USING EXPLANATION FACILITIES IN HEALTHCARE

EXPERT SYSTEMS

Keith Darlington

London South Bank University, BCIM, 103, Borough Road, London, SE1 OAA, UK

Keywords: Explanation, Expert Systems, NHS Direct, Prodigy.

Abstract: A great deal has been written about healthcare expert systems in recent years. This paper examines a

particular feature of expert systems: namely explanation facilities. A limited explanation capability is an

integral part of a rule based expert system. The role of explanation in expert systems has been largely

ignored in healthcare literature, since the MYCIN system and its derivatives were developed in the mid

1980s. However, empirical research has shown that users are more likely to adhere to recommendations

made by expert systems when explanation facilities are available. Furthermore, explanation provision have

been shown to improve performance and aid the user with a better understanding of the subject domain as

well as result in more positive user perceptions of an expert system. This paper looks at the evolution of

explanation facilities in healthcare expert systems, and investigates user requirements for explanation

facilities in the healthcare domain.

1 INTRODUCTION

The medical expert system MYCIN (Shortliffe

1981) was amongst the first of a number of decision

support diagnostic systems developed in the late

1970s. Since this time, the use of expert systems as

an IT decision making aid in healthcare has grown

rapidly (Schank, Doney and Seizyk, 1988),

(Thornett 2001). For example, NHS Direct Hotline

uses an expert system in basic patient diagnostics.

NHS Direct has been at the forefront of 24-hour

health care in the UK - delivering telephone and e-

health information services direct to the public, and

is accessed by over two million people every month.

PRODIGY (Thornett 2001) is another example of an

expert system that is used in primary care in the UK.

Prodigy provides decision support to general

practitioners within consultations regarding drug

prescribing.

But many researchers have been sceptical about

expert system usage in healthcare. For example,

Delaney et al, (1999) believe that computerised

decision support systems have great potential but

have largely failed to live up to their promise.

However, Walton et al, (1997) presents evidence to

suggest that advice from a computer

will be more

convincing if presented simultaneously with an

explanation for that advice. In their evaluation of

CAPSULE, an expert system giving advice to

general practitioners about prescribing drugs, they

say that “Finding the most effective way

presenting the explanation is an important goal for

future

studies of computer support for prescribing

drugs”. Yet, an explanation component is a standard

feature of expert systems in that the systems

problem solving behaviour can be observed during a

consultation.

The inclusion of explanation facilities can

enhance performance of the decision making and

lead to greater adherence to the recommendations of

the expert system (Gregor and Benbasat, 1999),

(Arnold et al, 2006). Indeed, many studies have

demonstrated the importance of a system being able

to explain its own reasoning. For example, in a study

of physician’s expectations and demands for

computer based consultation systems it was found

that explanation was the single most important

requirement for advice giving systems in medicine

(Buchanan and Shortliffe, 1984). Also, according to

(Berry et al., 1995), explanation is seen as a vital

feature of expert systems – particularly in high risk

domains, such as medicine, where users need to be

convinced that a system’s recommendations are

based on sound and appropriate reasoning.

However, despite the importance attached to

explanations, few expert systems provide acceptable

223

Darlington K. (2008).

USING EXPLANATION FACILITIES IN HEALTHCARE EXPERT SYSTEMS.

In Proceedings of the First International Conference on Health Informatics, pages 223-226

 SciTePress

explanation. Surprisingly, research in explanations

has been largely ignored in healthcare expert

systems since the development of MYCIN. Mao and

Benbasat (2000) cite reasons why explanations in

expert systems had failed to appeal to the user: that

they were difficult to understand, and that they

ignored the needs of different users. The following

sections examine ways in which some of these

shortcomings have been overcome in the healthcare

domain, beginning with a look in some detail at one

of the first ever expert systems to incorporate

explanation – the medical expert system MYCIN.

2 THE MYCIN EXPLANATION

FACILITY

The explanation facilities in MYCIN (Shortliffe,

1981) were presented in a natural language that was

translated from the rules making the explanations

easier to follow for the user. Explanations were also

supplemented with certainty factors that numerically

expressed the degrees of certainty attached to

conclusions or outcomes. This meant that users of

MYCIN could get an understanding of the likelihood

of the advice given. However, there were many

shortcomings identified with MYCIN’s

explanations. The following sections describe these

shortcomings and how they were overcome.

2.1 Rule Trace

MYCIN is a rule based expert system – which

means that knowledge is stored in the form of rules

(Darlington, 2000). The explanation facilities provided

in MYCIN and the other first wave of rule-based

expert systems would have been a rule trace. This is,

essentially, a record of the system’s run-time rule

invocation history during a consultation.

2.1.1 Feedforward and Feedback

Explanations

A feedforward explanation provides the user with a

means to find out why a question is being asked

during a consultation (i.e., during the data input

stage). The feedforward explanation will retain the

manner in which input information is to be obtained

for use in finding a solution.

