Use of IT in Diabetes Care
(Re)-Load AIDA Frames / Menus
Reprinted with the kind
permission of Editrice Kurtis, publishers of Diabetes, Nutrition &
Metabolism: Clinical and
Application of computers in clinical diabetes care
ABSTRACT. This article critically appraises selected clinically related papers that recently appeared in a two-part Special Issue of Medical Informatics, the official journal of the European Federation for Medical Informatics. This has been devoted to the application of computers in clinical diabetes care. The 15 papers included in the Special Issues cover database systems (including telemedicine and smart-card based applications), algorithmic-based systems, decision-support prototypes, the use of models, and educational software. In this article the computing background to the work is overviewed, before the clinical need and potential cost-benefits of utilising information technology in clinical diabetes care are highlighted. The DIAMOND, DIABCARD, DIABTel, HumaLink, AIDA, ‘Packy & Marlon’, and ‘Learning Diabetes’ systems are reviewed. Concerns over evaluation methodologies are raised and it is suggested that such issues need to be addressed, before programs like these will see widespread utilisation and clinical acceptance. Although the Medical Informatics Special Issues should not be considered as in any way comprehensive in their coverage of clinical diabetes computing - it is hoped that the compilation of papers provided there - along with this critical appraisal - may offer a useful source of novel ideas - as well as, perhaps, a starting point for future research.
Diab. Nutr. Metab. 10: 45-59, 1997.
© 1997, Editrice Kurtis.
Academic Department of Radiology, The Royal Hospitals NHS Trust, St. Bartholomew’s Hospital, London and Department of Imaging, National Heart and Lung Institute (Imperial College of Science, Technology and Medicine), Royal Brompton Hospital, London, U.K.
Key words: Diabetes mellitus, information technology, computers, education, decision support, telemedicine, simulations, games.
Correspondence to: Dr. E.D. Lehmann, MR Unit, NHLI, Royal Brompton Hospital, London SW3 6NP, U.K.
Received 15 February 1997; accepted 30 April 1997.
1997 is the 75th anniversary of the publication of Frederick Banting and colleagues’ keynote paper in the Canadian Medical Association Journal (1). One year later Banting et al (2) wrote in the British Medical Journal:
"It is not always easy to adjust so that there is sufficient insulin to nullify post-prandial hyperglycaemia and yet insufficient insulin to produce dangerous lowering of the blood sugar".
Not much has changed in 75 years. At around the same time as Banting and Best were reporting their preliminary findings (3) the field of engineering was beginning to lay the groundwork of control systems theory. These seemingly unrelated endeavours were brought together when medical researchers realised that metabolic controls in the body could be described and analysed using the theories and techniques developed for feedback control systems in engineering (4).
Medical Informatics, the official journal of the European Federation for Medical Informatics, recently devoted a two-part Special Issue to advanced information technology (IT) initiatives that may be able to address the problem of adjusting the insulin dose, and controlling blood glucose (BG) levels, as well as assisting in the provision of modern-day diabetes care (5, 6). This review focuses on some of the more clinically relevant papers included in the Special Issues (5, 6), which follow previous journal issues and conference proceedings on the subject of computers in diabetes.
However, the Special Issues do not consist of presentations arising from a particular meeting, but rather are made up of papers specifically invited from recognised experts in the field. Other novel work - as yet unpublished in the medical computing / diabetes literature - was also sought for inclusion; the aim being to produce two volumes describing new and exciting work in this important area. The niche for such Special Issues arose from a systematic two-part review by Lehmann and Deutsch (7,8) of the application of computers in diabetes care, which appeared in Medical Informatics in late 1995. While researching this it became apparent that there was some very exciting and novel work which had not found its way into the literature. In certain cases new but as yet unpublished work was being done by recognised experts in the field. However, in other cases highly novel and pertinent research was being carried out by non-academic workers, often outside the academic centres of excellence usually associated with such ventures.
The papers selected for inclusion in the Special Issues can broadly be classified under the following headings: (i) databases, (ii) algorithms, (iii) decision support, (iv) models, and (v) education; each of these five topics being addressed by a trilogy of papers. The more clinically relevant of these articles will be discussed here. First, however, a brief overview of the computing background is provided and the clinical need for and potential cost-benefits of utilising IT in diabetes care are considered.
Endeavours to apply IT routinely in diabetes care have been attempted for many years. Significant advances have been made by a number of key researchers. However, despite the novel work undertaken it is fair to say that the impact of computers on diabetes care thus far has been relatively limited. Fig. 1 summarises the spectrum of IT applications relevant to clinical diabetes care. The acceptance scale indicates how widely such applications have been adopted into routine clinical practice. For example, BG meters (either with or without electronic memories) are well accepted tools. By contrast at the other end of the spectrum computerised decision support is not at all widely used, at present (9).
Clearly a number of earlier attempts at the ‘intelligent’ application of computers in diabetes care were not sophisticated enough. In many cases user interfaces were primitive and prototypes difficult to use. However such developments were not helped by the fact that many physicians with busy clinical practices did not see a need for, or have a desire to use, such software programs. Also system developers certainly can be faulted for not demonstrating through rigorous retrospective and prospective clinical trials the safety and efficacy of their applications.
THE NEED FOR IT IN DIABETES CARE
Things however appear to be changing. The impact of the Diabetes Control and Complications Trial (DCCT) (10) cannot be overemphasised. It is now apparent that existing standards of diabetes care are not sufficient. Tightening glycaemic control really makes a difference to patient wellbeing - yet the facilities are not available to provide the sort of intensive insulin therapy applied in the DCCT in routine clinical practice. The resources needed to achieve and maintain really tight diabetes control in the DCCT were enormous. Specialist staff were recruited to maintain contact with patients who were often seen fortnightly in hospital, and contacted even more frequently by telephone during the day as well as at night. Efforts were redoubled at times of minor illness or emotional upset. Team meetings took place weekly to maintain these standards. Excluding research costs, the annual clinical cost of intensive therapy in the DCCT was US$ 4000 per year for multiple daily injections and $ 5800 per year for continuous subcutaneous insulin infusions - approximately three times the cost of conventional therapy ($ 1700 per year). A large portion of this cost difference was reported to be related to the greater frequency of out-patient visits and the greater resources used in self-care (11). Clearly on a population-wide basis it will be hard to sustain three-times greater expenditure for diabetes care in routine clinical practice.
