Click here to find out about helping with AIDA
Click here to return to the AIDA diabetic software simulator program home page
Click here to access the AIDA diabetic software simulator program
The information presented at this site is for general use only and is not intended to provide personal medical advice or substitute for the advice of your doctor or diabetes specialist. If you have any questions about any of the information presented here, concerns about individual health matters or the management of your diabetes, please consult your doctor or diabetes specialist
AIDA Technical Guide

(Re)-Load AIDA Frames / Menus
by Dr. Eldon D. Lehmann and Dr. Tibor Deutsch


CONTENTS
PART 1 PART 2
AbstractParameter estimation
IntroductionModel validation
EducationDiscussion
For patientsLimitations of the AIDA approach
For students Disclaimer
Model description References
Model equations



PART 1

ABSTRACT

A clinical model of glucose-insulin interaction in insulin-dependent (type1)diabetes mellitus has been developed for patient and medical staffeducation. The model attempts to reflect the underlying (patho)physiology of insulinaction and carbohydrate absorption in quantitative terms such as insulinsensitivity, volume of glucose and insulin distribution and maximal rateofgastric emptying. The model's predictions allow a 24 hour simulation ofbloodglucose profiles for hypothetical patients to be generated. Themathematicsunderlying the model are described and the results of validation work totestthe model against real patient data are provided. It is concluded thatthemodel is not refined enough for individual patient simulation and as suchthesystem can only be applied as an educational / demonstration tool.

The purpose of this on-line document is to provide some backgroundinformationand technical details about how AIDA actually works. Furtherinformation,examples and validation results can be obtained from published articles inthediabetes / medical / computing literature [1-18] where full details of allAIDA'sfunctions and components are open to scrutiny.

INTRODUCTION

In 1993 compelling evidence was published from the Diabetes Control andComplications Trial (DCCT) [19] that maintaining tight blood glucosecontrol,certainly in patients with insulin-dependent (type 1) diabetes mellitus,candelay the onset and slow the progression of the later life complicationsusually associated with the disease. As we have entered the 21st century, onewayof achieving the goals demonstrated by the DCCT in routine clinicalpracticemay well be through the use of Information Technology (IT).

Many workers have suggested that IT may be able to assist in thetransferof knowledge and expertise from specialist diabetes centres to primarycarephysicians (general practitioners) and practice nurses as well as directlytopatients [see refs 20-26 for reviews]. The concept of providing a type 1diabetic patient with a hand-held electronic log which could store bloodglucose readings and give the patient advice about the next meal orinsulininjection is clearly an exciting one. However, the development of ahand-heldlog which could provide patient specific advice and be able to justify itsreasoning is a non-trivial problem which has not yet really been solved.Furthermore, such a system even if developed and validated would onlyreallybe of use in health-care systems which could afford the costs of the hightechnology involved. A complimentary approach, however, would be to usesimilar IT techniques as a way of teaching patients about insulin dosage,dietary and exercise adjustments in type 1 diabetes mellitus. Onceproperlyeducated with a deeper understanding as to how changes in theirtherapeuticregimen could affect their glycaemic control, motivated patients should beina better position to optimise their own therapy with guidance from theirphysician. Such educational application of these techniques may beparticularly pertinent because - rather than doing something for a patientallthe time - teaching them how to do it for themselves may lead to a muchmorelasting (and possibly better) result [27].

Anderson et al [28] in 1992 reported a survey of some 400 nurse anddietician members of the American Association of Diabetes Educatorsregardingtheir attitudes, use and knowledge of computers. They found that evendiabetes educators who used computers infrequently had a generallyfavourableattitude towards them. However, even those who reported frequent use ofcomputers did not view themselves as adequately skilled, even in the moststraightforward computer applications. Not surprisingly, the highest useofcomputers was for non-educational applications - confined mostly to wordprocessing and record keeping.

Anderson and co-workers [28] reached the conclusion that their data wereconsistent with a perception among U.S. patient educators that there are,atpresent, few computer applications relevant to the field of diabeteseducation, and even fewer with any degree of broad acceptance. Thissurveyindicated that the respondents (only 50% of the 800 members sentquestionnaires actually replied) felt that computers have yet to make amajorcontribution to the teaching and learning process in diabetes education.

Furthermore, patient educators failed to view themselves as adequatelyprepared for the creative use or development of computer applications -leading the authors to suggest that "the present role of computers insupportof patient education will not change significantly without encouragement,support and demonstrations of efficacy by health care institutions andprofessional organizations" [28].

Juge and Assal [29] have developed a computer assisted instructionprogram on hypoglycaemia which has been used in a number of Europeancountriesfor patients with type 1 diabetes mellitus. Their perception is thatpatientssuffering from a chronic disease, such as diabetes, require specificskillswhich are very different in nature to the theoretical knowledge theyusuallyreceive from different sources, including health-care providers. In ordertobe really useful, the authors [29] felt that educational programs forpatientsmust help them to cope with their disease and, as far as possible, takeintoaccount their concerns, fears and misconceptions.

Clearly the provision of education about diabetes mellitus should beneeds directed. Health-care professionals and students need education indiabetes care, while patients need training in diabetes self-care. Thelattercan be compared to learning to drive a car, whereas the former is moreakin tounderstanding the physical principles on which the internal combustionengineis based (but never actually needing to acquire the skills of driving).Society should not try to teach all diabetic patients the patho-physiologyofdiabetes mellitus; yet we must endeavour to teach this to health-careprofessionals and students. Computers undoubtedly can help in both theseprocesses [27].

