Don’t drive blind: Think smart Future directions in medicine 15TH May 2006 Dr Geoff Shaw Dept of Intensive Care Christchurch Hospital Clin Sen Lecturer Dept of Medicine CSM&HS University of Otago, NZ Senior Fellow Dept of Mechanical Engineering, University of Canterbury, NZ Doctors aren’t so smart Most treatments in medicine are delivered with ‘guess’ work Sometimes you get lucky, other times …. Doctors aren’t so smart, and patients aren’t so lucky WHY?...... Students are not rewarded for their critical thinking; good students are those that can reel off lists and differential diagnoses Most doctors had their mathematical skills innovative ideas and beaten out of them at medical school Only use physics /math skills of year 8 students!! Medical curriculum = do’s & don’ts …1st law in healthcare…… …2nd law in healthcare… How can I get out of this jam? Getting smart… Agitation-sedation control Blood glucose control Dynamic models of renal failure Lung disease models for improved ventilation Cardiovascular models for diagnosis and treatment Agitation control Managing severe agitation is a nightmare It’s “in your face” It’s distressing There’s no escape! Therefore we “nail” the patient as quickly as possible….. A typical patient in our ICU…. “Hey! I think he just moved! Add one more!” ‘Nightmare on ICU Street’ Re-enactment of typical behaviour Note level of force needed to restrain the ‘patient’ The impact…. Kress et al (N Engl J Med 2000;342:1471-7.) Daily Interruption Of Sedative Infusions In Critically Ill Patients Undergoing Mechanical Ventilation Duration of mechanical ventilation: 4.9 days vs 7.3 days Median length of stay: 6.4 days vs 9.9 days The impact…. Brattebo et al (BMJ 2002;324:1386–9) Effect of a scoring system and protocol for sedation on duration of patients' need for ventilator support in a surgical intensive care unit Duration of mechanical ventilation: 5.3 days vs 7.4 days Median length of stay: 8.3 days vs 9.3 days Just imagine…. If poor sedation management contributes to 10% extension in ICU length of stay: In Aust /NZ….$50-100 million is wasted and 6-7000 patients denied intensive care annually In the US…$1 B US wasted… ? …100,000 patients denied intensive care..? Yearly cost in US of sedatives & analgesics in ICU is $0.8-1.2B US [Fraser and Riker, 2001] Indeterminate number of patients are harmed or die ‘State of the art’; the problem of ‘control’ Each nurse is different Inconsistent Nurse Patient Infusion Pump Control changes with nurse Agitation and sedation assessments are subjective…. Modified Riker Sedation-Agitation Scale (SAS) SEDATION/AGITATION SCORE DESCRIPTION 3 Dangerous agitation Pulls at endotracheal tube (ET), Tries to remove catheters, Climbs over bedrail, Strikes staff, Thrashes from side to side 2 Very agitated Does not calm, despite frequent verbal reminding of limits, Requires physical constraints, Bites ET tube 1 Agitated Anxious or mildly agitated, Attempts to sit up, Calms down to verbal instructions 0 Calm and cooperative Calms, awakens easily, Follows commands -1 Sedated Difficult to arouse, Awakens to verbal stimuli or gentle shaking but drifts off again, Follows simple commands -2 Very sedated Arouses to physical stimuli but does not communicate or follow commands, May move spontaneously -3 Unarousable Minimal or no response to noxious stimuli, Does not communicate or follow commands Sedation and agitation are linked – what if you are difficult to arouse and agitated? Descriptions are vague and can easily satisfy more than one Summary of ‘State of the Art’ Many ICU’s don’t even do “over-customised care and over-sedation this! [Wiener-Kronish, 2001] Longer than necessary The result?? … stay…in hospitals for many many days … and many many $ Higher costs Reduced level of patient … primarily due to a lack of a consistent care measure of agitation Agitation Sedation Aims To investigate a semi-automated method of sedation that… Allows rapid titration to measurements of agitation