{bowker and star} situations vs standsards in long term whide scale decision manking the case of the...

8/3/2019 {Bowker and Star} Situations vs Standsards in Long Term Whide Scale Decision Manking the Case of the Inter Nation l Classic at Ion of Disseases

http://slidepdf.com/reader/full/bowker-and-star-situations-vs-standsards-in-long-term-whide-scale-decision 1/9

Situations vs. Standards in Long-Term, Wide-Scale DecisionMaking: The Case of the International Classification of

Diseases

Geoffrey Bowker Susan Leigh Star

Department of Sociology and Social AnthropologyUniversity of Keele

Keele STS SBG [email protected]

This paper presents a case study of he development andevolution of an organizational decision support system (ODSS)over a long periodof ime andwide geographical area. It usesthe design of he International Classification ofDiseases (ICD)to address issues oforganizationaldecision making in large,dispersed organizations. Special attention is paid to the tensionbetween standards and local contingency in such systems. Fourkinds of "wins" are proposed in this type of decision setting:contingency wins, standardization wins, delegation wins, andtranslation wins.

The study is based on a longitudinal analysis of he developmentof he system, currently administered by the World HealthOrganization (WHO). It looks at a number of trategies thedesigners of he system have attempted in balancing the needfora "universal" classification scheme with the pluralism entailed bydifferent medical specialties, different national medical cultures,a large, changing bureaucracy, and the evolution of computing.

Introduction

Individuals and groups make decisions in very different ways,involving different use of resources and negotiating processes.Moving up to the level of organizations requires yet another

change in resource configuration and negotiations.[l] This is

especially the case when discussing large, distributedorganizations. Discussions of group decision-making processeshave commonly characterized three possible outcomes: truthwins, status wins, and majority wins. This is a description of

outcome in relatively small group interaction, either face-to-faceor electronically mediated. Very large organizations makingcomplex decisions have such groups and processes embedded inthem, of course. But we think that there are as well processessui generis to much larger organizations, and which areimportant for decision making and problem solving. Theprimary kinds of processes we describe here are those where theorganization is attempting simultaneously to impose standardrepresentations, and to represent multiple viewpoints fordecision making.

Turoff and Hiltz describe this problem succinctly as it appears atthe group level: "A management or professional group cannotmaximize consensus and quality in its decision at the same time.It has to choose which goal is more important. Appropriate

structures can be provided to promote either egalitarianparticipation and lots of opinions or strong leadership and ahigher probability of reaching complete consensus." [2] Yet in

0073-1129/91/0000/0073$01.00 © 1991 IEEE73

large organizational decision-making processes, one groupcannot impose s trong enough leadership to produce consensus.Standardization may replace "leadership" in the attempt tomanage the difficulties imposed by pluralism, and yet nostandard or imposed representation can answer to the situated,contingent nature of problem solving at another locale, as wehave been learning from the new wave of cognitive science.[3]Elsewhere, we have discussed this as the problem of

organizational due process. [4] Taking off from the typology of

decision-making outcomes described above, we conclude thatthere are four kinds of processes that are used to resolveproblems of the tension between standardization and localsituation: contingency wins, standardization wins,delegationwins, and translation wins.

We draw ou r arguments and materials for this analysis primarilyfrom an historical case study of the development of theInternational Classification of Diseases (lCD). This is a listofcauses of death and disease currently administered by the WorldHealth Organization. It is about one hundred years old, and hasbeen revised nearly every ten years since the nineteenth century.The ICD is distributed as a book (or sometimes embedded inmedical record-keeping software) to public health offices,hospitals, and bureaux of vital statistics throughout the world. It

contains numbers which correspond to causes of death orillness, and algorithms for arriving at those numbers in complexcases involving more than one disease or cause. In a sense, theICD is the backbone of a sophisticated organizational decision

support tool. On the basis of data collected using the ICDsystem, decisions are made about allocation of medicalresources, whether and how to control epidemics or endemicillnesses, and whether there are shifts in population based on

infant mortality rates, etc.We

were only able to locate a copyof

Anne Fagot-Largeault's magisterial discussion of the ICD afterthis paper was written. She emphasizes causation rather thanorganization; but is fully aware that the two arecomplementary. 5]