EDUCATION

For patients

The advent of home self-monitoring blood glucose (SMBG) measurements inthelate 1970s brought with it the hope of improved glycaemic control forpatientswith type 1 diabetes mellitus. By the end of the 1980s the optimismencouraged by initial reports of improved control using SMBG had faded.Inthe majority of patients it appeared that the introduction of SMBG had notresulted in improved diabetic control [30,31].

SMBG data are conventionally analysed by the physician visuallyscanningthe patient's logbook. As a consequence, clinical decisions are oftenbasedon inadequate assessments of the available data. The ability ofhealth-careprofessionals to efficiently access and analyse SMBG data is crucial ifallthe potential benefits of such data collection are to be fully realised[32].However, the problem runs deeper than this. Patients also need to be abletointerpret their own data - and act on them accordingly. Page and Peacock[31]make a powerful case for the fact that, on its own, SMBG will not lead toimproved glycaemic control without appropriate assessment and modificationofthe treatment regimen. The problem may be that not enough educationalefforthas gone into teaching patients how to 'close the loop' and adjust theirowninsulin injections, diet and lifestyle on the basis of their SMBG data.

Accurate and immediate feedback regarding performance is essentialto thelearning and maintenance of any skill. The provision of such rapidfeedbackis well suited to currently available computer technology. However, atpresent patients often perform SMBG month after month, producing largevolumesof data, without any appreciable or appropriate feedback from healthprofessionals [32]. Eventually, many can become disenchanted anddiscontinuemonitoring [33].

Many prospective studies of the impact of standard diabetespatienteducation have found that while knowledge is increased, metabolic controlimproves little, if at all [34-36]. The lack of formalised insulinadjustmentmethods may be a major reason for this - explaining why many diabetescontrolprograms fail to demonstrate significantly better metabolic control intheirpatients [37,38]. By contrast, using computer-based interactive educationinout-patient clinics, some improvement in glycaemic control has beendemonstrated in patients with type 1 diabetes mellitus [39].

So where do we go from here? That computers can 'close the loop'forin-patient and peri-operative blood glucose control in diabetes mellitushasbeen well documented. The problem is how to achieve similar ambulatorycontrol through educational means. Providing diabetic patients with theknowledge and confidence to 'close the loop' and act on their SMBGmeasurements is probably the single most important area in which computer-based simulation and education techniques can be applied with existingcomputer technology, today. The need for such systems is clear becauseevenamongst interested and motivated patients, knowledge of the principles ofinsulin self-adjustment at present appear limited [40].

In the words of the DCCT investigators "Because the resourcesneeded arenot widely available, new strategies are needed to adapt methods ofintensivetreatment for use in the general community at less cost and effort" [19].Itis against this background that the current educational versions ofAIDA(v4)and AIDA on-line have been produced.

For students

In parallel with use for patient education such a system may also havepotential for the teaching of medical and nursing students, as well asotherhealth-care workers. Michael and Rovick [41] have highlighted some of thefeatures of computer-based simulations which make them so attractive asteaching tools for developing problem-solving expertise:

Time will tell whether such a teaching tool - in the diabetes field - willprove of value for the education of medical and nursing students aboutinsulindosage and dietary adjustments in type 1 diabetes mellitus; this possiblybeing a novel way of providing a larger number of suitably trainedhealth-careworkers to disseminate the benefits of the DCCT trial more widely.

The concept underlying the current release of AIDA (v4) andAIDAon-lineis to allow patients, their relatives, students and health-care workers toexperiment with insulin dosage and dietary adjustments, and perhaps as aresultof their experience gain an increased knowledge and a deeper understandingofthe interplay between the processes involved.

MODEL DESCRIPTION

The glycaemic response of an insulin-treated diabetic patient goes throughtransitory phases leading to a steady state glycaemic profile following achange in either the insulin regimen or diet. The purpose of theAIDAmodelis to simulate these steady state glycaemic and plasma insulin responsesindependently of the initial values from which the simulation is started.Theaccompanying figures to this text provide a schematic showing theanatomicalbasis of the model which assumes a patient completely lacking endogenousinsulin secretion - i.e. an insulin-dependent (type 1) diabetic patient.

The model contains a single glucose pool representingextracellularglucose (including blood glucose) into which glucose enters via bothintestinal absorption and hepatic glucose production. Glucose is removedfromthis space by insulin-independent glucose utilisation in red blood cells(RBCs)and the central nervous system (CNS, see figure) as well as by insulin-dependentglucoseutilisation in the liver and periphery; the latter taking place mostly inmuscle and adipose tissue. Hepatic and peripheral handling of glucose inthemodel are dealt with separately. Glucose excretion from the extracellularspace takes place above the renal threshold of glucose as a function ofthecreatinine clearance (glomerular filtration) rate.

By separating the hepatic and peripheral handling of glucose inthe modelit is possible to assign different case scenario specific insulinsensitivityparameters to glucose-insulin interactions in the liver and periphery.

As shown schematically in the figure, peripheral glucose uptake takes place as a function of bothinsulin andplasma glucose levels; the former enhancing glucose utilisation accordingtothe peripheral insulin sensitivity parameter, Sp, which has a normalisedvaluebetween 0 and 1. Sp multiplied by the insulin level gives the effectiveinsulin level responsible for the control action.

