Text, competence and logic: An exercise

Text, Competence and I~gic: An Exercise G. N i g e l G i l b e r t a n d C h r i s t i a n H e a t h

University of Surrey

ABSTRACT' Professional medical practice, like other organizational conduct, relies upon records which document transactions between members and their clientele. Medical practitioners employ a set of conventions providing for the systematic recording and interpretation of medical record cards that forms a social organization underlying the records cards' ordinary usage. In this paper we examine these conventions and develop a computer program which captures elements of their structure and use. By doing so we illustrate one way in which sociological analysis can contribute to the design of 'intelligent systems.' We also suggest that the emerging discipline of Artificial Intelligence might find recent developments in sociology pertinent to its concerns.

Introduction For at least two decades there has been a growing interest within the social sciences in developing intelligent systems to capture or reproduce elements of h u m a n thinking and behavior. Much of this concern with artificial intelligence has emerged within cognitive psychology, yet essentially addresses topics and problems pertinent to sociology. As yet, however, sociologists have shown little interest in the subject and researchers in artificial intelligence have largely failed to recognize the importance of recent sociological ideas for their own work. This essay attempts to bridge part of this gap by applying some techniques and tools drawn from artificial intelligence to an analysis of social organization. The particular social organization with which we shall be concerned is a form of text or discourse: the medical record cards kept and used by medical practitioners in primary health care, general practice. These medical biographies provide the profession with a significant resource Address correspondence to: Department of Sociology, University of Surrey, Guitdford, Sur~ rey, GU2 5Xtt, England.

Qualitative Sociology, 9(3l, Fall 1986

215

© 1986 Human Sciences Press

216

QUALITATIVE SOCIOLOGY

in their day to day dealings with their clientele; the records inform decisions and play a crucial role in the organization of treatment. Doctors, as part of their professional competence, employ practices which provide for the systematic documentation and comprehension of the information in the records; these practices constitute a social organization which underlies the ordinary usage of the medical records. Our aim is to explicate the community of practices used by medical practitioners, by developing a model which captures elements of this social organization. The model, expressed in the form of a computer program, has been designed to simulate aspects of a h u m a n competence, n o t to reflect actual cognitive processes (Searle, 1981). In addition, we hope to show how attention to social organization is a prerequisite to developing practical systems for use within organizational settings, such as primary health care. There is both theoretical and empirical value in attempting to program aspects of practitioners' competence in using medical record cards. For example, we thought that developing a program based on findings derived from ethnographic research might lead to a more precise conception of the phenomena under study and might assist clarification of empirical issues. We also wished to explore the issue of h u m a n competence in rule use and defeasibility (cf. Hart, 1968; Rawls, 1955) in a program that approximates an important aspect of intelligent behavior. Thus the work reported here explores the links between sociology and the study of artificial intelligence. Medical Record Cards Medical record cards consist of descriptions of consultations; every consultation warrants a description or single entry in the records. The records are stored and made available to the doctor whenever he or she consults with a patient. If the patient moves to a different region and registers with a new doctor, processes are set in motion to transfer the records to the new location. Even after the death of a patient, his or her records are kept for a short time in case any contingencies arise for which the records might be of assistance. The importance of the records ties both in their being cumulative descriptions of consultations and resources in the organization of professional medical conduct. Collection i shows examples of entries found in medical record cards? At first glance the entries seem brief, almost crude and one might wonder why much trouble is devoted to their upkeep. Certainly, for social scientists and members of the medical profession who attempt to use the records for research purposes, many difficulties arise, not least their

Text, Competence and Logic: An Exercise

217

Data collection 1. Fragments from medical record cards: 1

28/9/75

c vomited x 2 in night Maxolon 10 b.d. (300m)

2

9/1/74

v Pain now appears to be allright Chest X-R Cert 1/52

3

12/1/76

c Dies 12.30 a.m.

4

25/1/77

c ~sore throat' ~slight cough' Apsin 125 m.g.

5

3/12/76

c Dog bite RfTetanus Toxoid 0.5ml

6

22/4/76

c Cold Flu also Rheumatism Cert 1/52 Paracetamol

7

29/2/75

c ~feeling tired' depressed. Librium (30) (5 m.g.)

8

10/2/73

c Tonsillitis. Apsin (30) (250 m.g.)

9

14/3/74

c Ted up'

10

13/2/73

c feeling sick depressed. Valium (15) 5 m.g.

11

26/11/70 c conjunctivitis Albacid 10% 3/12/70 c eye now appears virtually normal Neospurin

12

10/6/74 6/7/74 2/8/74

c schizo. Cert 4/52 c Cert 4/52 c Cert 4/52

13

12/4/75

c ~weepy', tired depression. Valium (10 m.g.) (30) cert. 1/52 c Valium (10 m,g.) (30) cert. 1/52

19/4/75

~inconsistency of i n f o r m a t i o n ' a n d the Clack of detail' (cf. B i t t n e r a n d Garfinkel, 1967). F u r t h e r m o r e , the contents are relatively in~lpermeable to change. E v e n w h e n specific a t t e m p t s are m a d e to improve t h e qual-

