Adaptive tuning of basal and bolus insulin to reduce postprandial hypoglycemia in a hybrid artificial pancreas

Navid Resalat, Joseph El Youssef, Ravi Reddy, Jessica Castle, Peter Jacobs

Research output: Contribution to journalArticle

Abstract

Objective: We introduce an adaptive learning algorithm to better adjust postprandial basal and pre-meal bolus insulin for reducing postprandial hypoglycemia in a hybrid artificial pancreas (AP). An AP uses a control algorithm and sensed glucose to automate the delivery of insulin to people with type 1 diabetes (T1D). A hybrid AP requires the person to dose insulin in advance of a meal. Insulin sensitivity is dynamic in people with T1D, making it challenging for an AP to maintain euglycemia. Adaptive approaches to meal dosing can help prevent postprandial hypoglycemia. Methods: An adaptive learning postprandial hypoglycemia-prevention algorithm (ALPHA) is introduced. One implementation of ALPHA adjusts the rate of postprandial insulin (ALPHA-BR) proportionally in response to prior postprandial episodes. This is achieved by an adaptive aggressiveness factor applied to postprandial basal insulin. The second implementation adaptively updates the pre-meal bolus insulin by changing the insulin-to-carbohydrate ratio (ALPHA-ICR), also proportionally in response to prior postprandial hypoglycemia. Both implementations were evaluated within an AP on an in-silico T1D virtual population of 99 subjects with circadian insulin sensitivity variations and 30% errors on meal estimations. Twenty real-world 4-day meal scenarios were given and glycemic outcomes were compared with an AP with no adaptation. Results: Out of the 99 in-silico subjects, 23 of them experienced postprandial hypoglycemia leading to greater than 1% overall time in hypoglycemia. Of these 23 subjects, we evaluated the benefit of using ALPHA-BR and ALPHA-ICR to prevent postprandial hypoglycemia. ALPHA-BR yielded substantially fewer percent time in hypoglycemia compared to AP (0.54% vs 1.92%, p < 0.001) and fewer rescue carbs per day (0.36 vs. 1.29, p < 0.001). For the control algorithm evaluated, it yielded an average aggressiveness factor of 0.72 for reducing postprandial basal insulin. ALPHA-ICR slightly reduced time in hypoglycemia compared to AP (1.77% vs. 1.92%, p = 0.09). Conclusion: Incorporating adaptive meal dosing into an AP can help reduce postprandial hypoglycemia, and the reduction is primarily due to changes in postprandial insulin delivery rather than pre-meal bolus. Significance: Adapting postprandial insulin can lead to substantial reduction in postprandial hypoglycemia and the adaptive algorithm presented can be used both to tune an algorithm prior to a study and to adapt to individuals during real-time usage.

Original languageEnglish (US)
Pages (from-to)247-254
Number of pages8
JournalJournal of Process Control
Volume80
DOIs
StatePublished - Aug 1 2019

Fingerprint

Insulin
Adaptive Learning
Tuning
Diabetes
Medical problems
Adaptive algorithms
Adaptive Algorithm
Control Algorithm
Carbohydrates
Glucose
Learning algorithms
Percent
Learning Algorithm
Dose
Person
Update

Keywords

  • Adaptive control
  • Hybrid artificial pancreas
  • Type 1 diabetes

ASJC Scopus subject areas

  • Control and Systems Engineering
  • Modeling and Simulation
  • Computer Science Applications
  • Industrial and Manufacturing Engineering

