Multimorbidity is when someone has multiple long-term conditions (e.g., having both diabetes and asthma). It is associated with a wide range of adverse events, including dying early and having a worse quality of life. Multimorbidity is becoming more common because of population aging, and also because people are more likely to survive illnesses like heart attacks and stroke and so must live with long-term conditions. Treating people with multimorbidity is challenging because there are complex interactions between their conditions and the medications used to treat them. It would be useful to predict unexpected health problems (like falls and bleeding) caused by the conditions someone has, and/or the medicines they take. In this study, we will develop methods for detecting people who are likely to have unexpected health problems. This will support the choice of the most appropriate treatment and preventive care for an individual. To this end, we will use state-of-the-art network science and Artificial Intelligence methods applied to health records to improve the prediction of health problems. This work will help us target treatments at the people most likely to benefit from them.
Multimorbidity presents a challenge to health systems internationally [1]. People with complex multimorbidity and polypharmacy face high risks of adverse events (AEs) including acute kidney injury, bleeding, delirium, and falls. These AEs can result from their health conditions, their treatments, or a combination of both. The occurrence and severity of AEs are determined by the presence of predisposing factors, which stem from the complex interaction between morbidities, medications, and individual characteristics like frailty. Most healthcare systems are primarily designed to treat one condition at a time [2], and likewise prediction tools almost always predict single events (e.g. new cardiovascular disease, fragility fractures). However, with increasing multimorbidity, clinicians and patients often must choose treatments based on trading off risks of multiple outcomes. Better anticipatory management of AE predisposing factors has considerable potential to improve outcomes. This study aims to use graph-based models to build an integrated representation of information from health records and medical expert knowledge, and make associated predictions to support clinical decisions and intervention design.
The study will have two main objectives:
1. To develop predictive models to measure the risk of experiencing adverse events, using graph-based modelling of electronic health records.
2. To improve our understanding of morbidity accrual via temporal graph models of longitudinal disease trajectories.
The study cohort will be a population sample of adults registered with a primary care practice. Using Clinical Practice Research DataLink (CPRD) Aurum, Hospital Episode Statistics (HES) admission data and Office of National Statistic (ONS) mortality data, we will build a network of inter-connected entities (people, morbidities, medications, AEs) to serve as a knowledge base for predictive modelling. To maximise clinical utility, we will adopt models with greater interpretability and explainability of predictions compared to traditional methods, including graph neural networks, graph attention networks and multi-layer networks.
Outcomes examined will be: (1) Any emergency hospital admission, and ambulatory care sensitive (also known as potentially preventable) emergency hospital admissions; (2) Complications of medical care associated with drugs (where explicitly coded as such in hospital discharge records/ICD-10); (3) Bleeding including upper and lower gastrointestinal, intracranial, haematuria, epistaxis, haemoptysis and other rarer sites (recorded in GP, hospital, or mortality records irrespective of any attribution to a drug; applies to 4-12 as well); (4) Acute peptic ulceration including complications such as perforation; (5) Acute kidney injury including subsequent non-recovery and development or worsening of chronic kidney disease; (6) Falls; (7) Major osteoporotic and low impact fractures; (8) Delirium; (9) Fluid and electrolyte disturbance; (10) Hypotension; (11) Heart failure; (12) Dysrhythmia; (13) Hypoglycaemia; (12) Subsequent excess diagnosis of new long-term conditions.
Bruce Guthrie - Chief Investigator - University of Edinburgh
Paola Galdi - Corresponding Applicant - University of Edinburgh
Atul Anand - Collaborator - University of Edinburgh
Guillermo Romero Moreno - Collaborator - University of Edinburgh
Imane Guellil - Collaborator - University of Edinburgh
Jacques Fleuriot - Collaborator - University of Edinburgh
Jake Palmer - Collaborator - University of Edinburgh
Lauren DeLong - Collaborator - University of Edinburgh
Luna De Ferrari - Collaborator - University of Edinburgh
Marcus Lyall - Collaborator - University of Edinburgh
Nazir Lone - Collaborator - University of Edinburgh
Sohan Seth - Collaborator - University of Edinburgh
Stewart Mercer - Collaborator - University of Edinburgh
Valerio Restocchi - Collaborator - University of Edinburgh
HES Admitted Patient Care;ONS Death Registration Data;Patient Level Index of Multiple Deprivation;Practice Level Rural-Urban Classification