Research Questions
1. Can causal machine learning (ML) models integrate multi-modal data-demographics, biomarkers, genetics, MRI, ECG, and other measures-to produce more accurate, stable, and interpretable predictions of cardiovascular disease (CVD) and dementia risk, compared to traditional models?
2. What are the causal relationships between key multi-modal predictors and the risks of CVD and dementia?
Research Objectives
1. To develop causal ML models for predicting CVD and dementia using UK Biobank’s rich multi-modal dataset.
2. To identify and interpret causal pathways linking predictive factors-such as lifestyle, imaging, biomarkers, and genetics-to cardiovascular and cognitive outcomes.
Scientific Rationale
CVD and dementia are major global health burdens with shared modifiable and genetic risk factors. These conditions often co-occur and are connected by overlapping vascular, metabolic, and inflammatory mechanisms. Traditional risk models typically rely on limited data and capture associations rather than causality, limiting accuracy and clinical utility.
Recent advances in causal ML allow for integration of diverse data and disentanglement of causal effects from correlations. The UK Biobank offers an exceptional resource with detailed demographic, clinical, biochemical, ECG, genetic, and imaging data, including both cardiac and brain MRI. This enables the development of interpretable, stable, and generalisable models for early prediction.
Our study will leverage this resource to build causal ML models for risk prediction and to examine the heart-brain connection. Joint analysis of cardiac and brain MRI will explore how cardiovascular health influences cognitive decline. These insights will support targeted prevention and enhance understanding of disease mechanisms.