doi:10.1093/brain/aww199

BRAIN 2016: 139; 2694–2705

| 2694

Delayed seizures after intracerebral haemorrhage Alessandro Biffi,1,2,3 Abbas Rattani,4 Christopher D. Anderson,2,3,5,6 Alison M. Ayres,2 Edip M. Gurol,2,7 Steven M. Greenberg,2,7 Jonathan Rosand2,3,5,6 and Anand Viswanathan2,7

Late seizures after intracerebral haemorrhage occur after the initial acute haemorrhagic insult subsides, and represent one of its most feared long-term sequelae. Both susceptibility to late seizures and their functional impact remain poorly characterized. We sought to: (i) compare patients with new-onset late seizures (i.e. delayed seizures), with those who experienced a recurrent late seizure following an immediately post-haemorrhagic seizure; and (ii) investigate the effect of late seizures on long-term functional performance after intracerebral haemorrhage. We performed prospective longitudinal follow-up of consecutive intracerebral haemorrhage survivors presenting to a single tertiary care centre. We tested for association with seizures the following neuroimaging and genetic markers of cerebral small vessel disease: APOE variants "2/"4, computer tomography-defined white matter disease, magnetic resonance imaging-defined white matter hyperintensities volume and cerebral microbleeds. Cognitive performance was measured using the Modified Telephone Interview for Cognitive Status, and functional performance using structured questionnaires obtained every 6 months. We performed time-to-event analysis using separate Cox models for risk to develop delayed and recurrent seizures, as well as for functional decline risk (mortality, incident dementia, and loss of functional independence) after intracerebral haemorrhage. A total of 872 survivors of intracerebral haemorrhage were enrolled and followed for a median of 3.9 years. Early seizure developed in 86 patients, 42 of whom went on to experience recurrent seizures. Admission Glasgow Coma Scale, increasing haematoma volume and cortical involvement were associated with recurrent seizure risk (all P 5 0.01). Recurrent seizures were not associated with long-term functional outcome (P = 0.67). Delayed seizures occurred in 37 patients, corresponding to an estimated incidence of 0.8% per year (95% confidence interval 0.5–1.2%). Factors associated with delayed seizures included cortical involvement on index haemorrhage (hazard ratio 1.63, P = 0.036), pre-haemorrhage dementia (hazard ratio 1.36, P = 0.044), history of multiple prior lobar haemorrhages (hazard ratio 2.50, P = 0.038), exclusively lobar microbleeds (hazard ratio 2.22, P = 0.008) and presence of 5 1 APOE "4 copies (hazard ratio 1.95, P = 0.020). Delayed seizures were associated with worse long-term functional outcome (hazard ratio 1.83, P = 0.005), but the association was removed by adjusting for neuroimaging and genetic markers of cerebral small vessel disease. Delayed seizures after intracerebral haemorrhage are associated with different risk factors, when compared to recurrent seizures. They are also associated with worse functional outcome, but this finding appears to be related to underlying small vessel disease. Further investigations into the connections between small vessel disease and delayed seizures are warranted.

1 2 3 4 5 6

Division of Behavioral Neurology, Department of Neurology, Massachusetts General Hospital, Boston, MA, USA J. Philip Kistler Stroke Research Center, Massachusetts General Hospital, Boston, MA, USA Program in Medical and Population Genetics, Broad Institute of Harvard and MIT, Cambridge MA, USA School of Medicine, Meharry Medical College, Nashville, TN, USA Center for Human Genetic Research, Massachusetts General Hospital, Boston, MA, USA Division of Neurocritical Care and Emergency Neurology, Department of Neurology, Massachusetts General Hospital, Boston, MA, USA 7 Division of Stroke, Department of Neurology, Massachusetts General Hospital, Boston, MA, USA

Received February 26, 2016. Revised May 16, 2016. Accepted June 20, 2016. Advance Access publication August 6, 2016 ß The Author (2016). Published by Oxford University Press on behalf of the Guarantors of Brain. All rights reserved. For Permissions, please email: [email protected]

Delayed seizures after ICH

BRAIN 2016: 139; 2694–2705

| 2695

Correspondence to: Alessandro Biffi, MD, Division of Behavioral Neurology, Department of Neurology, Massachusetts General Hospital, 100 Cambridge St., 20-2023, Boston, MA 02114, USA E-mail: [email protected]

Keywords: intracerebral haemorrhage; seizure; epilepsy Abbreviations: ADL = Activity of Daily Living; CSVD = cerebral small vessel disease; CT-WMD = CT-defined white matter disease; ICH = intracerebral haemorrhage

Introduction Intracerebral haemorrhage (ICH) is the most severe form of acute stroke, accounting for 15–25% of all strokes, and 50% of stroke mortality (Badjatia and Rosand, 2005; Poon et al., 2014). It is thought that the acute haemorrhage represents the cumulative expression of long-standing cerebral small vessel disease (CSVD), mainly caused by hypertensive damage (arteriolosclerosis) and/or cerebral amyloid angiopathy (Pantoni, 2010). The acute phase of an ICH is often complicated by seizures (referred to as early seizures), likely reflecting the disruptive effect of the haematoma and surrounding oedema. Survivors of acute ICH are also at high risk for long-term sequelae, including late seizures. Prior investigations into the nature of late seizures did not discriminate between recurrent seizures (occurring after early seizures) and newly diagnosed delayed seizures (Bladin et al., 2000; Passero et al., 2002; De Reuck et al., 2007; Rossi et al., 2013). Recently, a clinical score for late seizure risk prediction after ICH was proposed, following the aforementioned approach of jointly investigating all forms of post-ICH late seizures (Haapaniemi et al., 2014). However, this score’s predictive performance in a validation cohort was noted to be suboptimal, raising the possibility of a more complex biological substrate for late seizures after ICH than previously postulated. We hypothesized that delayed seizures are associated with different risk factors, compared to recurrent seizures, and that established neuroimaging and genetic markers for underlying CSVD increase delayed seizure risk, reflecting the importance of small vessel disease burden in influencing long-term seizure risk. Furthermore, we sought to clarify whether recurrent and/or delayed seizures are associated with higher mortality and worse functional outcome after ICH. To test these hypotheses we analysed prospectively collected data for ICH survivors enrolled in the Massachusetts General Hospital (MGH) ICH Longitudinal Study. We performed univariable and multivariable analyses to identify risk factors for late seizures, and conducted formal heterogeneity testing to identify variables specifically conferring risk for recurrent versus delayed seizures. We focused on testing associations between delayed seizures and

established CSVD markers. Finally, we explored whether late seizures (recurrent versus delayed) are associated with worse functional outcome after ICH.

