AI Models Show Promise in Predicting Type 2 Diabetes Risk: Study
Artificial intelligence (AI) models have demonstrated significant potential in predicting the risk of type 2 diabetes mellitus (T2DM), according to a recent study published in the journal npj Digital Medicine.
The prevalence of diabetes is rapidly increasing worldwide, making the development of accurate prediction models crucial for identifying individuals at high risk and implementing targeted prevention strategies. AI-based models offer the capability to personalize disease-preventive methods and focused therapies.
To assess the applications of AI in diabetes risk prediction, researchers conducted a comprehensive review of relevant studies. They systematically searched databases for longitudinal studies published between January 2000 and September 2022 that utilized AI-based models for predicting T2DM in human subjects.
Out of the initial 1105 records identified, 40 studies were included in the final analysis. The majority of these studies were published in the past four years and featured diverse populations from different countries, including China, Finland, California, and Kuwait.
The most commonly used data modality was electronic health records (EHRs), which encompassed various factors such as sociodemographic data, family history of diabetes, lifestyle factors, and biomarkers like glycemic traits and lipid levels. Additionally, multi-omics data, such as genetic variations and microbiota information, were frequently incorporated into the models, while medical imaging data was less utilized.
Most of the studies employed unimodal AI models, although some used multimodal approaches. The multimodal models consistently outperformed the unimodal models, exhibiting superior predictive accuracy. Classical machine learning algorithms, including decision trees, linear regression modeling, random forest classifiers, and support vector machines, were commonly employed in these studies.
However, there were limitations encountered at different stages of AI model construction, including challenges associated with data availability, model building, evaluation, and clinical translation. While internal validation was performed in the majority of the studies, only a few conducted external validation. Model calibration was also lacking in most studies.
In terms of interpretability, around half of the studies utilized methods to identify the risk predictors. Fasting blood glucose, body mass index (BMI), age, and serum triglyceride levels emerged as the most frequently documented predictors. Metabolomic markers and imaging-based biomarkers, particularly for diabetes-related retinal disease, were also identified in some studies.
As AI models show promise in predicting the development of T2DM, further validation and assessment through clinical trials and prospective research are necessary to fully understand their potential benefits. The integration of AI into healthcare should be a collaborative effort between AI models and human expertise, aiming to accelerate advancements and improve clinical outcomes for patients and healthcare professionals.