Spectral Exponents of the Twelve-Lead ECG Reveal the Anatomy of Cardiac Conduction Disorders and a Bifurcation Between Aging and Disease

The aperiodic (1/ f β) component of the electrocardiogram carries information about broad- band cardiac electrical dynamics, yet it remains largely unexplored as a quantitative marker of conduction system integrity. Here we apply Irregular-Resampling Auto-Spectral Anal- ysis (IRASA) to 412,730 twelve-lead ECG recordings across three datasets spanning three continents—PTB-XL (Germany, n = 21,799), Chapman-Shaoxing (China, n = 45,152), and CODE-15% (Brazil, n = 345,779)—to extract the spectral exponent β, a single number per lead quantifying the balance between order and disorder in the cardiac electrical signal. We report five findings. First, each diagnostic category deviates from the healthy reference in a specific direction and magnitude, with conduction disturbances showing the strongest di- vergence (Cohen’s d= 0.99). Second, the twelve-lead β-vector encodes the spatial anatomy of the conduction system: a classifier distinguishes complete left from complete right bun- dle branch block with AUC= 0.982 on the discovery cohort (PTB-XL), and this result repli- cates exactly on two independent external cohorts (AUC= 0.982 on Chapman-Shaoxing; AUC= 0.979 on CODE-15%), despite substantial variation in absolute β-values across recording equipment (βNORM = 1.52–2.75). Third, within a curated healthy cohort (PTB- XL), aging and pathology drive β in opposite directions—healthy aging flattens the spec- trum (β decreases, ρ=−0.179) while overt conduction disease steepens it (β increases), cre- ating a bifurcation from the healthy operating point. Beat-to-beat morphological variability independently confirms age-related complexity loss (ρ= 0.165, p < 10−19). Fourth, among ECGs labeled normal by cardiologists, β distinguishes truly clean recordings from those with subclinical conduction findings (AUC= 0.754), with the direction of β-shift inverted relative to overt disease. Fifth, β does not independently predict all-cause mortality beyond established diagnoses (adjusted HR= 1.02, p= 0.83 in CODE-15%), establishing it as a di- agnostic marker of conduction anatomy rather than an independent prognostic biomarker. These results demonstrate that the spectral exponent captures a cross-population invari- ant of cardiac electrophysiology: the geometry of the twelve-lead β-vector is determined by conduction system anatomy and is reproducible across equipment, ethnicity, and ge- ography. Keywords: electrocardiography, spectral exponent, 1/ f noise, IRASA, cardiac conduction, aging, PTB-XL, external validation, cross-population invariance.