A deep learning approach to diabetic blood glucose prediction

TL;DR

This paper presents a deep learning model for predicting blood glucose levels 30 minutes ahead using clinical data, demonstrating improved performance over shallow networks and introducing a domain knowledge-based approach for efficient representation.

Contribution

It introduces a deep learning method for cross-patient blood glucose prediction and shows how domain knowledge can create a parsimonious, effective model.

Findings

Deep learning outperforms shallow networks in glucose prediction

Model generalizes across different patients without re-calibration

Domain knowledge aids in constructing efficient representations

Abstract

We consider the question of 30-minute prediction of blood glucose levels measured by continuous glucose monitoring devices, using clinical data. While most studies of this nature deal with one patient at a time, we take a certain percentage of patients in the data set as training data, and test on the remainder of the patients; i.e., the machine need not re-calibrate on the new patients in the data set. We demonstrate how deep learning can outperform shallow networks in this example. One novelty is to demonstrate how a parsimonious deep representation can be constructed using domain knowledge.