Spatial Neglect and Delirium After Stroke

Delirium assessment and prevention has a tremendous impact on hospitalization outcomes and
health care costs. Delirium is a multi-component syndrome characterized by an acute reduction
in cognitive functioning, affecting awareness, thinking, attention, and memory. Stroke
survivors, representing 17% of the US population aged 65 and over (CDC, 2012), are at major
risk for developing delirium (up to 50% incidence in right-hemisphere stroke). Further, about
50% of right-hemisphere stroke patients experience spatial neglect, impairing safety and
recovery. This study investigates a potential neural mechanism explaining the high incidence
of both delirium and spatial neglect after right-hemisphere stroke. The study hypothesis is
that the brain networks for arousal and attention, comprised of ascending projections from
the mesencephalic reticular formation and integrating right-dominant dorsal and ventral
cortical and limbic components, may be affected in these disorders. The study assesses
magnetic resonance imaging (MRI) and behavioral data in right hemisphere stroke survivors. It
is predicted that impaired activity and structural integrity of the brain networks for
arousal and attention will correlate with behavioral signs of delirium and spatial neglect.
The study will be conducted at two sites recruiting an acute (n=45) and a subacute (n=30)
patient samples. The findings of this study have the potential to impact stroke care by
providing a critical biomarker and behavioral profile of post-stroke delirium. This may alert
clinicians to initiating preventive care and targeted interventions in patients who are at
high risk of hospital morbidity and loss of independence.

Study Hypotheses Hypothesis 1: It is hypothesized that a lesion-deficit analysis will reveal
that in patients with delirium and/or spatial neglect, right-brain areas within the
attention, orientation, and arousal networks are affected by stroke lesions.

Hypothesis 2: Because it is hypothesized that both disorders stem from a dysfunction of
common brain networks, it is expected that spatial neglect severity will be a predictor of
delirium severity, controlling for relative lesion size and stroke severity Hypothesis 3:
Both spatial neglect and delirium severity will be correlated with functional connectivity
among the brain areas within the attention, orientation, and arousal networks.

Hypothesis 4: Based on evidence from prior work in post-surgical delirium patients, neglect
and coma patients, it is hypothesized that the integrity of white matter connections among
brain regions comprising the cortical and subcortical networks for attention, orientation,
and arousal is inversely correlated with the severity of delirium and spatial neglect.

Planned Analyses:

Hypothesis 1:

To examine the critical network components affected in post-stroke delirium, the study will
examine participants' structural scans. The structural scans recorded in the study will
include a T1-weighted and a T2-weighted FLAIR image for 30 subacute stroke par-ticipants. In
addition, clinical scans will be obtained for the 45 acute stroke participants, resulting in
75 total brain scans. Structural lesions will be mapped using a semi-automated lesion mapping
in MRIcron package, where each brain voxel will be scored in a binary fashion as lesioned or
non-lesioned. Three-dimensional lesion masks will be fed into VLSM 2 software implemented in
Matlab (Voxel Lesion Symptom Mapping). For the pseudo-continuous outcomes, such as the
Behavioral Inattention Test (BIT) and Confusion Assessment Method-Severity (CAM-S) scores, a
t-score will be computed testing whether lesion in a given brain voxel predicts higher or
lower severity score. For the binary outcomes, such as delirium diagnosis, for each voxel a
Leibermeister measure will be computed testing whether a given lesioned voxel predicts
delirium vs. no delirium across all participants. The analysis will serially travel through
the brain voxels and conduct a test in each voxel. The voxels will be thresholded such that
only those that co-occur in at least 5 patients will be considered. In a typical lesion study
this will result in hundreds of voxels considered across participants. The final results are
corrected for multiple comparisons using False Discovery Rate of p<.05.

Hypothesis 2:

This hypothesis seeks to establish if any association exists between the presence and
severity of post-stroke delirium and spatial neglect. To test this hypothesis a regression
analysis will be conducted using the BIT score as the predictor and the CAM score as an
outcome. The analysis will control for NIH stroke scale score and lesion size, as well as
age. In both sets of analyses raw assessment scores will be converted to percent for scale
uniformity.

Hypothesis 3:

A seed-based functional connectivity analysis will be carried out, testing for a correlation
among time-courses of selected brain regions as a function of CAM-S and BIT scores. 16
regions of interest (ROI) will be used to conduct this analysis within the arousal and
attention networks. A whole brain ROI-to-voxel analysis will also be carried out to capture
any unexpected associations with delirium and neglect severity. This analysis and data
preprocessing will be done using CONN toolbox, a Matlab-based cross-platform software for the
computation, display, and analysis of functional connectivity in fMRI. First, realignment
will be performed between all successive brain volumes and the 1st volume of the series.
Slice-timing correction will be applied to account for the time difference in the interleaved
acquisition of brain slices. Structural segmentation will be performed to create masks of
white matter and cerebrospinal fluid (CSF). These masks will be used in estimating a nuisance
regressor representing physiological noise. The functional scans will be normalized (aligned)
to the structural scans and the atlas template. This is done for group comparisons and to
allow ROI definition using an anatomical atlas. Functional images will be smoothed with a
6mm-radius kernel. Next, the contribution of motion outliers (large movements), continuous
motion (roll, pitch, and yaw), and physiological noise will be regressed from the data.
Finally, ROI-to-ROI and ROI-to-voxel connectivity will performed within each brain scan to
allow for 2nd-order comparisons (i.e., contribution of delirium and neglect severity scores
to functional connectivity, controlling for covariates).

