Pre-Clinical Early Warning of Sudden Cardiac Death via Dynamic Takens Phase-Space Invariants and Bayesian Self-Calibrating Mahalanobis Embeddings

Abstract. Sudden Cardiac Death (SCD) from malignant ventricular tachyarrhythmias accounts for approximately 15–20% of natural deaths worldwide. Conventional ECG indices (scalar amplitudes, linear HRV) typically remain nominal until seconds before collapse. This work presents a deterministic, self-calibrating phase-space framework for retrospective early-warning analysis of ventricular tachyarrhythmic precursors from a single surface ECG lead. Using Takens delay embedding with dynamic autocorrelation lag selection and a Bayesian-regularized Mahalanobis distance on three topological invariants (planarity residual Π, precession angle Δθ, log orbital angular momentum), the method is evaluated on an aggregate multi-center pool of N=77 verified fatal in-recording collapse episodes from open PhysioNet databases (SDDB, VFDB, CUDB), achieving 79.2% aggregate sensitivity and 92.2% specificity under duration-stratified horizons. Secondary analyses cover ischemia precursors, microvolt T-wave alternans quantification, and 4-year longitudinal risk in the MUSIC heart-failure cohort. Limitations (retrospective design, moderate N, heterogeneous recording durations) and the need for prospective validation are explicitly stated. Intellectual property & reproducibility. In line with statutory exclusions under Argentine Patent Law (Ley 24.481 Art. 6: mathematical methods, software per se, diagnostic methods), the production inference engine and C99 firmware are protected as industrial trade secret (Ley 24.766 / TRIPS Art. 39). Independent verification of the reported metrics is available via a public live blackbox auditor: https://topological-cardiology-auditor-kit.onrender.com (OpenAPI: /docs). The SPIRIT-AI / CONSORT-AI prospective protocol (TOP-SCD) is included in this deposit.