Depression and cardiovascular disease comorbidity A multimodal investigation using machine learning, network analysis, and casual inference

Depression and cardiovascular disease (CVD) are leading causes of global disease burden that frequently co-occur, leading to worse prognosis and increased mortality. Despite extensive research, the mechanisms linking these conditions remain poorly understood, partly due to depression's clinical heterogeneity and limited use of integrative analytical approaches. This thesis aimed to identify symptom-level and biomarker mechanisms linking depression and CVD, quantify their longitudinal associations, and improve predictive models for incident CVD using multimodal and multi-cohort analyses.
Data from four large population-based cohorts (UK Biobank, Young Finns Study, NESDA, and Lifelines) were analyzed, encompassing over 350,000 participants from young adulthood to late middle age. Analyses integrated NMR-based plasma metabolites, inflammatory markers, depressive and anxiety symptomatology, childhood adversity, and traditional CVD risk factors. Tripartite and mixed graphical networks, machine learning, causal discovery algorithms, Mendelian randomization, and longitudinal mediation were used to examine metabolite-mediated pathways and identify key symptom-level predictors of CVD.
Across cohorts, shared metabolic markers such as alpha-1-glycoprotein acetyls, monounsaturated fatty acids, glucose, and omega-3 fatty acids, and inflammatory biomarkers including AGP, TNF-alpha, IL-6, and HDL diameter were identified linking energy-related depressive symptoms (appetite changes, fatigue, sleep disturbances) to CVD risk. These findings were validated in independent datasets, confirming both symptom-specific and metabolite-mediated pathways. Incorporating symptom-level and early-life adversity data into predictive models improved CVD risk prediction beyond traditional factors, achieving AUCs up to 0.88, with a parsimonious set of indicators capturing the added predictive value.
This integrated symptom and biomarker-level approach enhances understanding of the biological and psychological mechanisms underlying depression-CVD comorbidity, identifies potential targets for prevention and intervention, and demonstrates the translational potential of multimodal analyses for improved risk stratification and personalized medicine.