Moving Toward Artificial Pancreas -...
Transcript of Moving Toward Artificial Pancreas -...
Moving Toward Artificial Pancreas
Moshe PhillipInstitute for Endocrinology and DiabetesNational Center for Childhood DiabetesDiabetes Technology CenterSchneider Children’s Medical Center of Israel
AACE May 2017Texas, USA
Disclosure Information
• Grants/research support: Medtronic, Novo Nordisk, Roche, Eli Lilly, Merck, Sanofi,
Bristol Myers-Squibb, Kamada, Lexicon and Pfizer.
• Honoraria or consultation fees: Sanofi, Medtronic, Novo Nordisk, Eli Lilly and
Pfizer. Advisory board : Sanofi, Medtronic, AstraZeneca, Eli Lilly and Pfizer.
• Stock shareholder: NG Solutions Ltd., DreaMed-Diabetes Ltd., Nutriteen
Professionals Ltd.
model predictive control (MPC)proportional-integralderivative control (PID)
fuzzy logic control (FL)
Steil GM, Rebrin K, Darwin C, Hariri F, Saad MF. Diabetes 55:3344-50, 2006
Cobelli C et al. Diabetes 2011;60:2672-82
Atlas E, Nimri R, Miller S, Grunberg EA, Phillip M.Diabetes Care. 2010;33:1072-6
Controller Design
Fuzzy Logic Applications
Medicine
• Digital image processing as a diagnostic tool
• Evaluation of cardiac functions, ECG analysis
• Warning system in Intensive care unit
Other
• Air conditioning, washing machines ,mono-rails, elevators
• Video game engines
• Special effects
What is Fuzzy –Logic ?
The Fuzzy –Logic Washing Machine
The degree of dirt
The type of dirt
The color
The required duration of washing time:
The weight
very dirty
If…..
oily
white
heavy
&
&
&
Then….
2 hours
Current BG is: ______ 𝒎𝒈/𝒅𝒍180
If:
AND
Past trend is: ______ 𝒎𝒈/𝒅𝒍
𝒎𝒊𝒏𝒖𝒕𝒆
-1.5
Future BG is: _____𝒎𝒈/𝒅𝒍
𝒎𝒊𝒏𝒖𝒕𝒆?
AND
Then
Goal: to reach a BG level of 90-110mg/dl
The Fuzzy-Logic Controller- The Physician’s Perspective
• Deal with approximate reasoning rather than precise
• Simple way to arrive at a definite conclusion based upon vague, ambiguous and imprecise data
• Aim to solve problems by mimicking how a physician would make decisions, only much faster
• Making it easier to implement a non linear system then in conventional control
• Tests show superiority of Fuzzy Logic controller over conventional controllers handling disturbances (meal, stress)
• It has the ability to learn and adapt
Fuzzy –Logic Approach & Decision Making
Emulate the line of Reasoning of Diabetes Caregivers
Grant P, Medical Engineering & Physics, 2007Pagliaro L, Intern Emerg Med, 2007
IF the Queen covers less squares than does the opponents AND the opponents Queen is closer to the center of the board than mine THEN capture opponents Queen using same or lesser-valued piece, if possible
May, 1997
Future NIH pivotal trials
Advances in Artificial Pancreas
BostonCambridgeVirginiaIsrael& IDC
February 7th, 2017
Cambridge Group - MPC Artificial Pancreas
Day and night closed-loop in young people with type 1 diabetes
Closed loop platform (FlorenceM system)
Advances in Artificial Pancreas
Main Protocol: N=240 , 6-month RCT, ratio 2:1 to Closed-Loop Control vs. Sensor-augmented pump therapy, 8 sites US & 3 EU