In a recent survey, 96% of outpatient visits for primary care of patients with diabetes were made to general and family practitioners and internists. Such health-care professionals did not provide the care in the DCCT. Also the vast majority of primary care physicians lack the training and resources to offer DCCT-like intensity of care (12). Furthermore for the bulk of primary care physicians who have only a few patients needing DCCT-like care, it is simply not cost effective to provide the necessary staff or allocate the time required to deliver this intensity of service (13).
The findings of the DCCT come at a time when major health reforms are underway in an attempt to reduce the spiralling costs of health-care provision in the USA, as well as in other countries. In 1992 the direct costs of all forms of diabetes mellitus and its complications in the USA were estimated at $45 billion (14), while in 1996 in the UK the cost of non-insulin dependent (type 2) diabetes mellitus was estimated at £2 billion, or 8% of National Health Service hospital expenditure (15). Therefore the time may be right for the utilisation of IT in diabetes care. Indeed it is difficult to envisage the widespread delivery of improved glycaemic control, such as that seen in the DCCT, without the extensive use of IT. In this respect it is not a case of replacing doctors, nurses or other health-care professionals - rather there is a need for the volume of work undertaken to increase - yet the staff required to do this simply are not available. While the capital costs for developing, validating and utilising computer systems to help with this clinical workload may be high - the savings which could accrue are substantial (16,17). So in purely financial terms the case for IT in diabetes care may be compelling - if systems can be developed which really work and generate clinical benefit.
Initiatives to reduce the suffering associated with diabetes mellitus include the St. Vincent Declaration (18) which is in part looking to establish monitoring and control systems using IT for quality assurance of diabetes health-care provision and for laboratory and technical procedures in diabetes diagnosis, treatment and self-management. Now seven years on a multitude of database applications are available to monitor and document the level of care provided to patients with diabetes (19,20). While it could be argued that too many different database programs are being developed - and therefore perhaps efforts are unnecessarily being duplicated - hopefully at least common datasets are starting to be being agreed upon (21) so that regardless of the application being used - data will eventually become compatible and therefore accessible locally, regionally and nationally for research and audit. However, much work needs to be done to ensure data security and safeguard patient confidentiality - especially as data start to be transferred between primary and secondary care and between hospitals and regions.
While databases may not be viewed as the most exciting area of medical informatics research - they are an area where a clinical impact is being routinely made, now. Also perhaps one of the failings of earlier semi-‘intelligent’ decision support prototypes rests with the lack of stable, computerised medical record systems which could ‘feed’ the decision-support software with data. The need for decision support programmers to design and code their own individual small databases unnecessarily prolonged the development cycle, increased compatibility problems, and made the wider dissemination of the prototype systems for evaluation by others all the more difficult.
In part 1 of the Special Issue Kopelman and Sanderson (22) provide an introductory overview of some of the requirements of database systems in diabetes care - before demonstrating how their own application, DIAMOND, addresses these. Particular attention is devoted to the St. Vincent Declaration and the ways in which databases can assist in the audit required for this, as well as to help ascertain clinical outcome measures. The following paper by Engelbrecht et al (23) overviews a particular database application - a diabetes smart card - highlighting some of the requirements specific to this form of patient-held database. Results of a preliminary evaluation of the DIABCARD in a hospital setting are also presented. Interestingly acceptance of this new technology by patients appears to be greater than that by medical staff.
Efficient data analysis and decision making both depend heavily on the frequency and quality of communication between patients and health care professionals. The trilogy of database-related papers is concluded with an overview by Gomez et al (24) of their DIABTel system. This focuses on an alternative distributed database architecture - offering a telemedicine based approach for diabetes care (Fig. 2) which aims to: [i] improve patient communication with hospital-based diabetologists between clinic visits, [ii] allow doctors to assess the patient’s condition on a more frequent basis (e.g. every week), and [iii] provide patients with a service of ‘supervised autonomy’ to increase independence without decreasing the necessary continual support and supervision of the doctor. Clinical evaluation studies of this approach are planned, and the concept and systems architecture are to be developed further in the new telematics T-IDDM project (25). Separate telemedicine experience from other centres suggests the DIABTel objectives to be realistic. Fallucca et al (26) recently reported elsewhere on the application of the Diva system (27) in a cohort of pregnant diabetic women. Use of the telemedicine software led to tighter BG control, in particular faster optimisation of BG profiles, better pre-breakfast and pre-lunch BG concentrations and fewer hypoglycaemic episodes (27) - results which if confirmed in large-scale evaluation studies would be most encouraging.
ALGORITHMIC-BASED DECISION SUPPORT SYSTEMS
This author started his research interest in the application of computers in diabetes care in an academic Department of Medicine in a University teaching hospital where algorithms were frowned upon, because they could not explain their reasoning. This is a valid criticism, and certainly a limitation of algorithmic-based approaches. As a result the author embarked on the development of a series of prototypes linking knowledge-based systems (KBS) with linear models, and compartmental models, utilising various means of data interpretation and feature extraction. A retrospective on what is now a 7+ year voyage has recently appeared elsewhere (28). None of the KBS or model-based approaches were sufficiently robust or safe to be used as insulin-dosage advisors and having reviewed the literature (7,8) and considered ‘why not’ it has become apparent that algorithmic-based approaches do offer a computational robustness and practicality difficult to provide, at present, by other means.
