Prediction Models for Type 2 Diabetes Progression: A Systematic Review

Abstract

Diabetes, especially type 2 diabetes (T2D), is a chronic disease affecting millions of people worldwide. The increasing prevalence of T2D, coupled with the complex interplay between genetic, environmental, and lifestyle factors, presents a major challenge for effective disease management. The traditional methods for predicting T2D progression and determining appropriate treatment strategies are often subjective and less accurate, resulting in treatment delays. Therefore, artificial intelligence (AI) based prediction models become crucial, as they offer a more objective and data-driven approach to T2D management. By leveraging advanced statistical techniques and machine learning algorithms, AI-based prediction models can better identify patients at high risk for T2D progression and predict responses to different treatment options. This can ultimately lead to improved outcomes for patients suffering from T2D. Therefore, this paper aims to review the existing research articles published from 2018 to 2022 using a systematic literature review (SLR) approach. From 40 selected articles, a taxonomy of the most common techniques for developing a prediction model in diabetes progression is drawn in three approaches: mathematical, machine learning (ML), and deep learning (DL). In addition, the best practices of dataset characteristics, pre-processors, and evaluation metrics of the existing algorithms are also provided, focusing on the context of diabetes progression prediction. The findings found that the majority of the selected papers employed ML, specifically the RF model, proven to have superiority in performance. This review also discusses current challenges faced in building prediction models for diabetes progression and proposes future research directions to overcome these challenges. The promising directions drawn include 1) incorporating feature reduction or importance tools to explore the relationship between variables, 2) developing an interpretable predictive model to provide analytical results that are understandable to clinicians, and 3) validating the model with multiple large-sample size datasets and seeking clinical advice from experts. Authors