Materials and methods Study design and terminology The central hypothesis of the present study focuses on the possibility of biological and clinical heterogeneity in postICH late seizures. We specifically postulated that seizures occurring in the acute ICH phase (or recurring after and acute-phase seizure event) are primarily caused by the disruption of cortical networks by the haematoma via its structural damaging properties. In contrast, seizures manifesting for the first-time in a delayed manner (i.e. beyond the acute ICH phase), may be attributable to the more subtle cortical damaging effects of underlying CSVD, acting slowly but progressively over time. To test this hypothesis, we chose to perform a parallel, comparative analysis of seizure risk among ICH survivors presenting to our institution and followed longitudinally over time. We initially separated seizure events after ICH into early seizures (occurring within 7 days from ICH symptoms’ onset) and late seizures (occurring beyond 7 days from the initial ICH event), in agreement with recommendations formulated by the International League Against Epilepsy for definition of acute symptomatic seizure (Beghi et al., 2010). Late seizures were then further classified as recurrent (if patients previously experienced an early seizure) or delayed. Because different criteria for definition of late seizures were used in other studies (Passero et al., 2002), we conducted sensitivity analyses using cut-offs at 15 and 30 days post-ICH. These additional analyses (Supplementary Tables 1 and 2) returned essentially identical results to those presented below.

Patient recruitment and baseline data collection All participating individuals were enrolled in the MGH ICH Longitudinal Study, an ongoing single-centre longitudinal cohort study of ICH (Biffi et al., 2010a, 2015). Initial screening criteria for inclusion in all planned analyses were: (i) age 518 years; (ii) admitted to MGH from January 2006 to December 2013; (iii) diagnosed with primary ICH, i.e. not related to trauma, conversion of an ischaemic infarct, rupture

2696

| BRAIN 2016: 139; 2694–2705

of a vascular malformation/aneurysm, or a brain tumour (Fig. 1). Dedicated study staff collected all pre-enrolment baseline data via review of existing medical records and billing information, combined with a structured standardized inperson interview (Biffi et al., 2015). As the primary goal of this study was to investigate post-ICH seizures, we excluded all subjects diagnosed with seizure disorder prior to index ICH (diagnostic information derived from medical records, billing information, interview as per above). Pre-ICH dementia was identified by administering to reliable informants the 16-item (short) version of the Informant Questionnaire on Cognitive Decline in the Elderly (IQCODE). As per reported normative data any subject with an average score of 43.3 was diagnosed with pre-ICH dementia (Jorm, 2004). Pre-ICH functional dependence was defined as inability to perform any Activity of Daily Living (ADLs) or Instrumental Activity of Daily Living (IADLs). All subjects required an index ICH admission CT scan (for diagnostic confirmation and haematoma localization), obtained within 24 h of onset of symptoms. The study protocol was approved by the MGH Institutional Review Board. Written informed consent was obtained from all study participants or their surrogates.

Definition and capture of seizure events ICH survivors were considered to have experienced a seizure event if at least one of the following criteria were met: (i) electrographic evidence of seizure on EEG with or without corresponding clinical symptoms; (ii) witnessed clinical evidence of transient neurological phenomena, deemed by the attending neurologist of record to represent manifestations of electrographic seizures after direct clinical examination of the patient and review of pertinent medical records; or (iii) unwitnessed clinical symptoms deemed consistent with electrographic seizure by an epilepsy specialist, following direct clinical examination of the patient and review of pertinent medical records. Seizure events were then categorized as early, recurrent or delayed as described above. Study staff interviewed patients by phone (see below) and inquired about new seizure events, as well as new neurological symptoms (including cognitive, behavioural, motor and/or sensory abnormalities). Additional data to augment and/or confirm findings from patient interview were obtained via: (i) look-up of relevant seizure and epilepsy-related billing codes; (ii) systematic review of electronic medical records based on an established protocol; and (iii) request and review for external medical records, triggered as appropriate by phone-based interviews (Biffi et al., 2010a, 2015). In the absence of definitive EEG evidence of seizure disorder, clinical diagnosis was deemed acceptable only if formulated by the attending neurologist (for witnessed clinical events) or epileptologist (for unwitnessed clinical events). For all diagnostic purposes, EEG findings consistent with interictal epileptiform discharges failing to meet criteria for electrographic seizure(s) were not considered sufficient to formulate an electrographic diagnosis of seizure disorder (i.e. in the absence of pertinent witnessed or unwitnessed clinical findings). Initiation of antiepileptic drugs alone did not qualify for a diagnosis of seizure disorder. In case of an alternative diagnoses for previously observed transient neurological symptoms (e.g. transient

A. Biffi et al.

ischaemic attacks, psychogenic non-epileptic spells) being formulated later in follow-up, a seizure disorder diagnosis was retained only in the presence of definitive EEG evidence of electrographic seizures.

Genetic and neuroimaging data acquisition and interpretation We determined APOE genotype for subjects consenting to blood draw and genotyping by analysing DNA extracted from blood samples, according to previously published methods (Biffi et al., 2010b). ICH location (lobar, deep, cerebellar, or multiple locations/mixed) was assigned based on consensus review of index ICH CT scans by study staff (Biffi et al., 2010a, 2015). CT-defined white matter disease (CT-WMD) was quantified using a previously validated 4-point scale, separately grading severity of anterior and posterior white matter disease (van Swieten et al., 1990; Biffi et al., 2010a). MRI with axial gradient-echo images was performed in a subset of patients within 90 days of onset of symptoms, according to previously described methods (Smith et al., 2004). For patients with available MRI scans, we quantified white matter hyperintensity (MRI-WMH) volume, as previously described (Rost et al., 2010a,b, 2014). On available MRI scans we also counted cerebral microbleeds, and classified them by location in lobar and non-lobar, as previously described (Biffi et al., 2010a). All imaging analyses were performed and results recorded by study investigators without knowledge of participating subjects’ clinical and/or genetic information.

