Participants
The older adults were retrospectively recruited from SMG-SNU Boramae Medical Center for Dementia from January 2012 to December 2020. The participants underwent both neuropsychological assessment of dementia and a Magnetic Resonance Imaging (MRI) scan. This study was conducted under the Declaration of Helsinki, and the protocol was approved by the Institutional Review Board of SMG-SNU Boramae Medical Center (IRB No.10-2020-295). The informed consent was waived due to retrospective nature of the study by Boramae Medical Center ethics committee. The current study included older adults with Mild Cognitive Impairment and Alzheimer’s disease. All participants received the Korean version of the Consortium to Establish a Registry for Alzheimer’s Disease neuropsychological battery (CERAD-K) 22 which provided normㄴative decisions regarding the presence of cognitive impairment. The clinical diagnosis of MCI and dementia of AD was based on the core clinical criteria of National Institute on Aging-Alzheimer’s Association workgroups (NIA-AA) guidelines 23,24. Subjects suspected or diagnosed with dementia types including vascular dementia, Lewy body dementia, and frontotemporal lobe dementia were excluded. Those identified or suspected with significant neurological conditions including stroke, traumatic brain injury, meningioma, hemorrhage, normal pressure hydrocephalus, delirium, intellectual disabilities, schizophrenia, and bipolar disorders were excluded.
Finally, a total of 750 older adults who met the screening criterion were analyzed, with a mean age of 75.9 years (SD = 7.4, range: 49 - 97), mean education of 8.0 years (SD = 5.0, range: 0 - 23), a higher sex ratio of females (61.9%) (Table 1). The participants with MCI had a larger proportion in the total dataset (AD, n = 358; MCI, n = 392). We confined our analysis within the severity staging of ‘mild’ impairment level considering the feasibility of the neuropsychological test (Clinical Dementia Rating sum of box score ≤ 9; Mean = 3.1, SD = 1.9).
Delusion Symptoms
Neuropsychiatric Inventory (NPI) was used to assess informant and clinician rating of delusion symptoms. NPI assesses the presence and severity of multiple neuropsychiatric symptoms of behavior and socioemotional regulation 25,26. The NPI was based on the semi-structured interview administered to the patients’ informants or caregivers, if available, and rated by clinical psychologists. Delusion item consisted of questionnaires assessing the presence of false beliefs that typically includes themes of persecution (e.g., others are trying to harm or abandon), theft (e.g., others are stealing one’s property), infidelity (e.g., spouse is having affair), and misidentification (e.g., not living in one’s own house; strangers are residing in one’s home). The item was rated from 0 to 3 scores across severity levels (0: No symptom, 1: Symptoms causes mild distress, 2: Symptoms are intractable and cause distress, 3: Symptoms are present with major distress).
In order to confirm the association with brain structure coinciding with a higher-order factor, an identical analysis was conducted with a sum score that covers main behavioral symptom items (i.e., agitation/aggression, disinhibition, irritability). The current dataset showed a moderate correlation between the sum of behavioral symptom items and a delusion item (r = 0.43, p < 0.001). Furthermore, the correlation between the hallucination item and brain measures was additionally examined to infer the associating characteristic of delusion symptoms.
Cognitive Function
Cognitive functions were measured with the Korean version of the Consortium to Establish a Registry for Alzheimer’s Disease neuropsychological battery (CERAD-K) and Mini-Mental State Examination 22. The Korean version of MMSE (Mini-mental status examination) consisted of orientation, memory, attention, and language comprehension items 27. The CERAD-K battery measures multiple domains of cognitive function and facilitates the diagnosis of MCI and dementia. The battery contains the following subtests: Verbal fluency (the number of correct animal words generated in 60 seconds), Boston Naming Test (correctly named words when confronted with picture drawings), Word List Recall (immediate, delayed), Word List Recognition (subtraction of the number of false positives from the number of true positives), and Constructional Praxis (copy, reproduction), and Trail Making Test (TMT; total time spent to complete connect numbers or letters sequentially). TMT A / B measured the total time spent to complete the tasks. The test administration had set the maximum time limit at 360 s (TMT-A) and 300 seconds (TMT-B) based on administration instruction in CERAD-K 28. The score was interpolated as the maximum time limit (360s or 300s) in the cases when the TMT was aborted or not feasible due to the following reasons: exceeded the time limit, unable to understand the rule, or committed more than five errors.
Five of the cognitive measures included the domain-wise sum of subtest scores. The scores were set to have zero means and divided with standard deviations. Each mediating domain of cognition and its composition subtests included language (Boston naming test, Animal fluency), episodic memory (immediate recall, delayed recall, recognition of word list learning), executive/speed (TMT A and B; inverse of log-transformation), visuospatial (constructional praxis copy and reproduction) and general cognition (MMSE).