Minimises over sedation Has a simple user interface Has minimal set of rules Is acceptable by nursing staff Stores data for clinical records and audit -1 -1 -1 -1 -1 -1 -1 -1 +2 0 +3 +1 2 2 2 2 2 1+1+1 8 1+1+1+1 1 1 8 16 Development Computerised paper sedation sheets: n 1 1 y [ y i xi ] n 6 i n 4 yi = Background rate xi = Bolus doses given for agitation n = Number of hours 62 y female with multi-organ failure 17 y male with multi-trauma 30 nurses' ratings of the Infuse-Rite Before using Infuse-Rite After using Infuse-Rite Highly acceptable 10 9 8 7 6 5 4 3 2 Totally unacceptable 1 Pain & Agitation Ease of Personal discomfort control Admin safety control Sedation Time Patient Legal control efficiency safety safety Sedation consumption & cost Comparison of Christchurch vs Wellington sedation cost from July 1, 2002 – Nov 30, 2003 Overall both units are estimated to have a similar case mix, with slightly higher proportion of cardiac surgical patients in Wellington, 40% vs 30% The patients… 45.4 1419 1486 Christchurch Wellington Patients 50.2 Christchurch Wellington Vent hours/ patient Patient sedation dose 2413 145.4 69.0 496 Christchurch Wellington Midazolam (mg) Christchurch Wellington Propofol (mg) Hourly dose 3.0 47.7 1.6 11.9 Christchurch Wellington Midazolam (mg) /vent-h Christchurch Wellington Propofol (mg) /vent-h Midazolam & Propofol $113,175 $32,205 Christchurch ($) / year $107.89 $32.15 Wellington Christchurch Wellington ($) /patient over 3.5 stays Administration data from Infuse-Rite 3 years data from 1070 patients (Mar ‘02- Oct 05): 4,121,640 minutes of recorded data (7y 10m) total use 120,141 boluses / infusion rate changes ~36,000 additional boluses for procedures >156,000 doses / infusion changes without error Insights into agitation control… We hypothesize that loss of agitation control is a key driver of over sedation.. Period change Morphine (mg) First compute the mean, the maximum and minimum of the peak to trough time period t1 t 2 t 3 t 4 t 5 t 6 6 tmax t1, tmin t 3 t A measure of stability is then given by: Time (days) max{tmax t , t tmin} Hurst Exponent Measure of surface smoothness. 0<H<0.5 suggests a predictable oscillatory pattern of sedation given for agitation. An increase in H suggests patient is less responsive to sedation; extra doses are required and the system becomes more chaotic Loss of control Summary…. Minimised over sedation and eliminated drug error Better consistency and control of agitation. High degree of acceptance by nursing staff Sedation cost by avoidance of expensive agents Accurate medico-legal record for patient file Data storage for research and audit The Future…? Pharmacokinetics Concentration of drug in Central compartment Models kinetics of drug infusion and distribution Models transportation to effect site, and drug elimination U Cc K1Cc Vd C p K3C p K2Cc Concentration of drug in Peripheral compartment Standard twocompartment Pharmacokinetics Infusion rate Volume of distribution Agitation Dynamics Rate of change of agitation depends upon relative magnitude of stimulus compared to cumulative effect of Stimulus sedation Cumulative effect of sedation t A w1S w2 KT C p ( )e KT ( t ) 0 Agitation Index Drug Concentration in Peripheral compartment d Patient+Nurse Simulator Cc K1Cc U Vd C p K3C p K2Cc U K p A Kd A t A S C p ( )e K4 ( t ) d 0 Filters variability of nurse input Records infusion data What Data? Recorded Data = Standard & Procedural Boluses+ Background rate Modelled vs recorded data Recorded Simulated Improvements….. Morphine Vc Midazolam dCc ( K CL K ce K cp )Cc PAU K ec Ce K pc C p dt Vp Ve dC p dt Vc dCc ( K CL K ce )Cc PBU K ec Ce dt Ve dCe K ec Ce K ce Cc dt K pcC p K cp Cc dCe K ec Ce K ce Cc dt Pharmacokinetics ( ) UA C e, morph C50, morph EComb U A UB U 50 ( ) E0 [ Emax ( ) E0 ] ( ) U A UB 1 U 50 ( ) UB C e, midaz C50, midaz t dA w1S w2 KT EComb ( )e KT ( t )d dt 0 Pharmacodynamics Novel application of standard PK & PD Novel Equation & Application Morphine Midazolam 