Hemorrhagic Transformation in Acute Ischemic Stroke: A Quantitative Systematic Review

The prevalence and risk factors of hemorrhagic transformation (HT) after acute ischemic stroke HT have not been adequately delineated. We performed a systematic review and meta-analysis to identify English-language prospective observational MEDLINE and EMBASE-listed reports of acute ischemic stroke with HT published from 1985–2017. Studies that used the ECASS-2 definitions of hemorrhagic transformation subtypes, hemorrhagic infarction (HI), and parenchymal hematoma (PH) were included. Patients treated with intravenous thrombolysis with tissue plasminogen activator (IV-tPA) were compared with those who did not receive thrombolysis. A total of 65 studies with 17,259 patients met inclusion criteria. Overall, HT prevalence was 27%; 32% in patients receiving IV-tPA vs. 20% in those without. Overall PH prevalence was 9%; 12% in IV-tPA treated patients vs. 5% in those without. HT was associated with a history of atrial fibrillation (OR 2.94) and use of anticoagulants (OR 2.47). HT patients had higher NIHSS (Hedge’s-G 0.96) and larger infarct volume (diffusion-weighted MRI, Hedge’s-G 0.8). In IV-tPA treated patients, PH correlated with antiplatelet (OR 3) and statin treatment (OR 4). HT (OR 3) and PH (OR 8) were associated with a poor outcome at 90-day (mRS 5–6). Hemorrhagic transformation is a frequent complication of acute ischemic stroke and is associated with poor outcome. Recognition of risk factors for HT and PH may reduce their incidence and severity.


Introduction
Hemorrhagic transformation (HT) occurs frequently in patients with acute ischemic stroke. While HT may be part of the natural disintegration process of the infarcted tissue [1], it is unclear to what extent it may exert its own deleterious effect independent of those attributed to the infarcted tissue [2].
There are two accepted classification schemes for HT assessment; one is based strictly on radiological criteria, while the second combines both clinical and radiological variables. The former was used as a safety end point for intravenous thrombolysis with tissue plasminogen activator (IV-tPA)-related HT in the European Cooperative Acute Stroke Trials (ECASS) [3][4][5]. The ECASS classification categories are no HT, HT into hemorrhagic infarction (HI), manifesting as small petechial hemorrhage along the margins of the infarct, and parenchymal hematoma (PH), manifesting as confluent hematoma. HI and PH are subdivided into type 1 and 2 for the milder and more severe forms, respectively (without HT, HI, PH, and PH2). Notably, PH2, the most severe form of HT, is classified as blood clots exceeding 30% of the infarct area with significant space-occupying effect. In cases of HT classed as ECASS-I, PH2 was associated with both neurological deterioration (OR 32.3) and increased 3-month mortality (OR 18) [4] compared with patients without PH2.
The second HT classification system is based on clinical presentation; (a) symptomatic vs. (b) asymptomatic. Unfortunately, different definitions of symptomatic HT have been adopted, introducing biases regarding prevalence, risk factors, and prognosis [5,6]. ECASS-III required concurrent hemorrhage on CT scan and neurological deterioration, with an increase of >4 points in the National Institutes of Health Stroke Score (NIHSS) to diagnose symptomatic HT [3]. Subdividing HT into symptomatic vs. asymptomatic is problematic because it is difficult to attribute neurological deterioration solely to HT. Furthermore, the effect of asymptomatic HT on outcome remains controversial [7,8].
We aimed to characterize the prevalence, risk factors, and prognosis of HT in patients with acute ischemic stroke by conducting a systematic review of studies that reported HT. As rates of HT are higher in patients treated with IV-tPA [9], we specifically investigated interactions between possible predictors of HT in IV-tPA treated and untreated subpopulations. Additionally, we examined rates of HT subtypes in patients of East-Asian ethnicity, since they are more prone to suffer from HT [8,10].

Materials and Methods
This study was registered with the International Prospective Register of Systematic Reviews; PROSPERO 2017 (Registration number CRD42017074806) and adheres to the PRISMA guidelines for preferred reporting in meta-analyses [11]. Two authors (A.H. and J.P.) identified potentially relevant studies and independently extracted data. For purposes of quality assurance, a third author (O.R.B.) extracted information from ten studies and validated the results.