Longitudinal follow-up Trained study staff contacted and interviewed by telephone patients and/or their caregivers every 6 months, and inquired about: (i) new seizure events as per above; (ii) cognitive function via administration of the Modified Telephone Interview for Cognitive Status (TICS-m) test (Brandt et al., 1988; de Jager et al., 2003; Barber and Stott, 2004; Knopman et al., 2010; Seo et al., 2011; Pendlebury et al., 2013); (iii) functional status, as defined by performance on ADLs, IADLs, and the modified Rankin Scale (mRS); (iv) ICH recurrence (confirmed on neuroimaging); (v) interval health history changes, including new medical diagnoses and changes in medication regimens; and (vi) death events (Biffi et al., 2015). We augmented and validated phone-based data collection with review of medical records at our centre, as well as from external emergency departments, outpatient primary care and specialist practices, as previously described in detail (Biffi et al., 2015).

Statistical analyses Variable definition and handling Age at index ICH was analysed as a continuous variable. Patients’ sex, race, and education were analysed as static categorical variables. APOE genotype was analysed using two categorical variables indicating number of "2 alleles (0, 1, or 2) and number of "4 alleles (0, 1, or 2) with the "3/"3 genotype serving as reference (Biffi et al., 2010b, 2011a). CT-defined volumes for index ICH (and intraventricular component, if any) were analysed as continuous variables. CT-WMD was

Delayed seizures after ICH

BRAIN 2016: 139; 2694–2705

| 2697

Figure 1 Participants’ enrolment and eligibility criteria flow chart. Flow-chart summarizing sequential application of eligibility/exclusion criteria leading to definition of final study population. Solid bordered boxes in the centre report number of patients fulfilling eligibility criteria at each stage, thus being evaluated further for study inclusion. Double-lined bordered boxes at the bottom indicate participants selected for analyses mentioned in the ‘Results’ section. Criteria for eligibility at each stage are listed in grey background boxes on the right. Dashed lines connect to dashed boxes listing criteria for exclusion from analysis and number of participants excluded as a result.

expressed as an ordinal variable (0–4), indicating increasing severity (Biffi et al., 2010a). cerebral microbleeds were analysed using an ordinal variable with previously validated cut-offs (0, 1, and 52), with separate variables to capture cerebral microbleeds location (lobar versus deep) (Biffi et al., 2010a). MRIWMH volumes were log-transformed for normality and analysed as a continuous variable as previously described (Rost et al., 2010b). Seizure events were recorded at time of diagnosis as dichotomous variables for time-to-event analyses (see below). We quantified functional outcome using: (i) three separate dichotomous variables for mortality, incident dementia, and loss of functional independence (unable to perform independently any ADL or IADL); and (ii) a composite dichotomous functional outcome variable, indicating presence of any of the aforementioned negative outcomes (i.e. mortality, dementia, functional

dependence). We defined incident dementia for all outcome analyses based on: (i) relevant ICD-9 codes entered by attending physicians; and/or (ii) subjects having TICS-m scores 520 based on previously published normative data (Barber and Stott, 2004; Pendlebury et al., 2013). All changes in outcome variables were recorded at time of diagnosis for time-to-event analyses (see below).

Statistical models We used time-to-event univariable and multivariable analyses to identify risk factors associated with recurrent and delayed seizure risks. Risk factors associated with recurrent or delayed seizure risk were first assessed in univariable analyses using log-rank tests. Candidate independent variables for multivariable modelling included all those with P 5 0.2 for association

2698

| BRAIN 2016: 139; 2694–2705

with the outcome of interest. Multivariable analyses employed Cox regression models (separate models for recurrent versus delayed seizure risks). Medication exposures were treated as time-varying variables in all Cox regression analyses. Of note, variables associated with late seizure risk in either recurrent or delayed statistical models were included in both analyses for comparison purposes. After variable selection, a minimal model was generated by backward elimination of non-significant variables (P 4 0.05). Again, variables associated with either recurrent or delayed seizure risk were included in both multivariable analyses for comparison purposes (even if failing to achieve P 5 0.05 for association in the parallel analysis). Heterogeneity of effects for association of risk factors with recurrent versus delayed seizures after ICH were evaluated for statically significance using the metareg function, part of the meta package for the R statistical program. We evaluated different Cox models’ performance in predicting late seizure risk after ICH using Harrell’s C (Harrell et al., 1982), and compared them using the likelihood-ratio test (LRT) method as implemented in the coxph function of the survival package for the R statistical program. We addressed multiple testing burden by adopting the false discovery rate (FDR) method as developed by Benjamini and Hochberg (Keselman et al., 2002). All P-values reported are adjusted for multiple testing with the FDR methods. All significance tests were two-tailed, and significance threshold was set at P 5 0.05 (after FDR adjustment). All analyses were performed with R software v 3.2.0 (The R Foundation for Statistical Computing).

Results Study participants A total of 978 patients age 5 18 years presented to our centre and were diagnosed with primary ICH during the prespecified enrolment time (Fig. 1). Of these, 872 met all eligibility criteria and were retained for analyses. Of note, a substantial proportion of excluded subjects (52/106, 49%) were deemed ineligible due to pre-ICH diagnosis of epilepsy. Baseline and follow-up characteristics for participating individuals are summarized in Table 1 (including univariable P-values for association with recurrent versus delayed seizure risk).

Early and recurrent seizures: diagnosis and incidence Among eligible subjects, 86 (10%) were diagnosed with early seizures, having manifested clinical and/or electrographic evidence of seizure events within 7 days of ICH. Of these, 26/86 (30%) were diagnosed by joint capture of EEG data and clinical symptoms, 18/86 (21%) by analysis of EEG data alone, and 42/86 (49%) by observation of clinical symptoms alone. ICH survivors that were alive at 7 days (685/872, 79%) were followed longitudinally for a median of 3.9 years. We found evidence of recurrent seizures in 42/86 cases (49%), with median time to recurrent

A. Biffi et al.

seizure of 8.7 months [interquartile range (IQR) 5.2–10.3 months]. Of these, 5/42 (12%) were diagnosed by joint capture of EEG data and clinical symptoms, 3/42 (7%) by analysis of EEG data alone, and 34/42 (81%) by observation of clinical symptoms alone. Among patients with recurrent seizures, 29/42 (69%) had a single seizure, 7/42 (17%) had two recurrent seizures, and 6/42 (14%) had three or more recurrent seizures. All early seizure patients received treatment with antiepileptic drugs at time of seizure diagnosis. Among recurrent seizure patients, 37/42 (88%) were still receiving antiepileptic treatment at time of the first recurrent seizures.

