1987-LB: Limitations of Artificial Intelligence Research in Predicting Type 2 Diabetes Macrovascular Complications—A Scoping Review

Le résumé fourni par la source

Over 500 million people were estimated to live with diabetes in 2021, of which 96% were type 2. This leads to various complications, among which are macrovascular diseases (e.g., stroke and coronary heart diseases). As complications increase the 5-year mortality of patients, curbing their progression through early prediction and intervention is key. This scoping review explores the characteristics of current research on how artificial intelligence, including machine learning algorithms, has been utilized to predict diabetes macrovascular complications. In adherence to PRISMA-ScR guidelines, we systematically searched PubMed, Google Scholar, Scopus, IEEE Xplore, EMBASE, and Wiley for relevant literature up to 12 December 2023. Out of the 1,667 hits screened, 52 studies cumulating 7,510,245 people with type 2 diabetes are included. We found 43 studies from HICs/UMICs, in contrast to 9 from LMICs/LICs. 30 studies came from North America and Europe regions, while others from Asia and Australia. Of all macrovascular complications, cardiovascular diseases (e.g., coronary heart disease) have been the most investigated outcome. According to the features of the models, only 12 studies employed non-laboratory features as predictors, while the remaining studies applied solely laboratory (n=2) or mixed (n=38) features, signalling the lack of AI capability for history-taking and physical examination data alone, which are mostly available in low-resource settings. While artificial intelligence is promising in predicting diabetes complications, future studies should explore accessible features in low-resource settings and employ external validation. A systematic review and meta-analysis exploring the performance metrics of a variety of algorithms should be done. Disclosure A. Nur: None. D. Harbuwono: None.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.