Abstract / Summary
Objective: To examine what a dementia-associated electroencephalography (EEG) entropy difference permits us to infer about brain dynamics by testing its dependence on permutation-entropy (PE) representation. Methods: Four PE settings represented identical alpha-band data from 1,177 CAUEEG recordings. Nine duration/filtering specifications, phase reconstruction, controlled signals, age models, and 78 external participants (58 in the AD-control comparison) assessed measurement sensitivity. CAUEEG includes repeated patients without an available linkage key; its statistics describe recordings, and uncertainty estimates assume recording independence. Results: In the original analysis, dementia PE contrasts ranged from d=-0.690 to +0.705. Across nine specifications, sub-cycle PE remained negative (d=-0.729 to -0.576) and coarse-sampled PE positive (+0.625 to +0.712). Cycle-spanning and high-order contrasts remained small. Phase reconstructed 90.2% of ordinal patterns; similarly high agreement occurred after spectral phase randomization. The sample-entropy contrast fell from d=+0.523 to +0.197 with 2 s windows on the same recordings. The external PE profile retained its ordering but did not reproduce the CAUEEG sign reversal. Conclusions: The reversal is stable across the tested processing choices, but a stable contrast does not specify its biological meaning. Phase reconstruction supports a narrowband measurement account; patient-level precision and predictive generalization remain unresolved. Significance: Biological interpretation requires a justified connection between the signal distinctions retained by a representation and the proposed alteration in brain dynamics.