The Influence of Diffusion Weighted Imaging Lesions on Outcomes in Patients with Acute Spontaneous Intracerebral Hemorrhage

Abstract

Background/Objective:

Diffusion weighted imaging (DWI) lesions have been well described in patients with acute spontaneous intracerebral hemorrhage (sICH). However, there are limited data on the influence of these lesions on sICH functional outcomes. We conducted a prospective observational cohort study with blinded imaging and outcomes assessment to determine the influence of DWI lesions on long-term outcomes in patients with acute sICH. We hypothesized that DWI lesions are associated with worse modified Rankin Scale (mRS) at 3 months after hospital discharge.

Methods:

Consecutive sICH patients meeting study criteria were consented for an magnetic resonance imaging (MRI) scan of the brain and evaluated for remote DWI lesions by neuroradiologists blinded to the patients’ hospital course. Blinded mRS outcomes were obtained at 3 months. Logistic regression was used to determine significant factors (p < 0.05) associated with worse functional outcomes defined as an mRS of 4–6. The generalized estimating equation (GEE) approach was used to investigate the effect of DWI lesions on dichotomized mRS (0–3 vs 4–6) longitudinally.

Results:

DWI lesions were found in 60 of 121 patients (49.6%). The presence of a DWI lesion was associated with increased odds for an mRS of 4–6 at 3 months (OR 5.987, 95% CI 1.409–25.435, p = 0.015) in logistic regression. Using the GEE model, patients with a DWI lesion were less likely to recover over time between 14 days/discharge and 3 months (p = 0.005).

Conclusions:

DWI lesions are common in primary sICH, occurring in almost half of our cohort. Our data suggest that DWI lesions are associated with worse mRS at 3 months in good grade sICH and are predictive of impaired recovery after hospital discharge. Further research into the pathophysiologic mechanisms underlying DWI lesions may lead to novel treatment options that may improve outcomes associated with this devastating disease.

Introduction

Diffusion weighted imaging (DWI) lesions on magnetic resonance imaging (MRI) have been found in 11–41% of patients admitted with acute spontaneous intracerebral hemorrhage (sICH) [1–9]. The presence of DWI lesions in sICH has been associated with worse modified Rankin Scale (mRS) scores in prior sICH cohorts [5, 8]. However, many variables can influence sICH outcomes, including pre-morbid mRS, intensive care unit (ICU)-related complications [10–12], and withdrawal of life support [13]. Furthermore, patients with higher sICH scores and increased mortality may not survive after discharge to evaluate the influence of DWI lesions on functional recovery. These variables may confound the relationship between DWI lesions and sICH outcomes. Finally, pre-existing knowledge of DWI lesions may bias diagnostic and management decisions, hence influencing the natural course of these patients recovery. Therefore, a prospective blinded study is needed to better characterize the relationship between DWI lesions and functional outcomes in sICH.

The primary aim of this study was to prospectively determine the prevalence of DWI lesions in a defined cohort of primary sICH patients and assess their influence on long-term outcomes in a blinded manner. Although aggressive blood pressure (BP) reduction remains a leading hypothesis for DWI lesions in sICH, an exact pathophysiologic mechanism remains elusive. Hence, a secondary aim of this study was to examine whether BP reduction is associated with DWI lesions. Based on preliminary data [14], we hypothesized that DWI lesions would lead to worse mRS in sICH patients at 3 months. We further hypothesized that patients with DWI lesions experience a greater degree of BP reduction compared to those without DWI lesions.

Methods

Standard Protocol Approval and Patient Consent

This study was approved by the Rush University Medical Center (RUMC) Institutional Review Board and ethics standards committee. Written informed consent was obtained from all competent participants or surrogate health care decision makers in the study.

Study Population

All patients admitted to RUMC with a sICH were prospectively screened between September 10, 2011, and January 15, 2017. A diagnosis of primary sICH was made by consensus after review of the patient’s initial non-contrast computed tomography (CT) and/or CT angiography. Selected patients underwent further diagnostic studies with MRI and/or digital subtraction angiography to exclude occult vascular malformations. Primary sICH was defined as either due to hypertension or cerebral amyloid angiopathy using validated criteria [15]. All patients were managed by a multi-disciplinary team, according to a standardized protocol based on current guidelines [16].

Secondary sICH was defined as hemorrhage due to vascular malformation, coagulopathy, trauma, malignancy, or reversible vasculopathy. Patients with pure intraventricular hemorrhage (IVH) were excluded. Patients with withdrawal of life of support within 72 h, clinically unstable for MRI, or with contraindications to MRI based on our institutional safety protocol were not evaluated for enrollment. Inclusion criteria were: (1) primary sICH, (2) age > 18 years and < 80 years, (3) less than 24 h from last known normal to admission. Exclusion criteria included variables that may confound the prevalence of DWI lesions or outcomes after sICH: (1) prior clinical history of ischemic or hemorrhagic stroke [17], (2) history of arterial or venous thrombosis, (3) neurosurgical evacuation of the hematoma within 48 h [2], (4) acute hypoxemic respiratory failure, (5) pre-admission mRS > 2, (6) Glasgow Coma Scale (GCS) < 5, (7) associated traumatic brain injury, and (8) history of systemic cancer.

