Skip to navigation Skip to main content Skip to footer

Approved research

Predictive Models for Cardio-Metabolic Disease: a Machine Learning Approach

Principal Investigator: Dr Aidin Rawshani
Approved Research ID: 53308
Approval date: September 26th 2019

Lay summary

Cardiometabolic diseases, such as heart attack, stroke, diabetes and obesity are the leading cause of death worldwide. Previous research regarding risk assessment models for cardiometabolic conditions has mostly examined routine clinical predictors, using mainly conventional statistical models to understand the relationship between risk factors and complications. We intend to use machine learning methods such as ensemble methods and other prediction models to illustrate the predictive ability of various risk factors of cardiometabolic disease. Moreover, our goal is to explore which risk factors that lead to the development and progression of additional cardiometabolic risk factors. Phenotypic cluster groups will be identified and examined with various dimensionality reduction techniques and cluster analyses to investigate if any specific clusters are associated with increased risk of cardiometabolic disease. In these studies, we will also investigate predictors and their influence on markers of atherosclerosis and cardiac dysfunction according to imaging data. By means of various machine learning models, we intend to study risk factors and phenotypic cluster groups that pose a various risk of developing subclinical atherosclerosis, pathological carotid dimensions, as well as heart chamber dimensions and function. Previous research on this matter is relatively scarce, few studies have performed large-scale prognostic assessments of relative risk factor importance, using machine learning methods. Furthermore, we will examine genetic risk variants and their interaction with other risk factors of cardiometabolic disease. Our goal is to estimate the interaction between gene variants and different phenotypes or diabetes-related traits, using predictive machine learning algorithms. We will investigate how these factors influence the risk of developing cardiometabolic complications. To our knowledge, no prior study has performed such analyses in an informative registry with genotypic and phenotypic data along with radiological examinations. Our models will result in novel definitions of cardiometabolic disease and risk factor importance. By constructing powerful predictive tools, we may be able to identify novel biomarkers, interactions and elucidate the biological and biochemical pathways that underpin cardiometabolic disease. In additions, we hope that these models will result in more precise clinical decision management and effective risk assessment.