Comparison of Brain Age With Standard MRI Assessment for Dementia Risk Stratification in Subjective and Mild Cognitive Impairment

Abstract

Background and Objectives Early dementia risk-stratification is crucial for timely intervention. Brain-predicted age difference (brain-PAD) derived from structural MRI reflects the gap between computer-estimated age and chronological age and has been linked to clinical progression to dementia. We evaluate whether brain-PAD has prognostic value for progression to dementia in subjective cognitive decline (SCD) and mild cognitive impairment (MCI) and whether brain-PAD provides added value beyond visual rating scales. Methods This retrospective cohort study included patients with SCD and MCI with up to 10-year follow-up assessments from the Amsterdam Dementia Cohort and SCIENCe project. Brain-PAD was computed from baseline MRI using 2 pretrained methods. Cox proportional hazards models assessed the association between brain-PAD and risk of progression to dementia, controlling for age, sex, Mini Mental State Examination, and visual rating scales. We tested interactions between brain-PAD and baseline diagnosis to assess stage-specific value and evaluated the added value beyond amyloid status in a biomarker-available subset. Results The study included 799 patients (SCD 412, MCI 387; 39.4% female, 63.3 ± 8.0 years). During a median follow-up of 3.2 years, 235 (29.4%) progressed to dementia. Higher brain-PAD was associated with an increased risk of progression to dementia (hazard ratio [HR] 1.04, 95% CI 1.02–1.06) and improved model fit beyond visual ratings (χ2 = 12.24, p = 0.003). Following a significant interaction between brain-PAD and baseline diagnosis (p = 0.013), stratified analyses demonstrated brain-PAD was a stronger predictor in SCD (HR 1.09, 95% CI 1.03–1.15) than in MCI (HR 1.01, 95% CI 0.99–1.04). In SCD, brain-PAD improved model fit beyond visual rating scales (χ2 = 10.17, p = 0.003) and modestly increased discrimination (C-index +0.02). A data-driven brain-PAD cutoff of −2.6 years corresponded to a high negative predictive value (0.91–0.99) over 2–10 years. Although brain-PAD improved model fit beyond amyloid status (χ2 = 6.44, p = 0.018), it did not improve discrimination. In MCI, brain-PAD provided no additional value (χ2 = 0.88, p = 0.351). Discussion MRI-derived brain-PAD predicts progression to dementia in memory clinic patients with SCD or MCI. In particular, brain-PAD provided additional predictive value to visual rating scales in patients with SCD.