In Silico Trials of Surgical Interventions Using Routinely Collected Data

Background: Successful delivery of clinical trials is limited by assumptions that are required for design and planning, including event-rates, and treatment effect sizes. In-silico modelling of trials as part of the design process may overcome some of these limitations.

Aims: In-silico trial methodology was used. This involved the use of NHS Hospital Episode Statistics(HES) data to derive key trial parameters. This data was used to inform the design of novel, pragmatic clinical trials in cardiac surgery and provide data on trial feasibility and generalisability.

Methods: A research-ready-dataset using HES data was established. Statistical and machine-learning methods were applied to model treatment effects of the cardiovascular intervention in the target population as well as in high-risk patient subgroups. These analyses informed the assumptions required for trial design and provided additional information on feasibility and generalisability.

Innovation: The programme of work completed in this PhD was the first to emulate clinical trials for the purposes of assessing trial feasibility and informing the design of future clinical trials. It was the first study to use both statistical and ensemble machine learning methods in an attempt to mitigate the effect of both known and unknown confounders. It was the first study to explore the generalisability of a future clinical trial to both eligible and real world patients.

Value of Results: This project modelled the following clinical trials (1) Surgical or percutaneous coronary revascularisation in patients with heart failure (2) Surgical or percutaneous coronary revascularisation in different high risk populations (3) Comparing multiple arterial grafts to single arterial grafts to determine the optimal revascularisation strategy in women with coronary heart disease.