Predicting hemorrhagic transformation and its timing from maximum cerebral lesion diameter in nonlacunar ischemic strokes

Abstract Objectives We performed this retrospective cohort study to establish which factors are mostly indicative of the appearance of hemorrhagic transformation (HT) and of its time course in a sample of nonlacunar ischemic strokes. Materials and Methods In 402 patients with nonlacunar ischemic stroke (75.0 ± 12.7 years, 192 male), clinical, laboratory, and neuroimaging variables obtained during the first 3 days of hospitalization were compared between patients with and without HT at computer tomography scan. Results HT was documented in 129 patients (32.1%), including 36 with parenchymal hematoma (PH), after a median time of 6 days (range 1–27). Many variables were univariately associated with HT, but only 5 of them were confirmed in Cox regression (Hazard Ratio, 95% Confidence Interval): maximum cerebral lesion diameter (CLD) in cm (1.12, 1.06–1.18; p = .0001), hemoglobin in g/dl (1.16, 1.06–1.27; p = .002), blood glucose in mmol/L (1.10, 1.03–1.18; p = .007), prior anticoagulant therapy (1.82, 1.10–3.03; p = .02), and edema with mass effect (1.72, 1.08–2.75; p = .02). Thus, the most significant predictor was CLD. The overall risk of HT was minimum for CLD < 2 cm (1.5%), intermediate for CLD ≥ 2 and < 5 cm (22%), and maximum for CLD ≥ 5 cm (58%). The residual probability of having HT decreased progressively over time, and a simple formula is proposed to predict, from CLD in cm, when the probability of HT falls below 10%. Conclusions The main determinant of HT was CLD, a simple quantitative parameter that could prove useful, in particular, in deciding the timing of anticoagulation in cardioembolic stroke patients.


| INTRODUC TI ON
One of the main unfavorable evolutions of cerebral infarct is hemorrhagic transformation (HT). It is a relatively frequent event (on average 8.5% of the cases, ranging from 0% to 85% in different studies [Lindley et al., 2004]). Its clinical manifestations may vary from total absence of symptoms (mild "hemorrhagic infarct"), up to the progression of neurological deficit and patient's death (severe "parenchymal hematoma").
In patients with cardioembolic stroke, the assessment of HT risk is particularly relevant: even if no signs of bleeding are present at neuroimages, early anticoagulation in high-risk patients might favor subsequent important bleedings (Coull et al., 2002;Marsh, Llinas, Hillis, & Gottesman, 2013). Vice versa, delaying the treatment would uselessly expose low-risk subjects to increased cardioembolic risk. Many studies have dealt with this topic, with particular reference to the identification of the factors that may suggest avoiding thrombolysis (Strbian et al., 2014). In this case, the risk indicators should be immediately available on hospital admission. Instead, in view of deciding when starting oral anticoagulation, the risk indicators may be obtained during the first 3-4 days, as current guidelines suggest that before that time anticoagulation would be hazardous (Kirchhof et al., 2016;Paciaroni et al., 2015Paciaroni et al., , 2017Powers et al., 2018).
Despite this plethora of possible indicators of HT probability, the attempts to practically utilize these items of information have been rather scanty and not easily applicable in the clinical setting.
For example, Marsh et al. (2013) have proposed three parameters (age, cerebral infarct volume, and glomerular filtration rate) which, within the logistic equation, allow the calculation of the probability of HT (HeRS score). In a validation study (Marsh et al., 2016), the same authors decided to add three further parameters (serum glucose, white blood cell count, and warfarin use prior to admission) to obtain a satisfactory area under the ROC curve.
However, the difficulty of measuring infarct volume and calculating HT probability by the logistic equation makes this score hardly applicable in the clinical routine. More recently, Kalinin, Khasanova, and Ibatullin (2017) have proposed a simple scoring system, the Hemorrhagic Transformation Index (HTI), to identify acute ischemic stroke patients at high risk of having HT during the first 14 days of hospitalization (their sample also included lacunar strokes). The index is derived from four items: Alberta Stroke Program Early CT (ASPECT) score, initial NIHSS score, hyperdense middle cerebral artery sign, and atrial fibrillation on admission. The authors reported very good HT predictivity with their index, but they did not provide any indication on how this information should be used to decide the timing of anticoagulation in cardioembolic strokes. In addition, their index can be applied only to strokes in the middle cerebral artery territory.
Based on these premises, we performed a retrospective study in the hypothesis that one or more of the previously assessed variables, or some new factor, may be indicative of the appearance of HT and of its time course in nonlacunar ischemic strokes located in any vascular territory. In particular, we utilized a simple quantitative method to measure infarct size, namely its maximum diameter in cm. In addition, we propose a predictive method, easily applicable at patient's bed, which might prove useful in deciding the timing of anticoagulation in cardioembolic strokes.

