Active Insulin Infusion Control of the Blood Glucose Derivative J G Chase, Z-H Lam, J-Y Lee and K-S Hwang University of Canterbury Dept of Mechanical Engineering Christchurch New Zealand ICARCV 2002, Singapore
Silicon + Biology? Many biological and/or medical processes are effectively feedback control systems or can have their function replaced by feedback systems New technology creating new possibilities: –“BioMEMS” such as “wet sensors” and “gene chips”, are opening the path to real time physiological monitoring/sensing and actuation. –Wireless technology (LAN and PAN) for communication between active elements and/or monitoring technology (increased information flow). –Advanced embedded computers (DSP’s) and real time operating systems (RTOS’s) can now handle extensive calculations and operations required. Converging technologies enables the ability to monitor, control and/or replace dysfunctional physiological behaviour(s). Can the increased information from real-time sensing coupled with feedback control outperform the foreknowledge and intuition of an experienced diabetic??
3 Elements of Control Systems Sensing –Real-time sensing from GlucoWatch or similar technology at BW = 20 minutes or greater Computation –Modern embedded DSP’s are far more than adequate Actuation –Insulin pump All are existing and near-term technologies Must account for limitations of existing tech and determine the limits where practicality and feasibility occur together.
Diabetes Current Treatment = Manual Monitoring + Injection = Error Prone Diabetes is reaching epidemic proportions, treatment is dependent on unreliable individuals and has not changed significantly in 30+ years Ideal curve is flat! 2-3 hoursBack to Fasting Level Normal Time Blood Glucose Level over Basal Type I IGT, Type II GOALS : Automate the “5:95” (1 day every 3 weeks is “bad”) Account for variations in patient response, insulin employed, sensor bandwidth and actuator dynamics/limits.
System Model System model is constructed in MATLAB/Simulink. Three parts: one part for each equation in model and controller. Controller: input - G and dG/dt (GlucoWatch) and output u(t) (Insulin Pump)
Controllers Relative proportional controller (RPC). PD controller – controls slopes of incresing/decreasing blood sugar level rather than actual glucose concentration Two controllers, one proportional based and the other derivative weighted where Kp << Kd create two different approaches to control Shape control or Magnitude control
Why “Heavy Derivative” Control? Slope (derivative) is negative Glucose level is still positive. The slope is the highest here – Derivative control most active here Peak glucose level – Proportional control is most active here Derivative control - negative slope prevents further insulin injection when the glucose level is dropping and faster reaction to positive surge.
Control of Glucose Tolerance Test As sampling rate increases, the more effective the controllers become. Optimal control: G is very nearly flat as desired RPC – BW = 20 min
Insulin Infusion Rates for GTT PD controller minics what a diabetic would usally do, a routine optimised over 70 years of clinical treatment. Insulin rates are sharper and nearer injections as sensor BW drops. RPC – sensor BW = 20 min
A More Difficult Test 1000 calories in 4 hours over five “meal” inputs of glucose which is rapidly absorbed Inputs vary in magnitude from 50 – 400 calories Inputs occur in two groups of rapid succession at t = 0, 10, 30 minutes and at t = 210 and 300 minutes –The last meal is 40 calories from 980 – 1020 calories so the full absorption of about 1000 calories occurs by 4 hours quite easily. Controller has no knowledge of glucose input except in optimal case –Input knowledge is not currently practicable in any way for this system in general The goal is to “hammer” the system and see if it breaks!
Control of Glucose Inputs Glucose excursions shrink with sensor BW Optimal control very nearly flat as desired Simple PD control emphasizes derivative over proportional inputs by 100
Normal and Diabetic Glucose Response Response of a normal subject to Glucose Input (orange) PD controller developed is slightly better than normal subject by 7-25% on peak value and 1+ hour in return to basal glucose level
Insulin Infusion Rates for Glucose Inputs Insulin rates are sharper and nearer injections expected as sensor BW drops Lower insulin rates less effective control as might be expected.
u(t)=Uo(1+Kp(G/G b )) Relative Proportional Control Comparison Relative proportional control more robust to Hypoglycemic behaviour
PD Controller against Sensor Lag GlucoWatch™ (glucose sensor) has 20 minute sensor lag PD Controller ROBUST against 20 minute sensor lag The peak is slightly increased, but less hypoglycemic response (RPC)
PD Controller against Sensor Failure Sampling bandwidth = 20 minutes PD controller ROBUST against 20 minute failure Hypoglycemia induced for 60 minute failure
Summary & Conclusions Bergman equations found to very suitable for control systems approach Feasibility of automated insulin infusion is shown in simulation Basic tradeoffs between sensor BW and control efficacy delineated Derivative control or “control of slopes” seen to be the most effective form of feedback so far versus proporational dominated or relative proportional. Insulin inputs with derivative control trending towards matching those of “optimized” insulin injection regimes followed by diabetics.
Ongoing Future Work = First Known Trials GlucoCard error = 7% Kidney Failure Dialysis Machine 67 year old Female High fluid levels 3 rd day in ICU Hyper-insulinemic and Hyper-glycemic