Things obviously however do rather depend on the purpose of seeking computer-based assistance. If users require to analyse why a patient may be going hypoglycaemic overnight - or what the contribution of the glucose counter-regulatory process (Somogyi effect) might be - clearly algorithms may not be of great benefit. However most doctors do not look to IT to help them in this way. It is rare for a competent diabetologist or endocrinologist not to know how to improve a given patient’s glycaemic control. This obviously may not be the case for general primary care physicians - but to manage patients with diabetes non-specialists are unlikely to require such complex (physiological) analyses either.
The basic problem appears to arise because clinicians have insufficient time to see patients frequently enough - from the control systems perspective the under-sampled system with infrequent BG monitoring and even less frequent visits to a doctor can impact significantly on patients’ diabetes care. Also not having enough trained health-care professionals (nurses, educators, dieticians, etc) to motivate patients to make the necessary adjustments themselves can cause problems.
Algorithms by their very nature cannot cope with situations not explicitly stated. However they can cater for a large variety of situations - and if they can offer 90% coverage this could provide significant practical benefit to a considerable number of patients. While intellectually it may not be very satisfying to utilise algorithms which are not comprehensive - and while algorithms may be one of the less intellectually taxing methodologies in diabetes computing - and therefore perhaps not of such great interest to academics - they do seem to warrant closer attention.
In part 1 of the Special Issue Albisser et al (29) describe their algorithmic, telemedicine-based HumaLink system (formerly called TeleDoc (8)). Having measured their BG levels patients can access the HumaLink system from any touch-tone telephone, whether at home or while travelling, 24 hours a day. Patients are encouraged to call HumaLink at each BG measurement, if possible, and particularly when an insulin dosage is to be recommended. After identifying themselves with their own unique personal identification number, patients follow verbal instructions from the speaking HumaLink system and key in their BG measurement(s) along with any other information about illness, hypoglycaemia, changes in activity, unusual carbohydrate intake and medication. The system verbally verifies each entry before accepting it into the patient’s personal diabetes database.
The HumaLink system then relays instructions in accordance with an individualised treatment plan programmed by the caller’s physician. Following the computer’s provision of verbal advice down the telephone line the system requires patient confirmation of the instructions HumaLink has delivered by verbally requesting that patients key-in the new insulin doses.
HumaLink can operate in ‘manual’ or fully automated advisory mode. In ‘manual’ recording / documenting mode the computer logs the patient’s readings and a physician reviews the data before leaving a verbal message for the patient on the system. The next time the patient telephones HumaLink he / she will receive the physician’s advice about what therapy adjustment(s) to make. Furthermore, health-care professionals have the facility to activate a virtual recorder and using a microphone can leave messages for individual patients regarding specific instructions. All interactions by the health-care professional are documented and become part of the patient’s legal medical record. A facility is also provided to forward courtesy reports automatically by fax from the HumaLink computer to the patient’s referring physician (8).
The fully automated advisory mode applies algorithms to modify insulin dosages within pre-defined limits set by the physician. Using trigger levels, guidelines, and instructions individualised for each patient, the system can also automatically react on behalf of the physician immediately when a crisis (such as hypoglycaemia) is reported or whenever BG control deviates from the targets set for that particular patient. Details of the advisory module which utilises a pursuit algorithm have been previously described by Albisser (30).
If the patient’s BG profile does not respond in the manner expected, this is flagged by the computer for the physician’s attention. The guidance provided and / or pre-set threshold levels are then reset in order to direct the pursuit algorithm in an alternative, more appropriate direction. These interactive capabilities, together with automatic safety limits built into the system, are intended to ensure patient safety. While it is possible to construct rules and algorithms to be conservative and therefore which should theoretically be safe - apprehensions always exist about computers deciding on patient therapy without human intervention. Clearly this is not a concern restricted to diabetes computing - but rather a much more general issue - with medico-legal implications (31) for the entire decision-support field of medical informatics (32).
With the HumaLink system such concerns are at least in part addressed by allowing periodic expert human clinical review of the data and possible human intervention, if required, via the central telemedicine computer. By contrast such supervisory input clearly cannot be provided at present with hand-held devices. The US Food and Drug Administration (FDA) Center for Devices and Radiological Health has determined that HumaLink is a medical device as defined under Section 201(h) of the US Federal Food, Drug and Cosmetic Act (29).
Algorithms cannot explain or justify their reasoning - either to patients or health-care professionals. This may account for the relative lack of widespread use of Albisser’s earlier Insulin Dosage Computer (30) which was effectively a ‘black box’ which clinicians and patients just had to trust. This limitation appears to have in part been addressed in the HumaLink system where the algorithms are generated for individual patients by their clinician, from a framework devised by Skyler et al (33). Therefore different algorithms can be used for different patients - although standard ‘templates’ are available.
How well does HumaLink manage in clinical practice? In part 1 of the Special Issue preliminary ‘b-testing’ experience is reported from 124 insulin-treated diabetic patients in two US centres, compared with 80 insulin-treated diabetic controls (29). Further studies in other centres are on-going. At baseline, before starting to use HumaLink, these patients had HbA1c levels of 10.1-10.2%. After six months HbA1c levels had fallen 1.0-1.3% in those patients actively using the system, but remained unchanged in the control group who received routine therapy from their local diabetologist (Fig. 3). Clearly it could be that patients who actively used the system were better motivated and more interested in their diabetes, compared with those who did not make use of HumaLink - although it is noteworthy that there were no significant differences in HbA1c levels between users and non-users at baseline. However to overcome this as a potential confounder fully randomised trials are required.
