Towards Enabling Virtual Clinical Studies with Longitudinal Bayesian Network Modeling

A multi-scale and heterogeneous data set , namely Alzheimer’s Disease Neuroimaging Initiative (ADNI) data has been used intensively for research purposes. Studies like these have the potential to understand and model diseases in a longitudinal manner. In this work we showed that it is possible to describe the entire course of a multivariate clinical trial with the help of a Bayesian Network (BN), which specifically takes into account that typically a considerable amount of patients drops out of a study before completion. As BNs are generative models our approach allows for simulating virtual patient cohorts, which has the potential to address principle concerns about data privacy of medical data as well as limitations of sample size. We performed rigorous comparisons of virtual vs real patients to demonstrate the principle feasibility of our approach. Moreover, we showed the possibility to simulate counterfactual interventions into a clinical study cohort. The idea of virtual patients proposed by this work holds promise to facilitate virtual clinical trials as well as increase the statistical power of trials.