3-Compartment Model 2-Compartment Model Excretion Infusion site Infusion site Blood flow Blood flow Peripheral Pharmacokinetics Brain Effect site Brain • Response surface modeling [Minto et al, 2000] • Dual sigmoid • Incorporates effect saturation EComb • Captures synergism • Non-linear representation of the concentration-effect relationship • Models the combined sedative effect of the drugs on the brain Morphine Effect-site Concentration Midazolam Effect-site Concentration Pharmacodynamics Primary Equation Cumulative effect of current and prior sedation Stimulus t dA KT ( t ) w1S w2 KT EComb ( )e d dt 0 Agitation Index Combined drug effect Rate of change of agitation depends upon relative magnitude of stimulus compared to cumulative effect of sedation Rudge, AD, Chase, JG, Shaw, GM and Lee, DS (2006). “Physiological Modelling of AgitationSedation Dynamics Including Endogenous Agitation Reduction,” Medical Engineering and Physics, In Press Quantification of Agitation t U K p A Kd A dA w1S w2 KT EComb ( )e KT ( t )d dt 0 How can we measure agitation? Use a wide variety of medical signals: Heart rate Respiratory rate BP (systolic and diastolic) Heart rate variability (HRV) BP Variability (BPV) Patient Motion (MOV via webcam) Use fuzzy mathematics to add in medical expertise and organise a large amount of data into something useful HRV Frequency Components VLF: 0.0033 – 0.04 Hz LF: 0.04 – 0.15 Hz HF: 0.15 – 0.4 Hz VLF/HF falls as variability drops in agitated patients BPV HF/VLF falls as systolic BP variability rises in agitated patients ECG – HRV : Healthy Subjects ~20 min color word test CWT designed to induce sympathetic responses similar to agitated ICU patients Red - the correct answer is “green” VLF/HF ratio measured: Under stress (agitation) HR rises; the heart beats more consistently (HF drops). Therefore, rises in VLF/HF ratio indicate sympathetic responses akin to agitation. ECG – HRV : ICU Patients SAS = 1 SAS = 2 VLF/HF ratio measured: 100 minute sample shows two assessed times of significant agitation matched by two peaks in the HRV VLF/HF ratio BPV for an ICU patient SAS HF/VLF ratio measured: Identifies agitation which correlates with nurse-assessed Riker SAS (inverse to heart rate variability) Quantifying Agitation – Fuzzy Systems Fuzzy logic is good for modeling uncertain system dynamics Rules intuitive & based on medical experience Moving mean tracking over 1-20 minutes (can be tuned) Rules tuned on healthy and then applied to ICU Fuzzy logic rules applied on measured data and result is a [0,100] output. Quantify – Healthy & ICU subjects 4.4 input signal: 4.2 HRV HRV with fuzzy rules HRV 4 (VLF/HF) 3.8 3.6 3.4 0 12.5 25 37.5 50 62.5 75 1 Agitation level 0.8 HRV with fuzzy rules BPV with fuzzy rules 0.6 Fuzzy Rules 0.4 0.2 0 0 12.5 25 CWT START 37.5 50 62.5 CWT END 75 CP Results are scaled [0, 1] for a simple comparison and metric Digital imaging of motion with fuzzy rules SAS 1 2 1 Chase JG, Agogue F, Starfinger, C, Lam Z, Shaw GM, Rudge AD, Sirisena H; Quantifying agitation in sedated ICU patients using digital imaging. J of Computer Methods and Programs in Biomechanics and Biomedical Engineering. 2004;76(2):131-141. Combining many metrics How we might do things differently? Tight glucose control Hyperglycaemia is prevalent in critical care Impaired endogenous insulin production Increased effective insulin resistance Average blood glucose values > 10mmol/L not uncommon in some critical care units (over length of stay) Stress of condition induces hyperglycaemia Tight control better outcomes: Reduced mortality 27-43% (6.1-7.75 mmol/L) Reduced length of stay and length of mechanical ventilation Active Insulin Control: Evolution AIC 1 – 3 Development of Mathematical Model + 1st Trials Insulin-only AIC 4 Computerised Control Protocol Insulin + Nutrition AIC 5 New protocol with same (or better) control Easy to implement in clinical environment Comparison with international protocols AIC1 AIC4: Prior Art pG