JMIR Publications

ABSTRACT

Background:

The global prevalence of diabetes is one of the most pressing health concerns worldwide. Early detection is crucial for the effective management and prevention of complications. Various artificial intelligence (AI) techniques, including machine learning and deep learning, are employed for diabetes detection.

Objective:

This systematic review and meta-analysis aimed to evaluate the effectiveness and feasibility of AI-driven approaches for early diabetes detection in primary and preventive care settings.

Methods:

We followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses model to minimize bias and enhance accuracy. In October 2024, we searched two databases, PubMed and Google Scholar, using keywords such as ("Artificial intelligence" OR "machine learning") AND ("early diabetes detection" OR "diabetes prediction"). Data extraction focused on study design, population characteristics, AI model type, accuracy, comparison groups, outcomes (e.g., diagnostic accuracy), and follow-up periods. A meta-analysis was performed using RevMan to assess diagnostic precision, predictive value, and risk stratification capability.

Results:

The included studies focused on improving diabetes prediction through advanced machine learning algorithms, achieving up to 96.75% accuracy. The datasets used were diverse, including demographic, clinical, and behavioral variables. However, the studies also highlighted limitations, such as gaps in data completeness and external validation, with missing data being a recurrent issue.

Conclusions:

AI-driven methods show substantial promise in improving diagnostic precision and patient outcomes in diabetes management. Future research should address methodological gaps, ensure robust validation, and prioritize long-term effectiveness for real-world implementation.