As the liver both produces and utilises glucose depending on thebloodglucose and insulin levels we have modelled hepatic glucose handling intermsof the 'net hepatic glucose balance' which is computed as the sum ofgluconeogenesis, glycogen breakdown and glycogen synthesis data derivedfordifferent blood glucose and insulin levels from nomograms given in Guytonetal [42]. This representation of hepatic glucose handling was chosen inorderto avoid the use of non-physiologically based mathematical functions todescribe hepatic glucose handling. Table 1 shows how the net hepaticglucosebalance varies as a function of glucose and normalised insulin levels.Sh,the hepatic insulin sensitivity parameter, which also has a normalisedvaluebetween 0 and 1, allows computation of the effective insulin level whichcontrols hepatic glucose handling.

The net hepatic glucose balance for any arterial blood glucoselevelbetween 1.1 mmol/l and 4.4 mmol/l is computed by interpolation between thevalues shown on the nomogram. Capillary blood glucose values measuredusinghome monitoring blood glucose meters are approximately 25% lower than thearterial blood glucose levels which are given in [42].Hence, in the AIDAmodel we use the capillary blood glucose levels which correspond to thearterial blood glucose data given in Table 1.


Effective plasma insulin
(Sh * I / Ibasal)
AG <= 1.1 mmol/l AG = 3.3 mmol/l AG >= 4.4 mmol/l
0 291.6 160.0 78.3
1 194.6 114.6 53.3
2 129.3 66.0 -1.7
3 95.7 46.3 -54.3
4 85.0 22.6 -76.0
5 76.3 4.3 -85.0
6 69.0 -10.0 -92.0
7 62.0 -25.3 -97.3
8 52.0 -43.3 -101.0
9 48.0 -47.3 -104.0
10 41.7 -49.3 -106.7

Table 1. Net hepatic glucose balance (mmol/hr) as a function of thearterialblood glucose level, AG, and plasma insulin level, I; calculated fromGuytonet al [42]. Sh is a case scenario specific hepatic insulin sensitivityparameter which has a normalised value between 0 and 1.


The data shown in Table 1 are based on the steady state plasmainsulinlevel which is normalised with respect to a basal level, Ibasal. Notethatfor low blood glucose values there is an automatic compensatory increaseinhepatic glucose production (positive balance) and at high blood glucoselevelsthe net action of the liver is to take up glucose from the blood (negativebalance).

Glucose enters the portal circulation via first-order absorptionfromthe gut. The rate of gastric emptying which provides the glucose fluxintothe small intestine in the model is assumed to be controlled by a complexprocess maintaining a relatively constant glucose supply to the gut duringcarbohydrate absorption apart from the ascending and descending phases ofthe gastric emptying process.

The duration of the period in which glucose entry from the stomachintothe duodenum is constant and maximal has been defined as a function of thecarbohydrate content of the meal ingested. Thus the time course of thesystemic appearance of glucose is described by either a modifiedtrapezoidalor triangular function depending on the quantity of carbohydrate in themeal (See figure).

The function of the kidney to excrete glucose has been modelled intermsof two case scenario specific model parameters; the renal threshold ofglucoseand the creatinine clearance (glomerular filtration) rate(See figure).

The model contains separate compartments for plasma and 'active'insulin. Insulin is removed from the former by hepatic degradation whilethelatter is responsible for glycaemic control. The activation anddeactivationof insulin are assumed to obey first-order kinetics. The only insulininputinto the model comes from the absorption site following subcutaneousinjection.(See figure).

MODEL EQUATIONS

Four differential equations along with twelve auxiliary relations and theexperimental data from [42] constitute the model which is solved bynumericalintegration. The change in the plasma insulin concentration, I, is givenbythe following equation:

where ke is the first-order rate constant of insulin elimination, Iabs istherate of insulin absorption and Vi is the volume of insulin distribution.Thebuild-up and the deactivation of the 'active' insulin pool, Ia, is assumedtoobey first-order kinetics.

where k1 and k2 are first order rate constants which serve to describe thedelay in insulin action. The rate of insulin absorption is modelledaccordingto Berger and Rodbard [43].

where t is the time elapsed from the injection, T50 is the time at which50%of the dose, D, has been absorbed and s is a preparation specificparameterdefining the insulin absorption pattern of the different types of insulincatered for in the model (short-, intermediate- and long-acting).

A linear dependency of T50 on dose is defined as:

where a and b are preparation specific parameters, the values of which aregiven in [43] along with values for s. If the insulin regimen consists ofmore than one injection and / or components, Iabs becomes the sum of theindividual Iabs contributions resulting from the different multicomponentinjections.

The steady state insulin profile, Iss, corresponding to a givenregimenis computed by using the superposition principle assuming three days to beenough to reach steady state conditions:

i.e. the steady state response results from the composite effect ofinjectionsgiven for three subsequent days. It is evident that this summation is notneeded for short-acting insulin preparations (e.g. Actrapid) but it shouldbeused for other, longer acting, insulin preparations whose half time ofabsorption is higher, especially when larger doses are given.

Since the experimental data provided by Guyton et al [42] referstoequilibrium conditions, the insulin level equilibrated with the steadystateactive insulin is considered when computing the net hepatic glucosebalance and peripheral glucose uptake. In other words, at any time duringthe simulation, we have steady state Iss(t) and Ia,ss(t) values, but use:

as the insulin level responsible for the hepatic and peripheral controlaction, where I#eq(t) is the insulin level in equilibrium with Ia,ss(t).