SI Gmeasured t + • 4 years prior trials and research • Models mature • Adaptive Control • Short specific trials Update parameters u(t) pG and SI G pG G S I G Ge Gmodelled Q 1 GQ P(t ) u (t ) I nI I b V Overall AIC control system concept is well established The only ways to reduce glucose levels are: increase insulin (Q) which saturates decrease feed (P) G pG G S I G Ge Glucose = G Insulin = Q Feed = P Q 1 GQ P(t ) Insulin-only (AIC3) control of a patient Glucose level mmol/l Tight control target = 4-6 mmol/l Dextrose feed and Insulin input Insulin boluses Feed rate Time (minutes) Insulin-feed (AIC4) control of a patient Glucose level mmol/l Tight control target = 4-6 mmol/l Dextrose feed and Insulin input Feed rate Insulin boluses Time (minutes) Patient 5 = textbook case Wong, XW, Chase, JG, Shaw, GM, Hann, CE, Lotz, T, Lin, J, Singh-Levett, I, Hollingsworth, L, Wong, OS and Andreassen, S (2006). “Model Predictive Glycaemic Regulation in Critical Illness using Insulin and Nutrition Input: a Pilot Study,” Medical Engineering and Physics, In Press SPRINT Specialised Relative Insulin and Nutrition Table Optimises both insulin and nutrition rates to control glycaemic levels Developed through extensive computer simulation Ensures safe protocol before clinical implementation Simple interface for ease of use by nursing staff Combines the very tight control of computerised simulations with minimal implementation cost (no bedside computer required…) SPRINT Step 1 = Feed Rate Table Requires current glucose measurement and last hour change in glucose SPRINT Step 2 = Insulin Table If feed rate = 0 use only insulin wheel Requires current glucose measurement, last hour change and last hours insulin bolus Patient 5008 • Time = 163 hours • Mean = 5.4 mmol/L • 4-6.1 = 85% • 4-7.75 = 97% • Avg Feed = 85% • Avg Insulin = 3.4 U/hr Lonergan, T, LeCompte, A, Willacy, M, Chase, JG, Shaw, GM, Wong, XW, Lotz, T, Lin, J, and Hann, CE (2006). “A Simple Insulin-Nutrition Protocol for Tight Glycemic Control in Critical Illness: Development and Protocol Comparison,” Diabetes Technology & Therapeutics (DT&T), In Press Results Number of respondents Nursing survey: SPRINT 15 Very Good 10 Good Satisfactory 5 Poor 0 Ease of Use Quality Suitability Results 16,063 hours of control on SPRINT; 11,249 measurements 118 admissions Average APAPCHE II score = 21 (41% risk of death) Too high (hypoglycaemia) (hyperglycaemia) Number of measurements Too low 1500 2003 Retrospective Data (Doran, 2004) Mean Glucose = 8.1 Lognormal = outliers to high side Mean 1000 500 0 <4 4 to 6 6 to 8 8 to 10 10 to 12 2003 Numberofof Number measurements measurements 3000 12 to 15 15 to 20 20 plus Blood glucose [mmol/L] SPRINT Mean 2000 2000 Reduction in incidence of high blood glucose 1000 1000 0 0 <3 <4 3 to 4 4 to 6 4 to 5 to 5 6 6 to 7 6 to 8 7 to 8 8 to 10 8 to 9 to 2005 9 10 10 to 12 10 to 11 to 11 12 Normal distribution -- 90% in desired band 12 to 15 15 to 20 20 plus 12 to 13 to 15 to 18 to 20 [mmol/L] 13 15 17 Blood 20 glucose plus Tight control: Tight control: Tight control within target bands Areas under all fitted curves are equal Poor control: BG less than 2.5mmol/L = harmful!! 3.5% of simulated van den Berghe measurements less than 2.5mmol/L Poor control: 70% of simulated Krinsley measurements > 7.75 mmol/L 10% of SPRINT ICU measurements > 7.75 mmol/L 38% of simulated sliding scale measurements > 7.75 mmol/L Tight control 15.00 2003 retrospective data Avg BG Range Retroavg Retrorange Flatter is better Tighter is better 15.00 12.50 12.50 Blood 10.00 Glucose Average (mmol/l) 7.50 Blood 10.00 Glucose Average (mmol/l) 5.00 5.00 2005-06 SPRINT Avg BG Max Retroavg Retromax Flatter is better Tighter is better 7.50 R Sq Linear = 0.652 R Sq Linear = 0.283 R Sq Linear = 0.36 P < 0.05 P < 0.05 2.50 2.50 0.0 5.0 10.0 15.0 20.0 Blood Glucose Range (mmol/l) 5.0 10.0 15.0 20.0 Peak Blood Glucose (mmol/l) SPRINT is flatter and tighter in both cases (P < 0.05) R Sq Linear = 0.459 Outcomes: Tightness of glucose control: the first 118 admissions Average BG Average time in 4 -6.1 Average time in 4 -7 Average time in 4 -7.75 Percentage of all measurements less than 4 Percentage of all measurements less than 2.5 Average insulin bolus Average percentage of goal feed Average feed rate (assuming 1.06 cal/ml for feed) 5.9 60% 82% 90% 2.7% 0.1% 2.7 66% 51 1293 mmol/L All performance indicators agree with simulation and tight control! Protocol is safe – no clinically significant hypoglycaemia U Effective use of insulin and nutrition ml/hr cal/day Improved patient outcome: LOS >3 days . 