Principal Investigator: Dr Cedric Manlhiot
Hospital for Sick Children, Cardiovascular Data Management Centre, Peter Gilgan Centre, 686 Bay Street, Toronto
Ontario Canada M5G 0A4Tags: 30603, featured, Heart failure, Machine Learning, prognostication, visualization
Lead Collaborators: 1) Maria Kannu
Collaborating Institutions and Addresses:
University Health Network
Peter Munk Cardiac Centre
200 Elizabeth Street
Ontario M5G 2C4
1a: 2 clinical (#1-2)/2 methodological objectives (#3-4):
1. Develop risk prediction models for the development of heart failure (HF), including lifestyle factors, and develop interactive data visualization tools for patient education.
2. Develop prognostic models for HF to identify modifiable risk factors that can be used to individualize patient management.
3. Explore the incremental value of genetic polymorphisms in addition to non-genetic risk factors in HF prognostication.
4. Explore the use of machine learning in addition to classic probabilistic methods for prognostication with a particular focus on studying time-to-event outcomes and patterns of prediction failure.
1b: This work specifically addresses the UK Biobank’s stated purpose by aiming to improve prognostication for the development and management of heart failure and develop interactive, patient-focused, visualization tools for more effective patient education.
Ultimately we aim to prevent and/or individualize the management of heart failure by focusing on patient-centric, modifiable risk factors and better delivering the information to patients in a novel manner. Only by utilizing a large cohort of patients with comprehensive clinical, genetic and lifestyle assessments such as the one uniquely provided by the UK Biobank, can this be achieved in a meaningful manner.
1c: We will divide the UK biobank cohort into 3 groups using the intake assessment: those with heart failure (HF), those without HF but with signs/symptoms and those with no signs/symptoms of HF. For all groups, we will attempt to develop prognostic models determining whether, during follow-up, they will be hospitalized or die because of HF. Those models will be created using multiple different analytical techniques and different subsets of information; their performance will be compared and studied. We will then develop and test interactive data visualization tools to showcase these models to the public, patients and clinicians.
1d: The full cohort will be used given that all participants can be classified in one of the 3 study groups (described in 1c) based on the intake assessment and outcomes can be ascertained in all patients. Some of the prognostic models may use subset of data depending on risk factor availability (e.g. genetics, scans). A sub-analysis will be performed on patients with an intake cardiac MRI, this sub-analysis will focus on patients with sub-clinical HF, a high-risk but poorly characterized subgroup.