Three Latent Factors in Major Depressive Disorder Base on Functional Connectivity Show Different Treatment Preferences

The heterogeneity of major depressive disorder (MDD) complicates the selection of effective treatments. While more studies have identified cluster-based MDD subtypes, they often overlook individual variability within subtypes. To address this, we applied latent dirichlet allocation to decompose resting-state functional connectivity (FC) into latent factors. It allows patients to express varying degrees of FC across multiple factors, retaining inter-individual variability. We enrolled 226 patients and 100 healthy controls to identify latent factors and examine their distinct patterns of hyper- and hypo-connectivity. We investigated the association between these connectivity patterns and treatment preferences. Additionally, we compared demographic characteristics, clinical symptoms, and longitudinal symptom improvements across the identified factors. We identified three factors. Factor 1, characterized by inter-network hyperconnectivity of the default mode network (DMN), was associated with treatment response to antidepressant monotherapy. Additionally, factor 1 was more frequently expressed by younger and highly educated patients, with significant improvements in cognitive symptoms. Conversely, factor 3, characterized by inter-networks and intra-networks hypoconnectivity of DMN, was associated with treatment response when combining antidepressants with stimulation therapy. Factor 2, characterized by global hypoconnectivity without DMN, was associated with higher baseline depression severity and anxiety symptoms. These three factors showed distinct treatment preferences and clinical characteristics. Importantly, our results suggested that patients with DMN hyperconnectivity benefited from monotherapy, while those with DMN hypoconnectivity benefited from combined treatments. Our approach allows for a unique composition of factors in each individual, potentially facilitating the development of more personalized treatment-related biomarkers.