Notwithstanding the non-random nature of the recruitment, it is interesting to note that in this large cohort there were no reported events of serious hypoglycaemia (requiring assistance or hospitalisation) attributable to the system in either centre. Furthermore the prevalence of diabetes related crises (hyper- or hypo-glycaemia) was reported to fall approximately three-fold in the active use group (29).
Albisser and colleagues (29) are the first to report an evaluation of a decision support prototype in such a large number of diabetic patients. Even though their study is only preliminary and a safety and efficacy b-testing effort - the findings are substantial - and therefore certain aspects of the evaluation approach should be highlighted. The study had wide inclusion criteria - providing a heterogeneous patient population similar to that which would be found in routine clinical practice. While this may be more realistic than just studying highly selected subgroups of patients, in future studies it would be useful for the data or analyses to be stratified by patient type and status, to obtain some insight into the utility of the system in selected sub-groups. For example, the HumaLink approach might be better accepted, and therefore more utilised, by younger patients. However the current evaluation study does not permit such subgroup analyses.
Also, while the overall numbers of patients studied were large, and involved we are told in total 888 patient months of prospective follow-up, HbA1c data were only available at the end of the 6-month follow-up from 90 users and 77 non-users. Future reports should clearly aim to follow up a larger proportion of both users and non-users, and report HbA1c data on all the patients studied, lest biases be introduced by those subjects ‘lost’ to follow-up or for whom such repeat HbA1c data were not available. For example those patients who did not have such frequent blood tests and therefore missed their 6-month follow-up may have had worse glycaemic control as a result of being seen less often in clinic. Alternatively these patients may have derived less benefit from their use of the HumaLink system and therefore re-attended clinic less frequently. Whatever the reason - we cannot exclude that the active users for whom follow-up HbA1c data are not available represent the sub-group of the intervention cohort whose glycaemic control did not improve. If this is the case - loss of these patients to follow-up could significantly skew the results in favour of a positive outcome. Also the differential loss of patients to follow-up, at least 20% of the intervention group as compared with only 13% of the "control" group, may have introduced further biases into the analyses. Such considerations should be borne in mind by researchers planning evaluation studies with alternative decision-support prototypes, as well as for future studies involving the HumaLink system.
Furthermore, it is well recognised that glycaemic control can improve simply as a result of being enrolled in a study, or potentially in this case from the attention provided by telephoning HumaLink each time a BG measurement is made. Therefore fully randomised studies are very much needed.
However if improvements in HbA1c levels, such as those reported by Albisser et al (29), can be confirmed by further long-term, large-scale randomised clinical trials this will be a most encouraging development for patients with diabetes. By comparison, in the DCCT (10) the 2% difference in mean HbA1c between the standard and intensively treated groups was associated with a 60% reduction in risk for diabetic retinopathy, nephropathy, and neuropathy.
It would also be of considerable clinical interest to establish whether the patient benefits of using the HumaLink system persisted, even after discontinuing use - i.e. do patients learn from their telephone interactions with the computer, or does it only offer a ‘crutch’ on which they become totally dependent? As intimated above, medico-legal issues have always been a concern to developers of decision-support systems. Therefore it is interesting to note that the first law-suit regarding the HumaLink system has already taken place in the USA. This dealt with the denial by an insurance company of a patient’s claim for the medical care provided by the computer system. The ruling was in favour of the insurance company and the settlement was that no reimbursement for computer or telephone assistance would be provided. The insurance company did however offer unlimited, even daily, clinic visits for the patient - to be reimbursed without contest (Dr. A.M. Albisser, personal communication).
"Glucose measurements are futile if
not acted upon"
Despite the advanced technology which is being directed to the measurement and storage of SMBG data many patients, even in the 1990s, appear poorly equipped to alter their therapy on the basis of such data. When one considers that poor glycaemic control is associated with an increased later life risk of a plethora of devastating complications it seems surprising that more effort has not gone into educating diabetic patients about what to do with their BG readings.
For example, although essential to the process, it has repeatedly been shown that the isolated act of collecting SMBG data is not sufficient to improve metabolic control (34,35). While educational work may be considered less ‘glamorous’ than the latest artificial intelligence endeavours - education is a key means of improving understanding in medicine. As computer-assisted learning (CAL) techniques are revolutionising the way that people are taught, it is likely that CAL techniques applied to health-care education may provide powerful tools for the teaching and motivation of both patients and students. Furthermore with the massive expansion of the Internet over the past few years there at last appears to be a stable common platform which will greatly increase accessibility to on-line teaching from (remote) centres of excellence.
Is diabetes education already not widely available? The simple answer is ‘no’. In a recent survey of over 2400 patients to determine the proportion of adults with diabetes in the USA who had received diabetes education, it was found that over 41% of type 1 diabetic patients had never attended a class or program about diabetes (36). Furthermore even when education is made available, physicians often tend to explain the disease rather than ensure that patients acquire the reflexes and expertise that will genuinely enable them to manage their diabetes (37). Therefore there is clearly room for improvement, and perhaps IT can help. However, education is difficult if based only on verbal and written presentations of dry facts (38). Teaching materials using multi-media presentations may provide a partial solution. However, the aim should also be to teach diabetes self-management to patients in an intuitive and enjoyable way, so that the knowledge can be enduring. Clearly it is not ideal to learn about diabetes control solely from real life experiences because of the long time frame involved, aside from the possible dangers to the patient of hypo- or hyper-glycaemia. An interactive simulation of a diabetic patient could be one solution.
As Chao (4) has highlighted, in the same way that aircraft pilots and air traffic controllers are trained for routine and emergency procedures on airplane and air traffic simulators, it should be possible for diabetic patients to be trained for physiologically appropriate responses on a diabetes simulator in absolute safety, and in a relatively short space of time. In this respect management of error is a key component of the learning process. Identifying the error, getting patients to discover it for themselves, and asking how they would correct it, are essential steps in educating patients how to improve their glycaemic control (37).
