INTRODUCTION Premature, low-birth-weight infants in the neonatal intensive care unit (NICU) can lose blood glucose homeostasis due to immaturity of endogenous regulatory systems and the stress of their condition. Hyperglycemia occurs in 40-80% of premature, low birthweight infants. This condition has been linked to worsened outcomes in preterm neonates, including an increased risk of further complications, such as sepsis, increased ventilator dependence, retinopathy of prematurity, hospital length of stay and mortality Often, glucose restriction is used to control high blood glucose levels. However, this restriction also deprives the neonate of crucial energy required to promote growth. Continuous insulin infusion has thus been proposed as a solution to reduce plasma glucose concentrations and optimize nutrition in these small infants. The pathogenesis of hyperglycemia in critically ill adults and preterm may differ. The mechanisms responsible for hyperglycemia in preterm infants are related to immaturity of the glucose regulatory system, in addition to clinical stress. A model of the fundamental glucose regulatory dynamics in neonates can provide insight about the metabolic state of the patient. In-silico virtual trials were used to design optimal insulin therapy regimes for this vulnerable patient group, and piloted in 24-hour clinical trials. A. Le Compte, A. Lynn, J.G. Chase, G.M. Shaw, C. Pretty, K. Mayntzhusen, P. Docherty, J. Parente Tight Glycemic Control in the Neonatal Intensive Care Unit – Proof of Concept Pilot Trials MODELS AND METHODS A metabolic computer system model, clinically validated in adult intensive care patients and virtual trials using neonatal data, is used to provide tight control in very low birth weights infants. Seven clinical trials up to 24 hours each were performed based on initial blood glucose over 10 mmol/L to initiate insulin. Insulin infusions were modulated to hit a pre-determined target based on measurements every 2-3 hours (max 12 measurements/day). The overall goal was to control blood glucose in a 4-7 mmol/L band. A stochastic model ensured the risk of blood glucose below 4 mmol/L was less than 5% for each intervention. Ethics approval was granted by the Upper South Regional Ethics Committee. ( ) ( ) ( ) body frac G brain body END G I G m t V m CNS m P t P Q Q G S G p G * ) ( * * ) ( 1 . . . , − + + + − − = α & ( ) ( ) ( ) body frac G brain body END G I G m t V m CNS m P t P Q Q G S G p G * ) ( * * ) ( 1 . . . , − + + + − − = α & kI kQ Q + − = & kI kQ Q + − = & ( ) B t u k body frac I ex I I e m V t u I nI I ex I )) ( ( , * ) ( 1 − + + + − = α & ( ) B t u k body frac I ex I I e m V t u I nI I ex I )) ( ( , * ) ( 1 − + + + − = α & Pilot trial cohort clinical details. RESULTS: CLINICAL TRIALS Model-fitted insulin sensitivity during trials. Each line represents the hourly evolution of patient sensitivity to exogenous insulin. Within the relatively small study population, a 2.3x spread of dextrose infusion rates were used, and an 8.9x spread of insulin sensitivity was computed. In response, the controller used a 7.6x spread of median insulin infusion rates. Normoglycemia in a 4-7 mmol/L band was achieved in all cases. Median initial blood glucose was 10.1 mmol/L (Range: 7.4 – 14.4 mmol/L). Over all trials median blood glucose was 6.9 (IQR: 5.6 – 7.9, 90%CI: 4.6 – 11.2) mmol/L over 74 measurements. The minimum blood glucose was 3.8 mmol/L. Brain Other cells Insulin losses (liver, kidneys) Glucose Insulin Liver Blood Glucose Liver Insu l in sensitivity Insulin sensitivity Effective insulin Plasma Insulin Pancreas Brain Other cells Insulin losses (liver, kidneys) Glucose Insulin Liver Blood Glucose Liver Insu l in sensitivity Insulin sensitivity Effective insulin Plasma Insulin Pancreas Clinical BG results for pilot trial patients 2 (top panel) and 3 (bottom panel). Green areas denote stochastic model 5%-95% BG prediction confidence interval. TRIAL START TRIAL START Patient Gestational age at birth [weeks] Age at start of trial [days] Birth weight [g] Insulin usage before trial [hours] A 24.4 7 685 15.3 B 27.3 9 770 15.0 C 25.4 1 720 3.2 D 25.4 7 785 91.6 E 25.9 4 540 2.2 