A feedback explanation provides the user with a

record of problem solving action during a

consultation: i.e., how a conclusion was reached

when the data has been completely input. The

feedback explanations will present a trace of the

rules that were invoked during the consultation and

display intermediate inferences in getting to a

particular conclusion.

2.2 Strategic Explanations

Rule trace methods formed the basis of explanations

in MYCIN (Shortliffe, 1981), but Clancey (1983)

tried to adapt MYCIN from its diagnostic role to that

of tutorial role in a system called GUIDON. The

purpose of GUIDON was to provide a training

system for junior consultants. What was thought to

be a simple task turned out to be very difficult

because MYCIN did not contain knowledge which

explicitly contained strategic knowledge. This is

knowledge about how to approach a problem by

choosing an ordering for finding subgoals to

minimise effort in the search for a solution. For

example, the rule of thumb that alcoholics are likely

to have an unusual aetiology can lead the expert to

focus on less common causes of infection first –

thereby pruning the search space to find a solution.

The strategic knowledge in MYCIN was implicitly

incorporated in the problem solving rules. However,

Clancey realised that this knowledge needed to be

explicitly represented, so that it could become

transparent to students training to use the system.

The lessons of GUIDON led Clancey to develop a

follow up system called NEOMYCIN (Clancey &

Letsinger, 1981): this was a consultation system

whose medical knowledge base contained the

tutorial strategic knowledge.

2.3 Justification Knowledge

An expert system can only reason with the

knowledge contained in its knowledge base Thus,

the designer of a diagnostic medical expert system

would ensure the knowledge base contains enough

problem solving knowledge to ensure the system can

arrive at the correct conclusions. However, the rule

trace can only reconstruct a trace from what

knowledge is contained in the expert system

knowledge base. If the builder has not included the

knowledge to justify the knowledge in the rule-base,

then the system will not be able to justify the

existence of the knowledge? Clancey (1983) realized

the importance of this justification knowledge when

attempting to extend the MYCIN system to support

the training of junior physicians. Again, he found

that MYCIN failed to do this because it did not

contain justification knowledge. Justification

knowledge is often unavailable because the rules

which model the domain do not capture all the forms

HEALTHINF 2008 - International Conference on Health Informatics

224

of knowledge used by experts in their reasoning.

This is because builders of expert systems capture

“rules of thumb” shallow knowledge that only

enable the system to solve diagnostic problems

Empirical research has consistently shown that user

acceptance of expert systems increases for non-

expert users when this justification knowledge is

present, and that justification is the most effective

type of explanation to bring about positive changes

in user attitudes toward the advice giving system

(Ye & Johnson 1995).

2.3.1 Capturing Deep Knowledge to

Represent Justification Knowledge

Expert physicians would, of course, use rules of

thumb themselves in solving problems, but they

would also – as a result of their training and

experience – possess a deep theoretical

understanding of their subject domain. They may,

for example, use “rules of thumb” or heuristic

knowledge when performing a diagnosis. These

“rules of thumb” may be sufficient to enable the

physician to carry out a diagnosis, and therefore, this

is the knowledge that is captured in rule-based

expert systems because it is clearly much easier to

obtain and code, and is sufficient for problem

solving. However, this justification knowledge

would have to be explicitly captured by the system

designer if it were required for explanation.

3 USER REQUIREMENTS FOR

EXPLANATION

In the healthcare domain, user requirements would

vary according to employment categories which

may include physician (including junior physician),

nurses, administrators and also, as we will see later

in some application examples, patients.

Considering the expert physician vs. non expert

divide, research has shown that expert physicians do

make use of explanation facilities but their

requirements are very different to that of other

users. Experts tend to have a preference for

feedback rule trace explanations and are more likely

– than non-experts – to use explanations for

resolving anomalies (disagreement), verification

and exploring alternative diagnoses (Arnold et al

2006), (Mao and Benbasat, 2000). Non-experts such

as trainee physicians, on the other hand, are more

likely to use explanations for short and long term

learning. Moreover, (Arnold et al 2006), have

shown that non-experts tend to use both feedback

and feedforward justification explanations, as well

as terminological feedforward explanations, i.e.,

definitions of domain terms, etc., to assist during

the data input stage. Patients – or other non-experts

– using the NHS Direct system may benefit from

this type of explanation because these explanations

would facilitate learning of the subject domain.

Some examples of these user-centred applications

of explanation are briefly described in the next

section.

3.1 Some User-Centred Explanation

Prototypes in the Healthcare

Domain

A number of healthcare projects involving the use of

user modelling for explanation have been prominent

in recent years. OPADE (Carolis et al, 1996) is a

European Community Project funded expert system

for generating beneficiary centred explanations

about drug prescriptions that take into account the

user requirements. The main objective of OPADE is

to improve the quality of drug treatment by

supporting the physician in their prescription process

and by increasing compliance with the therapy

(Berry et al, 1995). OPADE supports two types of

user: those who directly interact with the system

such as general practitioners and nurses, and those

who receive a report of results – i.e., the patients.

