University of Manchester

Academic Institute

Glen Martin - Chief Investigator - University of ManchesterAlexander Pate - Corresponding Applicant - University of ManchesterDarren Ashcroft - Collaborator - University of ManchesterGregory Lip - Collaborator - University of LiverpoolIain Buchan - Collaborator - University of LiverpoolJamie Sergeant - Collaborator - University of ManchesterKatherine McAllister - Collaborator - National Institute for Health and Clinical Excellence - NICELaura Bonnett - Collaborator - University of LiverpoolMamas Mamas - Collaborator - Keele UniversityMartin O'Flaherty - Collaborator - University of LiverpoolMatthew Sperrin - Collaborator - University of ManchesterMichelle McDowell - Collaborator - Harding Center for Risk LiteracyNiels Peek - Collaborator - University of ManchesterRichard Riley - Collaborator - Keele UniversityThomas Lawrence - Collaborator - National Institute for Health and Clinical Excellence - NICETjeerd van Staa - Collaborator - University of Manchester

CPRD32

Individuals with multiple medical conditions are more likely to die earlier and have lower quality of life. There are an increasing number of people who are diagnosed with multiple conditions. Being able to predict the chance of someone developing multiple conditions would help guide decision-making. Clinical prediction models can predict the risk of chronic conditions, such as cardiovascular disease or type 2 diabetes. Normally, these models predict the risk of just one outcome at a time. However, people at a high risk of one disease may be at an increased of another (for example stroke and type 1 diabetes). It may be useful for patients and clinicians to know what groups of conditions patients are at a high risk of, rather than only considering the risk of each condition on its own. This information could help guide treatment to target prevention of multiple conditions rather than just one. The information could also be useful to policy makers, as treatment strategies known to prevent more than one disease could be recommended, if those diseases are known to both occur regularly.

The overarching aim of this study is to develop statistical methodology to enable the clinical prediction of multimorbidity. This will include evaluating currently available methodology, and developing new methodology where appropriate. This will be undertaken in two work packages (WP). WP1 will provide motivation for and highlight the importance of this project. We will show that the univariate Cox models which are currently used in clinical practice to predict the development of chronic conditions, are unable to predict the risk of more than one disease co-occurring. This motivates the need for multivariate techniques (shared frailty, copula models, marginal approach) and multistate models for the prediction of multimorbidity. WP2 will look at how to best predict multimorbidity using existing multivariate and multistate models, and seek to understand why such methods are not commonly implemented in practice. The work will be largely simulation based, with the CPRD data used in case studies to elicit the differences between the methods of interest. HES and ONS will be used to determine outcome events that do not appear in the primary care records, as would be done in univariate models that are currently used in practice.

18/05/2021

Release