Study Identification and Classification
An electronic search of the MEDLINE and EMBASE Ovid (Wolters Kluwer, Alphen aan den Rijn, The Netherlands) databases for English-language prospective observational studies published between January 1985 and May 2017 that reported rates of HT using the search terms "cerebral infarction/or brain isch*emia or stroke" and "h*emorrhagic transformation." Studies were included if they recorded baseline and follow-up brain imaging (CT or MRI) demonstrating hemorrhagic transformation in an adult population and used the ECASS definition of HT; conference abstracts or letters to the editor, reviews, case series with fewer than 10 patients, and retrospective studies (Supplemental Material) were excluded.

Each Study Was Classified According to the Following Criteria
IV-tPA treatment-According to the study protocol, each study population was classified as either IV-tPA treated or untreated. In a study including both subpopulations, data was extracted for each subgroup separately. Data from mixed populations of IV-tPA treated and untreated patients that did not differentiate the two subpopulations were only used for the total population included in the global assessment of each factor. Similarly, data from patients who underwent any endovascular procedure was included in the global assessment of each factor and not included in the IV-tPA treated and untreated analysis.
Imaging timing and modality-To assess the effect of imaging modality on the frequency of HT, studies were categorized based on whether CT or MRI was used in repeat scan after admission. Studies that included both imaging modalities were excluded from this specific analysis. To assess HT frequency with respect to timing of repeat imaging, studies were divided into series where the repeat scan after the index event was performed within 72 h or after a longer interval. Ethnicity-Studies published by an institution in an East-Asian country that included only the local population were designated as such and were compared to studies from non-East-Asian countries.

Data Extraction
For each study, demographic, clinical, and radiological data were extracted in accordance with PRISMA criteria [11]. Using predefined variables, two investigators (A.H. and J.P.) independently evaluated all studies and extracted the data. Data were collected by treatment (IV-tPA treated versus untreated patients) and according to the ECASS radiological classification (without HT, HI, PH, and PH2). In studies where results were given for a mixed population of IV-tPA treated and untreated patients, the available data were used only when calculating the effect of a certain variable on the total patient population. Supplemental Table S1 lists all studies included in the meta-analysis, and Supplemental Tables S3-S5 detail the specific studies used in each sub-analysis. For the sake of brevity, some of the studies used in the analysis may not be cited in the body of this paper.

Data Analysis
Two types of aggregate effect-size estimates, odd ratios (OR), and Hedges' g (HG), were calculated using continuous and categorical data. Categorical data were used to create the contingency table of analysis for OR (measure of association), and continuous data (i.e., mean, standard deviation, and sample size) were used to generate HG, a measure of group differences. The standard practice for interpretation of HG, as per Cohen's suggestions [12], is to stratify findings as a small effect (HG < 0.2), a medium effect (HG 0.2−0.8), or a large effect (HG > 0.8).
Continuous variables included age, low density lipoprotein (LDL) cholesterol level, admission glucose level (AGL), systolic and diastolic blood pressure, NIHSS score, and infarct volume measured on diffusion-weighted MRI (DWI).
Neuroradiological markers, patient body temperature, and renal impairment data were analyzed qualitatively.
For each variable, three group comparisons were made: HT vs. no HT, PH vs. no PH, and PH vs. HI. Data analysis was performed for IV-tPA-treated and untreated patients separately whenever possible.
Statistical analyses were carried out using the Comprehensive Meta-analysis software (Verson 2.0, Biostat, Englewood, NJ, USA) [13] including the random effect model and analysis of continuous and categorical data to report the aggregate prevalence estimate and the association markers with 95% confidence intervals (CI) for all variables. We examined for the presence of between-studies heterogeneity in the results by relying on the magnitude of the I-square statistics [14].

Assessment of the Risk of Bias
Two reviewers (A.H. and J.P.) independently assessed the risk of bias for individual studies according to widely accepted tools [15], noting methodology for participant selection, HT outcomes, blinding, loss to follow-up, methods for controlling confounding, and declaration of conflicts of interest.

