Prediction of Cardiovascular and Renal Complications of Diabetes by a multi-Polygenic Risk Score in Different Ethnic Groups

Abstract We developed a multi-Polygenic risk score (multiPRS) to predict the risk of nephropathy, stroke, and myocardial infarction in people with type 2 diabetes of European descent. The underrepresentation of non-European populations remains a major challenge in genomics research. Objective : To evaluate the ability of our multiPRS model to accurately predict these complications in patients of African and South Asian descents. Method : The multiPRS was developed using 4098 participants with type 2 diabetes of European origin from the ADVANCE trial. Its predictive performance was tested on 17,574 White British, 1,145 South Asian and 749 African participants with type 2 diabetes from the UK Biobank using different machine learning prediction models, including techniques tailored for imbalanced datasets. Results : Globally, linear discriminant analysis and logistic regression had the best performance to predict the risk of nephropathy, stroke, and myocardial infarction in people with type 2 diabetes for the three ethnic groups. Mondrian Cross-Conformal Prediction method when added to logistic regression improved the AUROC values and case detection, particularly in South Asians and Africans, while in White British, performance varied by phenotype. Conclusion : Logistic regression, when used as the underlying model within the Modrian Cross-Conformal Prediction framework, improved the prediction performance, with a confidence level, of diabetes complications and allows better translation of a multiPRS derived from European populations to other ethnic groups.