Patients presenting between February 2012 and February 2015 with an acute ischaemic stroke and taking NOACs at the time of the event were prospectively enrolled into the RASUNOA-pilot registry, an investigator-initiated, multicentre, observational cohort study (ClinicalTrials.gov, NCT01850797). Overall, 38 neurology departments with certified stroke units across Germany participated in the registry. Inclusion criteria were age ≥18 years, acute ischaemic stroke, and current therapy with a NOAC (i.e., Apixaban, Dabigatran or Rivaroxaban) at the time of stroke onset. For the present analysis, we excluded (i) patients without any available brain imaging, and (ii) patients with transient ischaemic attack (as defined by presentation without neurological deficit according to the National Institute of Health Stroke Scale (NIHSS) at the time of admission and absence of an acute ischaemic lesion on brain imaging [6]). Due to the observational character of the registry, all diagnostic and treatment decisions were left to the discretion of the attending physicians. Patient characteristics including demographic information, clinical data and laboratory parameters were prospectively collected using a standardized case report file. The modified Rankin scale score (mRS) was used for functional assessment prior to the stroke (pmRS), at admission, and at hospital discharge. Moreover, 3-month outcome was assessed by a structured telephone follow-up. The individual stroke risk in patients with atrial fibrillation was calculated using the CHA₂DS₂VASc-score (excluding the index event), and the individual bleeding risk using the HAS BLED-score, respectively, with the item “labile INR” set to zero [7,8].

The analysis was done stratified for patients not undergoing thrombolysis and/or thrombectomy (no-RT) and patients who received recanalisation therapy (RT), respectively (a focused analysis of the RT group without IV-only treated patients has been previously published [9].

Cranial CT and MRI examinations were assessed for signs of ischaemic infarction, as well as intracranial haemorrhage by an experienced reader (JP). Assessment was performed on scans obtained at admission, and on follow-up scans carried out until day 7 if available. Regarding diagnosis and classification of haemorrhage, findings were reviewed by a board-certified neuroradiologist (MW). Disagreement between the two readers was resolved by consensus. Readers were blinded for patient details. HT was defined as any level of hyperdensity within the area of low attenuation (CT), or hypointensities on T2^* or susceptibility weighted (SWI) sequences [10] in areas corresponding to DWI/FLAIR hyperintense regions on MRI. If rapid attenuation of initial hyperdensities on CT scans was observed, this was categorized as contrast staining rather than hemorrhage [9].

Haemorrhagic transformation was classified in accordance with the ECASS-I-criteria as either haemorrhagic infarction (HI1 and 2), or parenchymal haemorrhage (PH1 and 2) [11]. Accordingly, HI1 was defined as small petechiae along the margin of the infarct, HI2 as more confluent petechiae within the infarct but without space-occupying effect, PH1 as a blood clot not exceeding 30% of the infarct area with some mild space-occupying effect, and PH2 as blood clots exceeding 30% of the infarct area with significant space-occupying effect (Figure 1B) [11]. Moreover, we recorded intracerebral haemorrhage remote from the infarct, which was defined as small or medium-sized blood clots (PHr1) or large confluent dense blood clots (PHr2). Evidence of intraventricular or subarachnoidal haemorrhage was also captured. Symptomatic intracranial haemorrhage (sICH) was defined following the ECASS-II definition [12]. The extent of infarction was quantified using the Alberta Stroke Program Early Computed Tomography Score (ASPECTS; range 0 to 10 with 1 point subtracted for any evidence of ischaemic change in each defined region on axial cuts). In case of posterior circulation stroke, the pc-ASPECTS score was calculated [13,14]. For quantification of the site and size of ischaemic infarction the classification by Paciaroni and colleagues was used [2]. Accordingly, infarct size was classified as (1) small (lesion in the anterior or posterior circulation <1.5 cm), (2) medium (lesion in a cortical superficial branch of middle cerebral artery [MCA], or lesion involving the MCA deep branch, or lesion in internal borderzone territories, or lesion in a cortical superficial branch of posterior cerebral artery [PCA], or lesion involving the PCA branch or lesion in a cortical superficial branch of anterior cerebral artery [ACA]), (3) large anterior (lesion involving complete territory of MCA, PCA or ACA or lesion involving 2 cortical superficial branches of MCA or lesion involving a cortical superficial branch of MCA associated to the MCA deep branch, or lesion involving more than 1 artery territory [e.g., MCA associated to ACA territories]) or (4) large posterior (lesion involving brain stem or cerebellum >1.5 cm) [2]. Cases without definitive evidence of acute cerebral infarction on available scans were added to the Paciaroni category 1.

All analyses were descriptive and exploratory. Continuous variables were described by mean and standard deviation (SD) or median and interquartile-range (IQR) for categorical variables, absolute and relative frequencies were reported. The Shapiro-Wilk test was used to ascertain distribution of data. Post-hoc, groups of patients with haemorrhagic transformation (HT) and without (non-HT) were defined. Fishers-exact test was used to compare proportions in baseline and radiological characteristics, dosing categories (standard [apixaban 5 mg BID, dabigatran 150 mg BID, rivaroxaban 20 mg QD] vs. low drug dose[apixaban 2.5 mg BID, dabigatran 110 mg BID, rivaroxaban 15 mg QD], and considering the actual dose, current renal function and age, adequate or overdosing) and outcomes, as appropriate. To compare continuous variables, t-test or non-parametric Mann-Whitney U test was used according to the skewness of the data. To calculate 95% confidence intervals (CI) of proportions, the adjusted Wald method was used [15]. Logistic regression analysis were performed to estimate odds ratio (OR) and 95% CI for potential risk factors of HT occurrence or unfavourable 3-months-outcome (mRS 3-6). In multivariable analysis the independent effect of HT on unfavourable 3-months-outcome adjusted for a subset of risk factors was assessed. Demographic variables and variables significantly related to outcome in the univariate analyses were included in the model under avoidance of possible collinearity. For variable selection the stepwise backward method with likelihood ratio test was used. Main characteristics of patients from the five best recruiting centres were compared to the standard recruiting ones in a sensitivity analysis. All statistical tests were two-sided, and P values of <0.05 were considered statistically significant. Analyses were conducted using IBM SPSS Statistics, version 23 (IBM Corp., Armonk, NY, USA). This study was performed consistent with the STROBE guidelines for observational studies.

