Abstract
Objective: This study aimed to investigate the relationship between alterations in lymphocyte subsets and resting-state functional magnetic resonance imaging (rs-fMRI) patterns in patients with comorbid major depressive disorder (MDD) and insomnia disorder (ID). Methods: A total of 114 patients with MDD, 108 with ID, 126 with comorbid MDD and ID, and 168 healthy controls (HCs) were recruited, all experiencing their first episode. Emotional and sleep quality were assessed using the 17-item Hamilton Depression Rating Scale (HAMD-17), self-rating depression scale (SDS), Hamilton Anxiety Scale, self-rating anxiety scale (SAS), Pittsburgh Sleep Quality Index (PSQI), and Insomnia Severity Index (ISI). rs-fMRI data and lymphocyte subsets were analyzed. Multivariate prediction models were constructed using correlation analysis, least absolute shrinkage and selection operator (LASSO) regression with 10-fold cross-validation, and logistic regression. Model performance was evaluated with calibration curves and receiver operating characteristic (ROC) analysis. Results: No significant differences were observed in age (p=0.552), sex distribution (p=0.248), education level, or anxiety scores among the four groups, whereas depression and insomnia scores differed significantly (all p < 0.0001). The MDD with comorbid insomnia (iMDD) group exhibited lower fractional amplitude of low-frequency fluctuations (fALFFs) in the right lingual gyrus and fusiform gyrus compared to the MDD, ID, and HC groups. Additionally, compared with HCs, CD3+ and CD4+ T cell percentages were elevated, while natural killer (NK) cell percentage was reduced, with the most pronounced alterations in the iMDD group. fALFF values were negatively correlated with CD3+ and CD4+ T cell percentages, but positively correlated with NK cell percentage. The fALFF in the right lingual gyrus, CD4+ T and NK cell percentage, SDS score, and ISI score were identified as key risk predictors. Multivariable prediction models for ID, MDD, and iMDD demonstrated robust calibration (e.g., calibration degree = 0.502), high discrimination (AUC for iMDD vs. HC = 0.991; MDD vs. ID = 0.821), and good clinical applicability. Conclusions: The identified risk predictors might facilitate individualized clinical decision-making for iMDD patients. While the multivariable prediction model demonstrated strong internal diagnostic accuracy, further external validation using independent cohorts is needed to confirm its generalizability.