This point was addressed in part 2 of the Special Issue by a paper which overviewed the application of diabetes simulators - in particular those based on compartmental models - for use in the education of health-care professionals, students, patients and their relatives (39). The purpose of these tools is to create a learning environment for communicating and training intuitive thinking when dealing with insulin dosage, dietary and lifestyle adjustments. As Biermann and Mehnert (40) have highlighted such educational simulators are intended "neither to compete with normal diabetes education nor with dose adjustment algorithm programs, but rather to support these ... The increasing number of young persons with computer experience suggests an increasing acceptance of computer simulation and learning programs. This is particularly important where patient management of the disease is demanded". While the issues of supporting intuitive learning together with those of validating educational interventions were addressed - the relative paucity of evaluation data for such software was also highlighted as a severe limitation of such work and a research area which needs a great deal more attention and medical collaboration (39).
One particular simulator, called AIDA (Fig. 4a), was overviewed in greater detail (39). The software was offered in June 1996 without charge as a freeware program on the World Wide Web (41, 42). Thus far over people have visited the Web site where the program is stored - http://www.2aida.org - and to date over copies of AIDA have been downloaded gratis (Fig. 4b). This software represents one of an increasing number of diabetes-related programs which are becoming available, or being widely distributed, via the Internet. A version of this program, on diskette, with a printed, bound manual has also just recently become available to health-care professionals in the UK from the British Diabetic Association (London, UK) (43).
Chao (4) has also previously suggested that a proper evaluation of educational simulators could test patients for proficiency in the management of a given simulated test subject before and after programmed instruction [for example, similar to that described by Dammacco et al (44)] but with the addition of the simulator. Computers could evaluate the performance of the user by computing the number of hypoglycaemic reactions experienced by the simulated test subject, the average and standard deviations of the recorded BG levels, and the average and maximum amount of insulin administered during treatment; such quantitative analyses being used to (hopefully) demonstrate improvements in knowledge and ability. Comparisons of runs made before and after simulator-assisted instruction could also potentially be used to evaluate the effectiveness of this approach to education. Furthermore it has been suggested that the use of computers might remove some of the subjectiveness from the evaluation process and provide an impartial method of determining performance (4).
The issue of evaluating educational interventions was also addressed in the following paper in part 2 of the Special Issue by Brown et al (45) who overviewed Packy & Marlon (Raya Systems Inc., California, USA), an interactive computer game for the Super Nintendo™ platform, which is designed to improve self-care motivation and behaviour among children with diabetes. Packy & Marlon is a role-playing game in which players manage the diet and insulin of two elephants who have diabetes (Fig. 5). To optimise the educational benefits of this game players can select insulin plans which match their own. The scenario is a diabetes summer camp which has been raided by rats. The two elephants, Packy and Marlon, need to defend themselves by blasting the attacking rodents with peanuts and water from their trunks. They also need to find food and supplies - remembering to eat healthily, regularly check their BG levels, and take their insulin! Not only does this game provide children with a fun way to learn about diabetes, but it is also intended to aid dialogue between children with diabetes and their non-diabetic friends.
Packy & Marlon was assessed in two US centres in a 6-month randomised controlled trial, in a cohort of 59 insulin-dependent (type 1) diabetic children. Half the cohort received the diabetes game to use at home as much as they liked, while the other half (the control group) received a video game with no health-care content. While significant improvements in HbA1c were not demonstrated the authors quite correctly highlighted that the patients were reasonably controlled at the start (mean baseline HbA1c : 8.3-8.5%) and therefore quite possibly the study ran into a ‘ceiling effect’. Clearly to overcome this problem further randomised-controlled trials with diabetic children with more usual (poorer) glycaemic control would be required. Notwithstanding this, benefits were reported in diabetes self-efficacy, communication with parents about diabetes and self-care behaviour in the children who received Packy & Marlon. Also there was a decrease in unscheduled urgent doctor visits, a finding which if confirmed by larger studies would be most encouraging for children with diabetes.
The final paper in the education trilogy by Day et al (46) overviewed an alternative type of learning tool - a multi-media based educational package for patients, carers and health-care professionals. One of the strengths of this system lies in its breadth of coverage - a whole host of practical information about diabetes and nutrition being provided in an easy to access interactive manner on CD-ROMs (Fig. 6). As a valuable feature - rather than being ‘lectured’ by health-care professionals - patients themselves have been filmed on video recounting their own experiences with their diabetes and these clippings - 2 ½ hours in total - form one of the mainstays of the system. As an example of the sort of information stored, a patient recounting his own experiences of going hypoglycaemic is provided. Intuitively this may well be more informative and a lot more interesting for other patients than being told by a doctor or nurse what it may feel like to have a ‘hypo’. An example display from Day and colleagues’ Learning Diabetes system is shown in Fig. 6. ‘Maintaining the balance’ is a screen which allows the user to experiment with a combination of food, drink, insulin injections and exercise to discover the effect of such inputs on BG levels. Using a ‘drag and drop’ process they can build up a day with any combination of these inputs that they wish, and at the press of a button they can reveal the BG profile predicted to be produced (46).
It is not possible to cover in this review all of the 15 papers included in the Special Issues (5,6). Instead attention has been directed to the more clinically relevant articles - especially those focusing on databases, algorithmic-based decision support, education - and the clinical evaluation of such systems. However, the diversity of approaches reported in the Special Issues (5,6), effectively to solve a single problem - maintaining glycaemic control - is intriguing (47). Some researchers propose dose-by-dose adjustments, while others prefer longer-term visit-by-visit therapeutic interventions. Some prefer data-driven paradigms while others depend on model-based approaches. Some have tried to use qualitative advisors, others algorithms, and yet others causal probabilistic network (CPN)-based approaches. Some rely on hand-held devices for the patient whereas others make use of public telephone networks. Furthermore as the trend for treatment, mainly driven by financial forces, is now towards self-administered management - education becomes an increasingly important component of the care offered to diabetic patients.
