Global Deep Forecasting with Patient-Specific Pharmacokinetics

2023-09-22

AI Summary

Long-term modeling in clinical blood glucose forecasting is challenged by inter-individual variability in medication timing and patient-specific pharmacokinetics (PK) during personalized therapy. Method: This paper proposes a hybrid neural architecture integrating globally shared representations with locally personalized modeling. Its core innovation is a differentiable PK encoder that explicitly incorporates interpretable, embeddable PK response dynamics into deep temporal models for the first time. Contribution/Results: Evaluated on both synthetic and real-world clinical datasets, the method improves prediction accuracy for hypoglycemic and hyperglycemic events by 16.4% and 4.9%, respectively, and reduces overall prediction error by 15.8% compared to purely individualized PK models. It significantly enhances model generalizability and clinical interpretability while preserving physiological plausibility.

Abstract

Forecasting healthcare time series data is vital for early detection of adverse outcomes and patient monitoring. However, it can be challenging in practice due to variable medication administration and unique pharmacokinetic (PK) properties of each patient. To address these challenges, we propose a novel hybrid global-local architecture and a PK encoder that informs deep learning models of patient-specific treatment effects. We showcase the efficacy of our approach in achieving significant accuracy gains in a blood glucose forecasting task using both realistically simulated and real-world data. Our PK encoder surpasses baselines by up to 16.4% on simulated data and 4.9% on real-world data for individual patients during critical events of severely high and low glucose levels. Furthermore, our proposed hybrid global-local architecture outperforms patient-specific PK models by 15.8%, on average.

Problem

• Forecasting healthcare time series data
• Addressing variable medication administration
• Incorporating patient-specific pharmacokinetic properties

Innovation

• Hybrid global-local architecture
• Patient-specific PK encoder
• Deep learning for healthcare forecasting