Agile development methodology of matched synthetic control for Bayesian trials in the era of genomic medicine

With the rapid advanced development in technologies that are accessible, wealth of biological and clinical data is generated in clinical trials; applying deep learning methods to integrate digital pathology features with omics data allow us fully to delineate these tumours and understand which molecular features may have contributed to the treatment response. The amount of multi-omics data generated provide valuable resource for trial methodologists to draw biological information and data to inform future trial designs. There will be continuous increased popularity of trials with pre-defined embedded correlative sciences, integral biomarkers, marker-enrichment designs, or marker-adaptive or treatment-adaptive designs. Randomised control arms allow for comparison of treatment arms without concern on confounding factors, a randomised study is not always feasible (or ideal), especially for rare cancer like sarcoma, or rare subgroup within breast cancer (e.g., basal-like tumour within estrogen receptor positive breast cancer) In such case, use of external controls to supplement single-arm data may be an attractive approach that can be explored.

We are developing an artificial intelligence-based integrated genomics clinical tool, called COUNTERPOINT, which weave tumour genomics, microenvironment and clinicopathological data, to predict disease outcomes in breast cancer. This system is expandable to draw upon new biological findings with the goal to accelerate discovery of molecular features with potential for clinical implementation. For example, drug response/clinical outcome could be used to infer clinico-phenotypic associations by incorporating data from co-clinical PDX or organoid models in collaboration with cancer biologists (Dr Huang, Dr Sadanandam, Dr Perou (UNC at Chapel Hill)). Biological connectivity could then inform innovative trial designs e.g. by incorporating therapeutic-specific biological pathway impact scores/patterns as endpoints in biological-response adaptive designs, or by creation of matched synthetic controls for Bayesian trials, in line with the FDA's Real-World Evidence Program guidelines (Huang, Jones, Yap, Cheang, MRC/NIHR Rare Cancer Research platform).

This PhD project will focus on review and apply agile software development methodology to create the roadmap of the creation of matched synthetic controls (based on biology) to design more efficient Bayesian trials based on exemplars from breast cancer (common) and sarcoma (rare cancer).