Results
A total of 2087 studies were identified; 65 studies (seven randomized clinical trials and 58 observational studies) enrolling a total of 17,259 patients met selection criteria and were included in this analysis ( Figure 1). The characteristics of all included studies are summarized in Supplementary Tables S1 and S2.
Similar rates of PH were diagnosed using either CT or MRI (9% for both). In contrast, higher rates of HI were diagnosed using MRI (20% versus 15%). Table S3 details which studies were included in each analysis.
Details regarding alcohol abuse was provided in four studies including 2289 patients, of whom 284 (12%) had a history of abuse [27,33,36,43] (Table 2). There was no association with either HT (OR 1.32, 0.92-1.88) nor PH (OR 1.02, 0.5-2.08) in a mixed population of IV-tPA treated and untreated patients in this sample.
The frequency of chronic statin treatment was given in six studies including 2734 patients; 420 of them (15%) were under treatment [23,26,28,36,43,45] (Table 2). Statin treatment tended to be associated with PH in the overall population (OR 2.15, CI 0.98-4.76) and significantly associated with PH in IV-tPA treated patients (OR 3.58, CI 1.41-9.05), with the caveat that only two studies were included in the IV-tPA subanalysis.
An association between body temperature and HT by ECASS classification was found in four studies; however, we were not able to aggregate the data for meta-analysis. One study [27] found that a higher body temperature in the first day post-stroke was associated with HT (OR 7.3, CI 2.4-22.6) in non-IV-tPA treated patients. One randomized clinical trial [46] compared outcomes with therapeutic hypothermia in patients who had a middle cerebral artery (MCA) clot recanalized by stent thrombectomy and/or IV-tPA, and found lower rates of HT in (14% versus 39%, p = 0.016) and less cerebral edema (no cerebral edema in 17% with treatment versus 54% without, p = 0.001). Two studies of patients with MCA occlusion treated with IV-tPA [40,47] examining the association of PH with increased body temperature found a trend that almost reached statistical significance [47] in the larger cohort but was not significant in the second study [40].

Stroke Mechanism
Overall rates of HT were substantially different in comparisons of cases with different proposed etiologies.
The DWI ischemic volume was available in five studies including 656 patients [16,20,23,24,31]. DWI ischemic volumes were larger on MRI studies performed on the first day of ictus in HT compared with non-HT patients (HG 0.83, CI 0.1-1.76). The aggregated data did not allow further subgroup analysis; however, one study [23] found that the severity of hypoperfusion on MRI perfusion studies is more predictive of HT than DWI volume. Similarly, three CT perfusion studies [41,52,53] found a direct strong correlation between hypoperfusion and HT.
Several radiological and laboratory biomarkers suggesting blood-brain barrier (BBB) disruption were reported in selected studies. Unfortunately, there were not sufficient data for any marker to enable meta-analysis. Neuroimaging signs of BBB disruption found on dual-energy CT studies performed following stent thrombosis found a strong association with HT (OR 4.5, 1.2-16.4) [54] and parenchymal enhancement on postcontrast T1-weighted MRI following IV-tPA were predictive and localized subsequent HT in one study [55]. Laboratory biomarkers suggesting BBB disruption in IV-tPA treated patients were found to be strong independent predictors of PH in multivariate analysis, including the neutrophil-to-lymphocyte ratio (OR 8.5, CI 2.7-26.9 for a ratio >10.6) [37], baseline matrix-metalloproteinase-9 (OR 9.6, CI 1.3-70.3) [56], and platelet-derived growth factor C (PDGFC), where a level >175 predicted PH with 90% sensitivity [47].

Outcomes
The 90-day mRS was available in four studies that included 1838 patients. Among them, 388 patients (21%) had an mRS of 5-6 at 90 days [7,25,31,57] (Table 4). This high mRS was associated with HT (OR 2.16, 1.7-2.75), and even more strongly associated with PH (OR 5.4, CI 3.2-9.1) in the overall group. When the association with a poor outcome was analyzed the subgroup of IV-tPA treated patients, there was a clear association with HT (OR 2.22, CI 1.7-2.92) and a stronger association with PH (6.25, CI 3.2-12.3). Conversely, in those who did not receive IV-tPA, neither HT nor PH were associated with a poor outcome. In contrast, based on data from three studies including 1197 patients, a 90-day mRS of 0-1 found in 434 patients (36%) [7,31,58] was less frequent in those with HT (OR 0.45, CI 0.28-0.71) and even less likely in those with PH (OR 0.35, CI 0.15-0.86).