Major Eligibility Criteria: Type 1 diabetes > 1 y, Use of pump > 6 mon, Age ≥14 y
Outcomes:Primary outcome: Reduction of time below 70 mg/dL & non-inferiority for time above 180 mg/dL
Secondary Outcomes: HbA1c, technology acceptance
The System – a wireless mobile AP:Sensor: G5, DexcomInsulin Pump: t:AP ,Tandem or Accu-Chek Spirit Combo ,Roche
inControl AP residing on a smart phone and inControl Cloud remote monitoring & alert
The International Diabetes Closed-Loop (iDCL) Trial
Virginia Group - MPC Artificial Pancreas
Advances in Artificial Pancreas
Boston Group - MPC Artificial Pancreas
Late 2017/Early 2018
Mid 2018
Advances in Artificial Pancreas
Advanced HCL – MD-Logic & Medtronic AP
A Crossover Study Comparing Two Automated Insulin Delivery System Algorithms (PID vs. PID + Fuzzy Logic) in Adolescents & Young Adults with T1D
7 Clinical Sites: 4 US (IDC-
Minneapolis, Yale - New Haven,
Joslin- Boston, U of FL- Gainesville)
& 3 Europe (Schneider Children’s -
Israel, Hannover - Germany,
Ljubljana – Slovenia)
Primary Outcome: The advanced HCL (PID +Fuzzy Logic algorithm) will significantly reduce the time spent >180mg/dL during the day (7AM - 11PM) compared to a 670G system using a PID algorithm
Advances in Artificial Pancreas
Closed-Loop Research Sites (Academic)
Dual Hormone Single Hormone
Boston
Medtronic Received FDA Approval For the World’s First Hybrid Closed-Loop System For People With Type 1 Diabetes
DreaMed Received CE Marking For the World’s First Closed-Loop Algorithm For People With Type 1 Diabetes
February 2015
Road to Closed Loop productsPATH TO THE ARTIFICIAL PANCREAS
Pat
h t
o C
lose
d L
oo
p
Threshold SuspendMiniMed® 630G
Predictive SuspendMiniMed® 640G
Hybrid Closed Loop
Towards PersonalizedClosed Loop
Automatically doses insulin with minimal mealtime and exercise inputs from the patient
Suspends delivery when the system predicts a low is approachingSuspends delivery
when a low occurs
Improving interface & meal announcement: small, medium, large meal bolus settings and set meal insulin delivery buttons
Pattern recognition
Additional sensor inputs: Activity, food, heart rate, sleep, free fatty acids
Detecting sensor or infusion set failure
Advanced HybridClosed Loop
Combining PID with Fuzzy in collaboration with DreaMed
PID for insulin delivery & MPC for safety
Suspension protocol based on actual values
Suspension protocol w/ predictive algorithm
*Investigational only. Not approved and not commercially available
jamanetwork.com
Available at jama.com and on The JAMA Network Reader at mobile.jamanetwork.com
Bergenstal RM, Garg S, Weinzimer SA, et al.
Safety of a Hybrid Closed-Loop Insulin Delivery System in Patients With Type 1 Diabetes
Published online September 15, 2016
Results
MODAL DAY SENSOR GLUCOSE (SG) TRACINGS
Median and IQR of SG values throughout day and night.
Gray band and dotted line: run-in phase.
Pink band and solid line: study phase.