Clinical Data Acquisition

Sociodemographic status, relevant past medical history, admission labs, ICU complications, admission imaging, and pre-admission mRS were collected using a standardized protocol with pre-defined criteria by a single person (RKG) blinded to the results of the MRI. When possible, past medical history was confirmed with the patient’s doctor. The location and volume of the intracerebral hemorrhage (ICH) on the admission CT were calculated according to the ABC/2 method [18]. The ICH volume on the follow-up CT was also recorded using the same methodology to evaluate for hematoma expansion. The extent of IVH was measured using the Graeb scale [19]. This information, along with the admission GCS, was used to calculate the ICH score [20]. Physiologic derange ment, as assessed by the Acute Physiology and Chronic Health Evaluation (APACHE) score IV, was calculated using the approach described in the original study [21].

Imaging Data Acquisition

Participants enrolled in the study had a non-contrast MRI scan of the brain. We targeted all scans to be completed within 72 h of admission. MRI for all participants was performed on a single 1.5 Tesla scanner (Magnetom Espree, Siemens Healthineers) using a pre-defined imaging protocol. DWI was acquired using single shot axial echo-planar imaging (TE = 78 ms, TR = 10,000 ms, acquisition matrix = 128 × 128, b-value = 1000 s/mm2, slice thickness = 5 mm, interslice gap = 1 mm, field of view = 24 cm). Maps of apparent diffusion coefficient (ADC) were derived from the raw diffusion weighted data. Susceptibility weighted images (SWI) (TE = 40 ms, TR = 49 ms) and axial T2 Fluid Attenuated Inversion Recovery (FLAIR, TE = 100 ms, TR = 9000 ms, and TI = 2500 ms) sequences were also acquired.

A DWI lesion was defined as a hyperintensity relative to surrounding tissue with corresponding hypointensity on ADC maps. Each focus of DWI abnormality was counted as a single lesion. T2-shine through effects were identified and not counted as lesions. DWI lesions contiguous to the hematoma bed or secondary to external ventricular drains (EVD) were also not considered abnormal. The DWI lesion location was categorized as supratentorial versus infratentorial, cortical versus subcortical, and contralateral to the hematoma versus ipsilateral to the hematoma versus midline. SWI lesions were categorized in terms of location. FLAIR was used to determine white matter disease burden according to the Fazekas scale [22].

Each MRI was reviewed by two board-certified neuroradiologists (MJ, SB) who were masked to the clinical data. Differences in final read were adjudicated by three-party consensus (MK).

Blood Pressure and Cerebral Perfusion Pressure Data Acquisition

Systolic BP (SBP) and diastolic BP (DBP) upon arrival to the emergency department (ED) were manually abstracted from the patients’ medical records. Mean arterial pressure (MAP) was not documented on any ED record and thus could not be obtained. Serial SBP, DBP, and MAP readings in the ICU were electronically abstracted for both automated cuff and arterial lines from the electronic medical record. In addition, cerebral perfusion pressure (CPP) was also abstracted in those patients with an EVD. Individual readings for each parameter were reviewed and values outside of physiologic range were excluded. All readings within a given hour for a single variable were averaged to give hourly values.

Outcomes Acquisition

All outcome data were obtained by examiners certified in the mRS and the National Institutes of Health Stroke Scale (NIHSS). The examiners were blinded to each patient’s clinical course and radiographic imaging. NIHSS and mRS were obtained at 14 days or discharge, whatever occurred first, by in-person assessment. Blinded mRS was also obtained at 3 months via validated telephone interview from the patient or primary caregiver [23]. Patients lost to follow-up were queried using the social security death index to determine whether a patient died prior to follow-up.

Statistical Analysis

Univariate analysis was performed to compare the demographic data, clinical variables, and neuroimaging variables in patients based on the presence or absence of DWI lesions, and with respect to 3-month outcomes. Continuous variables were compared with t test or Mann–Whitney U test. Categorical variables were compared with Chi-square test or Fisher’s exact test. Significant variables from the univariate analysis were included in the multivariable logistic regression analysis to further examine the factors associated with DWI lesions and outcomes.

A linear mixed model with orthonormal polynomials was applied to analyze BP and CPP data from ICU admission until mean time to imaging acquisition to investigate the relationship between BP reduction and DWI lesions [24]. We varied the degree of polynomials from 5 to 9 and used Akaike information criterion and Bayesian information criterion to determine the best model for our data. Predicted BP curves generated from the best model were plotted with 95% prediction intervals.

The generalized estimating equation (GEE) approach was used to investigate the effect of DWI lesions on dichotomized mRS (0–3 vs 4–6) longitudinally over time. The GEE method provides unbiased estimates of parameters for generalized linear models for longitudinal data [25]. Dichotomized mRS at discharge or 14 days and 3 months were included as longitudinal outcomes in the analysis. Variables associated with 3-month outcomes in logistical regression were included in the GEE model. A worst case scenario imputation was performed for missing 3 month outcomes by substituting a favorable outcome for the DWI positive group and an unfavorable outcome for the DWI negative group. All analyses were performed with SAS 9.3 (SAS Institute Inc, Cary NC, USA). A p value < 0.05 was considered significant.

