The Driver-Blindness Phenomenon: Why Deep Sequence Models Default to Autocorrelation in Blood Glucose Forecasting

2025-11-25

AI Summary

This paper identifies a pervasive “driver-blindness” problem in deep learning sequence models for blood glucose prediction: models over-rely on glucose autoregression while neglecting critical clinical drivers such as insulin administration, dietary intake, and physical activity. To address this, we propose Δ_drivers—a novel, quantifiable metric that formally evaluates a model’s utilization of multivariate driver information. We identify three root causes: architectural biases favoring temporal autocorrelation, data distortions (e.g., irregular sampling, missingness), and physiological heterogeneity across individuals. Accordingly, we design a physiology-aware modeling framework integrating causal regularization, physiologically grounded feature encoders, and personalized modeling. Experiments show mainstream models exhibit Δ_drivers ≈ 0, confirming severe underutilization of driver signals; incorporating physiological mechanisms significantly increases Δ_drivers, enhancing both predictive accuracy and clinical interpretability. Our work advances glucose forecasting toward mechanism-informed, clinically trustworthy decision support.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Sequential Decision Making
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Deep sequence models for blood glucose forecasting consistently fail to leverage clinically informative drivers--insulin, meals, and activity--despite well-understood physiological mechanisms. We term this Driver-Blindness and formalize it via $Delta_{ ext{drivers}}$, the performance gain of multivariate models over matched univariate baselines. Across the literature, $Delta_{ ext{drivers}}$ is typically near zero. We attribute this to three interacting factors: architectural biases favoring autocorrelation (C1), data fidelity gaps that render drivers noisy and confounded (C2), and physiological heterogeneity that undermines population-level models (C3). We synthesize strategies that partially mitigate Driver-Blindness--including physiological feature encoders, causal regularization, and personalization--and recommend that future work routinely report $Delta_{ ext{drivers}}$ to prevent driver-blind models from being considered state-of-the-art.

Problem

• Deep sequence models fail to utilize clinical drivers in glucose forecasting
• Models default to autocorrelation due to architectural and data limitations
• Performance gap between multivariate and univariate models remains near zero

Innovation

• Physiological feature encoders capture clinical drivers
• Causal regularization reduces autocorrelation bias
• Personalization addresses physiological heterogeneity issues