| Patients
We considered for inclusion in the study 627 patients with ischemic stroke consecutively admitted to our stroke unit from February 2011 to January 2014, within 24 hr from the onset of symptoms. Ischemic stroke was defined as a sudden and persistent focal neurological deficit not associated with signs of cerebral hemorrhage on first CT scan. Lacunar strokes are associated with ischemic lesions of small size that usually do not present HT. Moreover, since by definition the strokes originated in a small artery are not deemed cardioembolic, they do not need any anticoagulant therapy. For these reasons, small artery strokes, defined according to the Trial of ORG 10,172 in Acute Stroke Treatment (TOAST) criteria (Adams et al., 1993), were excluded from the study (N = 138). In addition, further cases of ischemic stroke without 2 CT scans performed, or without any visible cerebral lesion on second CT scan, were also excluded (N = 87). Thus, the participants were 402 (mean age 75.0 ± 12.9 years; 192 male), including cases of ischemic stroke from large artery, cardioembolism, other determined cause, or undetermined cause according to the TOAST classification ( Figure 1 and Table 1).
Because of the retrospective nature of this study, which included several patients deceased after the stroke, a written informed consent could not be obtained. However, the utilization for the present study of the data derived from medical records was approved by our joint university-hospital Ethics Committee, and the study was conducted in accordance with the WMA Declaration of Helsinki.
Aim of the study was the identification of HT determinants among a series of clinical, laboratory, and instrumental variables obtained during the first 3 days of stay (or, in patients with earlier HT, before HT detection). The patients under antihypertensive treatment, or with mean systolic blood pressure ≥ 140 mmHg or mean diastolic blood pressure ≥ 90 mmHg during the first 3 days of hospitalization, were considered hypertensive. The patients under antidiabetic treatment, or with fasting blood glucose ≥ 7 mmol/L on the day after admission, were considered diabetic. The patients under statin treatment, or with serum total cholesterol ≥ 5.18 mmol/L, were considered hypercholesterolemic. The patients who referred to drink any amount of alcohol in a nonoccasional manner were considered alcohol drinkers. The mean heart rate and systolic and diastolic blood pressure were calculated from the values obtained in the morning during the first 3 days. The maximum body temperature in °C during the first 3 days was also recorded (tympanic temperature was measured 3 times a day). Furthermore, previous major ischemic events, the occurrence of current or previous atrial fibrillation, treatment with antiplatelet or anticoagulant drugs prior to admission, and treatment with i.v. thrombolysis were recorded.
Stroke severity was assessed on admission by the NIHSS score (Lyden et al., 1999) and the Oxfordshire Community Stroke Project classification (Bamford, Sandercock, Dennis, Burn, & Warlow, 1991). The impairment of consciousness was assessed on admission by the Glasgow Coma Scale (Teasdale & Jennett, 1974). Finally, the etiology of stroke was established according to the TOAST criteria (Adams et al., 1993).

| Brain CT scan execution and assessment
In our hospital, routine neuroimaging for stroke is normally obtained by noncontrast CT scans, as suggested by current guidelines (Powers et al., 2018). Brain CT scans were performed by a LightSpeed scanner (General Electric Medical Systems), and the measurements on CT scans were performed by Kodak Carestream PACS Web Software (Eastman Kodak Company). All CT scans were reassessed and agreed by 2 expert neuroradiologists (L.F. and L.S.) who inspected the images together and recorded the following aspects: 1. Hemorrhagic transformation -HT was classified into 4 categories according to the European Cooperative Acute Stroke Study (ECASS) criteria (Fiorelli et al., 1999): HI 1 (hemorrhagic infarct with isolated petechiae), HI 2 (hemorrhagic infarct with confluent petechiae), PH 1 (parenchymal hematoma ≤ 30% of the infarcted area with some mild mass effect), and PH 2 (parenchymal hematoma > 30% of the infarcted area with mass effect usually causing midline shift).
These aspects were searched in all CT scans performed during the stay (between 1 and 27 days after admission). In this study, white matter lesions were defined as any degree ≥ 1 according to Van Swieten, Hijdra, Koudstaal, and Gijn (1990).

