Association of Lesion Location and Depressive Symptoms Poststroke

Julian Klingbeil, MD; Max-Lennart Brandt; Max Wawrzyniak, MD; Anika Stockert, MD; Hans R. Schneider; Petra Baum, MD; Karl-Titus Hoffmann, MD; Dorothee Saur, MD


Stroke. 2021;52(3):830-837. 

In This Article


Data Availability Statement

Anonymized behavioral and lesion data are available from the corresponding author on reasonable request.

Patient Recruitment

Informed consent for study participation was obtained from all participants or their legally designated surrogates. Institutional review boards reviewed and approved all study protocols. To include as many patients with PSD as possible within a representative sample of patients with stroke, we relied on 2 separate, prospective cohorts. Both cohorts were recruited from the stroke unit of the Department of Neurology, University of Leipzig Medical Center from November 2017 to November 2018 (cohort 1) and from January 2012 to December 2014 (cohort 2). Inclusion criteria for cohort 1 were first ever stroke confirmed by neuroimaging. Exclusion criteria were ages <18 or >90 years, evidence of previous strokes in neuroimaging, not speaking German, inability to perform behavioral testing, history of depression before stroke and other psychiatric (eg, sustained drug abuse, schizophrenia, anxiety disorders), or neurological disorders affecting the central nervous system (eg, Parkinson's disease, dementia) and other severe diseases. Out of 629 screened patients, 131 patients were enrolled in the study (for details, refer to Figure 1). The second cohort of 300 patients was first recruited with the aim of prospectively evaluating care from a stroke nurse team in the first 3 years after stroke. Inclusion criteria were stroke either proven by neuroimaging or assumed by the treating neurologist based on clinical presentation. Exclusion criteria were age <18, altered level of consciousness, history of dementia and unlikeliness of successful follow-up on-site (eg, patients from outside Leipzig and the neighboring districts). Because the studies had different inclusion criteria, for joint analyses, we selected only those patients from cohort 2 (N=193) who matched the inclusion and exclusion criteria of cohort 1. In total, cohort 1 and 2 consist of 324 patients who were tested in hospital within the first 4 weeks after stroke (t1). For 83.3% (n=270) of these patients, behavioral scores around 180 days after stroke (t2) were available. The main reasons for loss to follow-up were recurrent disease events, unavailability by telephone or lack of interest in continuing the study (for details, refer to Figure 1).

Figure 1.

Flowchart of the patient recruitment.
Patients were recruited in 2 separate prospective studies on the same stroke unit of the Department of Neurology, University of Leipzig Medical Center. For cohort 1, all eligible patients from the stroke unit were included between November 2017 and November 2018. For cohort 2, of all 300 patients recruited from January 2012 to December 2014, only those who matched the inclusion criteria of cohort 1 were considered for the joint analysis. HADS indicates Hospital Anxiety and Depression Scale; and PSD, poststroke depression.

Behavioral Testing

Hospital Anxiety and Depression Scale (HADS) scores were acquired both during the initial hospital stay (t1: 6±3.5 days, range 1–27) and again after 6 months (t2: 189.5±10.3 days, range 159–284) in a telephone interview or patient visit. The HADS consists of 14 questions, 7 related to depression (HADS, depression subscale [HADS-D]) and 7 to anxiety (HADS, anxiety subscale [HADS-A]), with a maximum score of 21 points for each subscale and an optimal threshold of above 7 to establish presence of anxiety or depression.[16] We used HADS-D(t2) both as a binary score with a cutoff value of >7 considered as indicative for depression (PSD+) and as a continuous measure for the severity of depressive symptoms in a subsequent analysis. Stroke-related disability was quantified with the National Institutes of Health Stroke Scale (NIHSS) and the Barthel-Index. Information on current medication (eg, antidepressants), psychological, or psychiatric treatment was obtained in an interview.