218

QUALITATIVE SOCIOLOGY

ity of the descriptions in the records, such as by providing larger folders, specific information sheets and summary pages, it is found t h a t the contents remain much the same (Heath, 1982). These difficulties are found when the records are used for purposes for which they were not designed. During the course of day-to-day medical work, however, the contents of the records are frequently employed without trouble and generate little complaint. Reading and writing the descriptions found in the records is an integral part of conducting professional activity; the descriptions are necessary for both the assessment and m a n a g e m e n t of illness. Taylor (1954) went so far as to say: One has reached the conclusion that the key to good general practice is the keeping of good clinical records. Time and again one has seen a quick glance through a well-kept record card provide either the diagnosis or an essential point in treatment. Doctors frequently read the medical records before the beginning of a consultation to familiarize themselves with a patient's more recent problems and then tune the t r e a t m e n t accordingly. Or, in the course of discussing a problem with the patient, a doctor may turn to the records to discover the facts concerning the patient's previous difficulties. On other occasions, the medical records may be used as a source of ~hints' and ~confirmations,' allowing the doctor to develop hypotheses or choose between competing diagnoses. In these, and a variety of other ways, the records provide a generalized set of resources for conducting everyday medical activity. Doctors rely upon the records; they expect them to conrain certain sorts of information and to be adequate for the uses to which they are so frequently put. Part of being a physician involves the ability to write adequate and suitable descriptions for the records. This process of description and retrieval, reading and writing the records, involves a set of conventions. The aims of this essay are to illustrate these conventions and develop a program which can cope with aspects of their social organization. The Mapping of Category Items Fragment 7 29/2/75

c Teeling tired' depressed. Librium (30) 5 m.g.

The first item is the date on which the consultation took place and the second refers to the location of the consultations (~c" stands for the consulting room and "v" refers to a visit to the patient's home). Next, we encounter Teeling tired' and ~depressed'-both items describing the

Text, Competence and Logic: An Exercise

219

complaint of the patient. The former refers to the patient's version of his complaint; for the doctor, the ~subjective' version. The latter refers to the doctor's assessment of the patient's complaint; the ~objective' version which sometimes involves a diagnosis. Finally, the entry contains an item which refers to the m a n a g e m e n t of the patient's complaint, in this case details of a prescription-~%ibrium (30) (5 m.g.)." Other management items may include details concerning referrals. The three classes of items within an entry in the records are, the occasion of the consultation, the patient's complaint and the management of the patient. Within these classes there are categories, for example, the patient's version of the complaint and the doctor's assessment (crudely, symptoms and diagnosis). The information which these classes and their constituent categories entail can regularly be found in record card entries. This is not to say t h a t each item of each category is always specifically recorded, though often it is, but rather t h a t such information is retrievable in the description (if the relevant phenomena occurred in the consultation). We shall return to this issue. An aspect of producing and comprehending the contents of medical records is the assumption of the writer or reader t h a t the entries are mapped in regular ways. There is a geography to the records. Entries not only follow one another in serial, sometimes sequential, order, but are also internally ordered, as in the example above where the three classes follow each other, as do their component categories. Mapping the category items in this way allows the reader to understand the sense of the particular items and the entry as a whole. Consider how we understand the entry in our example. ~Depressed" achieves its factual status as an assessment of the patient's complaint because it follows ~feeling tired" and precedes the category items for the m a n a g e m e n t of the complaint. Were we to introduce an item following ~depressed," ~paranoid," for example, and perhaps omit ~feeling tired," then ~depressed" would be understood as referring to the patient's version of his complaint. Items such as ~paranoid," ~flu," etc. can be employed in various categories on a range of occasions; they do not have a fixed and determinate sense. Their sense within a particular entry is generated in part through the ways in which they are mapped onto the medical records and organized with respect to the constituent items of the entry. Doctors produce descriptions of consulations to enable them to be read in certain ways. The geography of items is an essential feature of the production and recognition of the description. The first step in developing a program to deal with a description in the medical records involves the assignment of specific categories to items, a process described as ~parsing.' Parsing allows categories to be ascribed to terms drawn from a vocabulary. This vocabulary and the

220

QUALITATIVE SOCIOLOGY

categories r e l e v a n t to the e n t r i e s in medical records m a y be specified as a collection of assertions in the form: category ([item]). This notes t h a t "item" is a m e m b e r of category "category." For instance the a s s e r t i o n d o c t o r s a s s e s s m e n t ([depressed]) records t h e fact t h a t "depressed" is a possible doctor's assessment. 2 T h e v o c a b u l a r y would also include patients__version ([depressed]) to express t h e fact t h a t "depressed" is also a t e r m used by p a t i e n t s to describe t h e i r symptoms. T h e r e is no difficulty in assigning one word to two categories, j u s t as "watch" would be listed in a d i c t i o n a r y as b o t h a v e r b and a noun. T h e total list of items and t h e i r categories k n o w n to a g e n e r a l practit i o n e r is clearly quite long, a l t h o u g h finite. A short list, sufficient for t h e purposes of t h e e x a m p l e s in this paper, is shown in F i g u r e 1. In F i g u r e 1, the assertions in which t h e r e is no i t e m (e.g. "patients__ version ([]).") are included to indicate t h a t e n t r i e s m a y possibly not cont a i n a n y item of t h a t description. For instance, a n e n t r y m a y include no patient's version of the complaint. F r o m this point of view, a null i t e m is a possible p a t i e n t ' s version. In addition to a vocabulary, p a r s i n g r e q u i r e s a set of g r a m m a r rules. F o r the case of p a r s i n g E n g l i s h sentences, a n e x a m p l e of a g r a m m a r r u l e m i g h t be: "a s e n t e n c e can t a k e the f o r m of a n o u n p h r a s e followed b y a verb phrase." F o r the case of medical records, an i l l u s t r a t i v e r u l e m i g h t be: "an e n t r y can t a k e t h e form of a list c o n t a i n i n g a date, a location code, a p a t i e n t ' s version, a doctor's a s s e s s m e n t and some details a b o u t m a n a g e m e n t of the illness." Following the s t r u c t u r e of the a c t u a l entries, it is c o n v e n i e n t to reg a r d the doctor's e n t r y as a whole as a 'list' of i t e m s w r i t t e n in order and placed w i t h i n s q u a r e brackets, each i t e m s e p a r a t e d from the n e x t by a comma. Lists in g e n e r a l m a y t h e n be r e f e r r e d to using a 'variable,' n o t a t i o n a l l y d i s t i n g u i s h e d f r o m items by h a v i n g an initial capital letter. Thus, [22/5/76, c, sore throat, tonsillitis, apsin (30) (250 m.g.)] is a list r e p r e s e n t i n g a p a r t i c u l a r e n t r y , and

