Artificial intelligence with temporal features outperforms machine learning in predicting diabetes.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Public Health

Background:

• Type 2 diabetes is a growing pandemic with increasing complications despite available treatments.
• Early prediction and intervention are crucial for preventing diabetes and its adverse outcomes.
• Longitudinal electronic medical record (EMR) data offers potential for advanced diabetes prediction models.

Purpose of the Study:

• To evaluate the predictive performance of deep learning models against state-of-the-art machine learning models for diabetes prediction.
• To assess the impact of incorporating the time dimension of risk using longitudinal EMR data.
• To identify key predictors and patient characteristics associated with diabetes onset.

Main Methods:

• Utilized deep learning models, including LSTM, and compared them with other machine learning approaches.
• Analyzed longitudinal EMR data from over 19,000 patients from the Canadian Primary Care Sentinel Surveillance Network (CPCSSN).
• Evaluated model performance based on prediction accuracy, feature importance, data density, and patient visit history.

Main Results:

• Deep learning models achieved prediction accuracy exceeding 91% without overfitting, outperforming traditional machine learning methods.
• Key predictors for diabetes onset include fasting blood sugar, hemoglobin A1c, and body mass index.
• Increased training data density and patient visit history positively correlated with improved model performance.

Conclusions:

• Deep learning models, particularly LSTM, effectively leverage the temporal dimension in EMR data for accurate diabetes prediction.
• Early identification of high-risk individuals (overweight, middle-aged, hypertensive) is feasible, enabling timely therapeutic interventions.
• This research highlights the significant potential of AI in managing the diabetes pandemic through predictive analytics.