Statistical Analyses

Separate logistic regressions were performed to test associations with PSD+ as dependent variable. Age, sex, lesion volume, HADS-D(t1), HADS-A(t1), Barthel-Index(t1), and NIHSS(t1) were considered as independent variables. Subsequently, a multiple logistic regression with selection procedures was performed to identify factors, which independently contribute to PSD+. For this purpose, all factors with P<0.1 in the separate logistic regressions (Barthel-Index, NIHSS, HADS-D(t1), HADS-A(t1)) were entered into the multiple regression in the following order: (1) stroke-related disability (Barthel-Index, NIHSS) and (2) symptoms of depression and anxiety (HADS-D(t1), HADS-A(t1)). With each step, an enter selection procedure was used. Factors selected in the first step were valid for selection in the following step. Adjusted odds ratios with 95% CI of all factors maintained in the final model for PSD are reported. The pseudo-R[2] (Nagelkerke) of the final model was calculated to quantify overall variance explained. Analyses were conducted with SPSS 25 (SPSS Inc, Chicago, IL).

Brain Imaging and Preprocessing

Lesion analyses were conducted on pseudonymized clinical imaging acquired at the Department of Neuroradiology, University of Leipzig Medical Center. Lesion delineation was thus performed by 2 reviewers blinded to the patients' outcome. From all computed tomography and magnetic resonance imaging scans available for a patient, those scans which best depict the lesion (202 magnetic resonance imagings and 68 computed tomographies) were chosen by one of the authors (Dr Klingbeil). We also included computed tomography scans to keep more severely affected patients in the cohort who were unable to tolerate a magnetic resonance imaging scan. Lesions were delineated in native space with the semiautomated Clusterize Toolbox[17] and then manually edited using MRIcron[18] by another author (M.-L. Brandt). All lesion maps were supervised by a neurologist (Dr Klingbeil) and used for cost-function masking during normalization. Corresponding magnetic resonance imaging and computed tomography scans were normalized to MNI152 (Montreal Neurological Institute) space and resliced to 1 mm isotropic voxels using the Clinical Toolbox[19] for SPM12 (Wellcome Trust Centre for Neuroimaging, London, United Kingdom) running on MATLAB (R2018b, The MathWorks, Inc, Natick, MA). The resulting normalization parameters were also applied to the native space lesion maps, which were then used for further lesion analyses in MNI space. An overlap of all lesions is shown in Figure 2A, and an overlap of all lesions from patients with a HADS-D score >7 at t2 (PSD+) is shown in Figure I in the Data Supplement.

Figure 2.

Lesion overlay plots and binary mask of all voxels included in the voxel-based lesion behavior mapping (VLBM) analysis.
A, Lesion overlap of all 270 patients with stroke included in the VLBM analysis. Color indicates number of lesions from 1 (dark blue) to the maximum of 22 within the right insula. B, Lesion overlap of at least 5 lesions represents voxels that were analyzed with VLBM. Axial coordinates refer to MNI (Montreal Neurological Institute) space in mm. L indicates left. This figure appears in color online.

Voxel-based Lesion Behavior Mapping

Whole-brain VLBM was performed using NiiStat software.[20] Lesion size was included as covariate of no interest, and analyses were restricted to voxels damaged in at least 5 patients based on recommendations for VLBM.[21] To control the family wise error (FWE) rate, the null-distributions of the maximum z-score were obtained by 5000 random permutations.[22] Results were thresholded at p(FWE) <0.05 on the voxel-level. We tested for differences between the 2 groups (PSD+/PSD−) by means of 1-tailed Liebermeister-tests for binominal data. Continuous raw scores from the HADS-D(t2) (reversed and with z-skew= −1.90 after transformation) were entered into a regression analysis using a general linear model as implemented in NiiStat.[20] Anatomic labeling was performed with Harvard Oxford atlas[23] and the Anatomy toolbox.[24] For visualization, all VLBM results were overlaid on the MNI template available in MRIcron.