Information technology and computer-based decision ... - ScienceDirect

Computer Methods and Programs in Biomedicine, 32 (1990) 179-188 Elsevier

179

COMMET 01099

Information technology and computer-based decision support in diabetic management E.R. Carson 1,2, S. Carey 2, F.E. Harvey 2,1, P.H. Sonksen 2,1, S. Till 2 and C.D. Williams 2 t Department of Systems Science, Centrefor Measurement and Information in Medicine, City University, London ECI V OHB, U.K., and 2 Department of Endocrinology and Chemical Pathology, UMDS, St. Thomas's Hospital, London SEI 7EH. U.K.

This paper describes the application of computer-based techniques within an intelligent, knowledge-based framework to the management of diabetes. The objectives are to structure data collection and storage so that the relevant patient-specific data are collected and made accessible as needed, and to provide clinical decision support on either a day-by-day or longer timescal¢ as appropriate; these objectives rela~ig to both hospital c~-,ic aad general practice. For longer-term management, a prototype rule set (> 500 rules) has been developed (coded in Sigma PROLOG), validated and tested on patient data. The data collection programs (written in SCULPTOR) to feed the ruleset have been tested in the hospital clinic and compared with the resident data collection system for usability, and impact on the running of the clinic. Links between the data collection programs and the ruleset program have been written and tested. The computer system will also incorporate a module, combining knowledge-based advisory system and glucose/insulin model as patient simulator, that can be tested as a potential decision aid for adjusting insulin dosage on a daily basis. Diabetes mellitus; Decision support system; Patient data collection system; Diabetic expert system

1. Introduction Diabetes mellitus is one of the major non-communicable chronic diseases in Western societies, with an estimated incidence of more than 2~ in the U.K. population. It is the major cause of blindness in people under 65 years of age, and results in a range of complications which affect the circulatory system. In terms of patient care it is the management of the disease which dominates, since once diagnosed it is a life-long condition. Good management requires a reliable patient record to monitor the state of the disease, regular screening so that complications can be detected Correspondence: Ewart R. Carson, Department of Systems Science, Centre for Measurement and Information in Medicine, City University, Northampton Square, London EC1V 0HB, U.K.

whilst there is still time to effect successful treatment, and appropriate supervision of glycaemic control. In this paper the role of information technology as it is being developed and applied in the management of diabetes at St. Thomas's Hospital, London is described. There are two main objectives, first to structure data collection and storage in order that the relevant patient-specific data are collected and made accessible as and when necessary; this is important in both the hospital clinic and in general practice. The second objective is to provide clinical decision support for both hospital and general practice, on either a day-by-day or longer-term timescale as appropri,ae. Three inter-related dimensions of information technology have been employed m this programme of work. The first is the oevelopment of a data collection and data management system. Two

0169-2607/90/$03.50 © 1990 Elsevier Science Publishers B.V. (Biomedical Division)

180

implementations of this are described. The second of these has been linked to a large rule-based expert system. This is the second dimension, a rule-based system which provides guidelines for the management of the diabetic patient during a visit to the hospital clinic for tests and consultations. The third dimension focuses upon decision support for the adjustment of insulin therapy. This component combines a rule-based module for advising on modifications to the insulin regimen coupled to a glucose/insulin simulation model. This third dimension, which is described briefly below, is the subject of the paper by Deutscb et al. elsewhere in this issue [3]. Overall the systems being developed are intended to provide guidance on which data need to be collected in the care of diabetic patients, and explicit advice, based on these data, for the appropriate management of individual patients.

2. Data collection system - I The development of a computerised database for diabetic records at St. Thomas's Hospital began in the early 19708. Since 1974 structured questionnaires have been used to record the clinical details of diabetic patients. At that time such records were batch.entered into a mainframe computer. Petient

:

CB No : z2222

Smith.John

TypP of visit : F o l l o w up C l i n i c date : Age = Seen by = Weight ( k s ) : V.A. Right eye: L e f t eye : Blood pressure= Urine Glucose : Ketones : Protein : Blood : Blood Glucose : Hba! : Treatment type:

30/ 1/86 40 PHS 78 6/6 6/6 170/80 0 0 0 0 8.9 0.9 Insulin

TAB f o r next f i e l d .