Text, C o m p e t e n c e a n d Logic: A n E x e r c i s e

Figure

1: C a t e g o r i z a t i o n

of some items from medical

221

record

cards

location (['c']). location (['v']). patients~version patients version patients_version patients__version patients__version patients__version patients version patients__version patients version

([]). (['feeling tired']). ([~sore throat']). (['infected tonsils']). ([~fed up']). (['depressed'I), (['feeling sick']). ([~vomited x 2 in night']). ([slight cough']).

doctors__assessment doctors~assessment doctors assessment doctors~assessment doctors~assessment doctors assessment doctors assessment doctors assessment

([]). (['depressed']). ([~depression']). ([~tonsillitis% (['conjunctivitis']). (['cold']). ([~dog bite']). ([~schizo']).

treatment treatment treatment treatment treatment treatment treatment treatment treatment treatment treatment

([]). ([~librium (30) (5 m.g.)']). ([~apsin (30) (250 m.g,)']). ([~apsin (125 m.g.)']). ([~valium (30) (5 m.g.)']). ([~valium (15) (5 m.g.)']). ([~albacid 5%']). ([~neospurin']). (['maxolon 10 b.d. (300m)']). (['tetanus toxoid 0.5 ml']). (['paracetamol%

certificate certificate certificate certificate

([]). ([~cert 1/52']). ([~cert 1/4']). (['cert 4/52']).

referral referral referral referral referral

([]). ([~chest XR']). ([~Brook centre']). ([~r/f A,A.']). ([%T G.C.']).

AnEntry is a v a r i a b l e , w h i c h m a y b e u s e d t o s t a n d for t h i s , o r a n y o t h e r s i m i l a r list.

222

QUALITATIVE SOCIOLOGY

Date Assessment are o t h e r variables, because t h e y begin w i t h a capital letter. To help in a s s e m b l i n g lists from t h e i r c o m p o n e n t items, a n d decomposing lists into components, the rule join ([A,B,C,D,E],[A],[B],[C],[D],[E]. m a y be used. This r u l e expresses the fact t h a t a list m a y be composed of the concatenation of the contents of five other lists, [A] to [E], in order. The r u l e for c o n s t i t u t i n g record card e n t r i e s m a y now be written: entry(AnEntry):-join (AnEntry,Date,Location,Complaint,Assessment, Treatment), date(Date), location(Location), patients_version(Complaint), doctors__assessment(Assessment), management(Treatment). H e r e , ~:-" is r e a d as ~if," and the c o m m a s b e t w e e n the t e r m s are r e a d as ~and." The rule as a whole m i g h t be i n t e r p r e t e d : the list, AnEntry, is an entry if AnEntry may be decomposed into the sublists, Date, Location, Complaint, Assessment and Treatment, and Date is an item belonging to the category '~date," and Location is an item belonging to the category ~location," and Complaint is an item belonging to the category '~patients version," and Assessment is an item belonging to the category ~doctors__assessment" and Treatment is an item belonging to the category ~management." The ~join" rule which earlier was p r e s e n t e d in t e r m s of lists [A], [B] etc. can be re-written: join([Date,Location,Complaint,Assessment,Treatment], [Date], [Location], [Complaint], [Assessment], [Treatment]). On a n y occasion it is likely t h a t w i t h i n a p a r t i c u l a r e n t r y c e r t a i n categories of items will not be recorded, for no o t h e r r e a s o n t h a n t h a t the r e l e v a n t event did not occur in the consultation. T h u s a n u m b e r of ~join" r u l e s which lay out the possibilities for o m i t t e d i t e m s in the record entries need to be considered. In F i g u r e 2 a set of ~grammar rules" for parsing the e n t r i e s which includes t h e various ~join" rules is shown. 3

Text, Competence and Logic: An Exercise

F i g u r e 2: G r a m m a r

223

rules for a medical record card entry

entry (AnEntry) :-join(AnEntry,Date,Locatio~,Complaint,Assessment,Treatment), date(Date), location(Location), patients version(Complaint), doctors assessment(Assessment), management(Treatment). join([Date,Location,Complaint,Assessment,Treatment], [Date]~ [Location], [Complaint], [Assessment], [Treatment]). join([Date,Location,Assessment,Treatment],[Date], [Location],[]~[Assessment],[Treatment]. join([Date,Location,Complaint,Assessment],[Date], [Location],[Complaint], [Assessment]~ []). join([Date,Location,Complaint,Treatment],[Date], [Location],[], [Assessment], [Treatment]). join([Date,Location,Complaint], [Date], [Location], [Complaint], [], []). join([Date,Location,Assessment], [Date], [Location], [], [Assessment], []). join([Date,Location,Treatment], [Date], [Location], [], [Assessment], [Treatment]). A f u r t h e r join rule also needs to be included fbr t h e category i t e m s in the m a n a g e m e n t class of the entry. T h e class can consist of t h r e e categories of item: a prescription, details of a medical certificate, a n d m e n t i o n of a r e f e r r a l to a specialist. A n y of these categories of i t e m m a y be omitted in a n ent1% The rule ~join3" in F i g u r e 3 h~s a similar function to t h e ~join" r u l e s for a n e n t r y as a whole, defining the v a r i o u s c o m b i n a t i o n s of m a n a g e m e n t i t e m found in t h e records. T h e s e r u l e s precisely and rigorously define the s y n t a x of a medical record e n t r y . T h e y can be s h o w n to be ~Horn Clauses,' a specific type

Figure 3 management([]). management(AListOfManagementItems):join3(AListOfManagementItems~Prescription,Certificate, Referral), treatment(Prescription), certificate(Certificate), referral(Referral). join3([Prescription,Certificate,Referral], [Prescription], [Certificate], [Referral]). join3([Prescription,Certificate], [Prescription], [Certificate], []). join3([Certificate, Referrat],[],[Certificate],[Referral]). join3([Prescription], [Prescription], [], []). join3([Certificate], [], [Certificate], []). join3([Referral], [], [], [Referral]).