Assuming a single compartment for extracellular glucose, thechangein glucose concentration with time is given by the differential equation:

where G is the plasma glucose level, Gin is the systemic appearance ofglucose via glucose absorption from the gut, Gout is the overall rate ofperipheral and insulin-independent glucose utilisation, NHGB is the nethepatic glucose balance, Gren is the rate of renal glucose excretion andVgis the volume of distribution for glucose.

Assuming a classical Michaelis-Menten relationship between glucoseutilisation and the plasma glucose concentration, with a constant Km suchthat insulin concentration is reflected in different values of the maximalrate of the transport process, we can write [43]:

where c is the slope of the peripheral glucose utilisation vs insulinlevelrelationship, GI is the insulin-independent glucose utilisation and GX isareference glucose level. The NHGB value at any combination of G and I#eqhasbeen derived from the data summarised in Table 1 using Sh * I#eq as theeffective insulin level. The amount of glucose in the gut, Ggut,followingthe ingestion of a meal containing Ch millimoles of glucose equivalentcarbohydrate is defined as:

where kgabs is the rate constant of glucose absorption from the gut intothesystemic circulation and Gempt is the rate of gastric emptying. Theduration of the period Tmaxge for which gastric emptying is constant andmaximal (Vmaxge) is a function of the carbohydrate content of the mealingested:

where Vmaxge is the maximal rate of gastric emptying and Tascge and Tdesgeare the respective lengths of the ascending and descending branches of thegastric emptying curve which have default values in the model of 30 mins(0.5 hrs).

However, for small quantities of carbohydrate (below approximately10 g) such values cannot be used because there will never be time for thegastric emptying curve to plateau out. In such cases Tascge and Tdesgearedefined as:

giving approximately a triangular function(See figure). Equation (11) is only usedwhenthe quantity of carbohydrate ingested falls below a critical level(Chcrit)which is defined as:

Using linear interpolation the rate of gastric emptying for mealscontainingCh millimoles of carbohydrate greater than Chcrit, can therefore bedefined,according to the time elapsed from the start of the meal, t, as follows:

Glucose input via the gut wall, Gin, can be modelled by:

Values for these model parameters which have been derived from [42,43]are given in Table 2. All parameters except Sp and Sh are assumed to becase scenario independent.

As shown in the figure, the rate of renal glucose excretion, Gren, in the model is definedas:

for blood glucose values (G) above the renal threshold of glucose (RTG)where GFR is the glomerular filtration (creatinine clearance) rate.Defaultparameter values in the model have been set for RTG and GFR at 9.0 mmol/land100 ml/min respectively. These default values are used for all casescenariosexcept where renal dysfunction is to be simulated and the clinicalparametersare chosen. The renal excretion of glucose (Gren) is zero for bloodglucosevalues below the renal threshold of glucose [Equation (15b)].


ke = 5.4 l/hr insulin elimination rate constant
k1 = 0.025 /hr parameter for insulin pharmacodynamics
k2 = 1.25 /hr parameter for insulin pharmacodynamics
Ibasal = 10 mU/l reference basal level of insulin
Km = 10 mmol/l Michaelis constant for enzyme mediated glucose uptake
GI = 0.54 mmol/hr/kg insulin-independent glucose utilisation per kg body weight
GX = 5.3 mmol/l reference value for glucose utilisation
c = 0.015 mmol/hr/kg/mU * l slope of peripheral glucose utilisation vs insulin line
kgabs = 1 /hr rate constant for glucose absorption from the gut
Vmaxge = 120 mmol/hr maximal rate of gastric emptying
VI = 0.142 l/kg volume of distribution for insulin per kg body weight
Vg = 0.22 l/kg volume of distribution for glucose per kg body weight

Table 2. Case scenario independent model parameter values calculated from[42,43].


It is noted that the insulin and glucose parts of the model are onlylinked by equation (8) and when computing the net hepatic glucose balanceas a function of G and I#eq. This means that the plasma and 'active'insulin profiles as well as the glucose absorption profiles for any mealcanbe computed separately. This characteristic of the model is utilised whenimplementing the system for computer simulations.


PART 2

PARAMETER ESTIMATION

Please note: Parameter estimation facilities are *not* available forAIDA on-line. They are are only available in the downloadablerelease of AIDA (v4) for the PC.

The algorithm used for parameter estimation of values for the hepatic andperipheral insulin sensitivities (Sh and Sp) determines estimates whichgivethe best 'fit' between the observed and predicted data. Fit is assessedusingdata-trend sensitive least squares criteria to calculate the root meansquare(RMS) deviation between the observed and predicted blood glucose data setsatthe observed blood glucose time points. RMS values are calculated usingtheequation:

where d is the difference between each pair of observed and predictedbloodglucose readings, n is the number of pairs of blood glucose values and npis the number of parameters in the parameter estimation procedure (2; Shand Sp). In determining the fit hypoglycaemic episodes in AIDAv4 areassigned a blood glucose value of 1.0 mmol/l. Parameter values for whichthereis a conflict of trends between the two data sets in any time period areassigned a very poor fit by using an empirical, 'penalty' score for suchcases.For example if the observed data shows a marked decrease in a given periodwhile there is an increase in the simulated glycaemic profile in the sameperiod, a penalty score is associated with the blood glucose level at theendof this period although the absolute deviation might be minor. Followingparameter estimation if the best fit obtainable is greater than 3 mmol/lthen the user is informed that it is not possible to fit the model to thedatasufficiently accurately to permit individual case scenarioparameterisationand simulation to be performed. A best fit value < 3 mmol/l was found, byinspection, to be the upper limit of acceptable parameter estimation. It should be apparent that the 'brute force' enumeration algorithmappliedfor parameter estimation precludes application of the model and fittingroutine for *individual* patient parameterisation and simulation.