From 1983, however, they have been entered into a multi-user MicroAPL Spectrum microcomputer by doctors interactively during diabetic consultations. The records are written in APL, the system running under the 'Mirage' operating system. The structured linear questionnaire contains 224 data items the majority of which are coded. Free text in fixed-length format is included to provide for registration details, clinical problem lists and details of medication prescribed. The design of the system involved close collaboration between physicians and programmers with the result that a high degree of user acceptability has been achieved. For example the number of key strokes required for data entry has been reduced by incorporating the selection of options from lists, by making extensive use of default values and by the adoption of a hierarchical branching structure to the questionnah'e. Instructions and help messages are included on each data entry screen (see for example, Fig. 1). The data entry programs include error traps in the form of logic checks, range checks and limited permissible options. This system enables the records of patients new to the clinic to be entered on-line at the time of their first visit. An immediate print-out of the first visit report is produced with a copy being filed in conventional patient notes and a copy being given

Follow up 12/ 5/87 42 CDW 80

0.8 lO.! insulin

F! F2 reg f u l l

Fol low up | 8/ 5/89 44 PHS 79 8/6 8/6 160/160 0 0 ! 0 It .5 q.7 Insul ~n

F4 F5 F6 F? manage s u r ~ a r y Its= l e t t e r

DOB =

2/ 3/1945

Fol low up I 7 / ! 0/8q 44 PH$ 78

6.8

9.9 Insul in

F8 F9 F10 o t h e r next ma~n

Fig. 1. Data entry screen from APL data collection system.

181 print total 2~0: 230: 220: 2]0~ 200~ 190~

htst 4533

dla~a~e records O O0 0O0(3 00000 00000000 00000000 00000000 0 00000000 O0 00000000000

180:

170: 160~

OOOOOOOOOOOO

150:

140: 1301 120~ II01 I00: 90: 80: 70:

0000000000000

OOOOOOOOOOOOOOOO

0 0

O 00000000000000000 0 OOOOOOOOOOOOOOOOO 0 O00000000000000OO 0000000000000000000

0 t) 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 O()O00OOo000000{}O00000000000 60~ t)t)O (JO0oO000000000000000000000000 .r~O ', OO(K)O00OO0000000000000(}OOO000000OO | 0 ', (100( ~0(~(}t)OO(~O0000000000OO000000(.)O00( ~(I0 3 0 ', 00( It)()(lO00000f}{.)O0 O( i'O0000000000r.}O()O O 0 0 0 0 O 20 ', I}{)()(R){)t)()(I{)f)I)(}UO(IOOOOO(JI)I.)(}f)OOO{)OOOOOOO(}OEIU 10 '.()()O()tl()O()O0000(}O0000(}O( )(}OOO000{3UOOOOOD(}(}O000 t~ ....................................................

OO

!

I I l I l 2 ' - ' 2 2 2 3 3 3 3 3 | 4 14 I 5 5 5 5 5 6 6 6 6 6 7 7 7 7 7 8 8 8 8 8 9 9 ' J 9 9 0 "_"|fiH(I21F, HO2 4GRtq2 4 6 H 0 2 1 6 8 0 " 4 b 8 0 2 4681)" 1 6 8 0 2 4 6 8 0 2 1 6 8 0

Fig. 2. Example of statistical analysis of data using the APL system. The histogram indicates the distribution of the 4533 patient records according to the age at which diabetes was diagnosed.

to the patient for transmission to their general practitioner. In a similar fashion, there is on-line entry of records of patients at follow-up clinics with again immediate print-out of the follow-up visit report. In addition to improved communications, both within the hospital and with the general practitioner, the system has also allowed implicit and explicit algorithmic-based prompting for scheduled screening procedures, e.g. visual acuity, blood pressure, urine analysis, and the collation of a database containing chronological records of over 4500 patients [5]. These data can be easily retrieved using a uniquely flexible interactive interrogation procedure and used for clinical audit and epidemiological research [6,7]. An example of the type of statistical analysis that can be performed is shown in Fig. 2.