The study was approved by the ethics committee of the Medical Faculty Heidelberg, Germany, and the ethics committees of each participating centre. Written informed consent by either the patient or a legal representative was mandatory for participation. The institutional review board at the primary study site in Heidelberg approved a waiver of consent for patients who died during initial hospitalization and for whom a legal representative could not be located.

Of the 290 patients with acute ischaemic stroke/TIA under NOAC treatment enrolled in RASUNOA-pilot, 231 patients were included in the haemorrhagic transformation main analysis group (patient flow in Figure 1A) and 32 patients who were treated with recanalisation therapies (RT; n=23 endovascular therapy alone, n=5 IV thrombolysis plus thrombectomy, n=4 IV thrombolysis alone) were analysed separately (Figure 1A).

Baseline characteristics of the no-RT group are presented in Table 1, and of the RT group in the online Supplementary Table 1. In the no-RT group, the mean age of included patients was 77 (SD 8.4) years, and 46.8% were female. Median onset to first brain imaging was 3.4 hours (IQR, 2.1-9.9; exact onset of symptoms was available in n=156). Most patients were affected by minor or moderate stroke (median NIHSS 3 [IQR 2-6]). Nineteen-percent suffered from large anterior or posterior strokes (Table 2).

Any follow-up imaging until day 7 was performed in 55.8% (129/231) of the no-RT group. The median time between scans was 49.6 hours (IQR, 22.4-86.0). Baseline characteristics of patients with follow-up imaging did not differ from those without (Supplementary Table 2).

Compared to patients with follow-up imaging and no recanalisation therapy (no-RT), RT-patients were more severely affected (NIHSS at admission 19 [IQR 14-22] vs. 4 [IQR 2-7], P<0.01) and suffered from larger baseline ischaemic lesions (ASPECTS 8 [IQR 7-10] vs. 10 (9-10, P<0.01; Supplementary Table 1).

At admission, HT was present in 9/231 patients of the no-RT group (initial-HT, 3.9%, 95% CI, 2.0 to 7.3; Table 2), and in none of the RT group. A strong trend towards a higher rate of concomitant antiplatelet therapy was observed in the initial-HT patients (33% vs. 9.5%; P=0.06; Table 1). In patients with follow-up imaging (n=129), 13 additional HTs were observed until day 7 (Table 2). In total, 18 cases of HT were registered among patients with available initial and follow-up imaging (early-HT, 18/129, 14.0%, 95% CI, 8.9 to 21.1).

There was no association between standard vs. low drug dose, or adequate or overdosing (considering actual dose and current renal function and age) and occurrence of initial-HT or early-HT in patients with AF. Univariate logistic regression revealed that infarct size had the strongest impact on the occurrence of early-HT (size small vs. medium OR 15.2, 95% CI 2.8 to 82.7; small vs. large OR 19.0, 95% CI 3.9 to 93.7, Table 3).

Symptomatic HT occurred in one patient in the follow-up-imaging study population (1/129, 0.8%, 95% CI, <0.1 to 4.7%; Table 2). This patient suffered from a large MCA-territory infarction but did not show signs of HT on the initial CT-scan. The NOAC was continued after the stroke and a PH2 bleeding developed on day 2. All other observed HT remained asymptomatic.

In an unadjusted analysis, HT was more frequent in patients treated with thrombolysis and/or thrombectomy compared to those without (n=13/32, 40.6% [95% CI, 25.5 to 57.8) vs. 18/129, 14.0% [95% CI, 8.9 to 21.1], P=0.001). No HT was observed in the patients treated with IV thrombolysis only (n=4). However, differences in HT-rates are likely due to higher stroke severity (Supplementary Table 1) in patients treated with recanalisation therapy (after adjustment for stroke severity according to the admission NIHSS, logistic regression for the occurrence on HT became non-significant).

Patients with early-HT were more severely affected at admission (median NIHSS 7 [IQR, 2-19] vs. 3 [IQR, 2-6], P=0.03; Table 1), and suffered more frequently from larger anterior and posterior infarctions than patients without early-HT (P<0.001; Table 2). Patients with early-HT had a significantly worse outcome compared to those without (median mRS 4 [IQR, 1-6] vs. mRS 2 [IQR 1-4], P=0.001; Table 1). However, in multivariate analysis, HT was not independently associated with unfavourable outcome (mRS 3-6) at 3-months (Table 4; for univariate analysis, see Supplementary Table 3). In-hospital mortality was not significantly increased among patients with HT.

In stroke survivors, OAC was resumed during the acute inpatient stay in the majority of patients (n=184/230, 80%; thereof 97% NOAC, 3% phenprocoumon). In patients with early-HT, OAC was resumed after occurrence of HT in all but one case. Median time to recommended resumption was significantly longer in patients with early-HT (15 d [IQR, 5-26]) compared to patients without (1 d [0-4]; P<0.001). Notably, the delay was at least partly mediated through larger infarct-sizes in the early-HT group (see above), as a generally longer time until resumption of OAC was observed in patients without HT but large infarct compared to patients without HT and a small infarct (10 d [IQR, 0-15] and 1 d [IQR, 0-2]; P=0.034; data not shown).