Forecasting personal health in an uncertain environment

Lead Research Organisation: University of Sheffield
Department Name: Computer Science

Abstract

Healthcare delivery takes place in a noisy and uncertain environment. Some patients present following their first symptom, whereas others may wait before presenting to a healthcare provider. Some patients comply with and respond to treatment, while others don't. Forecasting the health status of an individual is consequently difficult without the crude use of random numbers to determine risk. On the other hand, mechanistic models of human physiology show promise as research tools, but are difficult to apply effectively to healthcare problems because there is no framework to embed the variability in structure and function that is seen in individuals and human population. The challenge of this bold and ambitious proposal is to bring about a step change in both modelling for healthcare by developing new mathematical tools that embed uncertainty at every level, enabling models that describe the trajectory of an individual through healthcare to be informed by detailed mechanistic and multiscale models of organ systems. This project is unique in that it combines mechanistic models of physiology with the modelling personalised medicine with the integration of uncertainty at all scales.

We recognise that our vision is bold and ambitious. As a "way in" to this problem we have selected two specific exemplars where the project partners already have expertise and access to models, tools, and data. The first of these is the transmission of influenza within a population, and the second is a prevalent cardiac arrhythmia (atrial fibrillation). We will adopt a "middle out" approach, and will start at the patient (individual) scale where there are rich data for each exemplar. From this point we will work upwards to population scale, and downwards to the molecular scale. A conventional approach to model transmission of influenza would take into account contact between individuals, with a probability of infection passing from one to another. Our approach will be to unpick this probability, replacing it with a model of the likelihood of infection based on knowledge of the health status of an individual (do they smoke, do they have asthma, etc.), coupled with population variability (for example of airway geometry, a key determinant of how inhaled particles are transported to the mucus layer of lung airways). A conventional approach to modelling atrial fibrillation in the heart would take into account the geometry of the human atrium, combined with a trigger. Remodelling of the structure and function of the atria would be imposed on the model. Our approach will be to parameterise a model of atrial fibrillation using data from longitudinal studies, and use output from the model to determine the treatment that would have been most beneficial for that patient.

Among the many challenges faced by the healthcare system, clinical decisions informed by incomplete information can lead to low treatment success rates, inefficient use of resources and poor patient outcomes. This proposal seeks to enable clinicians to make better use of patient-specific data, informed by mechanistic modelling. We seek to push multiscale physiological modelling to a new level of sophistication, by incorporating uncertainty and variability systematically across different scales of organization and translating the outcomes to the clinic. Our ambition is for our approach to find utility across the healthcare sector: the clinician will be able to make better-informed forecast of outcomes for each patient; policy decisions will be informed by population models; pharma companies will be able to assess the efficacy of drugs on individuals; and the patient will enjoy improved outcomes.

Planned Impact

The beneficiaries of this research include the mathematics, healthcare modelling, and physiological modelling academic communities, as well as clinicians and healthcare professionals, manufacturers of healthcare devices and technology, and ultimately patients and the general public.

The main non-academic professional beneficiaries are clinicians and medical practitioners, as well as companies and businesses developing and manufacturing healthcare technology. In the long-term this may also include the pharmaceutical industry. These groups will benefit from this research because this project will contribute to the translation of modelling technology from a research tool to a widely accepted tools for use in clinical practice. Clinicians and other healthcare professionals will be able to deliver improved diagnosis interventions, and manufacturers of healthcare equipment could benefit from the competitive advantage provided by diagnostic or interventional systems that embed uncertainties.

As these ideas are deployed within healthcare, then patients and the general public will benefit from this research.
 
Description Throughout this research project, we have investigated different ways of quantifying uncertainty in two exemplar systems of infection by virus particles in the lung, and atrial fibrillation in the heart. We have found that it is possible to build emulators of heart cell models (models of models) that can be used to directly calculate how uncertainty in the model parameters results in uncertainty in the model output. We have also investigated how spatial disorder can be embedded in models of mucus spreading in the lung. We have developed a software toolkit for fitting emulators, released under a GNU license. This work has stimulated entirely new ways of thinking about mathematical modelling for biology and healthcare, and has contributed to subsequent development of models representing blood flow in the placenta and electrical activity in the heart, that take into account uncertainty and variability.
Exploitation Route Our findings will be relevant for other areas where predictive models are used within healthcare, and these findings have been disseminated through conference and workshop presentations as well as scientific papers. These publications are becoming well cited by others, and 3 papers have more than 40 citations as of February 2020. Follow on funding has been achieved from EPSRC (EP/P010741/1 and EP/P01268X/1), which is applying concepts of uncertainty quantification to models of the human heart, with the aim of streamlining interventional procedures in the clinic.
Sectors Healthcare

