This project aims to advance the development of AI models for stratified cardiovascular disease (CVD) risk assessment in women aged 40 to 60, using retrospective data from UK Biobank. It is part of the Horizon Europe-funded project CARAMEL, focused on sex- and gender-sensitive personalised CVD prevention during the menopausal transition. Traditional CVD risk scores are largely based on male cohorts and often fail to account for women-specific risk factors (e.g. hormone/inflammatory profiles) leading to underdiagnosis and suboptimal prevention. AI can improve personalised risk prediction and prevention tailored to individual women by learning complex patterns across diverse health data. However, building effective and generalisable models requires access to large-scale, multimodal, and longitudinal datasets making UK Biobank an ideal resource. This research will apply machine learning and deep learning techniques, including transformer-based models, temporal learning, and multimodal integration methods, to advance understanding of CVD onset, progression and risk prediction in midlife women. Models will be developed using EHR, clinical, omics (e.g. metabolomics), lifestyle and imaging (e.g. MRI, DXA, retinal scans) data. The scientific objectives are to: -Apply AI to longitudinal EHR data to predict CVD precursors and events, and to model disease trajectories; -Develop individual AI models using sex-specific, omics, and other data sources to identify novel biomarkers and improve CVD risk stratification; -Integrate all data modalities into multimodal AI models for comprehensive and personalised CVD risk prediction.
The research will address: Can multimodal predictors be combined to improve early CVD risk prediction and prevention in midlife women? Can AI models outperform traditional CVD risk scores in women aged 40-60? Can deep learning applied to imaging data uncover novel biomarkers that improve early CVD risk prediction beyond standard clinical assessments?