224

QUALITATIVE SOCIOLOGY

of proposition drawn from first-order predicate calculus (Kowalski, 1979) which has well understood properties. Although precise, the rules are flexible, in that they enable a very large number of entries with many different surface forms to be parsed under their guidance. For instance, an entry consisting a date, location, and the single item ~'fed up" (see Fragment 9) will be successfully parsed, as will a more complex entry with items representing all the categories. Moreover, as the rules stand, they permit certain entries to be parsed in more than one way. For instance, because '~depressed" may be either a patient's version or a doctor's assessment, and because either the doctor's assessment or the patient's version may be missing from any particular entry, the entry

[12/3/50, c., depressed, valium (15) 5 m.g.)] may be parsed to categorize ~depressed" as either a patient's version, or as a doctor's diagnosis. We shall see how this kind of ~ambiguity' may be resolved shortly. The rules of the grammar described above are not yet sufficient for the production or interpretation of entries in medical record cards. For example, they do not preclude the possibility of the program generating many medically nonsensical entries. Take for instance the following: [23/8/76 c, sore throat, depressed, neospurin, ref A/A] This includes a collection of items which fall within the appropriate categories, yet it describes a consultation which could not reasonably have taken place. As written, it reports a consultation in which the doctor provides a prescription of an eye cream, ~neospurin," for depression and refers the patient to Alcoholics Anonymous Cref. A/A"). The problem arises because the grammar rules place no constraint on how items from various categories may be combined to generate an entry. The program requires a certain body of information which relates together particular items and categories. So for example, ~sore throat" should be treated as a symptom of ~tonsillitis" rather than ~depression," and ~neospurin" should be associated with difficulties with the eye such as ~conjunctivitis." In associating items from different categories in this way, it is important to include the possibility of, for example, a particular diagnosis having more than one treatment, or a complaint having more than a single assessment associated with it. The facts in Figure 4 capture a few possible interrelations:

Tex% Competence and Logic: An Exercise

225

Figure 4 is__a~ymptom__of([Teeling tired'], ['depression']). is__a symptom _of([Tedup'], [Mepression']). is__a__symptora__off[~sorethroat'], [~tonsillitis']). is__a__symptom__of(['infectedtonsils'], [~tonsillitis']). is__a symptom__off[~depressed'],['depression']). is a symptom__of([Tedup'], []). is a treatment__for([~librium (30) (5 m.g.)'], [~depressien']). is a treatment~_for([~valium (15) (5 m.g.)'], [~depressed']). m~a__treatment__for([~valium (30) (5 m.g.)'], ['depressed'l), is~a~treatment__for([~neospurin'], [~conjunctivitis']). is__a treatment_.for([¢apsin (30) (350 m.g.)'], [~tonsillitis']). m a treatment_for([~apsin (125 m.go)'],[tonsillitis']). is__a treatment__for([~albacid 5%'], [conjunctivitis']). is__a treatment__for(['tetanus toxoid 0.5 ml'L [~dogbite']). ls__a treatment for([~paracetamol'],[~cold']). is__a treatment__for([~cert4/52'], [~schizo.']). T h e stock of i n f o r m a t i o n compiled in F i g u r e 4 is g a t h e r e d from ent r i e s in t h e medical record cards. It consists of associations m a d e in t h e records b y medical practitioners, r a t h e r t h a n t e x t book r e l a t i o n s found w i t h i n the ~professional' l i t e r a t u r e . It is not at all c e r t a i n t h a t this litera t u r e would include t h e details found in the medical records since t h e e n t r i e s f r e q u e n t l y e n t a i l ordinary, c o m m o n sense notions and relations. Because t h e concern h e r e is to explicate some aspects of o r g a n i z i n g a description in the records, it is the conventions of actual practice whlch are of relevance. U s i n g this ~knowledge,' the g r a m m a r rules of the previous section can be a u g m e n t e d to specify m e a n i n g f u l e n t r i e s m o r e closely. F o r e x a m p l e , well formed entries are typically those for which the doctor's assessment corresponds to t h e p a t i e n t ' s version of t h e p a t i e n t ' s complaint, t h a t is, t h e C o m p l a i n t is a s y m p t o m of the assessment. T h u s a fuller specifica~ tion of the rule for e n t r y is: entry(AnEntry) : - j ° i n ( A n E n t r y ' D a t e ' L ° c a t i ° n ' C ° m p l a i n t ' A s s e s s m e n t ' Treatment), date(Date), location(Location) patients version(Complaint) doctors__assessment(Assessment) management(Treatment) is__a__symptom__of(CompIaint,Assessment), is__a treatment~for(Treatment,Assessment). in which t h e last two r e l a t i o n s specify t h a t a well f o r m e d e n t r y is one which not only has a n a p p r o p r i a t e c o m p l e m e n t of items, but also has t h e p a t i e n t ' s version as a s y m p t o m of the illness t h a t the doctor has di-

