Transportability of a Released Deep-Learning Electrocardiogram Model for Atrial Fibrillation Across German and Chinese Cohorts: Reproducibility Package

This record contains the reproducibility package for a frozen, zero-training external evaluation of the released Ribeiro et al. six-output 12-lead deep neural network's atrial-fibrillation output in PTB-XL version 1.0.3 and the Chapman-Shaoxing/Ningbo ECG database version 1.0.0. The package contains analysis and inference scripts, tests, a frozen protocol and design lock, software requirements, derived cohort and label audits, complete prediction tables, bootstrap and statistical outputs, figure source data, publication figures, tables, and manifests. No PhysioNet ECG waveforms, waveform arrays, WFDB headers, released model binaries, inference batch caches, credentials, direct identifiers, or local-machine metadata are redistributed. Record-level tables retain only public de-identified source identifiers and fields needed for provenance, patient-cluster resampling, subgroup analyses, and the prespecified first-ECG sensitivity analysis. Users must preserve de-identification and comply with the cited upstream terms.