Discussion
Aggregated data in this meta-analysis revealed that IV-tPA-treated patients had higher rates of HT (32% versus 20%), PH (12% versus 5%), and PH2 (5% versus 3%) compared to those who did not undergo thrombolysis (p < 0.001 for all conditions). This finding is in agreement with a meta-analysis of individual patient of data from nine random control trials that found that IV-tPA significantly increased the odds of PH2 (OR 5.55, 95% CI 4.01-7.70, p < 0.0001) (6.8% versus 1.3%) [9]. Interestingly, IV-tPA-associated HT was more pronounced in the East-Asian patients, who showed higher rates of IV-tPA associated PH and PH2 compared to non-East-Asians (15% versus 12% and 9% versus 5%, respectively). This increased tendency of East-Asian populations to show higher rates of HT in response to IV-tPA treatment was also reflected in two randomized control trials in Japan and China [8,10], which showed higher rates of severe HT in the regular versus reduced IV-tPA dosage. Moreover, a large observational study found that a low-dose alteplase strategy was comparable to the standard-dose treatment in terms of effectiveness and safety [59].
Similar to previously reported literature [60], our results show that both CT and MRI have similar rates of PH in general and PH2 in particular, thus, allowing us to aggregate patients with PH seen on different imaging modalities without introducing a significant bias.
Understanding the underlying pathophysiological processes that lead to HT may give a plausible explanation to the different variables found to be associated with HT in general and with PH in particular. These processes include ischemia-induced metabolic changes, which, together with an inflammatory response [61] lead to BBB disruption [62]. BBB disruption, along with impairment of cerebral vasculature autoregulatory control, predisposes to blood extravasation when the ischemic tissue is eventually reperfused [63], resulting in HT [64]. Moreover, Matrix metalloproteinase (MMP) expression is related to blood-brain barrier disruption after cerebral ischemia in both IV-tPA treated and untreated ischemic stroke patients [65,66]. In addition, previous systematic reviews performed separately on IV-tPA treated and untreated patient populations, found a clear association between the size of the ischemic territory and prevalence of HT [67].
Among IV-tPA treated patients, PH patients were significantly older than patients without PH (HG 0.27, CI 0.04-0.5) However, this was not found in untreated IV-tPA patients (HG 0.12, CI −0.52-0.76). This difference may be because elderly patients could be more readily susceptible to BBB permeability under ischemia [68] and have a higher burden of cerebral microbleeds (CMBs) [69,70]. A systematic review found that in IV-tPA treated patients, the presence of CMBs increases the risk of symptomatic HT [71].
Male gender was not found to be associated with HT in general, but was associated with PH (OR 1.5, CI 1.07-2.11). One possible explanation could be the independently higher burden of cerebral microbleeds (CMBs) in men (OR 1.7, CI 1.3-2.3) [71].
Chronic hypertension was associated with PH only in the IV-tPA treated patients (OR 1.51, CI 1.1-2-07), possibly due to longstanding hypertension-induced vasculopathic changes such as Charcot microaneurysms and CMBs [72].
DM tended to be associated with HT (OR 1.23, CI 0.97-1.56) and was significantly associated with having PH rather than HI (OR 1.66, 1.05-2.61). Again, it may be attributed to higher burden of small vessel disease (SVD) in diabetic patients. Alternatively, higher glucose levels in diabetic patients may be associated with enhanced bleeding [73]. We found higher acquired generalized lipodystrophy (AGL) in comparisons of HT versus non-HT patients (HG 0.19, CI 0.09-0.3), and PH versus non-PH patients (HG 0.31, CI 0.15-0.48). In an adjusted meta-analysis of studies of IV-tPA-treated patients, higher AGL was associated with more symptomatic intracranial hemorrhage [73]. One possible mechanism is increased BBB disruption [74], which increases blood extravasation into the infarcted brain parenchyma.
Hyperlipidemia was negatively associated with HT in IV-tPA untreated patients. This seemingly protective effect was strong when both LDL level (HG −0.66, CI −0.29-1) and history of dyslipidemia (OR 0.53, CI 0.31-0.93) were examined separately. As seen with the known negative association between spontaneous intracerebral hemorrhage (ICH) and LDL level [75], it is unclear whether LDL itself plays a protective role or whether it is, instead, a biomarker for other mechanisms. A previous study found a body-mass index (BMI) >25 has an independent protective effect against HT (OR 0.39, 0.17-0.87) [76]. This paradoxical effect of obesity has been attributed to previous reports on higher levels of multiple coagulation factors [77] and a suboptimal response to antiplatelet medications [78,79] in obese patients.
