Stochastic models to enable tailoring of medications to patients with multiple morbidities
Lead Research Organisation:
University of St Andrews
Department Name: Computer Science
Abstract
Chronic non-communicable diseases (NCDs) constitute a major global burden and public health challenge. NCDs such as diabetes, cardiovascular disease, cancer, and chronic respiratory diseases are the leading cause of mortality worldwide.
Clinical pathways and guidelines documented by the National Institute of Health and Care Excellence (NICE) in the UK constitute an evidence-based knowledge base with recommendations on the medications and alternatives available to treat chronic conditions. However, guidelines often do not address the needs of patients with complex cases of chronic conditions where medications need to be chosen with care.
Patients with multimorbidity are subject to multiple treatment pathways and there is a risk of undetected adverse reactions to combinations of prescribed medications. In Scotland, over half of all people with chronic conditions have comorbidities. A recent survey indicates that medicines are implicated in 5-17% of hospital admissions, of which half are considered preventable. Furthermore, the majority of these medicine related occurrences are due to well-known adverse effects of commonly prescribed drugs.
This project looks at how can we use data to better understand and tailor care planning and treatment cognisant of multimorbidity burden for an individual and their predicted life expectancy and ability to benefit from treatment (that outweigh the risks).
Clinical pathways and guidelines documented by the National Institute of Health and Care Excellence (NICE) in the UK constitute an evidence-based knowledge base with recommendations on the medications and alternatives available to treat chronic conditions. However, guidelines often do not address the needs of patients with complex cases of chronic conditions where medications need to be chosen with care.
Patients with multimorbidity are subject to multiple treatment pathways and there is a risk of undetected adverse reactions to combinations of prescribed medications. In Scotland, over half of all people with chronic conditions have comorbidities. A recent survey indicates that medicines are implicated in 5-17% of hospital admissions, of which half are considered preventable. Furthermore, the majority of these medicine related occurrences are due to well-known adverse effects of commonly prescribed drugs.
This project looks at how can we use data to better understand and tailor care planning and treatment cognisant of multimorbidity burden for an individual and their predicted life expectancy and ability to benefit from treatment (that outweigh the risks).
Technical Summary
We propose to study three common chronic conditions: diabetes, cardiovascular disease and COPD, and their clinical guidelines. In order to find a combination of medications which result in the best outcomes for patients, we need to consider the patient's clinical history, current values and preferences, as well as have a model and treatment information for other patients with similar conditions.
For the later, we propose to use Scottish EHRs, specialist disease registry data (e.g., diabetes registry), the national prescribing dataset PIS, and primary and secondary care data in Scotland. This fellowship opportunity will explore combinations of machine learning, formal modelling techniques and automated reasoning approaches commonly used respectively in Theoretical Computer Science and in Artificial Intelligence, to construct (stochastic) models from guidelines and EHRs.
This will enable us to compute the optimal medication choices for a patient with one or more chronic conditions, given his/her recorded data and comparisons with similar patients. We will further investigate solutions in the presence of uncertainties and incomplete information. To evaluate the approach, we will develop and test predictive models iteratively using Scottish EHRs and replicate the processes on data from other centres involved in HDR UK, Canada and Brazil.
For the later, we propose to use Scottish EHRs, specialist disease registry data (e.g., diabetes registry), the national prescribing dataset PIS, and primary and secondary care data in Scotland. This fellowship opportunity will explore combinations of machine learning, formal modelling techniques and automated reasoning approaches commonly used respectively in Theoretical Computer Science and in Artificial Intelligence, to construct (stochastic) models from guidelines and EHRs.
This will enable us to compute the optimal medication choices for a patient with one or more chronic conditions, given his/her recorded data and comparisons with similar patients. We will further investigate solutions in the presence of uncertainties and incomplete information. To evaluate the approach, we will develop and test predictive models iteratively using Scottish EHRs and replicate the processes on data from other centres involved in HDR UK, Canada and Brazil.
People |
ORCID iD |
Marco Caminati (Principal Investigator / Fellow) |
Publications
Bowles J
(2019)
Automated Reasoning for Systems Biology and Medicine
Küster Filipe Bowles J
(2020)
Correct composition in the presence of behavioural conflicts and dephasing
in Science of Computer Programming
Bowles J
(2019)
A framework for automated conflict detection and resolution in medical guidelines.
in Science of computer programming
Bowles J.K.F.
(2020)
A formally verified smt approach to true concurrency?
in CEUR Workshop Proceedings
Title | Formally verified event structure enumeration software |
Description | Event structures are mathematical structures underpinning our approach to modeling our treatments of multimorbid patients. The software we produced outputs all possible such structures of a given cardinality, allowing us to perform simulations and computational experiments on them. Moreover, this software is verified using theorem provers. This means that the correctness of the output is proven to the highest standards available, granting that our work is based on the right data. |
Type Of Technology | Software |
Year Produced | 2018 |
Open Source License? | Yes |
Impact | The software is used internally in our project to test new ideas. |