Outcome and Sample Size Calculation

Based on the available literature in 2010, the average prevalence of DWI lesions in sICH was estimated to be around 25%. Similar prevalence estimates have been supported by more contemporary literature [26]. In our pilot data, the difference in proportion between patients with DWI lesion versus those without DWI lesion who had worse functional outcomes (mRS > 3) at 3 months was 0.318 [14]. Given these estimates, we calculated 120 patients would provide us with 85% power to detect a worse functional outcome which we defined as an mRS > 3.

Results

During the study period, 705 patients were admitted with a diagnosis of a primary sICH and could safely obtain an MRI. Thirty-six patients had withdrawal of life support and were excluded. An additional 9 patients were excluded due to a competing sICH intervention trial. Therefore, 660 patients were screened for enrollment during the study period (Supplemental Figure 1). From this cohort, 231 patients (35%) met the study’s pre-defined criteria, of which 139 (60.2%) provided consent. 121 were able to complete the MRI protocol. The remaining 18 were unable to tolerate the MRI due to hyperactive delirium or claustrophobia. A comparison of participants who completed an MRI versus non-consented subjects shows the groups were statistically significantly different for race (p = 0.0006) and total ICH score (p = 0.01) (Supplemental Table 1).

Table 1 shows the overall characteristics of the consented study cohort. The mean age was 56.5 ± 10.9 years; 59.5% (n = 72) were men, and 54.6% (n = 66) were African-American. A majority of subjects (n = 113, 93.4%) were diagnosed with hypertensive sICH. The mean ICH volume on admission was 9.3 mL (± 13.3), and the median ICH score was 1 (IQR 1). Most patients (81%) had a supratentorial hematoma. EVD were placed in 19 patients. Median time to MRI from admission for the cohort was 75.9 h (IQR 77.9). Interrater reliability for the presence of DWI between MRI readers yielded a kappa = 0.78 (95% CI 0.68, 0.89) indicating substantial agreement [27].

Table 1.

Clinical and imaging characteristics of the cohort and univariate analysis of patients with and without DWI lesions

