Premium
Mapping the contributing factors of depression in community elders
Author(s) -
Zhao Lei,
Yiu Brian,
Lam Bonnie Y.K.,
Biesbroek J. Matthijs,
Luo Yishan,
Shi Lin,
Mok Vincent C.T.,
Wong Adrian
Publication year - 2020
Publication title -
alzheimer's and dementia
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 6.713
H-Index - 118
eISSN - 1552-5279
pISSN - 1552-5260
DOI - 10.1002/alz.037441
Subject(s) - dementia , hyperintensity , neuroimaging , depression (economics) , diffusion mri , fluid attenuated inversion recovery , medicine , psychology , magnetic resonance imaging , physical medicine and rehabilitation , disease , radiology , psychiatry , macroeconomics , economics
Background The contributing factors of depressive symptoms in elderlies are not known quite well. This study aims to investigate the clinical and neuroimaging risk factors of depression in community elderlies. Method Clinical risk factors (Table 1) and MRI scans (including T1‐weighted (T1W), T2‐weighted, FLAIR, diffusion tensor imaging (DTI)) were acquired from 609 stroke‐ and dementia‐free elderlies. T1W images were processed with AccuBrain to quantify volumes of brain anatomical structures (Table 2). White matter hyperintensities (WMHs) were automatically delineated using AccuBrain with manual correction when needed. The generated WMH masks were projected to standard space to calculate regional WMH volumes (Table 2). Other small vessel disease features, including lacunes, cerebral microbleed and enlarged perivascular spaces (EPVS) were visually rated (Table 2). Peak width of skeletonized mean diffusivity (PSMD) was calculated on DTI sequence. These imaging features and clinical variables were used to predict geriatric depression scale (GDS) with support vector regression (SVR). We compared three prediction models here: (1) with clinical variables as predictors, (2) with imaging features as predictors, and (3) with the combination of clinical and imaging features as predictors. Different feature selection methods were attempted. Statistical inference was performed with permutations to investigate the significance of individual contributing factors. Result Representative characteristics of the participants were shown in Table 3. For the prediction of GDS, Model 2 with imaging features was inferior to Model 1 with clinical variables (p<0.001), and Model 3 that combined both types of features did no better than Model 1 (p=0.031). In the statistical inference for Model 1 and Model 2, the significant individual risk factors (p<0.01) included clinical variables such as disturbance of gait, hypertension and frequent urine (Table 4), and imaging features such as the WMH volumes within right superior cerebellar peduncle, retrolenticular part of internal capsule and cingulum (Table 5). Conclusion Combining clinical variables and MRI‐based features did not achieve better prediction of GDS than using either type of features alone. WMH in specific regions may have more independent contribution to depression than other MRI‐based imaging features. These findings may help to understand the risk factors of depression in community elderlies.