All Subjects
Adolescents
Adults
Bergenstal et al. JAMA 316:1407, 2016., EASD 2016, Oral
Noon MidnightMidnight
The future of diabetes management: Automated therapyFOR OPTIMIZED GLYCEMIC CONTROL
SAFETY SHIELD
* Investigational Device
2006
2007
2008
2009
2010
2011
2012
From the Idea to Regulatory Approved Closed-loop System
2003 - The Motivation & Idea
2006 - The DTC Center
2008 – The MD-Logic Controller
2009 - The DREAM Consortium
Diabetes wiREless Artificial Pancreas ConsortiuM
The DREAM way
DREAM 1 DREAM 3Camp
DREAM 44 Nights
DREAM2
Join Study
DREAM 46-14 weeks
DREAM 5Weekend
DREAM 52 weeks
Preclinical Swine Studies
Day & Night
Outpatient-Free livingPre-Clinical Feasibility & Inpatient Transitional
3 Days 2 weeks14 weeks4 nights 6 weeks1 nights
Learning Camp
1Atlas E et al. Diabetes Care, 20102Miller S et al. D. Technol Ther, 2011
3Nimri R et al. D. Technol Ther , 20124Nimri R et al. Pediatr Diabetes, 2013
7Nimri R et al. Diabetes Care, 20148Nimri R et al. Diabetes Obesity & Metab, 2016
2006 2008 2009 2010 2011 2012 2013 2014 2015 2016
Remote Monitoring
Evening & OvernightDay & Night
5Phillip M et al. NEJM, 20136Nimri R et al. Pediatric Diabetes, 2014
2011
2012
2013
2014
2015
2016
2017
First Outpatient Study at Diabetes Camp
Israel: 9-11 Oct 2011
Slovenia: 26-28 Nov 2011
Germany: 28-30 Jan 2012
Results: Glucose Control Over TimeG
luco
se [
mg
/dl]
Time [hh:mm]
MD-Logic Nights Control Nights
P<0.0001
IQR CGMMedian CGM
Median [IQR] Capillary Glucose
MD-Logic Overnight Studies
Site Duration No. Sign
*Inpatient 1 night 12
*Camp 1 night 56
*Home 4 nights 75
Home 6 weeks 22
*Home 3 days^ 47
Mean
• Age 10-55y• A1C 7-9.7%• *Multicenter• ^Only night
Studies Features
ITT (intension To Treat) analysis*Significant overall p-Value(using the Comprehensive meta analysis software)
.-80 -60 -40 -20 0 20
.
Favor Close-loop
Favor SAP
-20 0 20 40 60
Average BG * % within 70-180 mg/dl *
Favor SAP
Favor Closed-loop
.
-10 -8 -6 -4 -2 0 2 4Favor
Close-loopFavor
SAP
.
-60 -40 -20 0 20
.
Favor Close-loop
Favor SAP
Mean difference
% Below 60 mg/dl % Above 180 mg/dl *
closed-loopclosed-loop SAP SAP
MD-Logic Pooled Analysis of Free Living Overnight Studies
# of Nights
N=1033 N=976
Mean Overnight Glucose Levels
[mg/dl]
Histogram of Mean Overnight Glucose Levels, PP
139 (118,161)
1 Nimri R et al, Pediatric Diabetes 15: 20142 Nimri R et al, Diabetes Care 37: 20143 Nimri R et al, Diabetes Obesity & Metab 2016Unpublished data 3 months study
SAPCL
152 (120,190)
Factors predicting Closed-Loop Success
Rs -A Spearman's rank-order correlation was run to determine the relationship between 84 students' hA1C and percentage of readings between 70-140 mg/dl under CL treatment.PM - The Mann-Whitney U test was used to compare differences in time within 70-140 between male and female.
% of readings within 70-180 [mg/dl]
% of readings below 60 [mg/dl]
A1c (%) Age (years) Gender
6 7 8 9 10
20
40
60
80
100
20
40
60
80
100
20
40
60
80
100
010 20 30 40 50 60
6 7 8 9 10 10 20 30 40 50 600
1
2
3
4
0
1
2
3
4
0
1
2
3
4
Male Female
Male Female
2015
2016
2017
2018
2019
2020
2021
Glucositter CE-mark
Solid clinical evidence for safety & efficacy of the closed-loop system to become an integral part of diabetes management
Establishing
Quality system & Risk Management assurance (ISO 14971)Continuous maintenance
Hyperglycemia is Risky
The Challenges The MD-Logic Features
Enhanced Bolus Approach• Basal Bolus• Control to range & control to target• Interplay of basal & bolus• Continuous Corrections
Event Driven• Correction vs. Meal correction bolus is
feasible• Two engine of fuzzy logic
Learning & Adaptation• Personalized
Safety Features• Insulin on board• Safety layers
for insulin delivery
System “disturbances”• Post-Prandial glucose excursions
• Main factor to determine A1c• Limitation of current insulin • Announced meal “Errors”• Unannounced meal
• Physical Activity, Stress..• Unannounced exercise
Within & between days variability
“unexpected glucose responses”Home 24/7 challenges
Exercise
Postprandial Glucose Excursion
Basal - Bolus Approach
Duration No. Sign
4 nights 75
6 weeks 22
3 days^ 47
Mean
• Age 10-55y• A1C 7-9.7%• *Multicenter• ^Only night
• ITT (intension To Treat) analysis• *Significant overall p-Value
(using the Comprehensive meta analysis software)
Home Studies Features
.
