Predictive ability of current machine learning algorithms for type 2 diabetes mellitus: A meta-analysis

Area of Science:

• Biostatistics
• Medical Informatics
• Epidemiology

Background:

• Machine learning (ML) is increasingly explored for predicting type 2 diabetes mellitus (T2DM).
• Existing cohort studies suggest ML's potential, but its predictive accuracy for T2DM remains unclear.
• This meta-analysis synthesizes evidence on ML's ability to predict incident T2DM.

Purpose of the Study:

• To evaluate the predictive performance of machine learning algorithms for incident type 2 diabetes mellitus.
• To provide a quantitative assessment of ML's diagnostic accuracy in T2DM prediction.

Main Methods:

• Systematic literature search of MEDLINE and EMBASE for longitudinal studies (1950-2020).
• Inclusion criteria: studies comparing ML classification with actual T2DM incidence, reporting true positives, false positives, true negatives, and false negatives.
• Data pooled using hierarchical summary receiver operating characteristic and bivariate random effects models.

Main Results:

• 12 eligible studies were included in the meta-analysis.
• Pooled sensitivity was 0.81 (95% CI 0.67-0.90) and specificity was 0.82 (95% CI 0.74-0.88).
• Area under the curve (AUC) for the summarized ROC curve was 0.88 (95% CI 0.85-0.91), indicating good overall performance.

Conclusions:

• Current machine learning algorithms demonstrate sufficient ability to assist clinicians in predicting future type 2 diabetes mellitus risk.
• Caution is advised for individuals and clinicians when interpreting ML-based diabetes prediction results and modifying attitudes toward future risk.