0.82). Moreover, comparison of the MAST-I classification with those used by other authors (Table 4⇓) reveals that our HI type I and our intracerebral hemorrhage, HI type III, and sometimes HI type II correspond, respectively, to hemorrhagic infarction and parenchymal hematoma of most other studies.
In conclusion, the scale proposed by us showed an interrater reliability that can be considered good when five items were included and even better with three items. Thus, it could be a reliable tool for use in multicenter clinical trials. However, efforts should be made not only to establish a reliable classification but also to standardize the HT definitions among different researchers and to obtain a categorization of HT that could be a valid measure to predict short- and long-term outcome. Use of the scale for this purpose should be tested in future studies because its reliability does not guarantee its validity.
Selected Abbreviations and Acronyms
|MAST-I||=||Multicenter Acute Stroke Trial–Italy|
- Received July 1, 1996.
- Revision received October 15, 1996.
- Accepted October 21, 1996.
- Copyright © 1997 by American Heart Association
Fisher M, Zito JL, Siva A, De Girolami U. Hemorrhagic infarction: a clinical and CT study. Stroke. 1984;15:192. Abstract.
Hornig CR, Dorndorf W, Agnoli AL. Hemorrhagic cerebral infarction: a prospective study. Stroke. 1986;17:179-184.
Hart RG, Easton JD. Hemorrhagic infarcts. Stroke. 1986;17:586-589.
Moulin T, Crépin-Leblond T, Chopard JL, Bogousslavsky J. Hemorrhagic infarcts. Eur Neurol. 1993;34:64-77.
Ott BR, Zamani A, Kleefield J, Funkenstein HH. The clinical spectrum of hemorrhagic infarction. Stroke. 1986;17:630-637.
Lodder J, Krijne-Kubat B, van der Lugt PJM. Timing of autopsy-confirmed hemorrhagic infarction with reference to cardioembolic stroke. Stroke. 1988;19:1482-1484.
Fleiss JL. The measurement of inter-rater agreement. In: Statistical Methods for Rates and Proportions. New York, NY: John Wiley & Sons; 1981:212-236.
Wardlaw JM, Sellar R. A simple practical classification of cerebral infarcts on CT and its interobserver reliability. AJNR Am J Neuroradiol. 1994;15:1933-1939.
von Kummer R, Meyding-Lamadé U, Forsting M, Rosin L, Rieke K, Hacke W, Sartor K. Sensitivity and prognostic value of early computed tomography in middle cerebral artery trunk occlusion. AJNR Am J Neuroradiol. 1994;15:9-15.
Pessin MS, Estol CJ, Lafranchise F, Caplan LR. Safety of anticoagulation after hemorrhagic infarction. Neurology. 1993;43:1298-1303.
Okada Y, Yamaguchi T, Minematsu K, Miyashita T, Sawada T, Sadoshima S, Fujishima M, Omae T. Hemorrhagic transformation in cerebral embolism. Stroke. 1989;20:598-603.
Wolpert SM, Bruckmann H, Greenlee R, Wechsler L, Pessin MS, del Zoppo GJ, for the rt-PA Acute Stroke Study Group. Neuroradiologic evaluation of patients with acute stroke treated with recombinant tissue plasminogen activator. AJNR Am J Neuroradiol. 1993;14:3-13.
Mori E, Yoneda Y, Tabuchi M, Yoshida T, Ohkawa S, Ohsumi Y, Kitano K, Tsutsumi A, Yamadori A. Neurology. 1992;42:976-982.
Brott TG, Haley EC Jr, Levy DE, Barsan W, Broderick J, Sheppard GL, Spilker J, Kongable GL, Massey S, Reed R, Marler JR. Urgent therapy for stroke, I: pilot study of tissue plasminogen activator administered within 90 minutes. Stroke. 1992;23:632-640.
Hacke W, Kaste M, Fieschi C, Toni D, Lesaffre E, von Kummer R, Boysen G, Bluhmki E, Hoxter G, Mahagne MH, Hennerici M, for the ECASS Study Group. Intravenous thrombolysis with recombinant tissue plasminogen activator for acute hemispheric stroke: the European Cooperative Acute Stroke Study (ECASS). JAMA. 1995;274:1017-1025.