Developing an Artificial Pancreas The History & Future of Dose Safety Richard Mauseth, MD!! IMPROVING!!! LIVES.!CURING!!TYPE!1 DIABETES.
Controller Methods 1. Proportional-Integral-Derivative (PID) 2. Model Predictive Control (MPC) 3. Fuzzy Logic (FL)
People with diabetes respond very differently to the same conditions BG (mg/dl) 400 350 300 250 200 150 Exercise Trials Subj1 Subj2 Subj3 Subj4 Subj5 Subj6 Subj7 100 50 0 14:00 16:00 18:00 20:00 22:00 0:00 2:00 4:00 6:00 8:00 10:00 Time
The Fuzzy Logic Process Physicians Knowledge
Different Paths / Different Doses 300 Rising Blood Glucose 300 Falling Blood Glucose 250 250 200 200 150 150 100 Decreasing 100 Decreasing 50 Flat Increasing 50 Flat Increasing 0 11:00 AM 12:00 PM 1:00 PM 2:00 PM 3:00 PM 0 11:00 AM 12:00 PM 1:00 PM 2:00 PM 3:00 PM
Fuzzy Logic Dosing Matrix
The Fuzzy Logic Process in Detail
2002-2003 Boeing technology meeting Fuzzy logic versus others Evergreen Hospital IRB Q 15 minute venous glucoses Entered into computer given dose which was entered 2005 Continuous glucose sensing 2005 APS consortium
Data from 2003-2004 Studies Version 0.9 controller
2008 Aaron Kowalski at JDRF DTM poster Frank Doyle as UCSB In silico testing Ardy Johnson at JDRF
Differing response based on Personalization Factor (PF)
Differing response based on Personalization Factor (PF)
2010 Innovative Grant JDRF funded in bad economic time BRI IRB and FDA IDE approval Four part study Correction of blood sugar, purposely elevated post dinner Blood glucose overnight Blood glucose small and large meals Ten subjects- 24 hours Version 1.5 of controller
Innovative Grant Results 350" 300" "Sensor"BG"Avg" "YSI"BG"Avg" 250" BG#(mg/dL)# 200" 150" 100" 50" Version 1.5 controller 0" 20:00" 22:00" 0:00" 2:00" 4:00" 6:00" 8:00" 10:00" 12:00" 14:00" 16:00" 18:00" 20:00"
Timeline for FDA Funded Study 0 months - idea 4-6 months - write a grant, get institutional approvals for both academics and budget and submit 6-9 months - review a grant 10-12 months - to accept grant 12-15 months - FDA and IRB approval 15-24 months - to do study 24 months - review results 30 months - published results
2012 NIH - JDRF Software development- NIH Stress testing- JDRF
2012-13 Exercise Protocol
Exercise Study Results Version 2.0, 2.1 controller
2012-13 Pizza Protocol
Pizza Study Results Version 2.0, 2.1 controller
2011-2014 NIH Studies Software development Meal detection Sensor noise reduction Blood sugar prediction Remote monitoring Clinical- ad lib living in CRC 40 eight hour studies to compare day to day Version 2.2 of controller
2014 Future Guarded outpatient studies 30 subjects Randomized crossover 68 hours in facility, able to cook, exercise, order take out food, as they would at home Remotely monitored for safety Compare self control versus controller control Version 3.0 of controller
2014 Future - Medtronic AAGC System
2015 Future Outpatient clinical trials New sensors Integration of pump and sensor Remote monitoring of patients Automated warnings Faster insulin Learning Version 4.0 of controller
Our Team
JDRF AP Consortium Sites 18 sites worldwide running clinical trials, providing engineering resources, or doing both: Benaroya Oregon Stanford W. Ontario I.I.T. Rensselaer Montreal Harvard Cambridge Perth Melbourne UCSB/ Sansum Colorado Virginia Yale Montpellier Pavia/Padova Israel Jaeb Center for Health Research, Tampa, FL ConsorJum CoordinaJng Center
Backup Slides
Controller Technology CGM AP Controller Insulin The majority of AP controller teams utilize Proportional Integral Derivative (PID) or Model Predictive Control (MPC) technologies Other teams utilize fuzzy logic (FL) To calculate insulin doses, PID and MPC controllers employ a mathematical model of the system to be controlled, human glucoregulatory system. The models are based on a set of equations. The effectiveness of the dosing is dependent on the fidelity of the model. FL controllers calculate insulin doses based solely on how a clinical expert would interpret the CGM inputs. FL controllers make no assumptions about the system being controlled. The effectiveness of the dosing is dependent on the expertise of the clinicians and how that was codified.