3. Data collection system - I|

A further dimension in the application of information technology was the development of the rulebased expert system to provide advice to the clinician in the consultation with the diabetic patient. With a view to linking this rule-based system

(described below) to the data collection system, there was the need to ensure that the necessary data were being collected to meet its requirements in both initial and follow-up visits. Anah'sis revealed that the following necessary data for follow-up consultations were not being collected with the existing APL system: (1) medication taken by the patient immediately prior to the consultation. This current medication may frequently differ from that prescribed at the last appointment; (2) physical examination findings of eyes and feet; (3) investigations ordered at the last clinic visit; and (4) results of investigations ordered at the last visit. For the linked data collection system/rulebased system, a new version of the data collection system was written using the database management system SCULPTOR (Microprocessor Developments). This fourth-generation language runs under UNIX, maintains a keyed-file type database and provides appropriate screen entry and database reporting facilities. As well as having the aim of speeding up development, the use of SCULPTOR offered potential advantages in relation to having the diabetic system installed in general practice. ABIES, a widely used general practice software package (Abies Informatics) was also written in SCULPTOR and hence the specialist diabetic software could be linked more easily to the general purpose software, thereby enhancing its prospective marketability. The file structure of this SCULPTOR system was revised as compared to that of the APL system to allow a larger set of data to be stored chronologically, corresponding to successive follow-up consultations. The data are held in the following nine files: patient register; "full review care'; routine tests; eye examination; foot examination; treatments; problems; investigations ordered; and results of investigations [5].

4. Development of the rule-based expert system

Having devised an appropriate means of collecting and storing patient data, these data can be

182 processed and interpreted within an expert system t h a t h a s b e e n d e s i g n e d to assist t h e clinician in the m a n a g e m e n t o f a p a t i e n t c o n s u l t a t i o n . Initially t h e r u l e - b a s e d s y s t e m h a s b e e n d e v i s e d f o r u s e in t h e o u t - p a t i e n t clinic, b u t with t h e o b j e c t i v e that, in m o d i f i e d f o r m , it m i g h t b e a d o p t e d in g e n e r a l practice.

The elicitation of the knowledge which was incorporated into the rule-based expert system t o o k p l a c e o v e r a p e r i o d o f 2 y e a r s in a n e x t e n s i v e series o f s e s s i o n s i n v o l v i n g d i a b e t o l o g i s t s f r o m St. Thomas's and Guy's Hospitals, London, and with g e n e r a l p r a c t i t i o n e r s . D e t a i l s o f the p r o c e s s o f d e v e l o p i n g t h e r u l e s e t a r e given in W i l l i a m s et al.

TABLE 1 Ruleset areas and sub-areas (definition sub-areas in quotes) (adapted from [5]) Sub-area

Area

1. Content of visit

2. Potential diabetic emergencies (PDE) 3. Weight and glycaemic control 4. Treatment of noninsulin-dependent diabetes mellitus 5. Detection of complications

6. General advice

1:1 1 :2 1 :3 1 :4 1:5 1 :6 1:7 2:1 2 :2 2 :3 3:1 3:2 4:1

Eye examination Foot examination Blood pressure measurement Urinalysis, serum creatinine Check home monitoring results Weight measurement Blood glucose and HbA1 Subsequent data prompting ' Poor clinical condition' PDE management advice Glycaemia Body weight

5:1 5:2 5:3 5:4 5:5 5 :6 6 :1 6: 2 6 :3 6 :4 6: 5

' Retinopathy' ' Peripheral neuropathy' ' Peripheral vascular disease' 'Diabetic foot problem' ' Nephropathy' ' Hypertension' Foot care Intermittent claudication Treat hypertension Pre-hypertension warning Home blood glucose monitoring indications Indications for home urine testing

6 :6 7. Ordering investigations 8. Referrals

9. Next appointment Total

No. of rules 6 8 6 9 4 1

2 5 2 21 26 7

36

28 33 300

18 5 1

5 7 22

58

1

2 2 1

5 4

7:1

8:1 8 :2 8: 3 8: 4 8 :5 8:6 8:7 9:1

Total

15 3

Ophthalmologist Hospital Diabetologist Chiropodist Vascular surgeon Nephrologist Specialist nurse and dietitian

13 2 7 8 2 3 10

45 23 541

183 [5]. The ruleset is implemented in Sigma PROLOG.