Pharmaceuticals and Medical Biotechnology

 
Description The mathematical concepts underlying this award are detailed, but the main idea - that predictions from a computer model are subject to uncertainties in the inputs - is quite easy to communicate. A specific impact from this award has been to communicate these ideas to sixth form school students as part of outreach activities to raise awareness about he importance of maths and statistics for real world applications. The use of these techniques to benefit healthcare is a compelling case study, and of course would not be possible without the research that underpins these examples. The change in thinking arising from this award has also resulted in tentative economic impacts through conversations with industrial partners. The use of model-based approaches to guide healthcare interventions is an emerging technology. To be robust and reliable in the clinical setting, these approaches need to take into account noise, variability, and missing information in model calibration, and should assign confidence to model predictions. These ideas were the focus of this award, and we would expect to see more concrete economic impact in the future as they become more mainstream. The ideas at the core of this proposal are continuing to lead to new impacts at the interface of academia and industrial/clinical applications. They led to a month long work programme at the Isaac Newton Institute in Cambridge during 2019 (https://www.newton.ac.uk/event/fht/), which included a dedicated event for industry participants (https://www.newton.ac.uk/event/ofbw45/), and a follow up meeting will take place in June 2024. Uncertainty and probabilistic models have also fed into the ideas around digital health that form the basis of the EPSRC funded South Yorkshire Digital Health Hub (https://www.sheffield.ac.uk/sydhh), which aims to deploy digital health solutions to tackle health inequalities across South Yorkshire.
First Year Of Impact 2016
Sector Digital/Communication/Information Technologies (including Software),Healthcare
Impact Types Societal

 
Description EPSRC responsive mode
Amount £1,210,706 (GBP)
Funding ID EP/P010741/1 
Organisation Engineering and Physical Sciences Research Council (EPSRC) 
Sector Public
Country United Kingdom
Start 04/2017 
End 04/2021
 
Description In-Procedure Personalized Atrial Digital Twin to Predict Outcome of Atrial Fibrillation Ablation
Amount £1,534,181 (GBP)
Funding ID EP/W000091/1 
Organisation Engineering and Physical Sciences Research Council (EPSRC) 
Sector Public
Country United Kingdom
Start 03/2022 
End 03/2025
 
Description The SofTMech Statistical Emulation and Translation Hub
Amount £1,225,134 (GBP)
Funding ID EP/T017899/1 
Organisation Engineering and Physical Sciences Research Council (EPSRC) 
Sector Public
Country United Kingdom
Start 03/2021 
End 02/2025
 
Description The South Yorkshire Digital Health Hub
Amount £3,211,469 (GBP)
Funding ID EP/X03075X/1 
Organisation Engineering and Physical Sciences Research Council (EPSRC) 
Sector Public
Country United Kingdom
Start 08/2023 
End 08/2026
 
Description University of Manchester internal PhD studentship
Amount £1 (GBP)
Organisation University of Manchester 
Sector Academic/University
Country United Kingdom
Start 08/2013 
End 09/2016
 
Title UQ toolkit 
Description We have written and released software tools for undertaking emulation and uncertainty qualification in computer models and these have been used to examine models of electrical activity in the heart. 
Type Of Material Improvements to research infrastructure 
Year Produced 2016 
Provided To Others? Yes  
Impact None at present, but we expect to see academic impact over the next 24 months. 
 
Description Uncertainty in Computer Models @ Sheffield 
Organisation University of Sheffield
Country United Kingdom 
Sector Academic/University 
PI Contribution Providing exemplar problems that can be tackled using techniques developed under a previous EPSRC award for "Managing Uncertainty in Computer Models"
Collaborator Contribution Specific expertise in statistical techniques for uncertainty analysis.
Impact No specific outputs yet, although a paper describing our initial work is in preparation.
Start Year 2013
 
Title maGPy 
Description Uncertainty and sensitivity analysis using Gaussian process emulators 
Type Of Technology Software 
Year Produced 2020 
Open Source License? Yes  
Impact This software has underpinned a series of publications. 
URL https://github.com/samcoveney/maGPy