We develop and apply advanced Bayesian statistical inference tools and modelling approaches to analyse epidemiological and medical trial data sets, with emphasis on high-dimensionality and latent heterogeneity. Our methods can detect and map latent association or base hazard rate heterogeneity, correct survival analysis regression outcomes for overfitting, and optimise covariate selections. This enables objective data-driven stratification and more reproducible outcome prediction, and leads to optimized individual risk scores and treatment response scores.

Our research is carried out and applied to medical data in partnerships with teams in several international universities and medical centres. While our focus is on methodological innovation, our research themes are strictly driven by the needs of modern personalised (or 'precision') medicine.

Research Activities:

Bayesian latent class analysis of complex epidemiological data describing heterogeneous cohorts, with either time-to-event or ordinal class outcome variables

Bayesian latent class analysis of complex epidemiological data with informative censoring and competing risks

Bayesian data-driven retrospective patient stratification for clinical trial data with inhomogeneous treatment associations​

Generation of accurate personalised risk scores and treatment response scores

Overfitting corrections for regression parameters in survival analysis

Cross-validation based optimisation of covariates in Bayesian survival analysis, including covariate multiplexing

Bayesian discriminative analysis for very high dimensional data, based on full analytical evaluation of parameter integrals

Nonlinear dimension reduction and multivariate survival analysis for very high dimensional data​​

Generation of rigorously unbiased null models for proteomic or transcriptomic signalling networks, with tailored topological characteristics

Bayesian methods for medical image analysis

Traditional and adaptive medical trial design

Tumour infiltration biomarkers and immunotherapy modelling