4.1. Structure of the ruleset The component areas of the ruleset domain which eventually evolved were [5]: (1) recommended content of the consultation with the patient: - prompts for routine tests and scheduled screening procedures; (2) detection of possible diabetic emergencies; (3) determination of absolute categories and trends in body weight and glycaemic control; (4) adjustment of oral agents and diet in non-insulin-dependent patients; (5) detection of diabetic complications: - scanning recorded data and problem list for complications, examination of eyes, feet and urine; (6) general advice: - foot care, intermittent elaudication, hypertension treatment, recommended type of home monitoring; (7) recommended investigations; (8) referral to other agencies; (9) timing of the next visit to the clinic; (10) recommended content of the next visit. The various sub-areas which constitute the total ruleset, together with the number of rules contained in each, are presented in Table 1. The logical progression from data collection, through the use of definition rules and models and through management rules, to the generation of advice is shown in Fig. 3.

4.2. Example of rule structuring and progression An example of one of the definition rules in the ruleset is: Protein in urine if: Albustix greater than or equal to 1 + and patient not both female and menstruating. This 'definition rule' defines 'protein in urine' using Lata transferred from the data collection system, all these historical data concerning the patient under review being referred to as 'signal data'. In some instances only partial fulfilment of

:tion

rne~

Fig. 3. Schematic representation of the ruleset showing the logicalprogressionfrom data collectionto advicegeneration.

a rule's conditions may result in requests for data that are not routinely prompted for by the data collection system; data referred to as 'subsequent' data. In this example, the presence of 1 + protein in the urine of a female patient resulted in the ruleset query 'Is the patient menstruating?'. The subsequent data entered in response remain local within the ruleset. Signal data, subsequent data and medium level terms may form the condit;ons of subsequent definition rules at the next 'level' in the chain. For instance: Intermittent proteinuria if: 'protein in urine' in some but not all of last three tests and at least one positive test has proved sterile. The state of 'intermittent proteinuria' may be superseded in time by 'persistent proteinuria': Persistent proteinuria if: 'protein in urine' for the last three tests and overall time more than 6 months and no test has been infected. The presence of 'persistent proteinuria' with signal data satisfies the conditions of an 'outcome' rule at the final level:

184

Order serum creatinine measurement if: persistent proteinuria

and no creatinine result recorded in the last 3 months. Thus the rules are arranged in chains or strategies where signal and subsequent data trigger a series of definition rules culminating in practical advice offered by one or more outcome rules. At every consultation advice is offered in each outcome area of the ruleset, and in each sub-area. This ruleset does not reach a single conclusion, therefore, but several which are mutually exclusive and often negative. In diabetes, as in the management of many other chronic diseases, several aspects of patient management continue in parallel. Chains of rules in different outcome areas are not entirely separate and frequently inter-connect. For example, a single medium-level term may be used by several outcome rules. Similarly, a single term of signal data may be used in more than one outcome area. 4.3. implementation and user interface

The ruleset as described above was implemented in Sigma PROLOG. The APES (Augmented Prolog Expert System) shell was used to provide normal expert system functionality including explanation facilities, user interaction, and natural language templates. The interface comprises a series of menus corresponding to each of the ten phases described in Section 4.1. These menus display all of the expert system advice relevant to the current phase. The user is allowed to accept, reject or query any of the advice displayed. If a piece of advice is rejected, the user is required to provide a valid alternative and an explanation as to why the expert system advice was rejected. This is recorded for future analysis. If a piece of advice is questioned, an APES query is generated and executed. The APES shell then handles the interaction until the query is finished. The user interface also allows queries on 'negative advice', e.g. why a patient has not been referred to an ophthalmologist.

At the end of each phase, the current (possibly amended) advice is accepted or rejected by the user and the expert system proceeds to the next phase. A number of reports are also produced by the interface. These include a simple summary of all the advice for a particular patient, a log of any disagreement between the user and the system, and a log of the time taken to execute each phase of the system. In addition, all the ad~.'ice given can be reviewed at the end of a consultation through another set of menus that only permits querying of advice and not its acceptance or rejection. Entry to and exit from the expert system are via a main menu that allows the user to select the required function. 5. Validation and evaluation

An extensive programme ol' validating and evaluating the enhanced diabetic care system involving the components as described above has been carried out and is continuing. Three components of this study are" (1) a technical evaluation of the data collection systems and the rule-based expert system; (2) an operational study of the implementation of the data collection system in the weekly follow-up diabetic out-patients clinic; and (3) validation of the expert system ruleset. 5.1. Technical evaluation

The technical evaluation of the data collection systems and the rule-based expert system included assessing factors such as speed of operation and the number of keystrokes required per patient consultation. In terms of data collection systems, this study allowed comparison of the APL and SCULPTOR implementations. In this way an appraisal could be made of the costs and benefits of changing from the tried, tested and successful APL clinical records system to the SCULPTOR version which, when linked to the expert system, offers additional decision support in the out-patient management of a chronic disease. Sample results are shown in Figs. 4-8, with a more extensive analysis being reported in [5]. Fig. 4 depicts the overall time taken for each run of the

~'~

2oo{

$

80C

tn

70C

~

600 !