Some clinicians who promote educational interventions have argued against a ‘crutch’ for patients - which might increase reliance on a technological device while decreasing their ability to self-manage. However it should be apparent that not all patients are the same. While some are motivated and interested and will take the time and trouble to learn more about their diabetes and truly understand how to adjust their insulin doses on the basis of SMBG data, and therefore could clearly benefit from further education, others appear to have no such interest and providing them with a solution whether it be in the form of a hand-held device, or telephone access to a central computer may be what they require.
Clearly there will always be patients who require greater freedom. However such increased flexibility brings with it a requirement for increased complexity which may or may not be acceptable or usable by the vast majority of patients. For example the HumaLink system (29) offers the possibility of patient flexibility but at the ‘price’ of requiring frequent telephone calls. Not every patient will wish to have to telephone a central computer after each BG measurement or before each insulin injection - but some clearly could benefit from such support. Looking to the future, with the miniaturisation of electronics and the extension of mobile telephone networks, it is not inconceivable to consider a day when a BG meter will incorporate a mobile telephone and be able to automatically dial a central computer, transmitting data after each BG measurement. The central computer’s advice could be transmitted back and shown on the BG meter’s display. By having the possibility of human review and intervention at the central computer - many of the concerns about patients acting solely on the basis of advice from a stand-alone machine may be mitigated. When there are many ways of doing something in medicine, none of them generally manage the job very well. In diabetes-computing there have been a whole plethora of prototype decision-support systems - but few if any successful products. While things may be changing, as should be apparent from the diversity of approaches reported in the Special Issues, it would be over-simplistic to believe that a single computational technique will work for all patients. Different data requirements and demands on patients mean that different methodologies may be more or less suited to individual patients. Each approach, however, must be formally validated and clinically evaluated. Achieving this in practice is not without its difficulties.
CALL FOR EVALUATION
The testing of IT prototypes remains the subject of much debate, not just in diabetes-computing, but in medical informatics generally. For example whether an evaluation model based on a drug trial is appropriate still needs to be established. As overviewed in the Computers in Diabetes’96 Meeting Conference Report (48), which can be found in part 2 of the Special Issue (6), when a CPN-based diabetes advisory system was tested in 12 patients it ended up being evaluated as a decision system rather than a decision support system. It is obviously far from ideal if the process of evaluation itself affects the proposed use of the system under evaluation. Similarly, to rigorously evaluate the HumaLink system in a ‘blinded’ fashion could be problematical. Getting a control group to telephone a dial-in service with no diabetes-related content a number of times per day could be difficult to sustain over a long period of time. Obviously it would be possible to have a control group who only use HumaLink as a data collection device - recording their BG measurements - but not receiving any advice back. Conceptually this might allow the added benefit of the semi-‘intelligent’ component of the system - the algorithms - to be evaluated separately. Although once again it needs to be ascertained whether patients would sustain the use of such a system long-term, with all its demands, without any reward or perceived benefit.
It all clearly depends on whether one wants to show that a particular computational approach does some good or whether it is sufficient to demonstrate that the whole IT process is good for diabetes care. If telephoning a central computer and typing in BG values can make patients think more about their diabetes and improve their glycaemic control - and this can be demonstrated in a range of randomised studies in a wide variety of patients in different centres long-term - this would be a useful intervention. This is regardless of whether it is the algorithms or the act of telephoning which are improving BG control. In this respect we should perhaps not focus solely on evaluating the technology but rather the process. After all we do not want to lose sight of what we are trying to achieve, namely an improvement in glycaemic control and patient care.
Piwernetz et al (49) commented six years ago in an Editorial in this journal that: "The biggest challenge for each system and the unquestioned prerequisite is, however, its evaluation. In the field of evaluation more work needs to be done. This requires an agreed methodology for scientific evaluation of the impact of computer systems on health care. Feasibility and implementation studies cannot simply follow randomized, double-blind, cross-over designs, it is essential to apply more complex protocols". This appears to be as true now as it was when written. Unfortunately not much practical progress seems to have been made in the past six years. In this respect one of the most pressing issues which taxed referees when reviewing submissions for the Special Issues (5,6) was in what way should such systems be evaluated and how should the results be presented. Cynics might argue that by not exclusively using randomised, blinded controlled trials we are simply trying to move the goal posts to suite the needs of our computer systems. This author disagrees. While randomisation can easily be achieved, and therefore should be undertaken - blinded, placebo controlled trials are not feasible in every area of medicine (e.g. surgery). Such ‘gold standard’ evaluation procedures also may not be ideally suited for assessing currently available decision-support prototypes. While purists may demand randomised, blinded controlled trials to be convinced of a system’s clinical utility, it needs to be recognised that such demands may be difficult to address, for reasons other than whether a system offers substantial clinical benefit. Given this, there is a very real risk that such demands could unnecessarily restrict the wider utilisation of prototypes which might otherwise have promising clinical potential. Nevertheless concerns over evaluation methodologies will need to be addressed, before software - such as that reviewed here - will see widespread utilisation and clinical acceptance.
So what conclusions can be drawn about the application of IT in clinical diabetes care, as of early 1997? BG meters and insulin pumps are well established tools. The use of database software is also widely accepted now. Furthermore, quite a wide variety of educational games and programs for patient use at home are now available commercially. Some of these have started to be formally evaluated - but many have not. Decision support research has not yet come to fruition although promising approaches are starting to be tested. In this respect, most of the papers in the Special Issues (5,6) do not propose, at present, to manage complicated cases using computer-based decision making tools - i.e without human input. Rather the consensus seems to be to focus on handling reasonably straight-forward clinical cases with a computer - freeing up more of clinicians’ and health-care professionals’ limited time to devote to the more difficult cases (47).