	All patients N = 121	DWI lesion absent N = 61	DWI lesion present N = 60	p value*
Clinical data
Age (years), mean (SD)	56.5 (10.9)	56.8 (12.0)	56.2 (9.7)	0.77
Male, n (%)	72 (59.5)	37 (60.7)	35 (58.3)	0.79
Pre-morbid mRS, n (%)				0.27
0	105 (86.8)	51 (83.6)	54 (90.0)
1	15 (12.4)	10 (16.4)	5 (8.3)
2	1 (0.8)	0	1 (1.7)
Race, n (%)				0.50
African-American	66 (54.6)	32 (52.5)	34 (56.7)
Hispanic	26 (21.5)	12 (19.7)	14 (23.3)
White	28 (23.1)	17 (27.9)	11 (18.3)
Asian	1 (0.83)	0	1 (1.7)
Hx of hypertension, n (%)	102 (84.3)	53 (86.9)	49 (81.7)	0.43
Hx of hyperlipidemia, n (%)	38 (31.4)	22 (36.1)	16 (26.7)	0.27
Hx of atrial fibrillation, n (%)	1 (0.8)	1 (1.6)	0	1
Hx of coronary artery disease, n (%)	10 (8.3)	8 (13.1)	2 (3.3)	0.09
Hx of myocardial infarction, n (%)	7 (5.8)	5 (8.2)	2 (3.3)	0.44
Hx of diabetes, n (%)	23 (19.0)	13 (21.3)	10 (16.7)	0.52
Hx of PVD, n (%)	4 (3.3)	4 (6.6)	0	0.12
Hx of smoking, n (%)				0.53
Active smoker	35 (28.9)	19 (31.2)	16 (26.7)
Never smoked	53 (43.8)	24 (39.3)	29 (48.3)
Quit smoking 1–5 years ago	2 (1.7)	2 (3.3)	0
Quit smoking > 5 years ago	31 (25.6)	16 (26.2)	15 (25.0)
Pre-admission statin use, n (%)	19 (15.7)	14 (23.0)	5 (8.3)	0.03
Pre-admission daily anti-platelet use, n (%)	39 (32.2)	23 (37.7)	16 (26.7)	0.19
Admission GCS, median (IQR)	15 (2.0)	15 (1.0)	14 (4.0)	0.002
Admission NIHSS, median (IQR)1	8 (14.5)	6 (12)	11 (19)	0.07
Total ICH score (%)				0.01
0	48 (39.7)	33 (54.1)	15 (25.0)
1	52 (43.0)	21 (34.4)	31 (51.7)
2	19 (15.7)	7 (11.5)	12 (20.0)
3	2 (1.7)	0	2 (3.3)
Median (IQR)	1 (1)	0 (1)	1 (0.5)	0.001
Body mass index, mean (SD)	29.8 (6.9)	28.7 (5.9)	30.9 (7.8)	0.08
APACHE score, mean (SD)	40.6 (15.7)	39.2 (11.9)	42.1 (18.9)	0.31
Etiology of ICH, n (%)
Hypertensive	113 (93.4)	57 (93.4)	56 (93.3)
Possible/probable CAA	8 (6.6)	4 (6.6)	4 (6.7)
WBC (K/uL), mean (SD)	9.3 (3.1)	9.07 (2.7)	9.5 (3.5)	0.43
Initial platelet count, mean (SD)	242.6 (72.9)	240.8 (83.6)	244.4 (60.9)	0.78
International normalized ratio, median (IQR)	1 (0.09)	1 (0.07)	1 (0.10)	0.88
Initial partial thromboplastin time	27.1 (3.1)	27.6 (3.0)	26.7 (3.1)	0.98
Admission blood glucose (mg/dL), median (IQR)	126.0 (58.0)	121.0 (48.0)	138.0 (61.5)	0.02
GFR (mL/min)A, mean (SD)	72.8 (29.4)	69.0 (29.6)	76.7 (28.9)	0.15
High-density lipoprotein, mean (SD)35	44.47 (12.9)	43.21 (13.1)	45.66 (12.8)	0.38
Low-density lipoprotein, mean (SD)35	114.47 (29.4)	115.21 (30.5)	113.75 (28.7)	0.82
HgbA1C (%), median (IQR)6	5.60 (0.70)	5.60 (0.70)	5.70 (0.90)	0.64
Troponin I (ng/mL), median (IQR)2	0.02 (0.05)	0.02 (0.04)	0.02 (0.05)	0.65
Drug usage, n (%)9
Cocaine	9 (7.4)	2 (3.3)	7 (11.7)	0.16
Marijuana	9 (7.4)	5 (8.2)	4 (6.7)	1.0
Pneumonia, n (%)	14 (11.6)	5 (8.2)	9 (15.0)	0.24
Cardiac ejection fraction15, median (IQR)	62.5 (10)	62.5 (10)	67.5 (10)	0.11
≤ 55%, n (%)	6 (5.7)	5 (9.1)	1 (2.0)
> 55%, n (%)	100 (94.3)	50 (90.9)	51 (98.0)	0.21
Platelet transfusion before MRI, n (%)	4 (3.31)	1 (1.64)	3 (5.0)	0.36
Deep venous thrombosis, n (%)	5 (4.1)	3 (4.9)	2 (3.3)	1
Imaging data
ICH volume on admission CT (mL) Mean (SD)	9.3 (13.3)	6.9 (11.8)	11.7 (14.4)	0.046
ICH volume on follow-up CT (mL) Mean (SD)	10.7 (15.5)	8.0 (13.2)	13.3 (17.2)	0.06
Total Graeb scale, median (IQR)	0.00 (2.0)	0.00 (1.0)	0.00 (4.0)	0.31
Time from ED to ICU (h), median (IQR)	2.8 (1.1)	2.8 (1.4)	2.7 (1.0)	0.41
Time to MRI (h), median (IQR)	75.9 (77.9)	63.0 (53.9)	88.8 (85.8)	0.02
Hematoma location, n (%)				0.62
Infratentorial	21 (17.4)	10 (16.4)	11 (18.3)
Supratentorial	98 (81.0)	49 (80.3)	49 (81.7)
Mixed	2 (1.7)	2 (3.3)	0
PVH score				0.20
0	5 (4.1)	3 (4.9)	2 (3.3)
1	63 (52.1)	35 (57.4)	28 (46.7)
2	43 (35.5)	21 (34.4)	22 (36.7)
3	10 (8.3)	2 (3.3)	8 (13.3)
DWMH score				0.02
0	8 (6.6)	4 (6.6)	4 (6.7)
1	67 (55.4)	39 (63.9)	28 (46.7)
2	39 (32.2)	18 (29.5)	21 (35.0)
3	7 (5.8)	0	7 (11.7)
Sum score, median (IQR)	2.0 (2.0)	2.0 (2.0)	3.0 (2.0)	0.06
Microbleed location, n (%)2				0.24
None	34 (28.6)	22 (36.7)	12 (20.3)
Deep	17 (14.3)	8 (13.3)	9 (15.3)
Lobar	8 (6.7)	4 (6.7)	4 (6.8)
Mixed	60 (50.4)	26 (43.3)	34 (57.6)
Microbleed location, n (%)2				0.18
None	34 (28.6)	22 (36.7)	12 (20.3)
Bilateral cerebral hemisphere	64 (53.8)	26 (43.3)	38 (64.4)
Hemisphere contralateral to hematoma	10 (8.40)	6 (10.0)	4 (6.8)
Hemisphere ipsilateral to hematoma	8 (6.7)	4 (6.7)	4 (6.8)
Midline structure	3 (2.5)	2 (3.3)	1 (1.7)
BP data (mm Hg) Admission BP in ED
SBP, mean (SD)1	198.6 (34.1)	191.4 (31.9)	206.1 (35.0)	0.02
DBP, mean (SD)1	111.3 (25.2)	107.00 (24.4)	115.8 (25.5)	0.06
Change in BP from ED to ICU
SBP, mean (SD)1	− 21.8 (35.3)	− 8.2 (33.5)	− 5.5 (37.1)	0.26
DBP, mean (SD)1	− .6 (28.0)	− .7(21.0)	− 0.5 (33.9)	0.46