'.-

500

185

Q

q, . j ~ - " ¢ * @

@

100

400

,

30O

'

i

__~L_

J

J

7

N u m b e r of 13Qst v i s i t s

:.Z

Fig. 4. Total expert system (ES) run-time as a function of

number of past visits (adapted from [51).

23 77 *

/ '

.

.

~

[-3 Content ot tn,s v,slt ~:~ Potent,el EhobehC emergenoes [ ] Ccj/c,j!OtlOrl o| wor~ng del,n,boms f~:~].l, 84°/o{:3 Diobetl¢ COmphcotIOn5 s(reenil~g

*/o

.

/~

11 7 6 "/e

~

/ ~ 1 4 14°1.

Order,rig ,nvesttgct~o~

~

ot ne~t vls,t Content of next v,sit

6 31 * / o ' ~ _ _ ~ _ _ _ _ / / 377°/*

12

76°,*

Fig. 5. Proportion of time spent in the main functions in using the total expert system (ES) (adapted from [51).

~

o

#.~ 2°°/ie~S° e® ~ 6

•

8 , 9

, 12

_L 15

, _

18

t

21

,

24

,

,

27

30

Number of I:Xmt VI,~,ItS

Fig. 6. Pure expert system (ES) calculation time as a function of number of past visits (adapted from [51).

4 35%

~ • ,: ;:::,z 4o13,,o~ ;!: ~,';:J'::'/

I

I

I

I

I

3

6

9

12

15

18

_

1

1

21

24

27

30

NurrdDer o t post w s i t 5

of

number

2094°,'o

}'

71

, 3

"

I

Fig. 8. Time to load new patient data as a function of past visits (adapted from [5]).

X

1 0 0 t ._ 0

I" O/ 0

54 % T,me to i ef~o',e otd pcbent d~un

/

E3 I,me ~o m a c pct,ent aotc Pure E S c~l~ji~t,oI, t,ne

T~me speDt in , r t e r l ~ ¢ e menus

expert system as a function of past visits on record for each patient. F r o m the regression analysis carried out, the average expert system consultation was shown to be 5 min 40 s plus 13 s for each past visit of the patient on record. The proportions of time spent in each of the main expert system functions are shown in Fig. 5, indicating that only 43% of the total run time is spent executing the ruleset. Fig. 6 defines the pure expert system calculation time as a function of the number of past clinic visits. This represents the time that it would take to run the ruleset if all the conclusions made by the expert system were atttomatically ratified without reference to the user. The proportion of time spent in each of the ruleset areas is depicted in Fig. 7. The ruleset areas which increased execution time significantly as a function of number of past visits were: content of this visit; content of the next visit; and calculation of working definitions. The other areas did not result in execution times that varied significantly with increase in number of past visits. This is intuitively reasonable as it distinguishes between those ruleset areas relying heavily on historic data and those which rely mainly on current results. The other function that increased significantly with the number of past visits was the time to produce and load new patient data, as shown in Fig. 8.

5.2. Operaiional stud), of the implementation of the data collection system in the diabetic follow.up clinic Fig. 7. Proportion of

time spent in different areas of the ruleset (adapted from [51).

Over a period of several months the APL and S C U L P T O R systems have been used in parallel in

186 -

-I

so

PI APL

verslon

~2ci: .