We understand that this paper is an unusual one in theODSS/GDSS literature. We are speaking of decision-makingand p ~ o b l ~ - s o l v i n g tools and processes that span manyorganIzatIons and take place over very long periods of time. Inaddition, much of the ICD and tools like it are infrastructural tothe kinds of other decisions more commonly covered in theliterature. These are deeply buried in statistics, databases, orthe records of bureaucracies. We are far from recommendingspecific design strategies for building support tools for suchdecisions. Yet we believe that conceptualizing the role of suchtools is crucial to tbuilding a broader, more sociologicalunderstanding of decision-making itself. We are hereresponding in part to Sprague's call for general theories of

problem solving to inform the DSS/GDSS/ODSS literature. [6]

Sainfort, et. all similarly call for attention to problem solvingprocess, saying "it is almost impossible to determine the exac,



contribution of the DSS to the final outcome since it is almostimpossible to know what the same problem ~ o l v e r in the same

situation at the same time would have done without the DSS. [7]

In the I1!atter of long term, large scale decisions, such putativecontrol IS neve r really possible -- thus the attention toprocess iseven more critical.

Some of the most urgent problems facing large organizationsconcern the proper relationship between people, technologiesand tasks.[8] Those tasks are often highly distributed over timeand space. Large-scale software development projects, fore x ~ p l e , take place at many locations, and design and safetydecISIons made by and for groups which never mee t face-to

f ~ c e .. We d r a ~ h ~ r e . o n fields of sociology of technology anddistrIbuted artifiCIal mtelligence for useful concepts in~ n d e r s t a n d i n g organizational decision-making, including theIdea of boundary objects and the French Annalist approach to thestudy of lists and bureaucracies.

. The. Nature of Lists as Decision-Making ToolsLIst making has frequently been seen as one of the foundationalactivities of advanced human society. Goody argues that thefirst written records are lists (of kings, of equipment). [9]

Foucaul.t and T ~ r t have, in their different ways, claimed that theproductIOn of lIsts (of languages, races, the minerals animals)revolutionize? science in the nineteenth century and directly

to m o ~ e r n . SCIence. [10] Latour and others have proclaimed thatthe pnme Job of the bureaucrat is to compile lists, which can

~ e n be shuffled and ~ o m p a r e d one to another.[ll] What these

d i v e ~ authors have m common is that they have turned theirattentIOn away from dazzling end products in the various formsof Hammurabi's code, mythologies, the theory of evolution, the

welfru:e ~ t a t e and on. .They have instead looked at just whatwork IS mvolved m making these productions possible. In eachc ~ s e , they have dusted off the archives and discovered piles andplIes of lowly, dull, mechanical lists.

L . i s t - ~ ~ is foundational as a way of coordinating activitydistrIbuted m tIme and space, and of passing information backand forth across parts of an organization. Consider anapparently simple problem of coordination that children in manycl1l:tures. solve routinely: the treasure hunt. In this game a list of

objects IS made, usually by an adult, and teams of children areeach given an exact duplicate of the list The first team to bringback all iteffi:S on the list ~ i n s . Even a local, improvised listsuch as thIS entaIls all sort of Judgment calls: objects should be

e n o u g ~ to fmd to challenge the children's' ingenuity, but

n o ~ I m p o s s I ~ I ~ ; they should match the kids' resources (e.g., noobjects requmng use of a car to fmd). Typically they are thingsthat are odd but ~ o t impossibly rare -- a copy of the front page ofthe New York TImes from June 4,1964; a green high-heeled lefts h ~ . :reams may decide coordinate their internal work byasslgnmg each person an Item, or working in pairs, or movingas a group, and so on.