Delayed seizures: diagnosis and incidence Among those that were alive and seizure-free at 7 days (n = 599), we diagnosed 37 delayed seizure cases (6%), with median time to delayed seizure of 7.5 months (IQR 4.4–11.2 months). Of these, 5/37 (13%) were diagnosed by joint capture of EEG data and clinical symptoms, 5/37 (13%) by analysis of EEG data alone, and 27/37 (73%) by observation of clinical symptoms alone. Among patients with delayed seizures, 20/37 (54%) had a single seizure, 12/37 (32%) had two delayed seizures, and 5/37 (14%) had three or more recurrent seizures. All delayed seizure patients received treatment with antiepileptic drugs after the first seizure event. Based on the above reference data we estimated delayed seizure incidence at 0.8% per year [95% confidence interval (CI) 0.5–1.2%]. We present detailed information about temporal distribution of early and delayed incident seizure cases in Fig. 2.

Identification of risk factors for delayed seizures We performed univariable and multivariable time-to-event analyses to identify risk factors associated with delayed seizure risk (Table 2). We initially performed analyses including demographics, clinical and CT imaging data (available for all study subjects). In these univariable analyses we identified pre-ICH dementia, prior lobar ICH before index event, cortical ICH involvement, increasing ICH volume and CT-WMD severity as associated with delayed seizure risk (all P 5 0.05). In multivariable analyses, only pre-ICH dementia, prior lobar ICH before index event, cortical ICH involvement, and CT-WMD severity were independently associated with delayed seizure risk. By contrast, increasing ICH volume was not confirmed as an independent risk factor for delayed seizures. We subsequently expanded our modelling to include genetic (APOE) and MRI (MRI-WMH, cerebral microbleeds) data, which were only available for a subset of participants (n = 522, Fig. 1). All previously identified risk factors for delayed seizure risk were confirmed in these analyses. Additionally, we identified possession of 51 copies of the

Delayed seizures after ICH

BRAIN 2016: 139; 2694–2705

| 2699

Table 1 Subjects’ characteristics All participants

Early seizures

Late seizures Recurrent

n n Demographics Age, mean (SD) Sex, male Education, 510 years Medical history Hypertension CAD/MI DM ICH before enrolment event Lobar Deep Cerebellar Pre-ICH dementia Pre-ICH functional dependence Mood disorder Anxiety disorder Acute ICH characteristics ICH Location Lobar Cortical involvement No cortical involvement Deep Cerebellar Multiple locations Primary IVH ICH volume, median, cm3 (IQR) IVH volume, median, cm3 (IQR) Admission GCS 3 to 10 11 to 14 15 Neuroimaging data CT-WMD Overall Anterior Posterior MRI-WMH volume, median, cm3 (IQR)a Cerebral microbleeds (presence)a Any location Strictly lobar Strictly non-lobar Mixed Genetic data APOE "2, MAFa APOE "4, MAFa

%

872

n

%

P

86

n

%

Delayed P

42

n

%

P

37

71.0 (12.3) 462 523

53 60

71.5 (12.0) 41 48

0.84 48 0.76 56 0.65

72.0 (10.7) 21 23

0.77 50 0.54 55 0.82

73.4 (11.6) 18 21

0.15 49 0.36 57 0.78

619 174 166

71 20 19

59 16 19

29 8 9

6 1 2 0 0.3 0 12 7 14 12

1 0 0 4 5

69 0.88 19 0.73 21 0.69 0.33 2 0 0 10 0.49 11 0.38

27 8 6

52 17 3 105 122

69 0.39 19 0.80 22 0.67 0.24 1 0 0 7 0.21 13 0.36

4 1 0 5 6

73 0.52 22 0.43 16 0.47 0.002 11 3 0 14 0.044 16 0.52

131 61

15 7

14 0.50 5 0.88

6 3

14 0.64 7 0.28

5 3

14 0.83 8 0.62

354 266 88 399 110 6 3 18.7 (5.0 - 23.4) 7.6 (2.5 - 28.4) 131 174 567

13 5

0.011 41 44 51 31 37 43 10 7 8 46 34 40 13 8 9 0.7 0 0 0.3 0 0 23.9 (6.5–26.7) 0.008 8.5 (2.9–31.3) 0.012 0.049 15 16 19 20 21 24 65 49 57

523 480 436 17.6 (7.5–33.2)

60 55 50

53 62 0.48 45 52 0.62 42 49 0.40 18.6 (8.1–29.3) 0.23

318 136 130 52

61 26 25 10

28 12 11 5

0.09 0.16

0.11 0.17

64 29 24 11

0.22 0.61 0.59 0.80 0.12 0.60

0.009 23 19 4 16 3 0 0 24.6 (6.1–29.3) 8.9 (3.0–34.5) 10 9 23

19 13 6 14 3 0 0 0.006 21.8 (5.3–24.3) 0.011 7.9 (3.0–27.0) 0.039 22 5 23 8 55 24

252 60 0.53 214.2 51 0.55 201.6 48 0.38 19.0 (7.9–28.4) 0.36

14 5 6 2 0.11 0.17

0.006

55 45 10 38 7 0 0

63 23 27 10

0.79 0.27 0.39 0.67 0.17 0.58

51 35 16 38 8 0 0 0.032 0.50 0.59 14 22 65

24 65 0.10 21 57 0.64 20 54 0.55 20.8 (8.2–32.4) 0.10

20 13 5 3 0.09 0.22

75 48 19 11

0.042 0.008 0.65 0.40 0.50 0.011

P-values represent results of univariable analyses for risk of early seizures (versus no seizure within 7 days of ICH) or late seizures (recurrent versus no recurrent seizures and delayed versus no delayed seizures). Those P-values that qualified variables for multivariable modelling based on statistical significance thresholding (i.e. P 5 0.05) are bolded for ease of identification. a Values reported for subjects with available MRI and APOE data (n = 522). CAD/MI = coronary artery disease and/or myocardial infarction; DM = diabetes mellitus; GCS = Glasgow Coma Scale; IVH = intraventricular haemorrhage, MAF = minor allele frequency; SD = standard deviation.