~

,

5.3. Validation of the ruleset

i~l S c u l p t o r ver-51on

-II

,

1330 ]"(30 1430 1500 1530 1800 1630 1700 Consuitat,on stQrt,ng t,me

Fig. 9. Percentage computer usage during the operation of the clinic (adapted from [5]).

the weekly diabetic follow-up clinic. Appropriate experimental designs described in [5] were adopted in order to be able to evaluate these data collection systems in operation, Measurement was made of: (1) the reliability of the data collection system; (2) the effects of the data collection system on the dynamics of the clinic; (3) the reactions of clinicians using the system; (4) the effects in terms of degree of stress and arousal involved in the patients; (5) the effects on the nature and style of consultations; and (6) the accuracy and completeness of the data recorded. A detailed account of results obtained is contained in [5]. One typical effect is shown in Fig. 9. This shows that whereas during the early part of the afternoon clinic there is an almost equally high use of both computer-based data collection systems, as the afternoon progresses the use of the SCULPTOR system fell away substantially. This stemmed from the longer time taken for the consultation with the SCULPTOR system leading to consultations becoming progressively later than the times at which they would have occurred on the basis of a mean consultation time of 15 min. Neglect of the SCULPTOR system was perceived as being one means of completing the scheduled patients' consultations within the time allotted for the clinic. Analysis of patient questionnaires revealed no detectable effect on reported levels of patient stress and arousal when using the SCULPTOR system as opposed to the routinely used APL system of data collection.

The ruleset was validated by quantifying: (1) the degree of consensus with the ruleset amongst expert diabetologists; (2) the agreement between management decisions made by doctors and management advice offered by the expert system using the same recorded data; and (3) the effect on expert system output of incomplete signal data. Circulation of the ruleset, apart from those rules concerned with dosage adjustment of oral hypoglycaemic agents, to ten consultant diabetologists in the U.K. elicited responses indicating a very substantial agreemer~" with the content and structuring of the ruleset [5]. The areas where most disagreement was apparent, indicating a possible need for rule classification or refinement, related to: level of glycaemic control, 'retinopathy', 'hypertension', treatment of hypertension and referral to ophthalmology. Preliminary analysis of the expert system recommendations with patient management actions taken by doctors indicated an overall level of agreement of 84%. These results were obtained from a small sample (28 cases) which clearly needs to be expanded, as is there the need to compare expert system recommendation and doctor action by making use of an arbitration panel of experts. Nevertheless the study does highlight the potential clinical applicability of such a system.

6. Insulin dosage adjustment The expert system described above i-a its current version incorporates advice on dosage adjustment of oral hypoglycaemic agents but not insulin. The development of an advisory system for daily insulin adjustment is the third dimension of the application of information technology to diabetic management. Initially this insulin advisory system is incorporated in a stand-alone 'metabolic prototype', the concept of which is depicted in Fig. 10. It is planned that this insulin advisory system will be incorporated in the next-generation 'larger' expert system.

187

edge-based insulin advisory module and in a simulation model for examining the proposed effects of advise on insulin or diet therapy, or predicted glycaemic control. In the knowledge-based advisory module, a qualitative reasoning methodology is adopted, extending earlier work by Deutsch et al. [3]. This uses the blood glucose measurements and the times of occurrences of hypoglycaemic episodes as indicators of the patient's carbohydrate metabolism and suggests alternative control actions related to diet. and/or the insulin component of the complex diabetes therapy. The advice is given in qualitative terms indicating the direction of adjustment(s) needed in order to improve the quality of glycaemic control. The reasoning method is based on a ruleset encapsulating the knowledge about insulin pharmacodynamics and the effect of food absorption on glucose metabolism. This advisory module has been implemented, and partly tested, on a PC-AT using MicroPROLOG and APES. Full details are given elsewhere in this volume [3]. The simulation module which can then be used to examine the effects of the suggested changes in

PATIEP~TtL } SPECIAL il INSULIN "DIETARIy EVENT GLUCOSE II INJECTION INFORMATION INFORMATION INFORII~TIONI I

B'.000

F

il

. . . .

COMPUTER r-

. . . . . . . . . . . . . . . .

T

]

fSUGGEST ALTERATIONS t I N PATIENT'SREGIME iTO STABILISEBLOOD GLUCOSEPROFILE

I

I I

I SUGGESTED SIMULATEEFFBCTS EFI~EI OF J D .E'I'EIULTION I ALl rr's BLOOD ON PATIENT'S GLUCOSEPROFILE. PROFI

I I

ii

Fig. 10. Metabolic prototype - the concept (from [4]).