30% Mortality % 25% SPRINT has decreased mortality by 32% 20% 15% 10% 5% 44 deaths in 169 patients 23 deaths in 118 patients 0% 2004-05 SPRINT P=0.04 Outcomes: Tightness of glucose control* SPRINT Mortality grouped by APACHE II APACHE II Number Mortality 0-14 20 5% 15-24 44 20% 25-34 23 26% 35+ 6 67% SPRINT Sepsis data Total sepsis patients Total sepsis LOS<3 Total sepsis LOS≥3 Mortality sepsis all Mortality sepsis LOS<3 Mortality sepsis LOS≥ 3 * Average APACHE II = 21 * Incomplete data * 2004-05 Number 104 200 48 7 2004-05 21 3 18 4 1 3 49% 13% 25% 19% 33% 17% 35.0% 37.0% 34.0% Mortality 1.9% 15.5% 45.8% 71.4% (% change) -46% -10% -51% Average APACHE II =18.3 http:/www.geocities.com/active_insulin_control Measuring Renal Function Currently: Creatine Creatinine “Activity” Another Way: Both Cleared via Filtration GFR Urine Gentamicin Creatinine and Gentamicin can be measured in plasma and clearance estimated (AUC) However, you need to know the production/input Creatinine production is not known or well estimated e.g. MDRD / Cockroft-Gault equation empirical models assume constant production and steady state Gentamicin doses are known and thus likely more reliable for AUC estimation “Gold standard” = inulin or radioactive iodine compounds Modeling Renal Function C = concentration 2 uCreatinine “Activity” u dC kGFRC Creatinineor Gentamicin dt Vdistribution GFR = KGFR*Vd Urine 1 Gentamicin kGFR Steps: 1. Find kGFR clearance from Gentamicin measurements (AUC and model) 2. Knowing kGFR find uCreatinine production i.e. without guessing what it is! 3. Knowing uCreatinine find Creatinine clearance (or just use Gentamicin clearance) Cohort: n = 16 patients with both Gentamicin and Creatinine data from ICU RESULTS: Model vs MDRD “Easy” Patients : kGFR = time varying and uCreatinine = con stant “Easy” Patients: uCreatinine = constant kGFR not constant GFR and u = patient specific Cannot find uCreatinine and kGFR with just Creatinine data alone. i.e. over 50% of variation is not explained by the MDRD model Poor correlation too much variation for good diagnosis “Hard” Patients: uCreatinine NOT constant kGFR not constant Assumes measured Gentamicin clearance as a “gold standard” MDRD steady state assumptions is not valid for ARF Assuming constant Creatinine production results in very large errors (50+%) with measured data. Conclusions Creatinine production is not always constant Limits its utility for measuring renal function Constant Creatinine production is still patient specific requiring some other form of measurement of renal function (which can be nonconstant) Can get Creatinine production from Gentamicin clearance But only works for “Easy” patients! Going Forward… An adaptive measure of renal function requires a substance in the body with known constant production Cystatin C? More constant, known production than Creatinine? A larger set of trial data would enable more concrete conclusions Modeling Lung Mechanics Objective: To develop a model of mechanics of ARDS lung under mechanical ventilation to determine the appropriate ventilator setting in clinical situation. Superimposed pressure Model of the ARDS lung has up to 60,000 compartments which inflate and deflate according to their threshold opening and closing pressures. TOP’s, TCP’s are determined by the unit compliance and superimposed pressure. Opening Pressure (modified from Gattinoni) Inflated 