2700

| BRAIN 2016: 139; 2694–2705

A. Biffi et al.

Figure 2 Post-ICH seizure diagnosis incidence over time. The incidence of newly-diagnosed seizure disorders at different time points following ICH. The dashed vertical line identifies the cut-off for definition of early versus late seizure (i.e. within or beyond 7 days from onset of ICH symptoms).

APOE "4 allele and of 52 lobar cerebral microbleeds as independent risk factors for delayed seizures (Table 2). Increased MRI-WMH was associated with delayed seizure risk in univariable analyses, but failed to meet statistical significance criteria in multivariable modelling (P = 0.10).

Comparison of risk factors for recurrent versus delayed seizures We conducted univariable and multivariable analyses to identify risk factors associated with recurrent seizure risk, and compared results with above presented findings for delayed seizures. In multivariable analyses we identified Glasgow Coma Scale (GCS) at presentation, increasing ICH volume and increasing intraventricular haemorrhage volume as independently associated with recurrence risk after an early seizure (Table 3). None of these factors were associated with delayed seizure risk; of all previously identified risk factors for delayed seizure, only cortical ICH involvement was also associated with recurrent seizure risk. Formal heterogeneity of effects testing identified prior lobar ICH before index event, CT-WMD severity, APOE "4, and possession of 52 lobar cerebral microbleeds as preferential risk factors for delayed seizures as compared to recurrent seizures (heterogeneity of effects P 5 0.05). Presentation GCS and increasing ICH volume were similarly identified as risk factors specific to early-onset seizures (Table 3). Pre-ICH dementia and increasing ICH volume, while showing differential association patterns, failed to meet statistical criteria for heterogeneity of effects.

Prognostic relevance of late seizures following intracerebral haemorrhage Late seizure risk assessment We sought to clarify whether separate modelling of late seizure risk as recurrent versus delayed events resulted in better predictive performance overall, thus further supporting the hypothesis of biological heterogeneity for late post-ICH seizures. We initially applied the recently proposed CAVE scoring system for late seizure risk, which incorporates cortical involvement, age 565 years, ICH volume 410 ml, and early seizures within 7 days of ICH in prediction of future seizure risk. The CAVE score had moderate performance in predicting late seizure risk (Harrell’s C 0.65). Incorporation of all risk factors listed in Table 3 without separation of late seizure risk in recurrent versus delayed did not improve predictive performance in our dataset (Harrell’s C 0.66, P = 0.88 for comparison with CAVE score performance). We subsequently modelled late seizure risk separately based on whether patients had experienced early seizures or not (i.e. based on diagnosis of recurrent versus delayed seizures). Inclusion of demographics, clinical and CT data in a Cox model accounting for the aforementioned diagnostic heterogeneity resulted in markedly improved predictive performance (Harrell’s C 0.83, 95% CI 0.77–0.88, P = 0.012 for comparison with CAVE score model). We then demonstrated further improvement in late seizure risk prediction by incorporation of CSVD genetic (APOE) and MRI markers (Harrell’s C 0.86, 95% CI 0.82–0.90,

Delayed seizures after ICH

BRAIN 2016: 139; 2694–2705

| 2701

Table 2 Univariable and multivariable modelling of delayed seizure incidence risk Variable

Univariable

Model 1 Age (per year) Pre-ICH dementia Prior lobar ICH before index event Cortical involvement (lobar ICH) ICH volume (per 10 cm3 increase) CT-WMD severity Model 2a Age (per year) Pre-ICH dementia Prior lobar ICH before index event Cortical involvement (lobar ICH) ICH volume (per 10 cm3 increase) CT-WMD severity MRI-WMH (per 10 cm3 increase) Exclusively lobar cerebral microbleeds (5 2) APOE "4 (5 1 copies)

Multivariable

HR

HR 95% CI

P

HR

HR 95% CI

P

1.01 1.49 2.25 1.61 1.80 1.88

1.00–1.02 1.12–1.98 1.25–4.06 1.17–2.24 1.08–2.99 1.10–3.31

0.11 0.006 0.007 0.004 0.022 0.022

1.01 1.36 2.50 1.60 1.50 1.85

0.99–1.03 1.01–1.83 1.06–5.91 1.03–2.49 0.84–2.35 1.07–3.29

0.23 0.044 0.038 0.044 0.30 0.031

1.01 1.38 2.15 1.69 1.65 1.77 1.55 2.79 2.03

1.00–1.02 1.09–1.75 1.20–3.84 1.15–2.48 1.06–2.56 0.94–3.35 1.06–2.27 1.27–6.12 1.15–3.60

0.13 0.009 0.011 0.008 0.026 0.081 0.025 0.011 0.016

1.00 1.46 1.90 1.70 1.57 1.75 1.42 2.22 1.95

1.00–1.01 1.09–1.95 1.19–3.01 1.16–2.49 0.77–3.20 1.10–2.78 0.93–2.16 1.23–4.01 1.11–3.42

0.23 0.011 0.008 0.007 0.22 0.018 0.10 0.008 0.020

a Includes Model 1 variables as well as APOE and MRI data, and included only subjects with imaging and genetic data available. Age, ICH Volume and MRI-WMH (Model 2) failed to retain significance in multivariable analyses and were not included in the final modelling; the corresponding results are shown for informational purposes only. P-values in bold are 50.05 after adjustment for multiple testing, and thus identify statistically significant associations; P-values in italics are 50.1, thus indicating trends towards statistical significance.