Chronic medications play a significant role in the pathophysiology of HT [80]. The finding that previous treatment with anticoagulation increased risk for HT in general and PH in particular seems natural. Interestingly, chronic antiplatelet use was not associated with HT (OR 1, CI 0.83-1.23) but was associated with PH (OR 2.25, CI 1. . We suggest that once bleeding occurs within the brain parenchyma, the ongoing disruption of platelet function facilitates the bleeding process. We presume the ongoing effect of antiplatelet treatment is further intensified in IV-tPA-treated patients, where there was a higher risk of PH (OR 3.15, CI 1.4-7.2). Our findings are in agreement with a large systematic review including 108,588 patients [81], which found that patients receiving long-term antiplatelet medications were associated with greater risks of developing symptomatic intracranial hemorrhage after IV-tPA. Surprisingly, a similar association was found for pretreatment with statins with PH in IV-tPA-treated patients (OR 3.58, CI 1.4-9). Our findings on this point are in contrast with previous large cohorts that did not show independently increased rates of symptomatic HT in IV-tPA-treated patients with a history of statin treatment [82,83]. Statin treated patients may have long-standing atherosclerosis-inducted arteriopathy that could be prone to bleeding. Unfortunately, our inability to adjust for individualized patient data hampers additional investigation of this finding.
Emergency room information is essential for HT risk assessment. Among IV-tPArelated patients, the HT group had higher systolic blood pressure (BP) values compared with those without HT (HG 0.34, CI 0.11-0.57). These differences were not found in patients who did not undergo thrombolysis. The association between higher systolic BP values and symptomatic HT was reported previously [84]. The underlying pathophysiology may include higher perfusion to vessels that lack autoregulation in the setting of IV-tPA. The role of BP in the acute management of patients without IV-tPA treatment deserves further investigation. We found qualitatively that higher temperature increases HT. Animal models have shown that hyperthermia increases BBB permeability, and consequently brain edema, in addition to increasing the fibrinolytic activity of IV-TPA [85].
Stroke severity, as assessed by the NIHSS, showed the highest degree of association with HT (HG 0.96, CI 0.48-1.45), but its association with PH was weaker (HG 0.45, CI 0.22-0.69). Similar findings were found for associations for ischemic volume DWI with HT (HG 0.76, CI 0.02-1.5) and with PH (HG 0.35, CI 0.07-0.62) suggesting that other predictors may play a significant role in PH occurrence.
Stroke mechanism impacts the rate of HT. AF was associated with HT and PH while large vessel atherosclerosis (LVA) was not. This finding is in agreement with a previous meta-analysis of IV-tPA treated patients [86]. The association for AF with PH could be attributed to higher rates of pretreatment with anticoagulation and a larger infarcted territory [87]. However, after adjustment for infarct size, several studies have shown the independent role of a cardioembolic mechanism for HT [57,82,88], and an even higher association with PH [36]. Perhaps the brain parenchyma in patients with LVA has been exposed to ongoing ischemia, leading to collateral flow buildup, and, therefore, is less subject to HT in contrast to abrupt occlusion and possibly recanalization, as seen in cardioembolic etiology.
In regard to functional outcome, PH was strongly associated with a poor outcome (mRS 5-6) in IV-tPA-treated patients (OR 6.25, CI 3.2-12.3), but not in patients who had no IV thrombolysis. Perhaps the pathophysiological mechanisms underlying PH differ between IV-tPA-treated and untreated patients; HT in the non-IV-tPA treated patients is a more natural process of disintegration of the ischemic tissue and takes place in a more gradual way. One large (n = 954 patients) observational study on IV-tPA treated [7] patients highlighted the independent deleterious effect of HT on prognosis.
Our work has several limitations. First, since we lack individual patient data, we can only show the aggregate of the available data and cannot adjust it for possible confounders. Second, HT was assessed in different timeframes and using different imaging modalities, thereby introducing a possible bias. Third, we did not assess the risk of bias for each of the studies included using the Newcastle-Ottawa assessment scale. This was an intentional choice to enable us to include studies from many international locations and that enrolled patients of diverse ethnicities. However, our work also has several possible strengths. It uses unified criteria to assess HT, it is up to date with the current literature and is of large scale, thereby minimizing the risk of bias. We hope this may allow the clinician to determine the likelihood of this event based on an individual patient's pre-stroke historical characteristics and initial clinical, laboratory, and imaging evaluations in the emergency department.

Conclusions
Hemorrhagic transformation is a frequent complication of acute ischemic stroke and is associated with poor outcome. Recognition of risk factors for HT and PH may reduce their incidence and severity.
Prospective studies to further characterize these variables in a longitudinal manner are warranted.