Rationale. Aortic dissection (AD) is rare but catastrophic; prevention requires identifying high-risk people years before onset. UK Biobank (UKB) provides the needed scale, repeat blood pressure (BP), lifestyle/biomarker data, registry linkage, genetics, and cardiac/aortic MRI to quantify modifiable determinants and build transparent risk tools.
Research question. In participants free of AD at baseline, which clinical, lifestyle, biomarker, imaging, and genetic factors predict first (incident) AD, and can they be combined into a well-calibrated, interpretable risk model?
Objectives. (1) Curate/validate the AD phenotype using ICD-10 I71.0 plus aortic repair procedure codes; exclude traumatic/iatrogenic events; test narrow vs broad definitions and a 2-year lag. (2) Quantify associations for BP level and visit-to-visit variability, smoking, alcohol, physical activity, EAT-Lancet diet, adiposity/body composition, CRP, lipids, HbA1c, eGFR, medications, socioeconomic indices, and family history/prior aneurysm. (3) Develop and internally validate a penalised Cox model (ML comparators: random survival forests/gradient boosting) with nested cross-validation; report discrimination (C-index/AUC), calibration (slope/plots, Brier) and net benefit (decision-curve); derive simple risk charts. (4) Test added value of polygenic risk scores (PRS) and aortic/left-ventricular MRI metrics. (5) Probe causality for SBP, smoking, LDL-C, BMI, type 2 diabetes, and kidney function via two-sample Mendelian randomization with pleiotropy-robust sensitivity analyses. (6) Assess fairness/robustness by sex, age, and ancestry; use multiple imputation for missingness; repeat analyses in White British and multi-ancestry samples.
Design. Prospective cohort excluding prevalent AD; outcomes from Hospital Episode Statistics, death registries, and primary care. All analyses conducted within UKB-RAP, exporting only aggregate, non-identifiable results.
