Type 2 diabetes prediction without labs: a systems-level neural framework for risk and behavioral network...

Area of Science:

• Systems biology
• Behavioral science
• Computational epidemiology

Background:

• Traditional Type 2 Diabetes Mellitus (T2DM) prediction relies on biochemical markers reflecting established disease.
• Non-clinical factors like stress, sleep, diet, and smoking are modifiable and detectable earlier.
• Existing research often overlooks the complex interdependencies of these factors in preclinical T2DM development.

Purpose of the Study:

• To develop a dual-analytic framework integrating survival models and artificial neural network (ANN) coherence analysis.
• To identify non-clinical predictors of incident T2DM.
• To examine the reorganization of behavioral networks across health states in the preclinical phase.

Main Methods:

• Utilized longitudinal UK Biobank data (n=15,774) with up to 17 years of follow-up.
• Employed Cox proportional hazards models for predictor screening and ANN coherence analysis for network mapping.
• Quantified predictor stability, direction, and systemic coherence within behavioral networks.

Main Results:

• Identified 18 significant T2DM predictors, including loneliness, psychiatric consultation, insomnia, irregular sleep, and processed food intake.
• Found protective effects for adequate sleep, oat/muesli consumption, and fermented dairy.
• ANN analysis revealed a breakdown in behavioral coherence, with mood-altering foods becoming distress-associated and emotional states driving diet.

Conclusions:

• T2DM risk arises from systemic reorganization within behavioral networks, not isolated factors.
• Early T2DM prediction is feasible using modifiable behaviors without laboratory tests.
• The framework offers insights for psychologically informed, personalized prevention strategies.