MODEL VALIDATION

In order to carry out a preliminary validation of the AIDA modelandpermit aquantitative assessment to be made of its predictive accuracy bloodglucosedata, insulin dosage information and carbohydrate intake meal-related datawere collected over a 5-6 day period from 30 insulin-dependent diabeticpatients attending diabetes out-patient clinics in the Erzsebet Hospital[Budapest] as well as in the Diabetes Centre Bogenhausen [Munich],IstitutoPatologia e Metodologica Clinica [Perugia], St. Thomas' Hospital [London]andHospital Ramon y Cajal [Madrid] [44].

The data for a particular patient were entered into the system fortheday before a change in the insulin injection or dietary regimen (definedasday 1). Parameter estimation was performed on these data and the valuesof Shand Sp determined in this way were used to simulate the effect of changesinthe therapeutic regimen for day 2. Parameter estimation could only beperformed on data from 24 (80%) of the 30 patients in the study; crossovertrends between observed and predicted data preventing the model from beingused with data from the other 6 patients.

Upon simulation of the new insulin dosage or dietary regimen theRMSdeviation between observed and predicted blood glucose profiles wasautomatically calculated. These calculations were performed at eachobservedblood glucose time point and a mean value for the RMS fit determined forthatparticular simulation. This process was repeated for all 24 patients overaperiod of 4-5 consecutive days yielding a total of 578 pairs of bloodglucosemeasurements for simulation over 94 days; data from the first day of thestudynot being used in order to allow theoretical steady state conditions toapply.The RMS values for the fit obtained ranged from 0.8 to 4.6 mmol/l with ameanerror (+/-SD) of 1.93 +/- 0.86 mmol/l [13].

DISCUSSION

The model presented here focuses on the adjustment of insulin and / or dietinthe insulin-dependent diabetic patient. In contrast to previouslydevelopedheuristic rule based expert systems and linear models for insulin dosageordietary adjustment [45-50] this model can be interpreted in physiologicalterms and is therefore more readily understandable to a health-care workerorpatient; the anatomical basis of the model (figure) further aiding itsinterpretationby providing explicit functions for different organs within the body.

In developing the model we have followed the principles usuallyassociated with the minimal-model approach, to find a concise mathematicalformulation to represent the major physiological systems with the fewestpossible parameters. As such the model has intentionally been keptsimple.For example we have not, at present, attempted to model the role ofketonesin the fasting type 1 diabetic patient, nor have we tried to model thechangein the renal threshold of glucose which takes place with age. We havealsoavoided modelling any of the counter-regulatory hormones when simulatinglowblood glucose levels. Other complicating factors such as glucosetransporterswhich help mediate insulin-independent glucose utilisation in certainmusclebeds within the body have not been modelled. With increasing complexitythenumber of parameters for the model increases and so do the difficulties ofdetermining their values in real life. Educationally there does not seemtobe much to be gained by such added complexity. Despite these limitationswebelieve this model has potential application as a teaching tool forpatientsand medical personnel regarding insulin dosage and dietary adjustments indiabetes.

The validation work has shown that a set of differential equationswithindividually tailored parameters cannot be used to model all patients inanyconditions. Hence the system should not be used for individual patientsimulation.

The current version of the system takes account of both insulintherapyand the dietary regimen. However, physical activity, stress and otherlifestyle related events are not, at present, included as it is assumedthatthey will remain relatively constant for the duration of the simulationperiod. Methods of incorporating such parameters into clinical models arecurrently being investigated.

Automated estimation of clinical parameters is a key requirementfor theaddition to the downloadable program of extra example case scenarios. Wehavedeveloped a parameter estimation approach for AIDA v4 which notonlyminimises the least squares difference between observed and predicted datasetsbut also assesses the direction of change in the data. In this way it ispossible for the computer to reject parameter values for which there is agood'traditional fit' as assessed by least squares criteria, but clearlycontradictory trends in the observed and simulated data. If no parametervalues satisfy both criteria then the computer informs the user that themodelcannot be fitted to the example case scenario data. Such a situationmightoccur, for example, if an attempt is made to fit the model to data whererebound hyperglycaemia follows a hypoglycaemic episode.

From a clinical viewpoint the work described here, allowing a usertoexperiment 'on-line' with common clinical situations, should be useful asaneducational tool for patients, their relatives, students and / or medicalstaff.Furthermore it should be very easy for a clinician or diabetic specialistnurseto experiment with various ideas in order to identify possible changes inthetreatment regimen for an individual patient. With the patients themselvespresent this might provide a very powerful educational opportunity fordemonstrating to patients the glycaemic effects of certain changes intherapy.We stress, however, that such experimentation should only be performed byasuitably qualified and experienced diabetes specialist.