The explanations that are generated are dynamic

(unlike static canned text explanations) in that a

“user model” is maintained containing the

characteristics of the user. A “text planner”

component plans the discourse during a consultation.

The text planner will build a tree containing the

discourse plan which will depend on the objectives

that are to be met by the user model. The

explanation is then delivered in natural language by

taking as input the tree generated by the text planner

and transforming it using text phrases into the

appropriate format.

HEALTHDOC (Marco et al, 1995) is another

user-centred expert system whose main purpose is to

generate reports for patients and materials for health

education. This system enables the production of

health-information and patient-education documents

that are tailored to the individual personal and

medical characteristics of the patients. Health-

education documents can be much more effective in

achieving patient compliance if they are customized

for individual readers. The documents are presented

textually and the text can be adapted to different

patients, because the system contains a database

USING EXPLANATION FACILITIES IN HEALTHCARE EXPERT SYSTEMS

225

containing information about the clinical data of

every patient – such as their personal and medical

characteristics.

4 CONCLUSIONS

This paper recognises the role of expert systems in

healthcare. However, one of their main features -

explanation facilities – has been largely ignored in

healthcare systems to date. Yet, empirical research

has consistently shown, in recent years, that users are

more likely to adhere to expert system

recommendations when explanation facilities are

available. Furthermore, explanation provision have

been shown to improve performance and aid the user

with a better understanding of the subject domain as

well as result in more positive user perceptions of an

expert system.

However, users will not use an explanation

unless it addresses their basic information needs.

This means that system designers must involve users

in the evaluation of explanation facilities to ensure

that they serve the needs of specific user groups. As

this paper has shown, providing designers submit the

effort, explanations can be tailored to the needs of

different users. Perhaps the time has come for

healthcare expert system designers, and users of such

systems to re-evaluate the potential of explanation

facilities.

REFERENCES

Arnold, V., Clark, N., Collier, P.A., Leech, S. A., Sutton,

S.G., 2006. The Differential Use and Effect of

Knowledge-Based System Explanations in Novice and

Expert Judgment Decisions. MIS Quarterly, Vol 30

No 1.

Berry, D., Gillie, D. T., and Banbury. S., 1995.What Do

Patients Want to Know: an Empirical Approach To

Explanation Generation and Validation. Expert

Systems with Applications, 8(4):419{428,

Buchanan, B., G., Shortliffe, E., H., 1984. Rule-based

Expert Systems: The MYCIN Experiments of the

Stanford heuristic programming project. Addison

Wesley, Reading, MA.

Carolis, B.D., Rosis, F.D., Grasso, F., Rossiello, A., Berry,

D., and Gillie, T., 1996. Generating recipient-centered

explanations about drug prescription. Artificial

Intelligence in Medicine 8, 123-145.

Clancey, W. J., 1983. Epistemology of a rule-based Expert

System: A framework for explanation. AI magazine,

20(3) pp 215-251.

Clancey, W. J., Letsinger, R., 1981. NEOMYCIN;

reconfiguring a rule-based expert system for

application to teaching. Pages 361–381 in W.J.

Clancey and E. H. Shortliffe (eds.), Readings in

medical artificial intelligence: The first decade,

Addison-Wesley, Reading, Massachusetts

Darlington, K., 2000. The Essence of Expert Systems.

Prentice Hall, London. Pearson Education.

Delaney, B.C., Fitzmaurice, D. A., Riaz, D., and Hobbs, F.

D., 1999. “Can computerised decision support systems

deliver improved quality in primary care?” Journal of

BMJ. November 13; 319(7220): 1281.

Gregor, S., & Benbasat, I., 1999. Explanations from

intelligent systems: theoretical foundations and

implications for practice. Management Information

Systems Quarterly, 23, 497}530.

Mao, J., Benbasat, I., “The Use of Explanations in

Knowledge-Based Systems: Cognitive Perspectives

and a Process-Tracing Analysis”, Journal of

Management Information Systems, Vol. 17(2), pp153-

180, 2000.

Marco, C.D., Hirst, G., Wanner, L., and Wilkinson, J.,

1995. HealthDoc: Customizing patient information

and health education by medical condition and

personal characteristics. In First International

Workshop on Artificial Intelligence in Patient

Education, Glasgow, UK.

Schank, M. J., Doney, L. D., Seizyk, J., 1988. The

potential of expert systems in nursing. J Nurse

Administration. June; 18(6):26-31. PMID: 3286842

[Pub Med - indexed for MEDLINE].

Shortliffe, E. H., 1981. Computer Based medical

consultations - MYCIN Published by Elsevier, USA.

Thornett, A., 2001. Computer decision support systems in

general practice. International Journal of Information

Management; 21:39-47.

Ye, L.R., Johnson, P.E., 1995. “The impact of explanation

facilities on user acceptance of expert systems advice”,

MIS Quarterly, Vol. 19(2), pp157-172.

Walton, R., 1997. An evaluation of CAPSULE, a

computer system giving advice to general practitioners

about prescribing drugs. J INF Primary Care.

HEALTHINF 2008 - International Conference on Health Informatics

226