Table 3 Comparison of risk factors associated with recurrent versus delayed seizures after ICH Variable

Recurrent seizures HR

Delayed seizure risk factors Pre-ICH dementia 1.35 Prior lobar ICH before index event 1.87 APOE "4 (51 copies) 1.11 CT-WMD severity 0.89 Exclusively lobar cerebral microbleeds (52) 1.49 APOE "4 (51 copies) 1.12 Recurrent seizure risk factors Admission GCS (58) 0.55 ICH volume (per 10 cm3 increase) 1.45 IVH volume (per 10 cm3 increase) 1.32 Shared delayed/recurrent seizure risk factors Cortical involvement (lobar ICH) 1.90

Delayed seizures

Heterogeneity of effects P-value

HR 95% CI

P

HR

HR 95% CI

P

0. 76–2.39 0.62–5.61 0.81–1.53 0.34–2.34 0.11–19.73 0.87–1.44

0.31 0.27 0.52 0.81 0.76 0.37

1.36 2.50 1.95 1.85 2.22 1.95

1.01–1.83 1.06–5.91 1.11–3.42 1.07–3.29 1.23–4.01 1.11–3.42

0.044 0.038 0.020 0.031 0.008 0.020

0.35–0.86 1.12–1.87 1.08–1.61

0.009 0.005 0.006

0.88 1.50 0.67

0.48–1.61 0.84–2.35 0.21–2.16

0.68 0.30 0.51

0.012 0.033 0.055

1.17–3.10

0.010

1.60

1.03–2.49

0.044

0.51

0.44 _0.001 0.041 0.009 0.022 0.033

IVH = intraventricular haemorrhage. P-values in bold are 50.05 after adjustment for multiple testing, and thus identify statistically significant associations.

P = 0.008 for comparison with CAVE score model, P = 0.043 for comparison with models without genetic/ MRI data).

Functional outcome risk assessment Cumulative incidence rates for functional decline (defined as incidence of death, dementia, or functional dependence for ADLs/IADLs) grouped by seizure diagnosis status are presented in Fig. 3. We found no increased risk for functional decline (or any of its defining sub-outcomes) in patients with early seizures (all P 4 0.05, Table 4). Conversely, we did

identify increased risk for global functional decline, incident dementia and functional dependence among delayed seizure patients, as presented visually in Fig. 3. To investigate whether delayed seizures increase risk for poor functional outcome after ICH independent of CSVD markers, we repeated all analyses after adjustment for CT-WMD, MRIWMH, cerebral microbleed burden and APOE genotype. In these additional analyses delayed seizures were no longer associated with poor functional outcome after ICH [hazard ratio (HR) 1.42, 95% CI 0.81–2.48, P = 0.24], nor with any of its defining sub-outcomes.

2702

| BRAIN 2016: 139; 2694–2705

A. Biffi et al.

Figure 3 Cumulative incidence of functional deterioration after ICH-based on seizure diagnosis. Graph illustrates cumulative incidence of combined functional decline endpoint (mortality, incident dementia or incident functional dependence) over time after ICH, separated based on presence/nature of post-ICH seizures. The number inside each bar represents percentage of overall study population. Number of patients remaining in follow-up at each time point is reported in the box below each set of values. At each time point subjects are assigned to a group based on current diagnostic status (no seizure history/early seizure history/delayed seizure history).

Discussion We presented evidence that late seizures occurring after ICH demonstrate clinical and biological heterogeneity. We identified largely different risk factors for delayed seizures following ICH when compared to recurrent seizure events in patients with a known history of seizure in the acute ICH phase. Delayed seizures are strongly associated with known clinical, neuroimaging or genetic risk factors for CSVD. On the contrary, we identified acute ICH characteristics (chiefly haematoma size and severity of neurologic deficit at onset) as predictors of recurrent seizure risk. Modelling this clinic-biological heterogeneity has immediate implications for late seizure risk assessment, as well as for evaluation of long-term functional outcome after ICH. Our findings are of immediate clinical relevance to ICH survivors, their caregivers and health-care providers. While a relatively uncommon event (8–10% incidence among ICH survivors at 2 years in prior studies, 9% in our study), late seizures represent a feared long-term complication of ICH that can affect quality of life and return to normal pre-ICH activities (Rossi et al., 2013). Counselling about late seizure risk after ICH requires reliable estimation of individuals’ risk. We provide evidence that separate

modelling of recurrent versus delayed seizure risk results in more accurate predictive capability for overall late seizure risk, which directly translates into more effective clinical counselling. We also did demonstrate that availability of genetic and MRI data substantially improves ability to stratify risk for late seizures. Future clinical studies of postICH epilepsy may greatly benefit from incorporating these markers in their risk assessments. Furthermore, in contrast to patients with early seizures, it appears that patients with delayed seizures were at increased risk for poor functional outcome. However, as our findings suggest that common underlying factors are associated with both delayed seizures risk and functional decline (APOE genetic variation, neuroimaging markers), further studies are required to better understand the underlying biological mechanisms. From a clinical standpoint, however, the close association between delayed seizures and functional decline warrants increased attention from patients, caregivers and healthcare providers alike. The results of our study also contribute to increase our biological understanding of post-ICH epilepsy. Early postICH seizures are presumed to represent the acute ‘mechanical’ effects of the haematoma on brain parenchyma (cortex specifically), resulting in increased epileptogenesis (Haapaniemi et al., 2014). Our results corroborate this

Delayed seizures after ICH

BRAIN 2016: 139; 2694–2705

| 2703

Table 4 Associations of early versus delayed seizures with post-ICH functional outcomes Outcome(s)

Combined functional outcome endpoint Mortality Incident dementia Loss of functional independence

