Examining Sensor Agreement in Neural Network Blood Glucose Prediction

Neural Networks With Gated Recurrent Units Reduce Glucose Forecasting Error Due to Changes in Sensor Location

BACKGROUND

Continuous glucose monitors (CGMs) have become important tools for providing estimates of glucose to patients with diabetes. Recently, neural networks (NNs) have become a common method for forecasting glucose values using data from CGMs. One method of forecasting glucose values is a time-delay feedforward (FF) NN, but a change in the CGM location on a participant can increase forecast error in a FF NN.

METHODS

In response, we examined a NN with gated recurrent units (GRUs) as a method of reducing forecast error due to changes in sensor location.

RESULTS

We observed that for 13 participants with type 2 diabetes wearing blinded CGMs on both arms for 12 weeks (FreeStyle Libre Pro-Abbott), GRU NNs did not produce significantly different errors in glucose prediction due to sensor location changes (P < .05).

CONCLUSION

We observe that GRU NNs can mitigate error in glucose prediction due to differences in CGM location.

Development and Validation of an Electronic Health Record-Based Risk Assessment Tool for Hypoglycemia in Patients With Type 2 Diabetes Mellitus

BACKGROUND

The recent availability of high-quality data from clinical trials, together with machine learning (ML) techniques, presents exciting opportunities for developing prediction models for clinical outcomes.

METHODS

As a proof-of-concept, we translated a hypoglycemia risk model derived from the Action to Control Cardiovascular Risk in Diabetes (ACCORD) study into the HypoHazardScore, a risk assessment tool applicable to electronic health record (EHR) data. To assess its performance, we conducted a 16-week clinical study at the University of Minnesota where participants (N = 40) with type 2 diabetes mellitus (T2DM) had hypoglycemia assessed prospectively by continuous glucose monitoring (CGM).

RESULTS

The HypoHazardScore combines 16 risk factors commonly found within the EHR. The HypoHazardScore successfully predicted (area under the curve [AUC] = 0.723) whether participants experienced least one CGM-assessed hypoglycemic event (glucose <54 mg/dL for ≥15 minutes from two CGMs) while significantly correlating with frequency of CGM-assessed hypoglycemic events (r = 0.38) and percent time experiencing CGM-assessed hypoglycemia (r = 0.39). Compared to participants with a low HypoHazardScore (N = 19, score <4, median score of 4), participants with a high HypoHazardScore (N = 21, score ≥4) had more frequent CGM-assessed hypoglycemic events (high: 1.6 ± 2.2 events/week; low: 0.3 ± 0.5 events/week) and experienced more CGM-assessed hypoglycemia (high: 1.4% ± 2.0%; low: 0.2% ± 0.4% time) during the 16-week follow-up.

CONCLUSIONS

We demonstrated that a hypoglycemia risk model can be successfully adapted from the ACCORD data to the EHR, with validation by a prospective study using CGM-assessed hypoglycemia. The HypoHazardScore represents a significant advancement toward implementing an EHR-based decision support system that can help reduce hypoglycemia in patients with T2DM.