Bold values indicate p < 0.05

Superscripts indicate number of missing values

BP blood pressure, CCA cerebral amyloid angiopathy, DBP diastolic blood pressure, DWI diffusion weighted imaging, DWMH deep white matter hyperintensity, ED emergency department, GCS Glasgow Coma Scale, GFR glomerular filtration rate, ICU intensive care unit, mRS modified Rankin Scale, NIHSS National Institutes of Health Stroke Scale, PVD peripheral vascular disease, PVH periventricular white matter hyperintensity, SBP systolic blood pressure, SD standard deviation

^*Comparison between absent and present

Among the overall cohort, 49.6% of patients with primary sICH had one or more DWI lesions (Fig. 1). The range of DWI lesions was between 1 and 50 lesions. The mean lesion burden was 6.5 (SD ± 12.2) with a median of 2 lesions. A single DWI lesion was found in 30 patients (50%), 2–10 lesions were found in 22 patients (36.7%), and > 10 lesions were found in 8 patients (13.3%). Table 2 describes the frequency of DWI lesions based on cerebral location in patients with a single lesion and those with multiple (> 1) lesions. Table 1 highlights the clinical and radiographic differences in sICH patients with and without DWI lesions. When compared to the DWI lesion absent group, the DWI lesion present group had: (1) lower pre-admission statin use (8.3 vs 23%, p = 0.03), (2) lower admission GCS (14 vs 15, p = 0.002), (3) higher median admission ICH score (1 vs 0, p = 0.001), (4) higher median admission blood glucose (138.0 vs 121.0 mg/ dL, p = 0.02), (5) larger ICH volume on admission (11.7 vs 6.9 mL, p = 0.046), (6) longer time to MRI scan (88.8 vs 63 h, p = 0.02), (7) higher deep white matter hyperintensity (DWMH) score (p = 0.02), and (8) higher SBP on ED admission (206.1 vs 191.4 mm Hg, p = 0.02). Atrial fibrillation was not captured on telemetry for any patient during their hospitalization. Two patients in the DWI negative group developed ventriculitis from EVD placement. Neither the degree of SBP nor DBP reduction from the ED to the ICU was statistically different between the two groups. Hourly SBP, DBP, MAP, and CPP in the ICU were not found to be different between the DWI lesion present and DWI lesion absent groups (p > 0.05) (Fig. 2).

Fig. 1.

MRI of an sICH patient with a remote DWI lesion. A right basal ganglia hematoma (open arrow) in 53-year old male African-American male with uncontrolled hypertension. In the left basal ganglia, the DWI sequence reveals a hyperintense lesion of restricted diffusion (closed arrow) with corresponding hypointensity on ADC consistent with an acute infarct

Table 2.

Frequency of DWI lesions based on number of lesions and cerebral location

	Single DWI lesion (n = 30)	Multiple (> 1) DWI lesions (n = 30)
Location, n (%)
Supratentorial	26 (86.7)	22 (73.3)
Infratentorial	4 (13.3)	0
Both	0	8 (26.7)
Location, n (%)
Subcortical	18 (60)	14 (46.7)
Cortical	12 (40)	3 (25)
Both	0	13 (43.3)
Location, n (%)
Ipsilateral to the hematoma	17 (56.7)	4 (13.3)
Contralateral to the hematoma	10 (33.3)	6 (20)
Midline structure	3 (10)	0
Bilateral hemispheres	0	20 (66.7)

DWI diffusion weighted imaging

Fig. 2.

Linear mixed model with orthonormal polynomial analysis of SBP, DBP, MAP, and CPP from ICU admission until MR imaging in study participants. The difference in predicted mean pressure between sICH patients with and without DWI lesions was not statistically significant (p > 0.05) for any BP reading. SBP, DBP, and MAP results were based on the entire cohort (n = 121). CPP results were based on 19 patients with an EVD. DBP diastolic blood pressure, DWI diffusion weighted imaging, EVD external ventricular drain, MAP mean arterial pressure, SBP systolic blood pressure

In the multivariable logistic regression model (Table 3), the presence of DWI lesions was associated with DWMH score (OR 2.112, 95% CI 1.146–3.893, p = 0.0166), admission blood glucose (OR 1.008, 95% CI 1.001–1.015, p = 0.0174), ICH volume on admission CT (OR 1.042, 95% CI 1.007–1.078, p = 0.0174), and pre-admission statin use (OR 0.232, 95% 0.065–0.832, p = 0.0249). All 121 patients were assessed at 14 days/discharge. Patients with DWI lesions were more likely to need a tracheostomy (16.7 vs 4.9%, p = 0.04) and gastrostomy (36.7 vs 9.8%, p = 0.0005) upon discharge. They were also less likely to gohomeand more likely to need either a subacute nursing facility or long-term acute care upon discharge. The presence of a DWI lesion was associated with a higher discharge NIHSS score (6.5 vs 4, p = 0.02), although the difference in dichotomized mRS (0–3 vs 4–6) between groups reached borderline significance (p = 0.052) (Table 4).

Table 3.

Multivariable logistic regression model for variables associated with DWI lesions

	OR (95% CI)	p value
DWMH score	2.112 (1.146, 3.893)	0.0166
Admission blood glucose	1.008 (1.001, 1.015)	0.0174
ICH volume on admission CT	1.042 (1.007, 1.078)	0.0174
Pre-admission statin use	0.232 (0.065, 0.832)	0.0249

CT computed tomography, DWMH deep white matter hyperintensity, ICH intracerebral hemorrhage

Table 4.