F 27.0 2 900 0.0 G 25.0 6 995 11.5 Controller implementation overview. mg/kg/min kcal.kg.day A 7.1 40.6 0.0 0.025 7.7 1.04 8.7 6.9 2.1 0.45 B 9.4 54.4 5.0 0.040 6.7 1.16 12.3 4.1 1.7 1.25 C 4.1 23.3 0.0 0.068 8.7 1.19 14.4 5.2 1.7 0.14 D 9.6 55.0 3.0 0.052 6.5 1.08 8.0 5.0 2.5 0.7 E 6.8 39.1 11.0 0.116 9.0 1.12 12.6 6.5 2.1 0.2 F 8.2 47.5 4.5 0.069 7.0 1.23 14.5 3.8 1.7 0.5 G 9.4 54.3 2.0 0.191 8.7 1.17 14.7 5.3 1.9 0.14 Patient Min. BG [mmol/L] Measurement period, hours Median insulin sensitivity [L/(mU.min)] Median insulin rate [U/kg/hr] Median dextrose rate Geometric BG mean [mmol/L] Geometric BG StDev [mmol/L] Max. BG [mmol/L] Total EBM [mL] Clinical glycemic variables during trials. Infusion rates are computed as hourly averages. BG mean and standard deviation are computed using lognormal statistics. EBM = Expressed Breast Milk. 0 5 10 15 20 25 0 2 4 6 8 10 12 14 16 18 20 Time [hours] BG [mmol/L] 0 5 10 15 20 25 0 2 4 6 8 10 12 14 16 18 20 Time [hours] BG [mmol/L] 0 5 10 15 20 25 0 2 4 6 8 10 12 14 16 18 20 0 5 10 15 20 25 0 2 4 6 8 10 12 14 16 18 20 Time [hours] BG [mmol/L] 0 5 10 15 20 25 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 x 10 -3 Time [hours] Insulin sensitivitiy [L/(mU.min)] 0 5 10 15 20 25 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 x 10 -3 Time [hours] Insulin sensitivitiy [L/(mU.min)] Per-patient BG concentration during computerised insulin dosing. The shaded region represents the 4-7 mmol/L target band. Stochastic model forecasts drove controller decisions for more insulin resistant and/or dynamic patients, preventing episodes of hypoglycaemia. 0 5 10 15 20 25 30 35 0 5 10 15 BG (mmol/L) Patient G 0 5 10 15 20 25 30 35 0 0.1 0.2 0.3 0.4 U/kg/hr 0 5 10 15 20 25 30 35 0 1 2 x 10 -3 L/[mU.min] Time [hours] Insulin sensitivity Insulin Study start 0 5 10 15 20 25 30 35 0 5 10 15 0 5 10 15 20 25 30 35 0 5 10 15 BG (mmol/L) Patient G 0 5 10 15 20 25 30 35 0 0.1 0.2 0.3 0.4 U/kg/hr 0 5 10 15 20 25 30 35 0 1 2 x 10 -3 L/[mU.min] Time [hours] Insulin sensitivity Insulin Study start •Change in insulin sensitivity contained •Insulin infusion rate adjusted •More aggressive control may have created hypo event for this patient •Controller tolerated period of higher BG to balance risk of changes in patient condition. •Change in insulin sensitivity contained •Insulin infusion rate adjusted •More aggressive control may have created hypo event for this patient •Controller tolerated period of higher BG to balance risk of changes in patient condition. Overall cohort BG prediction error was 7.6% (0.54 mmol/L) during clinical trials. Distribution of BG prediction errors revealed 69% and 84% of BG measurements within ±10% and ±20% of forecasted concentration. [%] [mmol/L] A 19.20% 1.47 33% 83% B 8.60% 0.52 31% 85% C 7.60% 0.77 80% 90% D 8.40% 0.53 33% 100% E 6.40% 0.48 100% 100% F 8.50% 0.72 50% 83% G 5.90% 0.44 92% 100% Whole cohort 7.60% 0.54 62% 92% Patient Median BG prediction error (absolute) BG within IQR forecast range BG within 5%-95% forecast range BG prediction accuracy and stochastic model prediction coverage. 0 0.5 1 1.5 2 2.5 0 2 4 6 8 10 12 14 BG prediction error [mmol/L] # predictions 0 10 20 30 40 0 2 4 6 8 10 12 14 BG prediction error [%] # predictions 0 0.5 1 1.5 2 2.5 0 2 4 6 8 10 12 14 BG prediction error [mmol/L] # predictions 0 10 20 30 40 0 2 4 6 8 10 12 14 0 0.5 1 1.5 2 2.5 0 2 4 6 8 10 12 14 BG prediction error [mmol/L] # predictions 0 10 20 30 40 0 2 4 6 8 10 12 14 BG prediction error [%] # predictions BG prediction errors during trials (expressed as absolute concentration in left panel, and percentage of measured BG concentration in right panel). Comparison with simulated trial results indicated level of control matched predictions, despite significantly insulin-resistant clinical cohort. Distribution of insulin sensitivity between clinical trial patients (n=7) and matched 24- hour simulated trials on retrospective cohort (n=7). Distribution of BG between clinical trial patients (n=7) and matched 24-hour simulated trials on retrospective cohort (n=7). brought to you by CORE View metadata, citation and similar papers at core.ac.uk provided by UC Research Repository