Cite this

@article{7ead56567a1d415a8d4731fe59484882,
title = "Adaptive tuning of basal and bolus insulin to reduce postprandial hypoglycemia in a hybrid artificial pancreas",
abstract = "Objective: We introduce an adaptive learning algorithm to better adjust postprandial basal and pre-meal bolus insulin for reducing postprandial hypoglycemia in a hybrid artificial pancreas (AP). An AP uses a control algorithm and sensed glucose to automate the delivery of insulin to people with type 1 diabetes (T1D). A hybrid AP requires the person to dose insulin in advance of a meal. Insulin sensitivity is dynamic in people with T1D, making it challenging for an AP to maintain euglycemia. Adaptive approaches to meal dosing can help prevent postprandial hypoglycemia. Methods: An adaptive learning postprandial hypoglycemia-prevention algorithm (ALPHA) is introduced. One implementation of ALPHA adjusts the rate of postprandial insulin (ALPHA-BR) proportionally in response to prior postprandial episodes. This is achieved by an adaptive aggressiveness factor applied to postprandial basal insulin. The second implementation adaptively updates the pre-meal bolus insulin by changing the insulin-to-carbohydrate ratio (ALPHA-ICR), also proportionally in response to prior postprandial hypoglycemia. Both implementations were evaluated within an AP on an in-silico T1D virtual population of 99 subjects with circadian insulin sensitivity variations and 30{\%} errors on meal estimations. Twenty real-world 4-day meal scenarios were given and glycemic outcomes were compared with an AP with no adaptation. Results: Out of the 99 in-silico subjects, 23 of them experienced postprandial hypoglycemia leading to greater than 1{\%} overall time in hypoglycemia. Of these 23 subjects, we evaluated the benefit of using ALPHA-BR and ALPHA-ICR to prevent postprandial hypoglycemia. ALPHA-BR yielded substantially fewer percent time in hypoglycemia compared to AP (0.54{\%} vs 1.92{\%}, p < 0.001) and fewer rescue carbs per day (0.36 vs. 1.29, p < 0.001). For the control algorithm evaluated, it yielded an average aggressiveness factor of 0.72 for reducing postprandial basal insulin. ALPHA-ICR slightly reduced time in hypoglycemia compared to AP (1.77{\%} vs. 1.92{\%}, p = 0.09). Conclusion: Incorporating adaptive meal dosing into an AP can help reduce postprandial hypoglycemia, and the reduction is primarily due to changes in postprandial insulin delivery rather than pre-meal bolus. Significance: Adapting postprandial insulin can lead to substantial reduction in postprandial hypoglycemia and the adaptive algorithm presented can be used both to tune an algorithm prior to a study and to adapt to individuals during real-time usage.",
keywords = "Adaptive control, Hybrid artificial pancreas, Type 1 diabetes",
author = "Navid Resalat and {El Youssef}, Joseph and Ravi Reddy and Jessica Castle and Peter Jacobs",
year = "2019",
month = "8",
day = "1",
doi = "10.1016/j.jprocont.2019.05.018",
language = "English (US)",
volume = "80",
pages = "247--254",
journal = "Journal of Process Control",
issn = "0959-1524",
publisher = "Elsevier Limited",

}

TY - JOUR

T1 - Adaptive tuning of basal and bolus insulin to reduce postprandial hypoglycemia in a hybrid artificial pancreas

AU - Resalat, Navid

AU - El Youssef, Joseph

AU - Reddy, Ravi

AU - Castle, Jessica

AU - Jacobs, Peter

PY - 2019/8/1

Y1 - 2019/8/1

N2 - Objective: We introduce an adaptive learning algorithm to better adjust postprandial basal and pre-meal bolus insulin for reducing postprandial hypoglycemia in a hybrid artificial pancreas (AP). An AP uses a control algorithm and sensed glucose to automate the delivery of insulin to people with type 1 diabetes (T1D). A hybrid AP requires the person to dose insulin in advance of a meal. Insulin sensitivity is dynamic in people with T1D, making it challenging for an AP to maintain euglycemia. Adaptive approaches to meal dosing can help prevent postprandial hypoglycemia. Methods: An adaptive learning postprandial hypoglycemia-prevention algorithm (ALPHA) is introduced. One implementation of ALPHA adjusts the rate of postprandial insulin (ALPHA-BR) proportionally in response to prior postprandial episodes. This is achieved by an adaptive aggressiveness factor applied to postprandial basal insulin. The second implementation adaptively updates the pre-meal bolus insulin by changing the insulin-to-carbohydrate ratio (ALPHA-ICR), also proportionally in response to prior postprandial hypoglycemia. Both implementations were evaluated within an AP on an in-silico T1D virtual population of 99 subjects with circadian insulin sensitivity variations and 30% errors on meal estimations. Twenty real-world 4-day meal scenarios were given and glycemic outcomes were compared with an AP with no adaptation. Results: Out of the 99 in-silico subjects, 23 of them experienced postprandial hypoglycemia leading to greater than 1% overall time in hypoglycemia. Of these 23 subjects, we evaluated the benefit of using ALPHA-BR and ALPHA-ICR to prevent postprandial hypoglycemia. ALPHA-BR yielded substantially fewer percent time in hypoglycemia compared to AP (0.54% vs 1.92%, p < 0.001) and fewer rescue carbs per day (0.36 vs. 1.29, p < 0.001). For the control algorithm evaluated, it yielded an average aggressiveness factor of 0.72 for reducing postprandial basal insulin. ALPHA-ICR slightly reduced time in hypoglycemia compared to AP (1.77% vs. 1.92%, p = 0.09). Conclusion: Incorporating adaptive meal dosing into an AP can help reduce postprandial hypoglycemia, and the reduction is primarily due to changes in postprandial insulin delivery rather than pre-meal bolus. Significance: Adapting postprandial insulin can lead to substantial reduction in postprandial hypoglycemia and the adaptive algorithm presented can be used both to tune an algorithm prior to a study and to adapt to individuals during real-time usage.