226

QUALITATIVE SOCIOLOGY

agnosed, and the prescribed t r e a t m e n t as a t r e a t m e n t relevant to the diagnosis. We shall refer to the grammar including both the relations in Figure 4 and this new extended rule for an entry as the 'augmented g r a m m a r rules.' These rules are constitutive rather t h a n regulative (Rawls, 1955); they define what m a y be recognized as a particular type of entry. They are also permissive, in t h a t they indicate t h a t an entry may be composed in this fashion, not t h a t this is the only possible definition of an entry. Interpreting the Rules These rules can be given several interpretations. They can be regarded as parsing rules t h a t allow one to identify the categories to which the items in the record card entries belong, this being the interpretation used above to introduce the notion of a grammar. Under this interpretation, the rules suggest 'readings' of the entries, by virtue of the rules' ability to assign items to categories. The rules might alternatively be interpreted as providing a definition of what constitutes a well formed record card entry. That is, they could be seen as a formalization of a constraint on entries: it is not the case t h a t record entries can properly be composed of any collection of medical terms, for a necessary (though not sufficient) condition for an entry to be seen as well formed is t h a t it may be parsed by the grammar. Thirdly, the rules m a y be interpreted as a procedure for generating well formed entries, t h a t is, as 'production rules.' For instance, suppose t h a t in some mechanical way, an item was selected from the list of'doctor's assessments,' and further items were chosen from each of the other categories (where ~no item' is included in the range of choice when appropriate). These selections could then be inserted into the g r a m m a r rules, setting, or "instantiating," the variables to the chosen items. From application of the rules, and instantiation of the variables, a list of items would be created which constituted a well formed entry. This way of using the rules is a procedural interpretation. It requires a specification of the order in which the rules are to be used, and how the variables are to be instantiated. The procedure sketched informally in the previous paragraph is a 'bottom-up' procedure: it starts with the lowest l e v e l - a selection of i t e m s - and checks to see whether the selec* tion is a ~legal' one according to the rules. A 'top-down' procedure is also possible. In this case, the method would start with the observation t h a t an entry is required (the 'goal'), and t h a t entries are defined in terms of the categories "date," "location" "patient's version" etc. The original goal, to find an entry, is thus converted into eight sub-goals. Each of

Text, Competence and Logic: An Exercise

227

these sub-goals may be investigated by using the appropriate rules. The management goal, for instance, will succeed if some list of items can be found in which a "prescription," a "certificate" and a ~referral" occur, each of these in turn being sub-goals satisfied only by items which are members of the appropriate categories. These procedures can be followed through by hand. Since they are clearly defined, and only involve the manipulation of symbols, they can also be programmed on a computer, with the advantages of speed and accuracy this gives. A programming system called Prolog (Clocksin and Meltish, 1981) will perform the required work. Protog is the name of both a system for manipulating logical propositions, and a computer programming language in which the propositions may be written. The grammar rules shown in Figure 2 are in the form of Prolog clauses, as are all the formal rules of this paper. ~ Using the Prolog system to provide the procedural control and the grammar rules to provide the logic, well formed entries may be generated. Thus the third, procedural, interpretation of the grammar rules corresponds loosely to doctors' procedures in writing entries. Intra Entry Defeasibility In describing a consultation in the medical records and in reading an entry, doctors attend to a strict economy of items. The descriptions are brief, but more importantly, they avoid repetition and overlap. An adequate description of a consultation relies upon a stringent collection of items, gathered with an eye to a strict t h e m e - a theme which can only be discovered if one reads the collection of items as a whole. The economy is achieved not through the mechanical use of a set of classes and their constituent categories, but rather as a result of the doctor's practical reasoning and sensitivity to the understandings of his or her colleagues. In describing a consultation, the doctor gathers a collection of items with consideration as to how they will be understood. The doctor relies upon his or her colleagues reading into the descriptions what happened in the consultation; certain references and interpretations are triggered. The adequacy of a description of a consultation relies upon what is recorded and retrievable in the description. Considering how it wilt be understood, a doctor can economize and manipulate the categories of items. To use H.L.A. Hart's (1958) term, the doctor may 'defease' a particular category item assuming it will be inferred from other items in the description as a whole. The following examples illustrate the defeasibility of category items and the inferential work of a reader.

228

QUALITATIVE SOCIOLOGY

Fragments 8, 9 and 10 10/2/73 c Tonsillitis. Apsin(30) (250 m.g.) 14/3/74 c ~fedup'. 13/2/73 c Feeling sick. Depressed. Valium (15) (5 m.g.) In the first example it may be observed that the doctor has not recorded any items to describe the patient's version of the illness. Within the class ~patient's complaint," we find only the doctor's assessment. However, any doctor on reading such an entry in the records could infer the relevant information concerning the patient's symptoms. The doctor could warrantably assume t h a t the patient had a sore throat and infected tonsils, given the assessment, tonsillitis, provided. Hence, although no details of the patient's symptoms were recorded by the doctor who wrote the description, such information is retrievable by any doctor perusing the entry. It is therefore unnecessary for the writer to include a category item for the patient's presentation of symptoms. The writer may defease the relevance, considering what any doctor could read into the actual description. The inference which could be made by a doctor to retrieve the patient's version of his or her complaint can be reproduced using the augmented g r a m m a r rules we have developed. In the rule for an entry, replacing the variables by" those items included in F r a g m e n t 8 gives a partly instantiated version, thus: entry(AnEntry) :-join(AnEntry, [~10/2/73'], [~c'], Complaint, [~tonsillitis'], [~apsin (30) (250 m.g.)']), date([~10/2/73']), location([~c']), patients version(Complaint), doctors__assessment([~tonsillitis']), management([~apsin (30) (250 m.g.)']), is a symptom__of(Complaint,[~tonsillitis']), is__a__treatment__for([~apsin (30) (250 m.g.)'], [~tonsillitis']). Adopting the procedural interpretation of this rule, consideration of the medical facts given in Figure 4 shows t h a t within the scope of this limited range of medical knowledge, the only candidates for Complaint t h a t could make this rule succeed are Complaint = [] and Complaint = [~sore throat'] and Complaint = [~infected tonsils']

