Uncertainty quantification and personalisation of lumped parameter models of the cardiovascular system

Personalised medicine, facilitated by the growing capacity to collect comprehensive patient data, aims to provide personalised therapies for each individual. Computational models are boosting the capacity to draw diagnosis and prognosis, and future treatments will be tailored not only to current health status and data, but also to an accurate projection of the pathways to restore health by model predictions. These models, which are based on physiology and physics rather than on population statistics, enable computational simulations to reveal diagnostic information that would have otherwise remained concealed and to predict treatment outcomes for individual patients. The inherent need for patient-specific models in cardiology is clear and is driving the rapid development of tools and techniques for creating personalised methods to guide pharmaceutical therapy, deployment of devices and surgical interventions. For computational models to have clinical utility, we must be able to provide a quantification of the risk associated with any predictions or interpretations which are made from the model. Lumped parameter models (LPM) represent the cardiovascular system as a series of electrical segments, each characterised by parameter values that offer insights into the associated health status. Given one can constrain the model with patient specific data such that the parameter values are updated, one obtains a digital representation of a patient’s cardiovascular system. This research investigates the crucial offline stage of uncertainty quantification for models, primarily employing sensitivity analysis, with the goal of achieving personalisation. As such, this work first examines the process of performing a global sensitivity analysis of a cardiovascular LPM, establishing some best practices to ensure accurate model interpretations. Then we assess the impact of the chosen outputs of a model when looking to quantify uncertainty. We demonstrate how variation in the chosen outputs can significantly impact one’s interpretation of a model. Following the enhanced understanding of the best practices around sensitivity analysis, we extend a method for obtaining a personalised subset of input parameters, acknowledging the inherent variability among individuals in medicine. Alongside this, a novel examination of the sensitivity indices is proposed within the context of personalised medicine to provide insight associated with the calibration of model parameters. Finally, we investigate a data assimilation technique and explore the method’s effectiveness for model personalisation. The primary outcome of this research is the development of an offline model personalisation workflow, designed to reduce the uncertainty associated with calibrating models to patient data to the greatest extent possible.