AB - Objective: We introduce an adaptive learning algorithm to better adjust postprandial basal and pre-meal bolus insulin for reducing postprandial hypoglycemia in a hybrid artificial pancreas (AP). An AP uses a control algorithm and sensed glucose to automate the delivery of insulin to people with type 1 diabetes (T1D). A hybrid AP requires the person to dose insulin in advance of a meal. Insulin sensitivity is dynamic in people with T1D, making it challenging for an AP to maintain euglycemia. Adaptive approaches to meal dosing can help prevent postprandial hypoglycemia. Methods: An adaptive learning postprandial hypoglycemia-prevention algorithm (ALPHA) is introduced. One implementation of ALPHA adjusts the rate of postprandial insulin (ALPHA-BR) proportionally in response to prior postprandial episodes. This is achieved by an adaptive aggressiveness factor applied to postprandial basal insulin. The second implementation adaptively updates the pre-meal bolus insulin by changing the insulin-to-carbohydrate ratio (ALPHA-ICR), also proportionally in response to prior postprandial hypoglycemia. Both implementations were evaluated within an AP on an in-silico T1D virtual population of 99 subjects with circadian insulin sensitivity variations and 30% errors on meal estimations. Twenty real-world 4-day meal scenarios were given and glycemic outcomes were compared with an AP with no adaptation. Results: Out of the 99 in-silico subjects, 23 of them experienced postprandial hypoglycemia leading to greater than 1% overall time in hypoglycemia. Of these 23 subjects, we evaluated the benefit of using ALPHA-BR and ALPHA-ICR to prevent postprandial hypoglycemia. ALPHA-BR yielded substantially fewer percent time in hypoglycemia compared to AP (0.54% vs 1.92%, p < 0.001) and fewer rescue carbs per day (0.36 vs. 1.29, p < 0.001). For the control algorithm evaluated, it yielded an average aggressiveness factor of 0.72 for reducing postprandial basal insulin. ALPHA-ICR slightly reduced time in hypoglycemia compared to AP (1.77% vs. 1.92%, p = 0.09). Conclusion: Incorporating adaptive meal dosing into an AP can help reduce postprandial hypoglycemia, and the reduction is primarily due to changes in postprandial insulin delivery rather than pre-meal bolus. Significance: Adapting postprandial insulin can lead to substantial reduction in postprandial hypoglycemia and the adaptive algorithm presented can be used both to tune an algorithm prior to a study and to adapt to individuals during real-time usage.

KW - Adaptive control

KW - Hybrid artificial pancreas

KW - Type 1 diabetes

UR - http://www.scopus.com/inward/record.url?scp=85067623828&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85067623828&partnerID=8YFLogxK

U2 - 10.1016/j.jprocont.2019.05.018

DO - 10.1016/j.jprocont.2019.05.018

M3 - Article

VL - 80

SP - 247

EP - 254

JO - Journal of Process Control

JF - Journal of Process Control

SN - 0959-1524

ER -