Cerebral lesion diameter (CLD) -
5. Hyperdense appearance of middle cerebral artery, suggesting possible thrombosis, visible on first CT scan.

| Laboratory variables
Routine laboratory variables were obtained from the fasting blood sample withdrawn on the first morning after admission ("the baseline"). The laboratory variables obtained on admission were disregarded, as they could potentially be altered by the acute phase.

| Statistical analysis
Normal continuous variables were described by mean ± SD, while the variables with nongaussian distribution were described by median and interquartile range. Means were compared with Student's t test, medians with Mann-Whitney's test and percentages with chi-squared test.
Multivariable analyses were performed by Cox proportional hazards regressions with backward stepwise procedure to obtain hazard ratios and 95% confidence intervals, with any HT or PH as TA B L E 1 Baseline clinical variables and risk factors in patients with and without hemorrhagic transformation of cerebral lesion
As shown in In addition (Table 2), the patients with any HT or PH had higher baseline values of white blood cells, hemoglobin, hematocrit, C-reactive protein, blood glucose and lower levels of HDL cholesterol (except PH patients), and, more frequently, a hyperdense appearance of middle cerebral artery. Finally, they also had larger ischemic lesions and more often mass effect.
The variables associated with HT or PH with p values < .05 were included as independent variables in two Cox regressions, where the dependent variables were, respectively, HT and PH (Table 3).
In these analyses, the patients with missing data were excluded. In addition, the analysis with PH as dependent variable did not include the patients with HI. At the end of the procedures, four variables remained significantly associated with both HT and PH: CLD, hemoglobin concentration (which was missing in two non-HT patients), blood glucose (missing in six non-HT and three HT patients), and mass effect. Preadmission anticoagulant therapy remained associated with HT only. Table 3 shows that the most significant determinant of any HT was CLD (p = .0001), with a 12% increase in HT probability for every cm of diameter. Table 4 shows that the relationship between CLD and frequency of HT was very close and progressive. The table also shows that for each lesion size the cumulative frequency of HT increased in relation to the time of CT execution. Six cases of HT were detected on admission (4.7% of all HTs, including 2 PHs), namely, before CLD and edema assessment and baseline blood sampling.
These patients were not excluded, not to alter the complete pic-  after stroke onset would be the best one to start anticoagulation.