Looking to the future, most current decision-support prototypes provide a feedback loop which is closed only at discrete times rather than continuously. Non-invasive glucose monitoring could help to change this - offering potentially much more frequent data sampling which could revolutionise the provision of modern day diabetes care. Furthermore what many diabetic patients - especially teenagers - fear most are not the later life complications of their diabetes, but rather quite understandably going ‘hypo’. Hypoglycaemia alarms to alert patients to the possibility of nocturnal hypoglycaemia, although still at the research stage, could be another way in which computers may possibly be of direct benefit to patients in the future.
How can further advances in diabetes-computing be ensured? There is a very real need for greater co-ordination of endeavours in this field. Clinically a multifactorial approach is required, combining concerted efforts at tightening BG control with improved patient education. Computers can help with both these processes. However, as should be self-evident from the foregoing, on its own measuring BG will not achieve improved glycaemic control. Active interventions are required.
In conclusion, the way forward in decision-support and education in diabetes care is likely to be through integrated IT developments built on collaboration. Special attention will also need to be devoted to evaluation issues. Although the Medical Informatics Special Issues (5,6) should not be considered as in any way comprehensive in their coverage of clinical diabetes computing - it is hoped that the compilation of papers provided there - along with this critical appraisal - may offer a useful source of novel ideas - as well as, perhaps, a starting point for future research.
The author thanks Professor Enrique Gomez (Madrid, Spain), Dr. Michael Albisser (Miami, Florida, USA), Steve Brown (Raya Systems Inc., California, USA), and Dr. John Day (Ipswich, UK) for their kind permission to utilise figures / data from their Medical Informatics papers. Very special thanks are also extended to Taylor & Francis, publishers of Medical Informatics, for their kind support of the diabetes-computing Special Issues - further details of which can be found here on the World Wide Web.
1. Banting F.G ., Best C.H., Collip J.B., Campbell W.R., Fletcher, A.: Pancreatic extracts in the treatment of diabetes mellitus: a preliminary report. Can. Med. Assoc. J. 12: 141-146, 1922.
2. Banting F.G ., Campbell W.R., Fletcher A.A.: Further clinical experience with insulin in the treatment of diabetes mellitus. Br. Med. J. 1: 8-12, 1923.
3. Banting F.G., Best C.: The internal secretion of the pancreas. J. Lab. Clin. Med. 7: 251-266, 1922.
4. Chao S.C.E .: A recording simulator for teaching the self-management of diabetes mellitus. MSc Thesis. University of Toronto, Canada, 1990.
5. Lehmann E.D. : ed. Special Issue: Application of information technology in clinical diabetes care. Part 1. Databases, algorithms and decision support. Med. Inf. 21: 255-378, 1996.
6. Lehmann E.D. : ed. Special Issue: Application of information technology in clinical diabetes care. Part 2. Models and education. Med. Inf. 22: 1-120, 1997.
7. Lehmann E.D. , Deutsch T.: Application of computers in diabetes care - a review. I. Computers for data collection and interpretation. Med. Inf. 20: 281-302, 1995.
8. Lehmann E.D. , Deutsch T.: Application of computers in diabetes care - a review. II. Computers for decision support and education. Med. Inf. 20: 303-329, 1995.
9. Lehmann E.D.: Application of information technology in clinical diabetes care - a Special Issue. Part 1. Databases, algorithms and decision support [Editorial]. Med. Inf. 21: 255-258, 1996
10. 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. N. Engl. J. Med. 329: 977-986, 1993.
11. The Diabetes Control and Complications Trial Research Group.: Resource utilization and costs of care in the Diabetes Control and Complications Trial. Diabetes Care 18: 1468-1478, 1995.
12. Harris M.I .: Testing for blood glucose by office-based physicians in the U.S. Diabetes Care 12: 419-426, 1990.
13. Eastman R.C ., Siebert C.W., Harris M., Gorden P.: Implications of the Diabetes Control and Complications Trial. J. Clin. Endocrinol. Metab. 77: 1105-1107, 1993.
14. Moy C.S .: ed. Diabetes 1993 vital statistics. New York, American Diabetes Association, 1994, pp. 43.
15. King’s Fund Policy Institute .: Counting the cost: the real impact of non insulin dependent diabetes. British Diabetic Association, London, 1996.
16. EURODIABETA .: Information technology for diabetes care in Europe: the EURODIABETA initiative. Diabet. Med. 7: 639-650, 1990.
17. Lehmann E.D. : The Diabetes Control and Complications Trial (DCCT): A role for computers in patient education? Diab. Nutr. Metab. 7: 308-316, 1994.
18. WHO/IDF Europe.: Diabetes Care and Research in Europe. The Saint Vincent Declaration. Diabet. Med. 7: 360, 1990.
19. Gorman C ., Looker J., Fisk T., Oelke W., Erickson D., Smith S., Zimmerman B.: A clinically useful diabetes electronic medical record: Lessons from the past; pointers toward the future. Eur. J. Endocrinol. 134: 31-42, 1996.
20. Flack J.R. : The impact of information technology on diabetes patient care. Diab. Nutr. Metab. 9: 145-159, 1996.
21. Piwernetz K., Home P.D., Snorgaard O., Antsiferov M., Staehr-Johansen K., Krans M., for the DIABCARE Monitoring Group of the St. Vincent Declaration Steering Committee.: Monitoring the targets of the St. Vincent Declaration and the implementation of quality management in diabetes care: the DIABCARE initiative. Diabet. Med. 10: 371-377, 1993.
22. Kopelman P.G., Sanderson A.J.: Application of database systems in diabetes care. Med. Inf. 21: 259-271, 1996.
23. Engelbrecht R., Hildebrand C., Brugues E., de Leiva A., Corcoy R.: DIABCARD - an application of a portable medical record for persons with diabetes. Med. Inf. 21: 273-282, 1996.
