The rising prevalence of type 2 diabetes mellitus (T2DM) and the related health complications emphasize the necessity for predictive models that enable early diagnosis and intervention. While numerous artificial intelligence (AI) models for predicting T2DM risk have surfaced, a comprehensive assessment of their progress and challenges has been lacking. This scoping review aims to delineate the existing body of literature concerning AI-driven models for T2DM prediction, adhering to the PRISMA extension for Scoping Reviews guidelines.
In a study published in the journal npj Digital Medicine, researchers (Farida Mohsen, Hamada R. H. Al-Absi, Noha A. Yousri, Nady El Hajj & Zubair Shah) performed a scoping review to analyze artificial intelligence (AI)-based models for type 2 diabetes mellitus (T2DM) prediction.
A systematic exploration of longitudinal studies was carried out across four databases: PubMed, Scopus, IEEE-Xplore, and Google Scholar. Out of the numerous studies screened, forty met the specified inclusion criteria and were subjected to detailed evaluation. The studies predominantly leaned towards classical machine learning (ML) models, primarily relying on electronic health records (EHR) as their principal data source, followed by multi-omics data, while medical imaging was the least utilized modality. Most studies favored unimodal AI models, with only a limited subset embracing multimodal approaches. Both unimodal and multimodal models demonstrated promising outcomes, with the latter exhibiting superior performance.
In terms of evaluation, nearly all studies incorporated internal validation, but external validation was conducted in just five cases. The majority of studies employed the area under the curve (AUC) as the primary metric for assessing discriminative performance. Strikingly, only five studies offered insights into the calibration of their models. Half of the studies employed interpretability methods to identify the key risk factors identified by their models. Although some studies identified novel risk predictors, the majority reinforced the significance of well-established ones.
This review furnishes valuable insights into the present status and constraints of AI-based models for T2DM prediction, while underlining the challenges associated with their development and potential clinical integration.