Early seizures

Delayed seizures

HR

HR 95% CI

P

HR

HR 95% CI

P

1.24 1.44 0.90 1.12

0.56–2.77 0.14–14.82 0.60–1.35 0.16–8.06

0.60 0.76 0.60 0.91

1.84 1.49 1.77 2.10

1.21–2.79 0.92–2.42 1.22–2.58 1.07–3.97

0.006 0.12 0.002 0.025

hypothesis, as haematoma size, cortical involvement and severity of neurological compromise were associated with risk of recurrent seizures. Of note, cortical involvement was associated with both recurrent and delayed seizure risk. The biological implications of these associations, however, may be different in each case. Direct cortical involvement has been associated with epileptogenesis in both ischaemic and haemorrhagic stroke (Gibson et al., 2014; Haapaniemi et al., 2014). However, in the setting of post-ICH delayed seizures, cortical involvement stems from the lobar location of the haematoma. Lobar haemorrhages have been strongly associated with cerebral amyloid angiopathy small vessel disease pathology in these populations (Biffi and Greenberg, 2011). As other cerebral amyloid angiopathyspecific markers (APOE "4, lobar cerebral microbleeds) were also associated with delayed seizure risk, it remains unclear whether these results reflect direct cortical ICH involvement or progressive small vessel damage related to cerebral amyloid angiopathy. The pathogenesis of delayed post-ICH seizures (i.e. in the absence of early seizures) is not as well understood. In line with proposed models for epileptogenesis following cerebral ischaemia, complex long-term changes of neural networks, white matter and vascular endothelium are presumed to represent the underlying biological mechanism for delayed post-ICH seizures (Menon and Shorvon, 2009; Pitkanen et al., 2016). Our analyses have consistently shown association between markers of CSVD and delayed seizure risk. A number of potential explanations can be offered to account for our findings. Progressive neuronal and white matter damage due to small vessel disease may amplify epileptogenic processes occurring at the site of ICH. In this hypothetical scenario, CSVD would act as a facilitating mechanism, rendering an individual more prone to seizures triggered by the acute ICH event. However, the lack of association between delayed seizure risk and haematoma structural characteristics (i.e. haematoma size) does not support this hypothesis. Alternatively, injury secondary to CSVD (both micro- and macro-structural) may act independently on delayed seizure risk (Gibson et al., 2014; Hanby et al., 2015). Finally, a combination of the two mechanisms described above cannot be ruled out. Of note, CSVD has been previously associated with late-life onset epilepsy in non-ICH patients (Maxwell et al., 2013; Gibson et al., 2014). Severe underlying CSVD, which is common among ICH patients, may therefore account for

Heterogeneity of effects P-value _0.001 0.92 _0.001 _0.001

the relatively high incidence of pre-ICH epilepsy among our study population. Our study has several limitations. First, despite the comparatively large size of our cohort, post-ICH seizures remain a relatively uncommon occurrence, thus leading to a limited number of events in our study sample. This may have potentially led to overfitting of our statistical models, with resultant overestimation of predictive accuracy. We therefore acknowledge that direct clinical application of our risk models will require further validation efforts. However, statistical overfitting alone is extremely unlikely to have generated the differential association patterns for recurrent versus delayed seizure risks. Second, in the absence of EEG confirmation of seizure diagnosis for all our patients, diagnostic misclassification is likely to impact our analyses to at least some extent. Potential sources of bias include incorrect diagnoses of seizures based on clinical findings alone and/or under-diagnosis due to subclinical seizure events occurring in the absence of concurrent EEG monitoring. These limitations are likely to have negatively impacted statistical power in our study (thus biasing us towards the null hypothesis), rather than generate false positive findings of differential risk factor profiles for recurrent versus delayed seizures. Ultimately, because of varying approaches and/or availability of EEG monitoring for ICH patients, our findings will benefit from further validation in a wide variety of settings. Third, genetic and MRI data were only available for a subset of patients in our study (albeit representing the majority of the study population). Given the high early mortality associated with ICH, this represents an unavoidable pitfall in clinical research studies of cerebral bleeding (Biffi et al., 2011b). We did, however, have CT scan data available for all subjects, including CTWMD severity. The consistent association of CT-WMD with delayed seizure risk is therefore unlikely to have been affected by the potential bias in data acquisition related to high ICH mortality rates. Finally, we have established associations between delayed seizure risk and genetic/neuroimaging markers, themselves putatively associated with underlying CSVD pathology. We are therefore unable to establish a definitive causal connection between delayed seizures and CSVD. Additional studies will be required to further explore this relationship. Our study also displays several strengths, including; (i) in-depth capture and incorporation of detailed clinical information; (ii) access to state of the art techniques in electronic medical records analysis

2704

| BRAIN 2016: 139; 2694–2705

tools to limit data missingness and loss to follow-up; (iii) standardized, validated methodology for neuroimaging and genetic data capture and analysis; and (iv) follow-up based on standardized procedures, capturing multiple relevant functional outcome endpoints of immediate clinical relevance. In summary, we provide evidence of clinical and biological heterogeneity for late seizures after ICH. Recurrent seizures are associated with acute ICH characteristics, chiefly haematoma size. Conversely, delayed seizures are associated with genetic and neuroimaging markers of underlying CSVD. Patients diagnosed with early seizures do not appear to be at higher risk of long-term functional decline after ICH. Delayed seizure patients demonstrate higher incidence of cognitive impairment and loss of functional independence following ICH. Genetic and imaging correlates of underlying CSVD appear to be the primary determinants of these outcomes in patients with delayed seizures. Our findings could be important in clinical counselling of patients and their caregivers, potentially leading to a more precise estimation of late seizure risk among ICH patients. Additional studies are warranted to investigate the biological mechanisms linking CSVD, haematoma characteristics, and epilepsy risk in patients with ICH.

Funding The authors’ work on this study was supported by funding from the National Institute of Health (R25 NS065743, R01 NS063925, R01 NS059727, K23 NS086873, P50 NS051343 and R01 AG26484). All funding entities had no involvement in study design, data collection, analysis, and interpretation, writing of the report and in the decision to submit the paper for publication.

Supplementary material Supplementary material is available at Brain online.

References Badjatia N, Rosand J. Intracerebral hemorrhage. Neurologist 2005; 11: 311–24. Barber M, Stott DJ. Validity of the Telephone Interview for Cognitive Status (TICS) in post-stroke subjects. Int J Geriatr Psychiatry 2004; 19: 75–9. Beghi E, Carpio A, Forsgren L, Hesdorffer DC, Malmgren K, Sander JW, et al. Recommendation for a definition of acute symptomatic seizure. Epilepsia 2010; 51: 671–5. Biffi A, Anderson C, Battey TW, Ayres A, Greenberg SM, Viswanathan A, et al. Association between blood pressure control and risk of recurrent intracerebral hemorrhage. JAMA 2015; 9: 904–12. Biffi A, Anderson CD, Jagiella JM, Schmidt H, Kissela B, Hansen BM, et al. APOE genotype and extent of bleeding and outcome in lobar intracerebral haemorrhage: a genetic association study. Lancet Neurol 2011a; 10: 702–9.

