Many people have more than one long-term disease, this is known as “multi-morbidity”. For instance, there are many people who have both heart disease and diabetes. There exist computer tools that help doctors to predict the risk that a patient will develop diseases in the future (for instance, heart disease). However, these tools always focus on single diseases. This is not helpful when we try to prevent and treat multi-morbidity. In this project we will develop tools that predict the risks that a patient will develop multiple diseases (and what those diseases are).
Working out how best to treat patients with multi-morbidity is not straightforward. Typically, each disease comes with its own course of treatment, but it is not clear how to combine these. Therefore, we will also develop tools that predict what would happen if a patient were given a particular treatment. Such a treatment could involve changing their lifestyle (e.g., stopping smoking), taking a particular drug, or a combination of these things.
People with multi-morbidity often end up taking many drugs (known as polypharmacy). This can be a problem because some drugs do not work when they are taken with other drugs. We will therefore also extend the computer tools so that they can predict what would happen if a patient, who is living with multi-morbidity and is taking multiple drugs, would change the drugs that they are taking.
We will focus on patients with diabetes, heart disease, and related diseases.
An increasing number of people live with multiple long-term conditions (called multi-morbidity), with associated reductions in survival, physical functioning, and increased healthcare utilisation. The management of multi-morbidity is often complex because most evidence was acquired in patients with isolated health conditions, and it is unclear to what extent that evidence generalises to patients with multi-morbidity and polypharmacy.
Clinical prediction models (CPMs) are increasingly used in the prevention and management of long-term conditions. These models, which are typically developed using supervised statistical learning methods aim to support clinical decision-making by predicting the future occurrence of an event. However, existing CPMs are not suitable for supporting the prevention and management of multi-morbidity because they operate in silos of clinical specialties, with separate models being developed for each event of interest; and interpreting CPMs as predictions under hypothetical interventions is incorrect, because there is no consideration of causality in the supervised learning methods that are used to develop them.
The objective of this study is to develop new clinical prediction methods that are suited for patients with multi-morbidity. Specifically, we aim to develop methods for prediction models for multi-morbidity trajectories using non-negative matrix factorisation in order to extract multi-morbidity patterns as well as topological data analysis to summarise the trajectories of patients within the identified multi-morbidity patterns. We will also predict the causal effects of interventions and how these are likely to change the course of patients within the identified trajectories. We will test the developed methods by applying them in patients with metabolic disease (obesity, type 2 diabetes, cardiovascular disease, chronic kidney disease, and non-alcoholic fatty liver disease) to predict causal effects of initiating new treatments and terminating existing treatments on the severity of their metabolic condition, progression of multi-morbidity, adverse cardiovascular outcomes, and death, in the presence of multi-morbidity and polypharmacy.
The outcomes considered in the analyses will be:
(i) death from cardiovascular causes (including cerebrovascular disease); death related to type 2 diabetes; death related to chronic kidney disease; death related to non-alcoholic fatty liver disease;
(ii) death from other causes;
(iii) major adverse cardiovascular events (MACE);
(iv) coronary artery disease;
(v) stroke (ischaemic and haemorrhagic [including subarachnoid haemorrhage]), transient ischaemic attack;
(vi) hospitalisation for unstable angina, atrial fibrillation/flutter, heart failure, stroke (ischaemic and haemorrhagic [including subarachnoid haemorrhage]), transient ischaemic attack, hypoglycaemia, ketoacidosis, amputation, acute kidney injury;
(vii) the following clinical interventions: revascularisation procedures (coronary and peripheral), ablation interventions, amputation, bariatric surgery, dialysis, kidney transplant, liver transplant;
(viii) the following complications: nephropathy, retinopathy, neuropathy, peripheral vascular disease, renal failure, liver failure
Code lists for outcome categories are provided in Appendices 1, 2, 3, 4 and 5.
Niels Peek - Chief Investigator - University of Manchester
Narges Azadbakht - Corresponding Applicant - University of Manchester
Abdelaali Hassaine - Collaborator - University of Manchester
Andrew Yiu - Collaborator - University of Oxford
Chris Holmes - Collaborator - University of Oxford
Farideh Jalalinajafabadi - Collaborator - University of Manchester
Glen Martin - Collaborator - University of Manchester
Matthew Sperrin - Collaborator - University of Manchester
Maurice O'Connell - Collaborator - University of Manchester
Oscar Clivio - Collaborator - University of Oxford
HES Admitted Patient Care;ONS Death Registration Data;Patient Level Index of Multiple Deprivation Domains;Practice Level Index of Multiple Deprivation Domains