LIMITATIONS OF THE AIDA APPROACH

Limitations of the current AIDA approach can be related both to conceptualproblems and problems arising from the current implementation. The modelitself is clearly not refined enough, not coping with important processessuchas exercise or stress - which greatly affect the lives of many diabeticpatients. Also it does not allow the simulation of transient conditions.Furthermore, current modelling of the glycaemic effect of food in terms oftheoverall carbohydrate content of the meal is a well recognisedsimplificationof a very complex physiological process. In this latter case, however,dataare simply not available in the literature about the glycaemic indices andabsorption times to peak of a wide variety of foods. For example, theglucoseabsorption profile following the ingestion of, say, 50 grams ofcarbohydratein a hamburger, on its own, may differ quite substantially from that whenanidentical quantity of carbohydrate is ingested as a meal of chips. Alsotheglucose absorption profile will be completely different, once again, ifthehamburger and chips are eaten together. Quite what happens to glucoseabsorption when tomato sauce and vinegar are added to this meal no onequiteknows!

Clearly, current modelable knowledge of the processes involved intheabsorption of food from the gut is severely limited, and this needs to berecognised in any attempts to utilise such models clinically for glycaemicprediction. Also, it is clear that the current parameter estimationtechniqueemployed within AIDA v4 is not, at present, refined enough since itprovidesno error estimates for parameter values. The lack of a flexible interfacetoexternal data collection devices and other pre-processing systems is alsooften commented upon - however as AIDA is intended purely as aneducationaltool, we do not want it to be too easy for patients to enter their ownregimenand blood glucose data!

We have shown in previous work that the hepatic and peripheralinsulinsensitivity parameters (Sh and Sp) estimated for one set of patient dataon oneday may not necessarily be accurate several days later. We have alsoshownthat the RMS deviation between the observed and predicted blood glucosevaluesbecame systematically worse as time progressed from the date of theoriginalparameter estimation [4,13]. Given this, at the risk of labouring thepoint,neither AIDA v4 nor AIDA on-line are intended for individual patientsimulation and should not be used for generating therapeutic advice[51-54].

For further information about how AIDA works we refer interestedreadersto published articles in the diabetes / medical / computing literature [1-18]where full details of all AIDA's functions and components are open toscrutiny.

DISCLAIMER

The AIDA freeware software is provided 'as is' and the authors disclaim all warranties relating to the software program, whether express or implied, including but not limited to any implied warranties of merchantability and fitness for a particular purpose, and all such warranties are expressly and specifically disclaimed. Neither the authors nor distributors shall be liable for any indirect,consequential, or incidental damages arising out of the use or inability to use this freeware software even if the authors have been advised of the possibility of such damages or claims. In no event shall the authors have anyliability for any damages regardless of the form of claim. The person using the software bears all risk as to the quality and performance of thesoftware.

AIDA v4 and AIDA on-line are prototype computer systems. They arenot finished pieces of work and therefore could contain bugs. Users should be aware that the simulations provided have not been formally validated and therefore could be erroneous.

While we have experimented with the use of parameter estimation techniques for fitting the model to individual patient data, the glycaemic predictions provided may in no way reflect an individual's glycaemic response. As such the authors hereby disclaim any liability for the use of this software.

Users are strongly advised to read the relevant scientific literature [1-18] to understand the limitations and assumptions underlying thiswork.

So that users can be informed of all updates and developments to the system (as well as bugs) they are cordially requested to register their use of AIDA, without charge.

To register all you need to do is enter your email id below and press the Submit button

or send a blank email note to: subscribe@2aida.org


REFERENCES

1. Lehmann ED, Deutsch T. A clinical model of glucose-insulininteraction. In: Computer Modelling, North Holland, Amsterdam, 1991; 101-111.

2. Lehmann ED, Deutsch T. A physiological model ofglucose-insulin interaction. In: Proceedings of the Thirteenth Annual International Conference of the IEEE Engineering in Medicine and Biology Society, Orlando, Florida, USA, 1991; 13(5): 2274-2275.

3. Lehmann ED, Deutsch T. A physiological model ofglucose-insulin interaction in type I diabetes mellitus. Journal of Biomedical Engineering 1992; 14: 235-242.

4. Lehmann ED, Deutsch T. Insulin dosage adjustment indiabetes. Journal of Biomedical Engineering 1992; 14: 243-249.

5. Lehmann ED, Deutsch T. Validation of a prototype computer system to assist in the management of insulin-treated diabetic patients. In: Proceedings of the Fifth International Conference on System Science in Health Care, Prague, Czechoslovakia, 1992; 5: 777-780.

6. Lehmann ED, Deutsch T. An interactive, educational modelforinsulin dosage and dietary adjustment in type I diabetes mellitus. In: SCAMC Proceedings, 16th Annual Symposium on Computer Applications in Medical Care, Frisse ME (ed), IEEE Computer Society Press, New York, 1992; 16: 205-209.

7. Lehmann ED, Deutsch T. AIDA: An automated insulin dosageadvisor. In: SCAMC Proceedings, 16th Annual Symposium on Computer Applications in Medical Care, Frisse ME (ed), IEEE Computer Society Press, New York, 1992; 16: 818-819.

8. Lehmann ED, Deutsch T. An integrated decision support systemto assist in the management of insulin-dependent diabetic patients: A case study. Diabetes Nutrition and Metabolism 1992; 5: 283-294.

9. Lehmann ED, Deutsch T. AIDA2 : A Mk. II Automated InsulinDosage Advisor. Journal of Biomedical Engineering 1993; 15: 201-211.

10. Lehmann ED, Deutsch T, Roudsari AV, Carson ER, Sonksen PH. Validation of a metabolic prototype to assist in the treatment of insulin-dependent diabetes mellitus. Medical Informatics 1993; 18: 83-101.