Text, Competence and Logic: An Exercise

229

The first reproduces the entry provided by the doctor; the second and third show the patient's versions t h a t might have been offered. However, because these symptoms are retrievable through the inferential competence of the reader, specifically citing them in the entry is unnecessary, and liable to provoke further inferences about why such ~redundant' information has been provided. The second example is slightly different. The item recorded C Ted up' ") and its inverted commas allow us to understand t h a t it is a s u m m a r y of the patient's presentation of the complaint. The absence of category items concerning either an assessment or a form of m a n a g e m e n t would allow any doctor to infer that, given the normal enquiries, no serious evidence could be found to w a r r a n t a particular assessment or treatment. Such an item, standing alone, would be very likely to lead doctors to assume t h a t the patient's problem was trivial; it failed to warrant medical intervention. In other words, by absenting particular category items the doctor allows the reader to infer what was thought about the case2 Again, the description of the item and the absence of other possible category items are carefully designed with consideration of what can be inferred. In the third fragment, we find an item recorded for both categories in the class, the patient's complaint. Given what has been suggested concerning the inferential work of readers, one might wonder why the doctor bothered to record 'Teeling sick." However, the kinds of information a reader could infer from ~depressed," as an assessment of the patient's complaint, such as tired, moody, perhaps even tearful, would not include 'Teeling sick." Hence, the doctor's knowledge of the inferential work which might be expected from a reader indicates that ~Teeling sick" has to be included in the entry if the presence of this symptom is to be conveyed to the reader. To manage this and similar entries, 6 the list of ~medical facts' can be extended by further relations: is a__symptom of ([Symptom], [Assessment]). is__a__treatment for ([Treatment], [Assessment]). If these are inserted in the rules after all the other relations for symptoms and treatments, a procedural interpretation will try them only after all the previous relations have failed to match. The variables, Symptom, Assessment, and Treatment will match a n y complaint, assessment or t r e a t m e n t item and in doing so will be instantiated to t h a t item. This has the apparently unfortunate consequence t h a t the augmented g r a m m a r will again accept any set of complaints, assessments or treatments, no m a t t e r how medically incomprehensible as a collec-

230

QUALITATIVE SOCIOLOGY

tion t h e y m a y be. T h e added r u l e s r e m o v e the c o n s t r a i n t s imposed by the r e l a t i o n s of F i g u r e 4. H o w e v e r , one can t a k e a d v a n t a g e of t h e fact t h a t only ~inconsistent' e n t r i e s will ever m a t c h t h e s e two new rules, all o t h e r s m a t c h i n g t h e preceding, more specific relations. A s u i t a b l e ~sign a l l i n g device' m a y be a p p e n d e d to the new rules. The clause ~write" is used in Prolog to display a m e s s a g e to the user, a n d ~nl" to finish a line of text. Thus, the following version of t h e two new r u l e s will signal t h a t the assessment does not correspond with the recorded complaint items, or t h a t the t r e a t m e n t does not seem a p p r o p r i a t e for the assessment: is__a__symptom

is

a

treatment

of([Symptom]):nl, write(Symptom), write (~ is not a recognised symptom of '), write(Assessment), nl. for([Treatment], [Assessment]):nl, write(Treatment), write( ' is not a recognised treatment for '), write(Assessment), nl.

U n l i k e the other rules, these two h a v e only a p r o c e d u r a l m e a n i n g ; t h e y h a v e no c o r r e s p o n d i n g d e c l a r a t i v e significance. I n t e r - E n t r y Defeasibility T h e description of a c o n s u l t a t i o n m a y not only be produced t h r o u g h c o n s i d e r i n g t h e overall i m p r e s s i o n it gives, b u t m a y also be composed w i t h r e g a r d to t h e description of p r e c e d i n g consultations. Fragments 11, 12 and 13 26/11/70 c conjunctivitis Albacid 10% 3/12/70 c eye now appears virtually normal Neospurin 10/6/74 6/7/74 2/8/74

c schizo cert. 4/52 c cert. 4/52 c cert. 4/52

12/4/75

c ~weepy', tired depression. Valium (10 m.g.) (30) cert. 1/52 c valium (10 m.g.) (30) cert. 1/52

19/4/75

Text, Competence and Logic: An Exercise

231

Above there are three collections of examples, each collection gathering entries from a particular region in a patient's medical records. Given what has been said, consider the difficulty for a doctor confronting an entry such as the following. 3/1/71

