We aim to determine whether minimal atrial structure helps interpret atrial electrical signatures in the 12-lead electrocardiogram (ECG), and whether combining ECG-derived features with left-atrial (LA) size improves modelling of incident atrial fibrillation (AF) beyond standard clinical risk factors.
Scientific rationale: AF arises from heterogeneous atrial substrate, yet ECG markers are often hard to interpret because similar ECG patterns can reflect different combinations of atrial size, conduction properties and comorbidities. Anchoring ECG features to a minimal structural measurement (LA size/volume) is a practical way to reduce ambiguity and quantify how much additional information is needed to make ECG-based phenotyping more identifiable, reproducible and mechanistically interpretable. UK Biobank is uniquely suited because it combines standardised resting ECGs, cardiac MRI imaging-derived phenotypes (including atrial volumes) and linked longitudinal outcomes at population scale.
Our primary analysis will be prospective (AF-free at baseline, modelling incident AF after imaging), while individuals with prevalent AF at baseline will be retained for secondary cross-sectional phenotyping and calibration comparisons relevant to external clinical cohorts. We will use (i) raw 10-second 12-lead ECG lead-traces (bulk XML) and ECG summary measures for quality control, (ii) cardiac MRI imaging-derived phenotypes (LA volume primary; optional right-atrial volume; LV ejection fraction/mass/end-diastolic volume as covariates), and (iii) cine/anatomy cardiac MRI DICOM to enable future derivation of additional biomarkers and geometry (excluding flow, tagging and tissue-mapping sequences). Incident AF will be ascertained from linked inpatient data and the death registry. Outputs will be aggregated results and reproducible code; no re-identification will be attempted.