11. Lehmann ED, Deutsch T, Carson ER, Sonksen PH. AIDA: AnInteractive Diabetes Advisor. Computer Methods and Programs in Biomedicine 1994; 41: 183-203.

12. Lehmann ED, Deutsch T, Carson ER, Sonksen PH. Combiningrule- based reasoning and mathematical modelling in diabetes care. Artificial Intelligence in Medicine 1994; 6: 137-160.

13. Lehmann ED, Hermanyi I, Deutsch T. Retrospective validation of a physiological model of glucose-insulin interaction in type 1 diabetes mellitus. Medical Engineering and Physics 1994; 16: 193-202 [Published erratum appears in Med Eng Phys 1994; 16: 351-352].

14. Lehmann ED, Deutsch T. Computer assisted diabetes care: a 6 year retrospective. Computer Methods and Programs in Biomedicine 1996; 50: 209-230.

15. Lehmann ED. Interactive educational simulators in diabetes care.Medical Informatics 1997; 22: 47-76.

16. Lehmann ED, Deutsch T, Broad D. AIDA: an educationalsimulator for insulin dosage and dietary adjustment in diabetes. British Diabetic Association, London, 1997. ISBN 1-899288-55-4.

17. Lehmann ED. AIDA - a computer-based interactive educational diabetes simulator. Diabetes Educator 1998; 24: 341-346.

18. Lehmann ED. Preliminary experience with the Internet release of AIDA - an interactive educational diabetes simulator. Computer Methods and Programs in Biomedicine 1998; 56: 109-132.

19. The Diabetes Control and Complications Trial Research Group. The effect of intensive treatment of diabetes on the development and progression of long-term complications in insulin-dependent diabetes mellitus. New England Journal of Medicine 1993; 329: 977-986.

20. Lehmann ED, Deutsch T. Application of computers in diabetes care -a review. I. Computers for data collection and interpretation. Medical Informatics 1995; 20: 281-302.

21. Lehmann ED, Deutsch T. Application of computers in diabetes care -a review. II. Computers for decision support and education. Medical Informatics 1995; 20: 303-329.

22. Lehmann ED. Application of information technology in clinicaldiabetes care - a Special Issue. Part 1. Databases, algorithms and decision support. Medical Informatics 1996; 21: 255-258 [Editorial].

23. Lehmann ED. (Ed.) Special Issue: Application of informationtechnology in clinical diabetes care. Part 1. Databases, algorithms and decision support. Medical Informatics 1996; 21: 255-378.

24. Lehmann ED. Application of information technology in clinicaldiabetes care - a Special Issue. Part 2. Models and education. MedicalInformatics 1997; 22: 1-3 [Editorial].

25. Lehmann ED. (Ed.) Special Issue: Application of informationtechnology in clinical diabetes care. Part 2. Models and education. Medical Informatics 1997; 22: 1-120.

26. Lehmann ED. Application of computers in clinical diabetes care. Diabetes Nutrition and Metabolism 1997; 10: 45-59.

27. Lehmann ED. The Diabetes Control and Complications Trial (DCCT): A role for computers in patient education? Diabetes Nutrition and Metabolism 1994; 7: 308-316.

28. Anderson RM, Donnelly MB, Hess GE. An assessment of computer use, knowledge, and attitudes of diabetes educators. Diabetes Educator 1992; 18: 40-46.

29. Juge CF, Assal JP. Designing computer assisted instruction programs for diabetic patients: how can we make them really useful?. In: Proceedings, 16th Annual Symposium on Computer Applications in Medical Care (Ed.), M.E. Frisse, IEEE Computer Society Press, New York, 16: 215-219, 1992.

30. Worth R, Home PD, Johnston DG, Anderson J, Ashworth L, Burrin JM, Appleton D, Binder C, Alberti KGMM. Intensive attention improves glycaemic control in insulin-dependent diabetes without further advantage from home blood glucose monitoring: results of a controlled trial. British Medical Journal 1982; 285: 1233-1240.

31. Page SR, Peacock I. Blood glucose monitoring: does technology help? Diabetic Medicine 1993; 10: 793-801.

32. Zimmet P, Lang A, Mazze RS, Endersbee R. Computer-based patient monitoring systems. Use in research and clinical practice. Diabetes Care 1988; 11 (Suppl. 1): 62-66.

33. Zimmet P, Gerstman M, Raper LR, Cohen M, Crosbie C, Kuykendall V, Michaels D, Hartmann K. Computerized assessment of self-monitoredblood glucose results using a Glucometer reflectance photometer with memory and microcomputer. Diabetes Research and Clinical Practice 1985; 1: 55-63.

34. Mazzuca SA, Morrman NH, Wheeler ML, Norton JA, Fineberg NS, VinicorF, Cohen SJ, Clark CM. The Diabetes Education Study: A controlled trial of the effects of diabetes patient education. Diabetes Care 1986; 9: 1-10.

35. Rettig BA, Shrauger DG, Recker RR, Gallagher TF, Wiltse H. A randomised study of the effects of a home diabetes education program. Diabetes Care 1986; 9: 173-178.

36. Bloomgarden ZT, Karmally W, Metzger MJ, Brothers M, Nechemias C, Bookman J, Faierman D, Ginsberg-Fellner F, Rayfield E, Brown V. Randomised controlled trial of diabetic patient education: improved knowledge without improved metabolic status. Diabetes Care 1987; 10: 263-272.

