Model-Based Blood Glucose Control for Neonatal Intensive Care

Type of content
Conference Contributions - Published
Publisher's DOI/URI
Thesis discipline
Degree name
Publisher
University of Canterbury. Mechanical Engineering
Journal Title
Journal ISSN
Volume Title
Language
Date
2008
Authors
LeCompte, A.J.
Chase, Geoff
Russel, G.
Hann, C.E.
Shaw, Geoff
Abstract

Background and Aims

Premature, low-birth-weight infants represent a significant proportion of the neonatal intensive care population. Glucose is a primary source of energy for the foetus, which depends upon the mother for supply and regulation.

When born prematurely, several of the body’s glucose supply and control mechanisms are under-developed. Hyperglycemia (high blood sugar levels) occurs in 40-80% of very low birth weight infants in the neonatal intensive care unit (NICU). This condition has been linked to mortality and morbidities including retinopathy of prematurity, osmotic diuresis, reduced immune system performance and sepsis. Insulin therapy can control hyperglycaemia and promote growth, but increases the risk of dangerous low levels of blood glucose. Model-based methods can provide tools to accurately balance the energy demands of the infant with the constraints of insulin usage whilst maintaining normal glucose levels.

The goal of this model is to provide a vehicle for real-time blood glucose control by accurately capturing the dynamic effect of insulin to provide dosing recommendations for attending clinicians. The model is used for two tasks: predict future blood glucose concentration for real-time control, and perform simulated trials to optimise control strategies.

Models and Methods

The glucose regulatory system model is based upon a similar model employed successfully in adult intensive care, and modified to account for differing physiology in the neonate.

Insulin sensitivity is the driving parameter in the model. Stochastic modelling and time-series analysis methods provide confidence bands for blood glucose predictions. Retrospective data is used to generate patient-specific, time-varying insulin sensitivity profiles via integral-based parameter identification methods. The profiles are used to generate “virtual patients”, which are used to simulate patient responses to glucose and insulin inputs.

Results

Retrospective data for 25 episode of insulin usage representing over 3,500 hours of patient data was used to validate the model in simulation. Median absolute prediction errors for the 25 virtual patients at 1 and 2 hour intervals were 6.3% and 10.4% respectively, with 90% of 1-hour-ahead blood glucose predictions within ±20% of the in-sample value.

Stochastic modelling of the insulin sensitivity parameter averaged 53% of blood glucose predictions within the 1.03 mmol/L wide 25%-75% confidence range, and 88% of measurements within the 2.88 mmol/L wide 5%-95% confidence range.

Simulations using basic controllers over the 25 “virtual patients” resulted in 64.5% of hourly simulated measurements within the target 4-6 mmol/L band, compared to 19% for retrospective hospital control. Mean blood glucose decreased 37% from 8.3 mmol/L to 5.2 mmol/L, and the standard deviation decreased 60% from 3.0 mmol/L to 1.2 mmol/L. Increased time in a target band and reduced standard deviation are robust measures of the tight control possible with model-based methods.

Conclusions Hyperglycaemia affects a large proportion of premature infants, and has been linked to worsened outcomes. A model that accurately captures the dynamics of neonatal metabolism can provide a vehicle for real-time blood glucose control and a platform to develop high-performance control algorithms in simulation. Reduced hyperglycaemia and tighter control in simulated results highlight the possibility for tight glucose control for improved outcomes in this fragile neonatal cohort.

Description
Citation
LeCompte, A.J., Chase, J.G., Russel, G., Hann, C.E., Shaw, G.M. (2008) Model-Based Blood Glucose Control for Neonatal Intensive Care. Auckland, New Zealand: NZBio 2008 Conference & Exposition, 30 Mar-2 Apr 2008.
Keywords
Ngā upoko tukutuku/Māori subject headings
ANZSRC fields of research
Rights