Advancing Precision Medicine in Type 2 Diabetes through Machine Learning: Treatment Comparisons and Risk Predictions

Abstract

Precision medicine holds significant promise for improving the management of type 2 diabetes, yet its implementation faces several challenges. This thesis aims to address existing gaps in precision medicine for type 2 diabetes, focusing primarily on precision treatment, precision prognostic, and precision diagnosis. Each study within this thesis contributes to advancing the field in these key areas.

Study I compares the outcomes and safety of GLP-1 receptor agonists and SGLT-2 inhibitors for type 2 diabetes in a real-world setting, using propensity score matching and inverse probability treatment weighting to evaluate their practical effectiveness and side effects. Study II examines the validity and clinical utility of a proposed subclassification system for type 2 diabetes, using clustering analysis to explore potential subgroups and assess their ability to predict adverse events and diabetic complications. Study III identifies significant predictors for the development of type 2 diabetes, using an XGBoost classification model on a study population of 450,000 participants from the UK Biobank to project incidence and refine future risk scores. Study IV identifies principal predictors of cardiovascular complications and mortality in individuals with type 2 diabetes, using an XGBoost algorithm to analyze over 400 predictors and assess their impact on the risk of major adverse cardiovascular events.

Study I revealed similar effects of GLP-1 receptor agonists and SGLT-2 inhibitors on several cardiovascular outcomes, although GLP-1 receptor agonists were more effective in reducing stroke. Study II found distinct characteristics within potential subgroups of type 2 diabetes but limited predictive value in foreseeing adverse events. Study III identified HbA1c, BMI, waist circumference, and blood glucose levels as significant predictors of type 2 diabetes. The XGBoost model achieved high accuracy for 10-year type 2 diabetes prediction. Study IV highlighted the importance of variables such as age, Cystatin C, and pulse pressure in predicting cardiovascular events in individuals with type 2 diabetes. The model demonstrated high accuracy in predicting major adverse cardiovascular events.

In conclusion, this thesis contributes to advancing precision medicine in type 2 diabetes, providing valuable insights and paving the way for more personalized and effective approaches to treatment and management.