Basal Difference Bolus Difference*
-5 -4 -3 -2 -1 0 1 2 3 4 5Favor SAP
Favor Close-loop
Favor SAP
Favor Close-loop
-5 -4 -3 -2 -1 0 1 2 3 4 5
. .
PMD-LogicControl
N.S10.5 ± 4.510.7 ± 5.74 night (N=75)
N.S10.1 ± 4.09.4 ± 3.36 weeks overnight (N=22)
N.S10.2± 3.311± 3.03 days, the overnight (N=47)
Total Insulin Dose [Units] 23:00 – 07:00 (Average ± STD)
Insulin
Glucose
Insulin absorption
Insulin action
Insulin Kinetics
Amount
Glycemic Index
Fat
Insulin-to-Carb ratio
Meal Content
Glucose level & trend
Insulin sensitivity
Insulin on board
Meal Time
Glucometer
Sensor interstitium/ plasma
Sensing Errors
Event Driven Treatment for Closed Loop Meal Challenge
Postprandial Glucose Excursion
*Atlas E et al, Diabetes Care. 2010 ;33:1072-6
Example of MD-Logic Full Closed-Loop Control
Glu
cose
(m
g/d
l)In
sulin
(U)
Time
Time
Meal
Detection(~20 min) *
Insulin
Absorption & Action(~60-100 min)
Default Basal
Legend:
CGM
YSI
Basal
Bolus
Sensor Delay
(~7 min)
Event Driven Treatment Makes the Difference
Meal Bolus
Correction Bolus
Hour of Day
No Bolus
Basal Λ
Bolus
SensorΛ
YSI Λ
Manual Meal Bolus
Λ Median [IQR]
70% BolusReduced – Bolus 70%138.2±39.4 mg/dl*
Glucose(mg/dl)
Normal Bolus130.5±35 mg/dl*
Unannounced175.4±42.7 mg/dl*
Over – Bolus 120%110.6±32.7 mg/dl*
*Mean ± SD
Glucose(mg/dl)
Basal(U/h)
Bolus(U)
Basal(U/h)
Bolus(U)
Event Driven Treatment Makes the Difference
N= 10
Learning & Adaptation - Controller Effort
Example 3
Example 1
Example 2
Data from 3 months overnight study
% of insulin dosing relative to open loop
15y, start A1c 9% (74.8 mmol/mol)
24y, start A1c 7.4% (57.4 mmol/mol)
45y, start A1c 7.1% (54.1 mmol/mol)
73 (44,89)% *
67 (33,95)% *
65 (43,87)% *
Since the beginning of the study (evening & night) *Time within range 70-180 mg/dl
0
10
20
30
40
50
60
70
21/5/13 26/5/13 31/5/13 5/6/13 10/6/13 15/6/13 20/6/13 25/6/13 30/6/13 5/7/13 10/7/13 15/7/13
D.A
L.L
S.T
S.S
Y.Y
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
Weeks of Study
Modified CorrectionFactor
Learning & Adaptation Data from the 6 weeks study
Closed Loop Challenges- Exercise
Inpatient Study
The PHYSI-DREAM study
Camp study
Home studies
Overnight glucose control during and after physical activity with
closed-loop system -Gluco Sitter TM in youth with type 1 diabetes
The PHYSI-DREAM study
Two-arms, crossover, open-label, randomized, in-hospital study Single site –Slovenia - Klemen Dovč. MD & Prof. Tadej Battelino. MD
Closed Loop Challenges- Exercise