0 Small Airway Collapse 10-20cmH2O Alveolar Collapse (Reabsorption) 40-60cmH2O Consolidation Model Components Unit Compliance Curve Based on Sigmoid Shape Very little change in volume from alveoli Threshold Pressure Distributions Threshold Opening Pressure (TOP) Threshold Closing pressure (TCP) Normally distributed Shifting distribution mean Model Validation Chase, JG, Yuta, T, Shaw, GM, Horn, B and Hann, CE (2005). “A Minimal Model of Mechanically Ventilated Lung Mechanics to Optimize Ventilation Therapy in the Treatment of ARDS in Critical Care,” IFMBE Proceedings, 12th Intl Conference on Biomedical Engineering (ICBME), Singapore, Dec 7-10, 4-pages, ISSN 1727-1983. How does it work? – Step 1 Step 1 Obtain clinical data with a few different PEEP Step 2 Fit model Step 3 Calculate parameters Step 4 Simulate Step 5 Determine optimal setting How does it work? – Step 2 Step 1 Obtain clinical data with a few different PEEP Step 2 Fit model Step 3 Calculate parameters Step 4 Simulate Step 5 Determine optimal setting How does it work? – Step 3 Threshold Pressure Distribution Mean Shift 4 TOP 3.5 3 Mean Step 1 Obtain clinical data with a few different PEEP Step 2 Fit model Step 3 Calculate parameters Step 4 Simulate Step 5 Determine optimal setting 2.5 TCP 2 1.5 1 0.5 0 0 2 4 6 8 PEEP [cmH 2O] 10 12 14 How does it work? – Step 4 Step 1 Obtain clinical data with a few different PEEP Step 2 Fit model Step 3 Calculate parameters Step 4 Simulate Step 5 Determine optimal setting How does it work? – Step 5 900 800 700 Tidal Volume [ml] Step 1 Obtain clinical data with a few different PEEP Step 2 Fit model Step 3 Calculate parameters Step 4 Simulate Step 5 Determine optimal setting 600 500 400 300 200 100 0 0 5 10 PEEP [cmH 2O] 15 20 ‘Strengths’ in using this approach….. Real time assessment of recruitment status which is dependent on PEEP, ventilation strategy, and disease Readily identifies TCP distributions optimization of PEEP Provides opportunity to simulate a ventilation strategy before application. TOP distribution characteristics Prediction of “overstretch”. E.g. Δ recruitment < % max rate ? Correlated with E2% or CT scan Limitations… Although flow resistive forces through the endotracheal tube are accounted for, the model assumes the pressure at the carina will reflect what is happening to alveolar units. Unforeseen resistive changes (eg major bronchial airway obstruction) could therefore lead to incorrect inferences about recruitment status Needs to be clinically validated Minimal model of heart & circulation Smith, B W, Chase, J G, Shaw, G M, Nokes, R and Wake, G C: Minimal Haemodynamic System Model Including Ventricular Interaction and Valve Dynamics,Medical Eng and Physics, 2004; 26(2):131-139 Experimental pulmonary embolism in 8 pigs Data kindly provided from a researchers in Belgium* Pressure and volume in left ventricle Pressure and volume in right ventricle *Dr Thomas Desaive - Institute of Physics, University of Liège, Belgium *Dr Alexandre Ghuysen - Hemodynamics Research Laboratory, University of Liège, Belgium Experimental pulmonary embolism in pigs Pressure and volume in left ventricle Pressure and volume in right ventricle Modelled cardiovascular dynamics Provide rapid diagnosis in real time optimised treatment Simulation of interventions removes guess work Improved understanding new paradigms and improved approach to the haemodynamically unstable patient Optimised therapeutics and safer drug delivery … ....is out there ….. ……It’s just a matter of seeing where you are going! Acknowledgements AIC1 AIC2 Jessica Lin & AIC3 AIC V: Cardiovascular Modelling: Aaron Le Compte Bram Smith Jason Wong & AIC4 Dunedin Tim Lonergan Mike Willacy Renal Function Modelling: Prof Zoltan Endre Annie Jo Assoc. Prof. Geoff Chase The Danes Franck Agogue Thomas Lotz Prof Steen Andreassen Christina Starfinger Dr Kirsten McAuley & Zhewy Lam Prof Jim Mann Maths and Stats Gurus Dr Dom Lee Dr Bob Broughton Prof Graeme Wake Dr Chris Hann