When lists are used to coordinate important work distributedwidely over time and space, and to inform more importantdecisions than where to look for an old green shoe, acorrespondingly complex organizational structure andinfrastructure evolves. The judgment calls are still there butinvolve multiple actors, both persons and organizations 'themselves. The decisions about division of labor remain butnow e n ~ l bureaucracies as well as local conventions. As all theauth?rs ~ I t ~ abo,,:e h a v ~ concl?<ied, large-scale decisionmaking IS ImpOSSIble WIthout lIsts. These in turn entrain wholeseries of substantive political and cognitive changes in the

classes they inventory.

74

The ICD is a particularly powerful list that has received littleanalytic attention from social scientists. We draw severallessons from our case study of the ICD:- first, there is a permanent tension between attempts atuniversal standardization of lists, and the local circumstances of

those using them;- second, this tension should not, and cannot, be resolved byimposed standardization, because the problem recurses;- third, rather, from the point of view of coordination,ad hoc

responses to standardized lists can be mined for their richinformation about the heterogeneous knowledge domain, andinformation technology built which will support those needs, not

subvert them;- fourth, making this sort of list is an example of the creation of

the sort of object which must satisfy members of worlds or

organizations with conflicting requirements. In its creation, andlater in its use, the compl ex list is a kind of knowledgerepresentation particularly useful for coordinating distributedwork, which often contains requirements of this sort. Some,

ourselves among them, would argue necessarily conflicting.[12]The Impact of the ICD

The ICD was one of the tools bound up in the origins of thewelfare state: the epidemiologists and government statisticianswho conce ived it were concerned with large scale public health

n:teasures d ~ n g the times of the cholera epidemics. [13] It hass I l e n ~ y (u!lIike its notorious c?usin in psychiatry, theclaSSIfication system of psychIatric diseases, DSM I-IV)

accompan ied all major epidemiological work this century.

The resource-allocation power of a classification of diseaseappears in the debate about Britain's mortality decline in the

nineteenth century. [14] Three interest groups have at differentt i m e ~ claimed p rimacy here using different classificationr e ~ g s - .and.a share f u n d i n ~ and recognition appropriate totheIr contrIbution: medical spec!a/ists who claimed new forms of

treatl?ent rid the C?untry of its major scourges (particularly TB);pubbc health offiCials who asserted the value of sanitation in thecities; laissez-faire economists who highlighted the general risein the ~ t a n d a r d .of living in a successful ~ o n o m y unburdened byexpenSIve medical welfare. The modern mterpretation of theoutcome hinges on a reading of the Tables of Mortality whichlisted causes of death by region. These show unequivocally thatthe new forms of treatment developed after the decline inmortality, but in step with local public health measures.

In this century, the I CD has played a similar key role in

determining the outcome of epidemiological, public health andeconomic arguments. We will look at the way it has been usedby different groups, constituting both a common and acustomizable object for these groups. We will look at thetension between the desire to standardize (so as to be able toperform bureaucratic functions such as comparison over timeand space, produce algorithms, compute etc) and the drive of

each in terested party to produce and use its own specific list.will also examine the tension between attempts to make a

uruversally standard list and the idiosyncrasies and localcircumstances of users.

In organizational decision making it is particularly important tounderstand communication processes, as several authors have

recentl y emphasized.[15] Maruyama calls this "multi

ocularity".[16] We have discussed this phenomenon under therubric of "boundary object", drawing originally on our studiesof science and technology.



