Abstract
Alzheimer's Disease Dementia (ADD) prognosis is an unmet medical need. Mitochondrial dysfunction is an early AD etiopathogenic factor. The present study analyzed mitochondrial DNA (mtDNA) methylation patterns in blood samples from patients with mild cognitive impairment (MCI) who progressed to ADD (P), MCI remained stable (NP), and Cognitively Normal (CN) individuals. Differentially methylated sites were identified in the D-loop region in both CN vs. NP and NP vs. P comparisons, even before β-amyloid positivity. A Random Forest model was developed using mtDNA methylation data combined with cognitive and risk factor features. Model's performance was assessed by cross-validation and tested on an independent set, achieving 84.4% accuracy in training and 83.2% (95% CI: 75.2%-89.4%) in testing. For identifying P patients, sensitivity and specificity were 95.1% and 70.7%, respectively. The AUC-ROC was 90.3%. The developed model demonstrates predictive capacity in distinguishing cognitive decline and stability in MCI individuals, independently of their β-amyloid status.