• Primary Endpoint : % of readings < 60 mg/dl during and following afternoon exercise (till 7:00 AM next day)
• Secondary Endpoints: % of readings < 70, >250 and
within 70-180 mg/dl
Inclusion Criteria N=20 , Mean ± SD
9 females
Age: 10 - 17 years 14.2±2 years
T1D > 1 year 8.3±3.2 years
CSII > 3 months 7.4±3.2 years
BMI 5° - 95°p.le 21.10±2.78 kg/m2
HbA1c < 9%(75 mmol/mol) 7.7±0.6 % (60.0±6.8 mmol/mol)
Insulin dose - 0.8±0.2 units/kg/day
VO2max 43.3±9.3 ml/kg/minPulse 186.6 ±10.2 /min
Study Population
The PHYSI-DREAM Study
Over-the Weekend (60h)
Day & NightMD-Logic Control
N=47* Median (IQR)
Age (years) 16.1(13.2,18.5)
Gender (M/F) 18/29
Body Mass Index -SD Score 0.2 (-0.2,0.8)
A1c % (mmol/l) 7.6 (7,8.1), 59.6 (53,65)
Diabetes duration (years) 9.4 (5,12.7)
Pump therapy duration (years) 5.4 (3.1,9.4)
Daily insulin dose (units/kg) 0.8 (0.7, 0.9)
Study PopulationTwo Sites – Germany & Israel
* including Pilot study
Primary Endpoint • Time within 70-180 mg/dl
Main Inclusion
• Type 1 Diabetes > 1y• 12 – 65 years old• 6.5 ≤ HbA1c ≤ 10%• Pump > 3 months• CGM use
Randomized, cross-over MD-Logic vs. SAP
Day & Night Real-Life Challenge
Over the Weekend MD-Logic Study ITT Results [60h]
0
20
40
60
80
0
20
40
60
80
P=N.S P=0.02 P=0.02
% Within 70-180 mg/dl % > 180 mg/dl% < 60 mg/dl
69.4
51.9
0.50.4
0
40
80
120
160
200
144
Glucose Levels [mg/dl]
SAP MD-Logic
• ITT analysis, (N=47) • Median (IQR)
% o
f Ti
me
163
0
20
40
60
80
25.3
42.2
P*ControlMD-LogicN=47 (Median & IQR)
0.13109.6 (96,129)118.4 (99.8, 152.7)Total Insulin Dose [Units]
0.000250.4 (44.5,68.8)44.5(36.3,68)Total Basal Insulin [Units]
0.00158 (44.2,69.4)69.2 (60,90)Total Bolus Dose [Units]
Day & Night MD-Logic Control
P=0.03
0 1 2 3 4 5
% of Readings
P<0.01
P=0.02
P<0.01
MD-Logic SAP
23:00-07:00
07:00-23:00
0 20 40 60 80 100
Within 70-180 [mg/dl]
NS
NS
NS
0 20 40 60
Above 180 [mg/dl]
P<0.01
P=0.02
P<0.01
Day & Night MD-Logic Control - PP results
Below 60 [mg/dl]
N=29
2016
2017
2018
2019
2020
2021
2022
Advanced Closed Loop with Automated Correction Bolus
CampJune 2016
Inpatient 36 HDec 2016
The Next Step on the Closed Loop Path
Feasibility study to evaluate Hybrid-Logic Closed Loop System
Time In Day
Glu
cose
Val
ue
[mg/
dl]
MD- Logic Boluses [Units]
Basal PID Rate [Units/Hour]
Unannounced Lunch - 95 gr Carb Meal
14 years, A1c 7.8%
Insu
lin
FDA Approved
What's Next ?
Goal : commercial availability in 2020
NIH funded study
• cross-over study
• 124 patients from 7 centers (USA&EU)
• 670 G vs. 690 G