c eye now appears virtually normal Neospurin

The entry contains an assessment of the progress of the patient's com~ plaint, not an assessment of what the trouble is. Not knowing the original problem with the eye, the doctor might find it difficult to comprehend what was going on. The treatment, Neospurin, provides some hints, but would leave the field open to blepharitis, styes and corneal ulcers, as well as conjunctivitis. Descriptions such as %ye now appears virtually normal, Neospurin," ~cert. 4/52," or "valium (10 meg.) (30) cert 1/52" include neither a patient's version nor a doctor's assessment of the complaint. Although the treatment specified indicates t h a t some assessment was made, the details are not included in the entry. The absence of particular items in an entry can suggest t h a t the doctor should look elsewhere for them, in particular to prior entries, to ascertain whether they have a relationship to each other and to the one under consideration. In the examples above the reader would turn to ~'conjunctivitis. Albacid 10%," ~'schizo cert. 4/52," and "depression." The reason for t u r n i n g back to previous entries is t h a t illnesses often take a number of consultations, each consultation dealing with the development of the same problem. An example is conjunctivitis in the first fragment above. Accordingly, in describing a consultation it is import a n t for a doctor to provide the reader with the impression of the course of the illness and its related consultations, t h a t is, a sense of its career. It is also considered unnecessary to repeat information, such as the diagnosis, which recurs in each consultation. By t u r n i n g back and considering previous descriptions, the doctor can find just the information needed. Thus the machinery which provides for the occasioned relevance of items, also provides for their coherence and thereby instructs doctors as to where the relevant information m a y be discovered. In describing a subsequent consultation the doctor designs the entry with regard to the previous entries dealing with the same complaint. The doctor can assess the additional information which is required to describe the subsequent consultations. As in the first fragment, changes in the state of the complaint and t r e a t m e n t details may be recorded. Thus describing a subsequent consultation involves a careful balance of accuracy and detail, and may have the consequence of requiring the

232

QUALITATIVE SOCIOLOGY

reader to turn back and elaborate the relevant previous entry or entries. The way in which the descriptions of a number of interrelated consultations are built provides the reader with the resources through which he or she can infer the sequential relationship between them. A reader of the records, confronting interrelated descriptions, elaborates the whole course of a particular illness, its beginnings, its development and its outcome (if it is complete). A proper understanding of the descriptions and the illness can only be achieved by elaborating all the relevant descriptions. Thus, it is necessary for the program to deal with the sequential character of the entries and the defeasibility of categories across entries in the records. To handle circumstances in which no category item is recorded for either the patient's version or the doctor's assessment, a rule can be added to specify t h a t previous entries should be consulted. With this would be coupled the stricture that the assessment in the previous entry should serve as a candidate assessment in the present entry. The rule would also have to include some stipulation concerning the time difference between the present entry and the previous one. A general version of the stipulation might be: ~'if the previous entry occurred less t h a n one month ago then treat the last entry as relevant to the present one and so on." A more specific characterization might involve types of illness and their recurrently associated duration. Extensions of this kind are illustrated in the following version of the ~'join" rule. join([Date,Location,Treatment], [Date], [Location], [], [Assessment], [Treatment]):previous__entry, illness(Assessment,StartDate), lasts(Assessment, Duration), TimeSinceIllness is Date - StartDate, TimeSinceIllness < Duration. This extended rule references three further new rules and relations: "previous__entry," ~'illness" and ~'lasts." The rule for ~previous__entry" requests the previous entry on the record card, and then uses the "entry" rule to parse t h a t entry (and in doing so may possibly need to consult its previous entry). previous__entry :nl, writeCPrevious entry is: '), read(PrevEntry), entry(PrevEntry).

Text, Competence and Logic: An Exercise

233

The "lasts" relation notes the maximum duration of particular illnesses. For example, the following records the fact that conjunctivitis is typically cured in less than 30 days: lasts(['conjunctivitis'], 30). One further refinement of the rules is necessary in order to accommodate inter-entry defeasibility. Previously obtained inferences need to be recalled when parsing the current entry; this may be achieved by adding an extra goal to the rule for an entry: entry(AnEntry) :-join(AnEntry,Date,Location,Complaint,Assessment, Treatment), date(Date), location(Location) patients__version(Complaint) doctors__assessment(Assessment) management(Treatment) is~a__symptom offComplaint,Assessment), is a treatment__for(Treatment,Assessment)~ remember(Assessment,Date). This goal, "remember(Assessment, Date)," remembers the assessment and the date of the consultation by invoking the rule: remember([Assessment], [Date]) :-assertz(illness(Assessment,Date)). "Assertz" is a Prolog goal which has the effect of adding a new rule or relation to the rule s e t - i n this case, a relation that records the illness the patient is suffering from and the date of the consultation, so that these facts can be recalled subsequently. The action of the expanded join rule should now be apparent. First, a previous entry is requested and parsed. The last goal invoked in parsing that previous entry is ~5"emember," which inserts a record of the illness and its date (the date of the previous consultation) into the rule set. The goal " p r e v i o u s - e n t r y " has now succeeded, and the next goal in the join rule is attempted. This sets the variable "Assessment" to the illness just recorded. The "lasts" goal is tried to obtain the normal maximum duration of that illness, and a check is made that the current entry is still within the time span of the illness. If all is satisfactory, the illness is returned as a candidate assessment to be considered with the t r e a t m e n t explicitly mentioned in the entry. As part of parsing the current entry, the rule is able to obtain deductions made on the basis of prior entries, and so reproduce a doctor's reading of that entry.

234

QUALITATIVE SOCIOLOGY Conclusion

Drawing from an ethnographic study of a particular form of t e x t - t h e medical record cards in primary health c a r e - t h i s brief essay develops a preliminary model for the constitution and elaboration of entries in the records; a model which provides for the computerization of anonymous and bureaucratic sources of information. The model parallels or mimics aspects of a set of narratives used by professional personnel in their ordinary practical actions of producing and recognizing entries in the medical records. It provides a set of rules or procedures which allow a doctor to constitute an ~adequate' or treasonable' entry in the medical records and a collection of resources for decomposing or retrieving the sense of textual items. Moreover, the model does not simply retrieve sets of particular items and information but rather co-ordinates items within an entry and interrelates items across entries. In so doing the model and its rules provide a systematic basis for the documentation and elaboration of information from medical consultations that is sensitive to the practical circumstances and ordinary competence of professional personnel. Given the significance of rules within sociological explanation, it is perhaps important to clarify the concept of the rule used within the model. As suggested, the rules captured in the model are designed to parallel or mimic narratives or procedures actually employed by professional personnel in describing and recognizing textual versions of medical consultations. The model is thus designed to capture publicly orientated to, socially organized sets of procedures. In the same way that following a rule can provide for the comprehension of an action, so the rules within the model capture aspects of the production and recognition of an entry in the records; the rules are constitutive of the %ureaucratic' description and interpretation of medical consultations. Moreover, the rules of the model are in no sense deterministic but rather flexible and defeasible with respect to the practical circumstances at hand. The rules are a particular rendition of some aspects of the competence in describing and interpreting medical consultations for the medical cards. In attempting to develop an ~inteltigent" system which captures aspects of public ~rules,' we hope to suggest one way in which sociology may be pertinent to Artificial Intelligence. In recent years, within sociology, there has developed a body of vigorous empirical analysis of human practical reasoning and the interactional structures of social actions and activities. Such work describes sets of systematic procedures used