The concept of the metabolic prototype involves establishing a patient database which combines blood glucose information, insulin injection information, information on special events (e.g. times of hypoglycaemic episodes and dietary information. This database is then used in a knowl-

DIREUTDATAENTRY

[

i .,cRoo, T I T|ONALDATA.] I

O0'A 1

,F--,'

,

i , ,,,

t P~u,,E~, l A l I E N I . ~'- . . . . . . . .

-]

,

H o .oooL 1

UEDICATEODATABASE

......................................

I

i'

]

It,,,

GRAPHIS '~ .oooLE ]

/

/

/[

f. . . . . . . . . . . . . . . . . . ,

i !

" I

"1 ~ . " - ~ l -I"

~ 1 ..... I ~

,.1=

W I UMERC | AL

J

f

I

-[

~RCH I VE • IRAI.;E OF

I SUGGES~lq,ONS

,

,,

I ~"~""

Fig. 11. Metabolic prototype '1990' (from [41).

']

188 insulin a n d / o r diet on the patient's blood glucose incorporates a non-linear model of glucose and insulin dynamics [1]. Current developments of the metabolic prototype, incorporating advisory and simulation modules, are leading to the configuration as shown in Fig. 11. Here KBS corresponds to the advisory module, whilst CAMIT and M I C R O D I E T are data collection and food analysis systems respectively [3]. The 'figure-of-eight' loop indicates the way in which advisory module and simulation could be used in an iterative manner until suggestions are provided which result in the stabilisation of the patient's blood glucose profile.

7. Conclusions This paper has described a role of information technology as it is being applied in order to enhance information management and to provide decision support for diabetic management at St. Thomas's Hospital. The first data collection system, implemented in APL, is being successfully apphed in the routine diabetic clinic since 1982. It has been shown to be welcomed by clinical staff and patients alike and, with its powerful statistical analysis capability, constitutes a unique approach to clinical audit and epidemiological research. The user's response to re~ent hardware failure has clarified their acceptance of the system as an integral part of m o d e m diabetes care. The second data collection system, implemented using SCULPTOR, as well as offering compatibility with commercially available general practice software, provides a convenient link to the expert system which makes available advice to the clinician in the organisation and operation of the patient consultation in the diabetic follow-up clinic. This system clearly requires refinement so

that it can function within the time constraints imposed upon clinic operation, something which will be facilitated with the rate of development of computer hardware. Preliminary results from using the expert system appear encouraging and medically consistent and more extensive evaluation is currently in progress. Incorporation of the insulin advisot~ module, currently being developed within a separate 'metabolic prototype', will add a means of extending the large rule-based expert system to cover the totality of diabetes care.

References [1] M.A. Boroujerdi, C.D. Williams, E.R. Carson, J.K. Piwernetz, K.D. Hepp and P.H. Sonksen,A simulation approach for planning insulin regimes, in: International Symposium on Advanced Models for Therapy of Insulin-dependent Diabetes (Serona SymposiumNo 37), eds. P. Brunetti and W.K. Waldhaust, pp. 41-46 (Raven Press, New York, 1987). [2] T. Deutsch, M.A. Boroujerdi, E.R. Carson, F.E. Harvey, P.H. Sonksen,G. Tamas and C.D. Williams,The principles and prototyping of a knowledge-baseddiabetes management system, Comput. Methods Programs Biomed. 29 (1989) 75-88. [3] T. Deutsch, E.R. Carson, F.E. Harvey, E. Lehmann, P.H. Sonksen, G. Tamas, G. Whitney and C.D. Williams, Com. puter-assisted diabetic management: a complex approach, Comput. Methods Programs Biomed 32 (1990) 195-214. [4] E.D. Lehmann, Systems Analysis and Systems Design tot the Metabolic Prototype 1990 (unpublished report). [5] C.D. Williams, F.E. Harvey, S. Carey, N. Morrish, H.W. McAughtrey, S.C. Bateman, S. Till, E.R. Carson and P.H. Sonksen. A technical and operational evaluation of a rulebased co-operative system for decision making in diabetes care (Department of Health Reporg, London, 1990). [6] C.D. Williams, S. Till, A.F. MacLeod and C. Lowy, Biothesiometry in diabetic patients, Diabetic Med. 5 (Suppl.) (1988) 30. [7] C.D. Williams, S. Till, P. Calder, S. Pack and P.H. Sonksen, Statistical analysis of a large computerised database container', diabetic outpatient records, Diabetologia, 30 (1987) 5c/A.