Hypothesis 4:

A probabilistic tractography analysis will be conduced using 16 ROI, defined a priori from
the literature. The analysis will be carried out on DTI data using diffusion tensor
tractography tools available in the FSL analysis suite (e.g.., Probtrax, FDT Toolbox). The
analysis is expected to identify tracts that connect regions within the attention and arousal
networks. It will estimate voxel-wise Fractional Anisotropy (FA) as a measure of fiber
integrity. FA is a scalar value between 0 and 1, measuring the principal diffusion direction
of water molecules. An FA of 0 indicates a perfect sphere, i.e., uniform diffusion. Places
where adjacent voxels have the same directional coherence of diffusion are probabilistically
assigned to tract locations. The values are then averaged to obtain a global FA for each
tract. Delirium and neglect severity scores will be regressed on FA values in an ANCOVA
(accounting for clinical/demographic/physiological/patient characteristics). If the ROI-based
approach is not successful, whole brain FA maps will be considered including areas outside
the brain networks of interest, for an association with delirium and neglect severity scores.

Sample Size Determination.

A pilot investigation was conducted to estimate effect size. 19 right-brain stroke patients
(12 females, 8 Caucasian, 8 African American, 3 Asian), aged 60 years (SD=17 years),
comprised of 1 chronic (> 6 months post-stroke) and 18 subacute (< 1 month post-stroke)
participated in the pilot study. R-squared value of .26 was obtained for the association
between delirium and neglect severity. This gives an effect size (Cohen's f2) of .35. With
this effect size, there is over 99% power to observe the effect, assuming a sample size of 75
patients and an alpha of .05 (Hypothesis 2)

To estimate power of our lesion deficit analysis (Hypothesis 1), data was simulated using
15%, 30%, and 50% as proportion of the participant sample in whom the same voxel will be
lesioned. Leibermeister approach was applied to compute p-values for testing the null
hypothesis of no association between voxel lesion status and delirium status. With 75
participants (Acute+Subacute Sample) there is over 80% power to detect an effect, assuming
that at least 30 percent of the sample have the same lesion, and the lesion is between 40-100
voxels, forming a tight cluster. These are reasonable assumptions given past studies.

In the pilot study, a general linear model that included as regressor delirium or neglect
severity score was applied to the 16X16 ROI connectivity measure matrix for each participant.
The resulting t-scores for the between-subjects effects were in the range of 4.61 to 7.88 for
the BIT measure and 7.07 to 9.31 for the CAM-S. These t-scores correspond to a large effect
size. Using a Cohen's f of .35 (large effect) to estimate power for linear multiple
regression, with 4 predictors, an alpha level of .05, and a two-tailed test, it is estimated
that there will be 87% power to observe this effect, given a sample of 30 participants (Acute
sample) (Hypothesis 3).

A previous study reported thalamus fractional anisotropy (FA) of .315 (.026) in delirious
compared to .333 (.023) in non-delirious patients. These values correspond to a large Cohen's
d effect size. Assuming that FA values in a similar range will be observed, there is
excellent power to detect these effects with a linear multiple regression model.

Based on average patient flow, is is estimated that up to 20 patients will be screened
monthly for study eligibility at each site. Based on recruitment rates of the pilot sample,
it is estimated that 10-20% of the screening sample will be consented. Factoring in a
conservative dropout estimate of 25-40%, it is more than likely that the proposed study
sample will be recruited. To account for patient attrition, the study will over-recruit the
target sample by 5 patients.

Estimated rates of delirium and neglect after right brain stroke are around 50% . Therefore,
around 50 of the study sample is expected to have neglect, delirium, or both. In addition to
using delirium/neglect status as a dichotomous variable, quasi-continuous severity scores for
each variable will be used, which will increase the power to detect associations between
these two disorders and brain lesion locations.

Inclusion Criteria:

- Able to give informed consent prior to beginning of testing; have 1st ever ischemic
stroke on the right side of the brain; speaks English; >18 years of age

Exclusion Criteria:

- all of the following will be excluded:

1. Participants with another brain disorder expected to produce severe
visual-spatial abnormalities (including malignant brain tumor, traumatic brain
injury with post-injury neurological problems, or Alzheimer's Disease)

2. Women knowing themselves to be pregnant will not be enrolled; pregnancy may
itself affect visual-spatial attention. However, women of childbearing potential
will not be tested for pregnancy before participating.

3. Persons consuming more than 10 alcoholic beverages weekly

For participants who will undergo an MRI scan the following will be excluded:

4. Pacemaker or other implanted electrical device incompatible with the MR
environment

5. Eye injury to the eye involving metal filings

6. Unable to undergo MRI due to doctor recommendation

7. Claustrophobia