A. Biffi et al. Biffi A, Devan WJ, Anderson CD, Ayres AM, Schwab K, Cortellini L, et al. Statin use and outcome after intracerebral hemorrhage: case-control study and meta-analysis. Neurology 2011b; 76: 1581–8. Biffi A, Greenberg SM. Cerebral amyloid angiopathy: a systematic review. J Clin Neurol 2011; 7: 1–9. Biffi A, Halpin A, Towfighi A, Gilson A, Busl K, Rost N, et al. Aspirin and recurrent intracerebral hemorrhage in cerebral amyloid angiopathy. Neurology 2010a; 75: 693–8. Biffi A, Sonni A, Anderson CD, Kissela B, Jagiella JM, Schmidt H, et al. Variants at APOE influence risk of deep and lobar intracerebral hemorrhage. Ann Neurol 2010b; 68: 934–43. Bladin CF, Alexandrov AV, Bellavance A, Bornstein N, Chambers B, Cote R, et al. Seizures after stroke: a prospective multicenter study. Arch Neurol 2000; 57: 1617–22. Brandt J, Spencer M, Folstein MF. The telephone interview for cognitive status. Neuropsychiatry Neuropsychol Behav Neurol 1988; 1: 111–18. de Jager CA, Budge MM, Clarke R. Utility of TICS-M for the assessment of cognitive function in older adults. Int J Geriatr Psychiatry 2003; 18: 318–24. De Reuck J, Hemelsoet D, Van Maele G. Seizures and epilepsy in patients with a spontaneous intracerebral haematoma. Clin Neurol Neurosurg 2007; 109: 501–4. Gibson LM, Hanby MF, Al-Bachari SM, Parkes LM, Allan SM, Emsley HC. Late-onset epilepsy and occult cerebrovascular disease. J Cereb Blood Flow Metab 2014; 34: 564–70. Haapaniemi E, Strbian D, Rossi C, Putaala J, Sipi T, Mustanoja S, et al. The CAVE score for predicting late seizures after intracerebral hemorrhage. Stroke 2014; 45: 1971–6. Hanby MF, Al-Bachari S, Makin F, Vidyasagar R, Parkes LM, Emsley HC. Structural and physiological MRI correlates of occult cerebrovascular disease in late-onset epilepsy. Neuroimage Clin 2015; 9: 128–33. Harrell FE Jr, Califf RM, Pryor DB, Lee KL, Rosati RA. Evaluating the yield of medical tests. JAMA 1982; 247: 2543–6. Jorm AF. The informant questionnaire on cognitive decline in the elderly (IQCODE): a review. Int Psychogeriatr 2004; 16: 275–93. Keselman HJ, Cribbie R, Holland B. Controlling the rate of type I error over a large set of statistical tests. Br J Math Stat Psychol 2002; 55 (Pt 1): 27–39. Knopman DS, Roberts RO, Geda YE, Pankratz VS, Christianson TJ, Petersen RC, et al. Validation of the telephone interview for cognitive status-modified in subjects with normal cognition, mild cognitive impairment, or dementia. Neuroepidemiology 2010; 34: 34–42. Maxwell H, Hanby M, Parkes LM, Gibson LM, Coutinho C, Emsley HC. Prevalence and subtypes of radiological cerebrovascular disease in late-onset isolated seizures and epilepsy. Clin Neurol Neurosurg 2013; 115: 591–6. Menon B, Shorvon SD. Ischaemic stroke in adults and epilepsy. Epilepsy Res 2009; 87: 1–11. Pantoni L. Cerebral small vessel disease: from pathogenesis and clinical characteristics to therapeutic challenges. Lancet Neurol 2010; 9: 689–701. Passero S, Rocchi R, Rossi S, Ulivelli M, Vatti G. Seizures after spontaneous supratentorial intracerebral hemorrhage. Epilepsia 2002; 43: 1175–80. Pendlebury ST, Welch SJ, Cuthbertson FC, Mariz J, Mehta Z, Rothwell PM. Telephone assessment of cognition after transient ischemic attack and stroke: modified telephone interview of cognitive status and telephone Montreal Cognitive Assessment versus face-toface Montreal Cognitive Assessment and neuropsychological battery. Stroke 2013; 44: 227–9. Pitkanen A, Roivainen R, Lukasiuk K. Development of epilepsy after ischaemic stroke. Lancet Neurol 2016; 15: 185–97. Poon MT, Fonville AF, Al-Shahi Salman R. Long-term prognosis after intracerebral haemorrhage: systematic review and meta-analysis. J Neurol Neurosurg Psychiatry 2014; 85: 660–7.

Delayed seizures after ICH Rossi C, De Herdt V, Dequatre-Ponchelle N, Henon H, Leys D, Cordonnier C. Incidence and predictors of late seizures in intracerebral hemorrhages. Stroke 2013; 44: 1723–5. Rost NS, Rahman R, Sonni S, Kanakis A, Butler C, Massasa E, et al. Determinants of white matter hyperintensity volume in patients with acute ischemic stroke. J Stroke Cerebrovasc Dis 2010a; 19: 230–5. Rost NS, Rahman RM, Biffi A, Smith EE, Kanakis A, Fitzpatrick K, et al. White matter hyperintensity volume is increased in small vessel stroke subtypes. Neurology 2010b; 75: 1670–7. Rost NS, Sadaghiani S, Biffi A, Fitzpatrick KM, Cloonan L, Rosand J, et al. Setting a gold standard for quantification of leukoaraiosis

BRAIN 2016: 139; 2694–2705

| 2705

burden in patients with ischemic stroke: the atherosclerosis risk in communities study. J Neurosci Methods 2014; 221: 196–201. Seo EH, Lee DY, Kim SG, Kim KW, Kim DH, Kim BJ, et al. Validity of the telephone interview for cognitive status (TICS) and modified TICS (TICSm) for mild cognitive imparment (MCI) and dementia screening. Arch Gerontol Geriatr 2011; 52: e26–30. Smith EE, Gurol ME, Eng JA, Engel CR, Nguyen TN, Rosand J, et al. White matter lesions, cognition, and recurrent hemorrhage in lobar intracerebral hemorrhage. Neurology 2004; 63: 1606–12. van Swieten JC, Hijdra A, Koudstaal PJ, van Gijn J. Grading white matter lesions on CT and MRI: a simple scale. J Neurol Neurosurg Psychiatry 1990; 53: 1080–3.