14 day/discharge outcomes and univariate analysis of patients with and without DWI lesions

	All patients (n = 121)	DWI lesion absent (n = 61)	DWI lesion present (n = 60)	p value*
Tracheostomy, n (%)	13 (10.7)	3 (4.9)	10 (16.7)	0.04
Gastrostomy, n (%)	28 (23.1)	6 (9.8)	22 (36.7)	0.0005
Discharge disposition, n (%)				0.03
Home	30 (24.8)	21 (3.4)	9 (15.0)
Acute rehabilitation	72 (59.5)	35 (57.4)	37 (61.7)
Subacute nursing facility	9 (7.4)	2 (3.3)	7 (11.7)
Long-term acute care	8 (6.6)	2 (3.3)	6 (10.0)
Dead	2 (1.7)	1 (1.6)	1 (1.7)
14 day or discharge NIHSS, median (IQR)	5 (12)	4 (9)	6.5 (19)	0.02
14 day or discharge mRS, median (IQR)
0–3	51 (42.2)	31 (50.8)	20 (33.3)	0.052
4–6	70 (57.9)	30 (49.2)	40 (66.7)

Bold values indicate p < 0.05

DWI diffusion weighted imaging, mRS modified Rankin Scale, NIHSS National Institutes of Health Stroke Scale

^*Comparison between absent and present

At 3 months, mRS assessments were completed on 119 patients (98.3%). In univariate analysis, DWI lesions were associated with worse functional outcomes at 3 months for each degree of mRS and when dichotomized to mRS 0–3 versus 4–6 (Fig. 3). Supplemental Table 2 shows the univariate analysis of variables associated with poor outcomes (mRS 4–6) at 3 months. In a multivariable logistic regression model for all variables associated with poor outcomes (mRS 4–6) at 3 months, we found the following predictors significant: (1) GCS (OR 1.448, 95% CI 1.036–2.025, p = 0.03), (2) admission NIHSS score (OR 1.243, 95% CI 1.119–1.381, p < 0.0001), (3) ICH score (OR 14.398, 95% CI 3.461–59.896, p = 0.0002), (4) diagnosis of healthcare-associated pneumonia (OR 12.981, 95% CI 1.690–99.731, p = 0.014), and (5) presence of DWI lesion (OR 5.987, 95% CI 1.409–25.435, p = 0.015) (Table 5). Assuming a worst case scenario with the two missing patients, our 3-month outcome results did not change. Using the GEE model, we found that the likelihood of good outcome increased in both groups from 14 days/ discharge to 3 months. However, patients with DWI lesions were less likely to improve compared to patients without DWI lesions (OR = 1.19, 95% CI 1.10–1.29 for DWI present group and OR = 1.43, 95% CI 1.29–1.60 for DWI absent group, p = 0.005). Given the prevalence of DWI in our cohort, our data were powered at 96% to detect our hypothesized difference in outcomes. A query of the social security death index did not reveal that death was a cause for lost to follow-up in any of our missing patients.

Fig. 3.

Univariate analysis of mRS at 3 months between DWI absent and DWI present groups. The difference in outcomes between both groups was statistically significant for each level of mRS (p = 0.02) and when dichotomized (diagonal line) between good (mRS 0–3) versus poor (mRS 4–6) outcomes (p = 0.001)

Table 5.

Multivariable logistic regression model for variables associated with mRS of 4–6 at 3 months

Variable	OR (95% CI)	p value
GCS	1.448 (1.036, 2.025)	0.03
Admission NIHSS	1.243 (1.119, 1.381)	< 0.0001
ICH score	14.398 (3.461,59.896)	0.0002
Diagnosis of HCAP	12.981 (1.690, 99.731)	0.014
Presence of DWI lesion	5.987 (1.409, 25.435)	0.015

DWI diffusion weighted imaging, GCS Glasgow Coma Scale, HCAP healthcare-associated pneumonia, ICH intracerebral hemorrhage, NIHSS National Institutes of Health Stroke Scale

In a sensitivity analysis, we excluded patients who obtained a cerebral angiogram prior to the study MRI given the risk of DWI lesions associated with diagnostic cerebral angiography [28]. A total of 7 patients, all of whom were in the DWI positive group, were excluded. Removal of these patients did not change the results for our outcome analysis with respect to the presence versus absence of DWI lesions at 14-days, 3 months, or with the GEE model. Furthermore, in univariate analysis we found that a higher DWI lesion burden was associated with a worse mRS of 4–6 at 3 months (p = 0.002, Supplemental Table 3). However, in the GEE model we did not find a difference in recovery between patients who had a single DWI lesion versus those with multiple (> 1) DWI lesions (OR = 1.18, 95% 1.07–1.30 for single group, and OR = 1.21, 95% CI 1.08–1.35 for multiple group, p = 0.79). The presence of multiple supratentorial lesions as opposed to a combination of multiple supratentorial and infratentorial lesions was associated with worse 3-month outcomes (p = 0.01, Supplemental Table 4). However, the location of a single DWI lesion did not influence our outcome results.

