Managing Diabetes Using Machine Learning and Digital Twins

Abstract

Diabetes is a major public health problem worldwide, and early diagnosis will remain pivotal for intervention and management. This Systematic Literature Review (SLR), therefore, attempts to explore the prospects of integrating Machine Learning (ML) and Digital Twins (DT) to enable diabetes treatment through prediction and patient-specific modeling. This SLR contributes to the body of literature by examining how ML and DTs are being applied in diabetes treatment, identifying the opportunities and challenges that exist, and determining which algorithms are most commonly used. In contrast to SLRs that have been reviewed previously, this study considers Digital Twin-based technological perspectives, along with algorithmic evaluations of ML models, to provide an overall view of the potential for combined use in diabetes care. Following PRISMA guidelines, 11 relevant studies were selected from major academic databases. The study identified Random Forests, Gradient-Boosted Decision Trees, K-Nearest Neighbors, Time Series and Structured Analysis, Regression-based algorithms, and Artificial Neural Networks as machine learning algorithms commonly used to predict diabetes risk. The integration of ML and DT for diabetes management enables the personalization of patient management through virtual representations, real-time monitoring of an individual's glucose levels, simulation of disease progression, and prediction of subsequent treatment steps for proactive and immediate decision-making. Through this collaboration, simulations of various situations are performed, and the interventions are optimized to correspond with unique human physiological profiles for better patient outcomes. Based on the results, policymakers must balance data quality and patient privacy.