24. Gomez E., del Pozo F., Hernando E.: Telemedicine for diabetes care: the DIABTel approach towards diabetes telecare. Med. Inf. 21: 283-295, 1996.
25. Bellazzi R ., Cobelli C., Gomez E., Stefanelli M.: The T-IDDM project: Telematic management of Insulin Dependent Diabetes Mellitus. In: Bracale M., Denoth F. (Eds.), Proceedings HT-95, Ischia, 1995, pp. 271-276.
26. Fallucca F ., Di Biase N., Sabbatini A., Borrello E., Sciullo E., Napoli A.: Telemedicine in the treatment of diabetic pregnancy. Pract. Diab. Intl. 13: 115-118, 1996.
27. Rosenfalck A.M., Bendtson I.: The Diva System, a computerized diary, used in young type 1 diabetic patients. Diabete Metab. 19: 25-29, 1993.
28. Lehmann E.D ., Deutsch T.: Computer assisted diabetes care: a 6 year retrospective. Comput. Methods Programs Biomed. 50: 209-230, 1996.
29. Albisser A.M ., Harris R.H., Sakkal S., Parson I.D., Chao S.C.E.: Diabetes intervention in the information age. Med. Inf. 21: 297-316, 1996.
30. Albisser A.M.: Intelligent instrumentation in diabetic management. CRC Crit. Rev. Biomed. Eng. 17: 1-24, 1989.
31. Day J.P .: Some considerations of legal liability concerning the use and future development of knowledge based or expert systems in diabetes care. Diab. Nutr. Metab. 8: 195-200, 1995.
32. Brahams D ., Wyatt J.: Decision-aids and the law. Lancet ii: 632-634, 1989.
33. Skyler J., Skyler D., Seigler D., Sullivan M.O.: Algorithms for adjustment of insulin dosage by patients who monitor blood glucose. Diabetes Care 4: 311-318, 1981.
34. Worth R., Home P.D., Johnston D.G., Anderson J., Ashworth L., Burrin J.M., Appleton D., Binder C., Alberti K.G.M.M.: Intensive attention improves glycaemic control in insulin-dependent diabetes without further advantage from home blood glucose monitoring: results of a controlled trial. Br. Med. J. 285: 1233-1240, 1982.
35. Daneman D. , Siminerio L., Transue D., Betschart J., Drash A., Becker D.: The role of self-monitoring of blood glucose in the routine management of children with insulin-dependent diabetes mellitus. Diabetes Care 8: 1-4, 1985.
36. Coonrod B.A ., Betschart J., Harris M.I.: Frequency and determinants of diabetes patient education among adults in the US population. Diabetes Care 17: 852-858, 1994.
37. Assal J.P.: The Diabetes Education Study Group of the European Association for the Study of Diabetes. 15 years devoted to improving patient management. Medicographia 17: 15-20, 1995.
38. Assal J.P., Ekoe J.M., Lacroix A., Liniger C.: Teaching patients about their disease and its treatment: therapeutic success, professional failure. In: Sakamoto N., Alberti K.G.M.M., Hotta N. (Eds), Recent trends in management of diabetes mellitus, Excerpta Medica, Amsterdam, 1987, pp. 63-74.
39. Lehmann E.D .: Interactive educational simulators in diabetes care. Med. Inf. 22: 47-76, 1997.
40. Biermann E. , Mehnert H.: DIABLOG: a simulation program of insulin glucose dynamics for education of diabetics. Comput. Methods Programs Biomed. 32: 311-318, 1990.
41. Lehmann E.D .: Diabetes moves onto the Internet. Lancet 347: 1542, 1996.
42. Lehmann E.D ., Deutsch T., Broad D.: AIDA: an educational simulator for insulin dosage and dietary adjustment in diabetes. 1996. Available from: http://www.2aida.org
43. Lehmann E.D. , Deutsch T., Broad D.: AIDA: an educational simulator for insulin dosage and dietary adjustment in diabetes. British Diabetic Association, London, 1997.
44. Dammacco F ., Frezza E., Torelli C., Pacifico A., Buono A.: Comparative study of teaching methods for diabetic children: superiority of computer-assisted instruction. In: Sakamoto N., Alberti K.G.M.M., Hotta N. (Eds), Recent trends in management of diabetes mellitus. (eds.), Excerpta Medica, Amsterdam, 1987, pp. 610-613.
45. Brown S.J. , Lieberman D.A., Gemeny B.A., Fan Y.C., Wilson D.M., Pasta D.J.: Educational video game for juvenile diabetes: results of a controlled trial. Med. Inf. 22: 77-89, 1997.
46. Day J.L ., Rayman G., Hall L., Davies P.: "Learning Diabetes" - a multi-media learning package for patients, carers and professionals to improve chronic disease management. Med. Inf. 22: 91-104, 1997.
47. Lehmann E.D. : Application of information technology in clinical diabetes care - a Special Issue. Part 2. Models and education [Editorial]. Med. Inf. 22: 1-3, 1997.
48. Lehmann E.D .: Computers in Diabetes’96. Med. Inf. 22: 105-118, 1997.
49. Piwernetz K., Massi Benedetti M., Bauersachs R., Sönksen P.H.: Foreword. Diab. Nutr. Metab. 4 (Suppl. 1): 1, 1991.
Return to Top of Page If you like AIDA, why not display our logo on your home page? For more information about linking to the AIDA Website please click here.
Return to AIDA Website Home Page AIDA is a freeware diabetes 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 9th December, 2000. (c) www.2aida.org, 2000. All rights reserved. Disclaimer. For the AIDA European Website, please click here. For the Diabetes / Insulin Tutorial, please click here.