Discussion

The results of this prospective blinded cohort study support our hypothesis that the presence of a DWI lesion in sICH is an independent risk factor for worse outcomes at 3 months. Contrary to our hypothesis, we did not find an association between SBP, DBP, or MAP reduction and the presence of DWI lesions. Supporting these results and unique to the literature, albeit with a small sample size, is the absence of an association between DWI lesions and reductions in CPP.

The etiology for the worse functional outcomes caused by DWI lesions noted in our and prior sICH cohorts remains unknown. Although dichotomized mRS at 14 days/discharge was of borderline significance in our study, patients with DWI lesions were more likely to have a worse neurologic examination and need a continued higher level of care based on discharge disposition. Our 3-month outcomes are consistent with results from several other cohorts that found worse death and dependence in sICH patients with DWI lesions [5, 8]. In their larger cohort of sICH patients, Kidwell et al. also found that DWI lesions were associated with worse mRS at 6 months in multivariable analysis. Finally, a single acute DWI lesion in sICH has been associated with a higher probability of combined ischemic stroke, recurrent intracerebral hemorrhage, and vascular death at follow-up of 38–47 months after discharge [6].

The total ischemic burden caused by a DWI lesion when compared to the relatively larger hematoma volume should theoretically have minimal impact on long-term functional recovery. Similar to our results, most studies found that the median number of DWI lesions in sICH was between 1 and 2 lesions. The total volume of ischemic burden from DWI lesions has been estimated to be only between 0.25 and 0.44 mL [2, 29]. However, these lesions on admission may be a marker for the presence of ongoing cerebral injury through recurrent infarcts. In one sICH cohort, the presence of DWI lesions was retrospectively examined in patients who had received an MRI of the brain on admission and at 1 month [4]. This study found that 87% of patients with sICH had a new DWI lesion at 1 month after admission. Radio-pathologic data suggest that a single subclinical DWI lesion may be indicative of a significantly higher burden (up to 5000/year) of microscopic infarcts on histology in sICH patients [30]. The combined data support the hypothesis that a single DWI lesion in sICH is a marker for progressive functional decline and worse outcomes in sICH patients. If DWI lesions are recurrent and lead to microscopic ischemic injury, they may be leading to worse mRS thru accelerated cerebrovascular disease. The results of our GEE model support this hypothesis by showing impaired recovery over time in sICH patients with DWI lesions. Further prospective data are needed on the long-term frequency and prevalence of DWI recurrence after sICH. Future research in this area may also benefit from the inclusion of cognitive outcome measures, which may be more sensitive than mRS in detecting subtle neurologic deficits over time.

We did not find an association between BP reduction and the development of DWI lesions. Several studies to date have suggested BP reduction as a likely cause for DWI lesions; however, this conclusion has several limitations. Firstly, analysis of BP reduction in other cohorts has been based on a linear estimation between two readings, usually the highest and lowest values. Acute BP reduction in the ICU during the management of sICH is nonlinear. Therefore, interceding values can influence the degree and rapidity by which BP reduction may appear during analysis. Secondly, there are limited functional imaging data to suggest that reductions in the pressure driving cerebral blood flow during acute BP reduction lead to tissue ischemia remote from the perihematomal region [31, 32]. Radiographi cally, the solitary DWI lesion found in most studies favors local thrombosis as opposed to hypoperfusion, which would likely result in multiple DWI lesions in a watershed distribution. Finally, intensive BP reduction has been associated with potentially improved outcomes after sICH [33]. If DWI lesions were related to BP reduction, one would expect worse sICH outcomes due to increased ischemic burden in patients randomized to more aggressive BP control. Our analysis of CPP is unique in the literature and provides supporting data that reductions in cerebral blood flow accounting for fluctuations in ICP do not influence DWI lesion prevalence.

Our results are similar to several other studies that have shown a link between DWI lesion and white matter hyperintensity (WMH) [4, 6, 29]. The radio graphic subdivision of WMH by Fazekas et al. into periventricular hyperintensity (PVH) and DWMH has been associated with distinct pathologies in several case series [34, 35]. Whereas PVH appears to be nonischemic in nature, DWMH shows evidence of necrotic tissue, enlarged perivascular spaces, and small vessel disease. The association of DWI lesions with DWMH in our sICH cohort suggests that DWI lesions are related to small vessel disease in the brain. Although most studies included a combined Fazekas score or volumetric analysis of the WMH, the possibility of the DWMH sub-score driving the association with the DWI lesions cannot be excluded. Several studies did not find an association between WMH and DWI lesions; however, a significant proportion of patients in these cohorts included sICH due to cerebral amyloid angiopathy which has a different underlying arteriopathy than atherosclerotic small vessel disease [1, 3]. One study found enlarged perivascular spaces, and not WMH, on MRI were associated with DWI lesions [7]. However, enlarged perivascular spaces, specifically in the basal ganglia, have been associated with WMH and lacunar strokes suggesting a common underlying pathophysiology [36].

