The role of dynamic environmental factors, genetic information, and gene-environment interactions in non-communicable chronic diseases and the development and validation of prediction models for risks
Approved Research ID: 95259
Approval date: January 11th 2023
The prevalence of non-communicable chronic disease (NCDs), such as diabetes and its complications, cardiovascular disease and dementia, is rapidly increasing and imposes a large economic burden globally. For the prevention and control of NCDs, it is essential to identify modifiable risk factors (diet and intake of nutrients, lifestyle, obesity, and socioeconomic factors etc.) and intervene these factors through behavioral changes and clinical treatments. The accumulation of gene-environment interaction data can provide us with novel insights into the precise personalized prediction and intervention of NCDs. However, this field is still largely unexplored.
The data of dynamic factors are obtained in longitudinal studies with a long follow-up, allowing researchers to better quantify the association between risk factors and NCDs. Machine learning and deep learning have made significant advances in medicine over the last decade, notably in the areas of computer-aided screening and triage, precision diagnosis, and decision support. Mendelian randomisation (MR) utilizing genetic variations as instrumental variables will be conducted to explore causal effect of exposures on the outcome. Sufficient sample size in UK Biobank and its extensive information on diet, lifestyle and genetics enable us to explore gene-environment interactions and the role of different risk factors in the pathogenic mechanisms underlying NCDs.
Specifically, we aim to 1) explore the association between dynamic modifiable exposures and NCDs using machine learning algorithm and Cox regression model based on the prospective longitudinal cohort; 2) examine the interaction between dynamic modifiable exposures and the genetic variations in relation to NCDs risk; 2) estimate the causal effect of modifiable exposures on NCDs using the multivariate MR and network MR analyses; 4) develop and validate the risk prediction models for new-onset NCDs using deep learning-based survival prediction algorithms based on time-varying variables and genetic information. According to our expectations, the project will last 36 months. The findings will eventually contribute to the improvement of the prevention, diagnosis and treatment of NCDs, as well as the realization of precision medicine, and will help to reduce the future burden of NCDs on individuals and society.