ABSTRACT NUMBER: ESOC2026A1120 DISENTANGLING PROGNOSTIC HETEROGENEITY IN VASCULAR COGNITIVE IMPAIRMENT: A BAYESIAN NETWORK ANALYSIS

Abstract Background and aims Vascular cognitive impairment (VCI) is a heterogeneous construct with variable risk factors, pathologies, and outcomes. This heterogeneity combined with attrition bias complicates prognostic modeling. Here, we demonstrate how a Bayesian network (BN) can jointly model interdependencies for future cognitive decline, cardiovascular events, and attrition across diverse cardiovascular patient populations and quantify added value of plasma biomarkers. Methods We analyzed data from the Heart–Brain Connection cohort (n=566; median age 68 IQR 62–74; 64% men) including memory clinic patients with vascular brain lesions, patients with heart failure (HF) or carotid occlusive disease (COD), and controls. A hierarchical BN (demographics→vascular risk→imaging→functional status→outcomes) was learned (K2 algorithm). We (i) derived conditional probability tables for Clinical Dementia Rating (CDR) increase and Major Adverse Cardiovascular Events (MACE) over 5 years; (ii) quantified incremental information of plasma biomarkers; and (iii) performed patient-level inference. Results Diagnostic group and baseline CDR were key determinants of both outcomes. Among baseline CDR=0 patients, cognitive decline probability was 0.55 (memory clinic) versus 0.06 (reference). In COD with Mini Mental State Examination (MMSE) 27–29, MACE probability was 0.44. HF with MMSE<27 showed highest dropout probability (≈0.79). Plasma biomarkers showed no direct dependencies with outcomes once incorporated with other clinical and imaging factors. Conclusions Diagnostic group and baseline cognition determined both outcomes; plasma biomarkers provided no independent information beyond conventional factors. This proof-of-concept demonstrates that BNs can clarify conditional dependencies in heterogeneous populations. The methodological framework may offer researchers an approach for understanding heterogeneity among patients. Conflict of interest This work is part of the Heart-Brain Connection crossroads (HBCx) consortium of the Dutch CardioVascular Alliance (DCVA). HBCx has received funding from the Dutch Heart Foundation under grant agreements 2018-28 and CVON 2012-06. LMO is a recipient of TAP-dementia, receiving funding from ZonMw (#10510032120003) in the context of Onderzoeksprogramma Dementie, part of the Dutch National Dementia Strategy. Figure 1 - belongs to Conclusions