Neuroimaging Analysis
The neuroimaging data were collected in the MR scanner (3 Tesla, Achieva, Philips Medical Systems, Best, The Netherlands) to acquire a high-resolution T1 anatomical brain image with a 3D T1-weighted turbo field echo sequence (TR: 9.3 ms, TE: 4.6 ms, flip angle: 8˚, voxel size: 1.0 × 1.0 × 1.0 mm, slice thickness: 1mm, 180 slices, image matrix: 224 × 224). We used a fully automated preprocessing procedure implemented in CAT12 r1450 (Computational Anatomy Toolbox; Structural Brain Mapping Group, Departments of Psychiatry and Neurology, Jena University Hospital, http://dbm.neuro.uni-jena.de/cat/) to apply a standardized analysis pipeline. First, a spatial-adaptive non-local means denoising filter was employed 29. Before image preprocessing, a weighted imaging quality rating (IQR) – an index that takes image spatial resolution, noise, and intensity bias into account – was automatically computed by CAT12 for each participant. The images with poor preprocessing performance were excluded (IQR > 2.3). Partial volume estimation was used to create a more accurate segmentation for the mixed tissue classes. Segmentation algorithms based on the adaptive maximum a posterior technique, implemented in CAT12, were used to classify brain tissue into gray matter, white matter, cerebrospinal fluid, and white matter hypointensities. All segmented GM images were visually checked for artifacts and motion effects.
For anatomical precision, surface-based morphometry by projection-based estimation of cortical thickness was conducted in the segmented images 30, which showed a comparable accuracy with other surface-based tools 31. The values were extracted from the CAT12 region of interest (ROI) analysis pipeline. The regional definition of cortical thickness was based on Destrieux’s automatic parcellation of gyri and sulci resulting which provides 148 regional thickness measures 32. For functional network labeling, gray matter density measures were extracted under modulated and spatially normalized segmentation GM images. All brain measures were residualized with the effect of sex and total intracranial volume to adjust for the nuisance effect. The resulting residuals were rendered uncorrelated with the covariates.
Principal Component Analysis
Principal Component Analysis (PCA) with varimax rotation was conducted to summarize the multivariate dimensions of cortical thickness. Varimax rotation was applied after PCA in ways that maximize the sum of variances. The resulting output items were loaded onto the specified number of components. This avoids PCA solution that typically results in components with intermixed loadings which leads to difficulty in interpreting the multivariate patterns. The resulting output is enhanced for a cleaner interpretation of the loadings while leaving an orthogonal basis of component scores.
The number of rotated components was determined with parallel analysis. This analysis compares the scree of eigenvalues of the observed data with that of a random data matrix of the same size as the original. The random data matrix was generated with 50 iterations. Components with higher eigenvalues than the randomly generated data were considered meaningful units of the principal components. This parallel analysis was conducted using the psych package 33. PCA with varimax rotation on the cortical thickness measures identified nine Rotated Components (RCs) (Figure 1) which explained a total of 56% variance of brain measures.
To provide functional interpretability of RCs, the association between component scores and regional gray matter density measures that was based on functional network parcellation were additionally provided. A cortical parcellation map assigns 400 regions to the 17 functional networks 34. It has been suggested that the covarying pattern of gray matter morphology partly indicates the large-scale influence of neurodevelopment and neuropathological changes 35,36. Based on the correlation between RC score and regional gray matter density measures, we labeled RCs with the highest correlating functional network regions.
Statistical Analysis
Partial correlation (Spearman’s method) between the NPI-Delusion rating and RC scores of cortical thickness was examined while controlling for sex and years of education. A total of nine RCs of cortical thickness were examined. Since the age and clinical diagnosis overlap with the main effect of interest, age-adjusted testing results were additionally provided.
To examine the overlapping basis of general cognition and delusion symptom, we explored the extent to which cognitive function mediate the relationship between cortical thickness pattern and delusional symptoms. The mediation effects on five of the mediation terms were explored to compare the specific contribution of cognitive domains separately. Based on the identified RCs that showed significant association with delusion, further mediation analysis was conducted. The strength of the mediation effect was tested with standard errors estimated at 1,000 bootstraps. The effect of education and sex was controlled on every mediating term and outcome variable term (delusion). The total mediation effects were primarily tested and additionally examined the specific domain effects.
The principal component regression method lumps multiple correlating features into large components, thus we additionally explored regionally confined effects of interest. We examined whether ROI-level analysis corresponded to the preceding multivariable findings. After identifying regions that show rank correlation with delusion symptoms across 148 regions of Destrieux’s parcellation (p < 0.01, uncorrected). In the ROI-level mediation analysis, regional mediation analysis assessed the relative strength of mediating effect within each ROI. The mediation strength was indicated by estimated mediation effect divided by the standard error (Z-values). All of the statistical analysis was conducted under R (4.2.1).
The freesurfer_statsurf_display MATLAB function was used to visualize the PCA loadings and ROI-level analysis results on cortical surface areas (Murdoch Children’s Research Institute Developmental, Imaging Group, 2017, https://chrisadamsonmcri.github.io/freesurfer_statsurf_display).