The ICD as a Boundary ObjectBoundary objects are those which inhabit different socialworlds, which are differently used by members of those worlds.Boundary objects are both plastic enough to adapt to local needsand constraints, yet robust enough to maintain a commonidentity across sites, times, or social worlds. A taxonomy of

different kinds of boundary objects was given in [17] and apreliminary discussion of the use of the concept in organizational

decision making in [181" Two of the types of boundary objects

we identified were repositories andforms. Repositories areordered piles of objects indexed in a standardized fashion, andare particularly good at handling heterogeneous units ofanalysis. They have the advantage of modularity. Forms arewhat Latour has called "immutable mobiles" -- they can be sent

and returfled over distance and time, capturing certain standardinformation and deleting many local contingencies from thedescription. They are advantageous for communication acrosshighly dispersed work groups. To develop this analysis further,let us first inventory the different classes of informationalconflicts involved with building up and using the list, andexamine the types of informational needs and structures involvedin each case.

International Conflicts. One of the values of a list like the ICD isthat it can be used in trans-national comparisons. This is usefulepidemiologically, in that it enables one to trace specificenvironmental and nutritional factors that might be involved inthe occurrence or spread of particular diseases. It also facilitates

the tracking of epidemics and the imposition of any necessaryquarantine measures.

These advantages can only be fully exploited if the various statesagree on the way information is collected and coded. However,a continuing problem has been that different countries have senttheir information in more or less promptly. In the 1920s, Franceand Portugal were notably slack. Further, once the informationarrives, it is often of variable quality - countries with large ruralpopulations rmding it difficult to give the same sophisticatedmedical treatment of each case as heavily urbanized, Westerncountries. [19] At one stage in the USSR, no attempt was madeto compute causes of death in places with less than 10,000inhabitants! [20] Different states also have different bureaucraticstructures - for example, in the nineteenth century, statisticswere run by a central service in Italy but were broken down byprovince in France.[21] The regulations for death certificateshave made an appreciable difference to the results of the ICD.

Thus in Germany in the 1920s there was no separation betweenthe civil statement of the cause of death and the cause of deathissued for statistical purposes. In Switzerland, on the otherhand, the statistical cause of death was confidential, making itmuch easier for doctors to cite causes that might distressrelatives (and upset insurance companies). When Hollandswitched over to the confidential system in 1927: "There was aconsiderable increase in Amsterdam of cases of death fromsyphilis, tabes, dementia paralytics, aneurism, carcinoma,diabetes, diseases of the prostate and suicide, while deaths frombenignant tumors and the secondary diseases such asencephalitis, sepsis, peritonitis etc showed a falling-off". [22]

Further, different cultures place differential emphasis onparticular causes of death. A recent caseof this has been thecurious case of Japan's low rate of fatal heart attacks. Atraditional reading of the list has suggested that this statistic isdue to nutritional or environmental factors peculiar to thatcountry - level of fat in the diet, etc. Recently some

epidemiologists have suggested that the 'cause' may well be thatheart disease is a very low status cause of death within Japanese

75

culture, suggesting a life of physical labour and a physicalbreakdown. Accordingly, what we would call heart attacks areoften classed as strokes, since an overworked brain is moreacceptable. When this is factored in, they suggest, there is no

discrepancy in Japan's figures. These national differences arecomplicated by the facts that some diseases present differently indifferent countries. AIDS is one such; malaria another. For thelatter, E.J. Pampana noted that: "At a first glance, malaria doesnot appear to have an international character at all; one could

almost say that no other disease is so strictly dependent on localconditions. Malaria might, in fact, almost be called anationalistic disease, because it takes from the country its verycharacteristics, as does its folklore. These very localistic aspectsof malaria epidemiology are the bricks with which the science of

malariology is built." [23]

Different national schools of medicine can also havedisagreements about such issues as simultaneous causes of

death. One WHO committee noted that there were indeed suchdifferences; and that if there were no agreement by 'reason',then countries would vary according to : "facts of pathology (or)clinical medicine, (or) public health importance". [24] Itrecommended that the different countries produce a table ofcontributory causes for comparison. The problem becameunwieldy. In the Census Manual of the International List ofCauses of Death there were 8300 terms, representing 34 millionpossible combinations. I f even half the terms could becombined, then an assignation of priority in all possible cases

would involve 61 volumes of 1000 pages each.[25]

