Identify treatment responders in patients with type 2 diabetes using a machine learning based dynamic cardiovascular risk assessment tool (ML-CVD) in clinical trials

Type 2 diabetes is a common health concern characterised by high blood sugar levels. It affects over 500 million adults worldwide and leads to serious health problems related to the heart. To manage diabetes, several types of medicine are now available. Sodium/glucose cotransporter 2 inhibitors (SGLT2i) help the kidneys get rid of extra sugar through urine and also protect the heart. Glucagon-like peptide-1 receptor agonists (GLP-1 RAs) help the body release more insulin after meals, lower blood sugar, and reduce appetite, which can also support weight loss and heart protection. Dipeptidyl peptidase-4 inhibitors (DPP4i) stop the breakdown of certain hormones that help the body make more insulin and reduce sugar release from the liver. Metformin mainly works by reducing the amount of sugar made by the liver and helping the body use insulin more effectively. Non-steroidal mineralocorticoid receptor antagonists (ns-MRA) block certain hormone signals that can damage the heart, lowering the risk of heart failure progression. Now clinical guidelines recommend special tools to figure out who might get cardiovascular events because of diabetes. The tools are designed to categorize patients into low, medium, or high cardiovascular risk, and the initiation of novel drugs is preferred for patients at high risk. However, these predictive tools weren’t always accurate enough for people with type 2 diabetes, especially for those already have heart diseases.
In our previous study, we created a new tool called ML-CVD (Machine Learning for CardioVascular Diseases). This tool used patient information and health updates over time to have a more precise prediction on the 5-year probability of heart disease than the traditional tools. It also helped to identify which patient benefit more from SGLT2i. But whether this model could predict the treatment effect of other drugs is unknown. In this study, we aim to validate the ML-CVD model in controlled clinical trials. We would estimate the risk score calculated by ML-CVD model and test the C-index (a popular statistic often applied to models with binary or survival outcome variables) of the model on predicting the risk in each trial to know if it's effective to predict risk for a broader patient group. Observing the change of the risk score, we would also identify those who might have more heart benefit from specific medications. The study could lead to better care for people with type 2 diabetes by providing more accurate risk assessments and identifying more beneficial treatments for each individual.
The study was transferred from the Vivli project ID00010038.