Decomposing a clinical EEG montage for low-channel cognitive triage: placement, age referencing and output design

Portable electroencephalography (EEG) may support cognitive triage where imaging and fluid biomarkers are unavailable, but its value should be judged against age. We analysed 1,272 Korean memory-clinic recordings across 42 simulated acquisition conditions. With channel-averaged features, EEG-only screening AUC spanned 0.075 across montages but at most 0.010 across the tested sampling rates and bit depths. Two temporo-parietal electrodes (T5–T6) scored 0.748, compared with 0.742 for all 19 (difference +0.006, 95% CI −0.013 to +0.025). In three external cohorts (257 recordings), their pooled Alzheimer's disease AUC differed from all 19 by −0.002. In exploratory within-cohort analyses of 41 frontotemporal dementia recordings, four temporo-occipital electrodes differed from all 19 by −0.159 (95% CI −0.255 to −0.063; Holm-adjusted P = 0.012). Adding EEG to age raised screening AUC from 0.786 to 0.827 (increment +0.041, 95% CI +0.026 to +0.056); the dementia increment was +0.067 (95% CI +0.045 to +0.090). For mild cognitive impairment (MCI), the increment was +0.017 and remained unresolved: Holm-adjusted permutation and DeLong P values were 0.025 and 0.061. Age-referenced deviations were less discriminative than EEG+age (AUC 0.662) and poorly calibrated. With 25 draws per sample size, the screening increment's 10th percentile first exceeded zero at n = 350. These findings are design hypotheses for prospective device studies.