| D ISCUSS I ON
This study has shown that, in patients with nonlacunar ischemic stroke, among numerous variables obtained during the first 3 days of hospitalization, the variable most closely associated with the appearance of HT was the size of cerebral lesion. It seems therefore reasonable to estimate the probability of HT and its timing starting from this parameter.
The frequency of HT in our study (32.1%) was higher than the average frequency reported in the literature (Lindley et al., 2004).
This finding is explainable considering that lacunar strokes were excluded from the study and that all the available CT scans were used to detect HT (up to 27 days after the acute event).
It is well known that the size of the ischemic lesion is an important factor favoring HT (Kerenyi et al., 2006;Lee et al., 2010;Marsh et al., 2016;Paciaroni et al., 2008). According to the current European guidelines on atrial fibrillation (Kirchhof et al., 2016), both the size of brain infarction and HT should be taken into account to decide when starting anticoagulation. However, the same guidelines only generically suggest "delaying" oral anticoagulation in the presence of "large/moderate" lesions, without specifying how long the delay should be and what large/moderate means.
In addition, NIHSS score is used as primary indicator of the timing of anticoagulation. NIHSS score is correlated, in principle, with the extension of the ischemic lesion. However, in some cases, low TA B L E 2 Laboratory and neuroradiologic variables in patients with and without hemorrhagic transformation of cerebral lesion NIHSS scores may be associated with large lesions (nondominant hemisphere, occipital lobe, and cerebellum), while on the contrary, there may be small lesions causing elevated NIHSS scores (internal capsule and brainstem). And, in fact, in our multivariable analysis, which also included CLD, NIHSS was eliminated. Moreover, several other factors (e.g., mechanical valve prosthesis, severe uncontrolled hypertension, or other determinants evidenced in this study) should be considered in deciding whether to anticipate or delay the timing of anticoagulation.
If CLD has to be used as one possible factor indicating the timing of anticoagulation, we should remember that the ischemic lesion is clearly visible on CT scan only after 24-72 hr from the onset of symptoms. This delay would be, however, compatible with the fact that initiating anticoagulation earlier than 3-4 days would be hazardous according to Paciaroni et al. (2015). A previous study from the VISTA registry (Abdul-Rahim et al., 2015) suggested that anticoagulation could be performed safely even earlier. However, the authors recognized that the highest risk of both stroke recurrence and symptomatic intracerebral hemorrhage was in the first 2 days.
Nonvitamin K antagonist oral anticoagulants (NOACs) are associated with a lower risk of intracranial hemorrhage than vitamin K antagonists. Thus, they might be prescribed earlier after an acute stroke.
This possibility is currently being assessed in three different randomized trials (ELAN, START, and TIMING; Åsberg et al., 2017;Ng & Whiteley, 2017). However, a large registry study on the same subject (Paciaroni et al., 2017) recently showed that the minimum composite rate of recurrence and major bleeding was obtained, again, when NOACs were started between 3 and 14 days after acute stroke.
After literature revision, Paciaroni, Agnelli, Ageno, and Caso (2016) have proposed an assessment of the probability of HT based on the size of the ischemic lesion (small, medium, or large), considering small infarcts those <1.5 cm, medium infarcts those caused by partial occlusion of cerebral arteries, and large infarcts those caused by complete occlusion of cerebral arteries. Other authors have used ischemic lesion volume, rather than its diameter, to assess the probability of HT (Marsh et al., 2013(Marsh et al., , 2016. This measurement is certainly more precise, but is more complex to perform, and therefore, it is not suitable for routine use. Instead, as we have shown in the present investigation, the simple measurement of CLD may provide useful information about the occurrence and timing of HT in everyday practice. Usually, edema with mass effect is associated with large ischemic lesions. However, our multivariable analysis showed that mass effect was associated with HT (especially PH) independently from CLD. On the other hand, mass effect may be a consequence, rather than a cause, of PH, as PH itself determines mass effect.
The existence of a direct association between hemoglobin and HT had never been reported previously. It is noteworthy that hemoglobin was the second parameter most strongly associated with HT after CLD. Thus, it seems that polycythemia may favor HT, while anemia would have a protective effect. The reasons for this are presently unknown, although some explanations can be proposed.
HT generally occurs during cerebral lesion reperfusion (Fiehler et al., 2005). Ischemia-reperfusion injury is mainly due to an oxidative mechanism (Kaminski, Bonda, Korecki, & Musial, 2002), also favored by reduced shear stress in the ischemic area (Kumar et al., 2018). Ensuing endothelial dysfunction increases vascular permeability (Lum & Roebuck, 2001), with possible blood extravasation (Lin et al., 2007). While hematological disorders and polycythemia are well-known (although sometimes unrecognized) causes of ischemic stroke (Arboix, Jiménez, Massons, Parra, & Besses, 2016), anemia with reduced blood density might favor the microcirculation, with increased shear stress, less endothelial dysfunction, and reduced permeability of the blood-brain barrier. In addition, the free iron released by erythrocytes exposed to hypoxia-reoxygenation may contribute to oxidant stress (Ciccoli et al., 2004). Thus, fewer red blood cells could mean less free iron and less oxidant stress in the ischemic area. Finally, ischemic and hypoxic preconditioning has been shown to protect the brain from more severe ischemic injuries (Arboix et al., 2004;Ran, Xu, Lu, Bernaudin, & Sharp, 2005;Steiger & Hänggi, 2007). Anemia might possibly exert a protective preconditioning effect, with reduced occurrence of HT.
Current guidelines do not recommend recoagulation in the acute phase of ischemic stroke, in the patients that were previously on oral anticoagulant therapy. Thus, oral anticoagulation is just withdrawn, and if warfarin was used, complete spontaneous recoagulation occurs after a few days. It is therefore understandable that these patients have an increased risk of HT, especially during the first few days of stay, as already shown by others (Marsh et al., 2016). The finding that hyperglycemia favors HT is also in agreement with several previous studies (Kerenyi et al., 2006;Kunte et al., 2012;Marsh et al., 2016;Paciaroni et al., 2009Paciaroni et al., , 2008Wang et al., 2014).
The main limitations of this study derive from its retrospective design, particularly from the fact that only the available data were collected and some data (especially radiological) were not obtained after prespecified intervals from stroke onset. However, all patients underwent at least two CT scans, and those who underwent three or more scans were numerous and equally distributed in the three areas of risk. In addition, performing multiple scans at regular intervals in all patients would be complicate and probably unethical.
Six patients had HT detected on first CT scan, before some radiological and laboratory variables could be obtained. These patients were not excluded, not to alter the description of HT incidence and timing in its entirety. It seems improbable, however, that these few cases may have significantly influenced the predictive model. The numbers in bold are those experimentally found. The other numbers are extrapolated. a See Figure 3 and text.
anticoagulant therapy in patients with PH, had been few, and this might have caused the nonsignificance of their association with HT. Another limitation comes from the possibility that the therapy administered during the first days of stay may have influenced the results. Finally, these results should be confirmed in a different cohort of patients.

| CON CLUS IONS
Although with limitations, this study has shown that, in patients with nonlacunar ischemic stroke, 4 already known parameters (infarct size, mass effect, hyperglycemia, and preadmission anticoagulant therapy) and one new parameter (hemoglobin concentration) collected during the first 3 days of hospitalization are independently associated with the appearance of hemorrhagic transformation. In particular, we have found a simple relationship linking the most relevant of these parameters, cerebral lesion diameter, to the timing of hemorrhagic transformation, which could be considered in deciding when starting or restarting anticoagulation in cardioembolic strokes.

ACK N OWLED G M ENTS
This work was self-financed. No sources of funding have to be acknowledged.

CO N FLI C T O F I NTE R E S T
The authors have no conflict of interest to declare.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.