Our areas of expertise include:
- Development and application of dynamic prediction models with clinical applications
- Classification techniques
- Longitudinal/multilevel data modelling
- Manipulation and merging of large datasets from multiple sources (e.g. primary care, hospital data, etc.)
- Bayesian computational techniques (Variational approximations)
- Feature selection.
Examples of applications include diagnostic and prognostic models, stratification/triage, identification of patients at risk of developing a given disease or condition over a fixed period of time. We apply our models to a wide range of areas, including diabetes, epilepsy, cancer, graft versus host disease and encephalitis, among others.
We have for instance developed a dynamic prediction model using longitudinal clinical records to identify patients with epilepsy who are pharmaco-resistant. We have also developed a risk-prediction model to classify patients with diabetes who are at higher risk of developing sight threatening diabetic retinopathy (primary care and hospital data from an initial cohort of ~20,000 patients were merged for statistical analysis) and a diagnostic model to identify babies with necrotising enterocolitis.