Text, Competence and Logic: An Exercise

235

a n d o r i e n t a t e d to by m e m b e r s of society in the p r o d u c t i o n a n d recognition of t h e i r practical actions. The work w i t h i n this field has direct bearing upon m a n y of the issues at the center of Artificial Intelligence, such as n a t u r a l l a n g u a g e u n d e r s t a n d i n g a n d the c o n t e x t u a l issues of h u m a n vision. D r a w i n g u p o n a n d d e v e l o p i n g this sociological work, we c a n begin to g r o u n d i n t e l l i g e n t s y s t e m s w i t h i n aspects of social o r g a n i z a t i o n a n d t h e r e b y solve a few of t h e p r o b l e m s w i t h i n Artificial Intelligence w h i c h h a v e evolved from a n over psychologistic notion of h u m a n cognition a n d b e h a v i o r (Gilbert a n d H e a t h , 1985). In 1980 Dr. Clifford M a y of the Royal College of G e n e r a l P r a c t i t i o n e r s s u g g e s t e d ~that a c o m p a t i b l e c o m p u t e r s y s t e m could (and should) be in w i d e s p r e a d use in g e n e r a l practice in five y e a r s a n d adopted by virtually all practices in t e n years." As far as we are aware, t h e s e r e m a r k s were m a d e p r i m a r i l y in r e l a t i o n to the c o m p u t e r i z a t i o n of medical record cards. C e r t a i n l y one can e n v i s a g e h o w useful c o m p u t e r s y s t e m s in g e n e r a l practice m i g h t be, not only in r e c o r d i n g a n d r e d o c u m e n t i n g inf o r m a t i o n , b u t also in p r o v i d i n g a l t e r n a t i v e t r e a t m e n t p r o g r a m s , offering differential diagnosis, a n d t h e like. W h a t e v e r s y s t e m is finally adopted, it is i m p o r t a n t t h a t it is sensitive to a n d derived from the day to d a y issues a n d practices of medical personnel. It is here t h a t s o c i a l ogy c a n m a k e a c o n t r i b u t i o n ; in a t t e m p t i n g to develop p r o g r a m s w h i c h are e m b o d i e d in local convention, o r d i n a r y practice and social organization.

Reference Notes 1. The fragments from medical record cards were taken from Heath (1982), who collected them during a two year observation and interview study of an urban general practice health center. 2. The ~__'character is used merely to clarity that ~doctors._.assessment' is one category name. 3. These rules, and those developed from them later in the paper, omit consideration of the parsing of the Date in the record card entries, so as not to complicate the exposition. 4. This paper only presents that part of Prolog required for an initial understanding of the rules; there is much more to Pralog than wiI1be evident from the paper. Clocksin and Mellish (1981) is the standard text on Prolog. Kowalski (1979) provides a deeper discussion of the relation of Prolog to logic. 5. It is possible that the inference that there is nothing much wrong with the patient depends both on observing that the presenting symptom, Ted up,' allows for a very large number of potential assessments, and on failing to find further information in the record card that would enable a more precise diagnosis to be identified. This kind of inference is of interest because it requires reasoning about other inferences (in this case, the number of inferences which can be made), that is, it is a metainference. 6. Other, more sophisticated methods of dealing with such entries are possible.

236

QUALITATIVE SOCIOLOGY

References Bittner, E. and H. Garfinkel 1967 ~Goodorganizational reasons for %ad' records." In H. Garfinkel studies in Ethnomethodology. Englewood Cliffs, NJ: Prentice Hall. Clocksin, W. F. and C. S. Mellish 1981 Programming in Prolog. Berlin: Springer-Verlag. Garfinkel, H. 1967 Studies in Ethnomethodology. Englewood Cliffs, New Jersey: Prentice Hall. Gilbert, G. N. and C. Heath (eds) 1985 Social Action and Artificial Intelligence. Atdershot: Gower. Hart, H. L. A. 1968 Punishment and Responsibility: Essays in the Philosophy of Law, Oxford: Clarendon Press. Heath, C. 1982 ~'Preserving the consultation: Medical record cards and professional conduct." Sociology of Health and Illness 4: 56-74. Kay, C. at al. 1980 Computers in Primary Health Care. Occasional paper No. 13 London: Royal College of General Practitioners. Kowalski, R. 1979 Logic for Problem Solving. New York: North-Holland. Levinson, S. C. 1983 Pragmatics. Cambridge: Cambridge University Press. Rawls, J. 1955 ~'Twoconcepts of rules." Philosophical Review 64: 3-32. Sacks, H. 1972 ~On the analysability of stories by children." reprinted in Turner, R. (ed) Ethnomethodelogy (Harmondsworth: Penguin, 1974. Searle, J. 1981 "Minds, brains and programs." In Haugeland, J (ed.) Mind design, Vermont: Bradford Books. Taylor, S. 1954 Good general practice. Oxford: Oxford University Press.