37. Albisser AM. Adjusting insulins. Diabetes Educator 1992; 18: 1-8.

38. Carlson A, Rosenqvist U. Diabetes care organization, process, and patient outcomes: effects of a diabetes control program. Diabetes Educator 1991; 17: 42-48.

39. Wise PH, Dowlatshahi DC, Farrant S, Fromson S, Meadows KA. Effect of computer-based learning on diabetes knowledge and control. Diabetes Care 1986; 9: 504-508.

40. Gill GV, Redmond S. Self-adjustment of insulin: an educational failure? Practical Diabetes 1991; 8: 142-143.

41. Michael JA, Rovick AA. Problem-solving in the pre-clinicalcurriculum: the uses of computer simulations. Medical Teacher 1986; 8: 19-25.

42. Guyton JR, Foster RO, Soeldner JS, Tan MH, Kahn CB, Koncz L, Gleason RE. A model of glucose-insulin homeostasis in man that incorporates the heterogenous fast pool theory of pancreatic insulin release. Diabetes 1978; 27:1027-1042.

43. Berger M, Rodbard D. Computer simulation of plasma insulin and glucose dynamics after subcutaneous insulin injection. Diabetes Care 1989; 12: 725-736.

44. EURODIABETA Deliverable 15, Andreassen S, Bauersachs R, Benn J, Carson E, Gomez E, Hovorka R, Lehmann E, Nahrgang P, del Pozo F, Roudsari A, Schneider J. Report on developed prototypes integrating KBS and other methodologies for insulin therapy advisory systems. Technical Report to the EEC Advanced Informatics in Medicine Exploratory Action, EEC-AIM, Brussels, 1990; 15.

45. Deutsch T, Carson ER, Harvey FE, Lehmann ED, Sonksen PH, Tamas G, Whitney G, Williams CD. Computer-assisted diabetic management: a complex approach. Computer Methods and Programs in Biomedicine 1990; 32: 195-214.

46. Deutsch T, Lehmann ED, Carson ER, Sonksen PH. Rules and models for insulin dosage adjustment. Diabetes Nutrition and Metabolism 1991; 4 (Suppl. 1): 159-162.

47. Lehmann ED, Deutsch T, Roudsari AV, Carson ER, Benn JJ, Sonksen PH. A metabolic prototype to aid in the management of insulin-treated diabetic patients. Diabetes Nutrition and Metabolism 1991; 4 (Suppl. 1): 163-167.

48. Lehmann ED, Roudsari AV, Deutsch T, Carson ER, Benn JJ, Sonksen PH. Clinical assessment of a computer system for insulin dosageadjustment. In: Lecture Notes in Medical Informatics, Adlassnig K-P, Grabner G, Bengtsson S, Hansen R (eds), Springer-Verlag, Berlin, 1991; 45:376-381.

49. Lehmann ED, Deutsch T, Roudsari AV, Carson ER, Sonksen PH. Acomputer system to aid in the treatment of diabetic patients. In: Computer Modelling, North Holland, Amsterdam, 1991; 90-100.

50. Lehmann ED, Deutsch T, Roudsari AV, Carson ER, Benn JJ, Sonksen PH. An integrated approach for the computer assisted treatment of diabetic patients on insulin. Medical Informatics 1992; 17: 105-123.

51. Lehmann ED. Application of computers in diabetes care. Lancet 1994; 344: 1010.

52. Lehmann ED. Diabetes moves onto the Internet. Lancet 1996; 347:1542.

53. Lehmann ED. Computers in Diabetes'96. Medical Informatics 1997; 22: 105-118.

54. Lehmann ED, Deutsch T. Compartmental models for glycaemicprediction and decision-support in clinical diabetes care: promise and reality. Computer Methods and Programs in Biomedicine 1998; 56: 193-204.


Graphical summaries of the AIDA model structure are available via these links: AIDA model graphics (with links to further graphs) and AIDA model graphics (more detailed / printable graph).

A wide range of general diabetes computing (and AIDA-related) full text references can also be accessed in either HTML format or as portable document format (PDF) files by clicking on this link here.


For more information about AIDA or AIDA on-line please contact:

Dr. Eldon D. Lehmann
Department of Imaging (MR Unit)
Imperial College of Science, Technology and Medicine
National Heart and Lung Institute
Royal Brompton Hospital
London SW3 6NP
United Kingdom

or via the on-line AIDA contact form

Source: Printed from the AIDA Website
from: http://www.2aida.org/technical

This AIDA Technical Guide is based on: J. Biomed. Eng. 1992; 14: 235-242;Diab. Nutr. Metab. 1994; 7: 308-316;Med. Eng. Phys. 1994; 16: 193-202

The primary reference describing the AIDA model is reference 3 above.



Return to Top of PageIf you like AIDA, why not display our logo on your home page? For more information about linking to the AIDA Website please click here.

AIDA Website homeReturn to AIDA Website Home PageAIDA is a freeware diabetic software simulator program of glucose-insulin action + insulin dose & diet adjustment in diabetes mellitus. It is intended purely for education, self-learning and / or teaching use. It is not meant for individual blood glucose prediction or therapy planning. Caveats

This Web page was last updated on 8th January, 2003.(c) www.2aida.org, 2000-2003. All rights reserved. Disclaimer. For the AIDA US Mirror Site, please click here. For the Diabetes / Insulin Tutorial, please click here.