The association between admission hyperglycemia (defined as blood sugar > 140 mg/dL) and DWI lesions in our study is unclear. Higher fasting blood glucose and stress-induced hyperglycemia have been associated with remote DWI lesions in other studies [37, 38]. In our cohort, there was no difference in prior medical history of diabetes or mean HgbA1c value between groups. It is possible that differences in admission glucose levels between the DWI positive and negative groups are a reflection of hemorrhage severity or degree of physiologic derangement. However, we did not find a difference in disease severity between groups as assessed by the ICH score, GCS, or APACHE-IV score.

There are several possible pathophysiologic mechanisms linking admission hyperglycemia and DWI lesions associated with sICH. In animal models, hyperglycemia appears to promote inflammation and neutrophil infiltration in and around injured cerebral tissue [39]. Similar to other types of brain injury, this inflammatory milieu is prone to both platelet and coagulation cascade activation that can lead to small artery thrombosis, particularly in the setting of pre-existing atherosclerotic disease [40, 41]. Hyperglycemia, however, may also be a protective physiologic response to sICH. Glucose is the primary fuel for cerebral metabolism, and injury leads to increased utilization of glucose when metabolic demands are increased. In positron emission tomography (PET) imaging of acute sICH patients, perihematomal glucose metabolism was increased for 2–4 days after sICH ictus [42]. Treat ment of hyperglycemia in other types of brain injury has led to metabolic crisis with increases in the cerebral lactate/pyruvate ratio measured via microdialysis [43]. Therefore, sICH patients with admission hyperglycemia may be more prone to developing ischemia with greater reductions in blood glucose with standardized ICU insulin protocols. Optimal glucose target in acute sICH patients requires further study.

Our findings of an association between larger hematoma volume and the presence of DWI lesions remain unclear. A greater hemorrhage volume has been associated with an increased white blood count elevation in the cerebrospinal fluid of patients with ICH with IVH [44]. Rupture of whole blood into the brain parenchyma causes the release of various blood components, including erythrocytes, leukocytes, and plasma proteins [45]. Hemolysis of erythrocytes, neutrophil activation of proinflammatory cytokines, and release of thrombin protein are some of the various pathways implicated in post-sICH inflammation. Inflammation in the brain has been associated with platelet and coagulation cascade activation, which can result in the thrombus formation and DWI lesions [46]. Disruption of the blood brain barrier allows for the spread of inflammatory cells and proteins throughout the cerebrospinal fluid, which would explain the presence of DWI lesions in cerebral tissue distant from the ictal hemorrhage.

Finally, we found that that pre-admission statin therapy was protective against the development of DWI lesions. We did not find a statistical difference between LDL or HDL levels between DWI lesion groups suggesting an alternative mechanism of action. Statins are known to have anti-inflammatory properties that can reduce the risk of plaque rupture and thrombus formation, especially in the setting of pre-existing small artery disease [47]. In addition to promoting atherosclerosis in disease vessels, systemic inflammation has been associated with platelet and coagulation cascade activation which can result in thrombosis [48]. DWI lesions are the end result of thrombotic events and may be a surrogate marker for an elevated systemic inflammatory state. There are limited prospective data on the role of statins in the patients with sICH. Several studies have found an association between statin therapy and lower LDL levels with an increased risk of sICH [49, 50]. However, others have also described improved functional outcomes, decreased risk of sICH, and reduction in all-cause mortality [51–54]. Given our findings, statin therapy may be beneficial in sICH patients prone to the development of new DWI lesions. Future randomized controlled trials may provide the needed evidence regarding the role of statins in sICH management.

This study does have several limitations. Firstly, we had a small sample size. However, by limiting potential confounders using a smaller cohort, we were able to better isolate the influence DWI lesions have on outcomes. If DWI lesions had a small influence on outcomes, a larger sample size would likely have been required to see any differences for our primary outcome at 3 months. Second, our results are reflective of a single-center cohort of sICH patients with relatively good grade ICH scores. Even non-consented subjects had higher ICH scores compared to participants that completed an MRI in our study. This limits the generalizability of our results. However, we would argue that outcomes for larger hemorrhages are likely driven by the volume of the hematoma, the resultant effects of elevated ICP, and the more complicated critical care course often experienced by these patients. Including such patients in our study would have limited our understanding of the role DWI lesions play on outcomes in more disabling or less survivable bleeds. Third, our BP analysis included some patients whose BP was managed only with intermittent automated cuff readings, as opposed to continuous arterial BP monitoring. Extreme BP values between intermittent measurements may be missed, and not included in our BP analysis. However, this study examined the degree of change and variability over time using all of the available BP data for each patient allowing for a more comprehensive BP analysis compared to prior studies. Finally, we had a higher proportion of African-American patients who are at greater risk of stroke, which may have increased our prevalence of DWI lesions. Similar to our data, however, more contemporary cohort studies have not found race to be a factor with respect to DWI lesion prevalence and outcomes. The strengths of our study include the use of a defined cohort with inclusion and exclusion criteria, standardized imaging protocol, blinded assessment of MRI imaging and outcomes, and polynomial BP analysis including CPP.

Conclusions

In conclusion, the presence of DWI lesions portends a worse outcome in patients with sICH at 3 months. Furthermore, the degree of BP reduction is not associated with the presence of these infarcts. Future research into the timing and pathophysiology of DWI lesions is needed. Therapies directed at the prevention of these lesions may be a novel approach to improving outcomes in patients with this devastating disease.