Finally, managing the ICD has been a charged issue ininternational politics. Originally, it was run by the French OfficeInternationale d'Hygiene Publique (OIHP); and was seen by theFrench government as a sign of their natural lead in internationalpolitics. Indeed, when the League of Nations started to gaincontrol of the production of the list, one British diplomat notedthat: "an influential clique in the French Foreign Office ismoving heaven and earth to retain the Office Internationalunaltered". [26] The United States became key actors when theyrefused to join the League, leaving the OIHP to mediate betweenthe US and the League of Nations. So doing, they tried to gainmore control by squeezing out the International Institute of

Statistics from their advisory role. The director of thatorganization complained that: "The new masters of the world arelaying down their law, without any consideration for the rightsof others and for an international organization that had received

universal respect to that time." [27]

Relationships between developed and underdeveloped countriesalso figure in the design of the ICD. For the former, with stateof-the-art computing capacity, some kinds of artificiallyintelligent systems could handle data with more flexibility anddetail than has previously been possible. However, the resultantlist is often not useful for third world members of the WHO,who lack computers capable of implementing the software.Even with computing power, this level of granularity is notalways considered necessary for countries where the vastmajority of deaths are caused by infant diarrhoea andcontaminated water. Until these issues are solved, who caresabout the incidence of rarer diseases? The question is notrhetorical - other member nations do care, since they want to be

able to trace their own epidemics (flu, AIDS etc) throughout theworld so as to get a picture of their etiology and development.



So international co-operation was hampered within each nationby the diversity of ways of recording and reporting causes ofdeath, by local cultures' rating of the prestige of certain diseases,by local medical cultures with respect to medical controversiesand by the different national characterof some diseases. It washampered between nations by the issueof who would controlepidemiology, and by the differing national medical needs. Thepublic health actors involved before this apparently simple,homogeneous list could even be compiled included governmentofficials, statisticians, anthropologists, medical analysts,epidemiologists and diplomats.

Government: The State vs. the Individual . Another series ofactors emerges whenwe turn to the relationship between thestate and the individual. We have already referred to theproblem of ensuring the confidentiality of death certificates - asequence of perforations in a piece of paper enabling vastly moreaccurate statistics for certain classesof disease.

The classification of death by suicide also comes under thisrubric. Early in this century, many doctors complained aboutthe detailed breakdown of this category, which had 'noprophylactic value'. Statisticians responded that the detailsshould be recorde d "for their sociological interest and for thepolice", and that the Justice Department rather than the medicalprofession should defme the sub-categories. When this wasdone, however, some moral and political distinctions wereinscribed directly into the list Thus: "In the case of collective

suicides, you have to count as many suicides as there are peopleover the age of majority. Minors have to be considered victims

of murder". (28) Similarly, death by starvation was said to be a'crime' if children suffered it; a 'misfortune' if an adult cause of

death.(29) Abortion was a nother similar case. When criminalabortion was defme d in an undifferentiated way as homicide(legal abortion had its own category), it was hard to get statisticsabout the distribution and safety of different procedures. Thedefinition of stillbirth was as well a political and religiousdecision varying by nation and Christian sect whether a foetusthat had never breathed (or had 'tried to breathe at leastthreetimes but failed') was to be recorded as a death. If so, it wouldboth contribute to infant mortality statistics and have a soul; ifnot, the mother would jus t be recorded in the morbidity tables assuffering a miscarriage. We are not arguing here that the listshould have been expunged of all moral and political categories.On the contrary, we will later suggest that this is not possible.At this stage, we merely wish to point out a new set of actorsenters into the scene around the very thorny questionof the

relationship between the state and the individual: politicalinterests, religious groups and ethicists.

Corif/icting Needs ofDoctors, Epidemiologists and Statisticians.The task of mling in the death certificates falls on the doctor. Heor she does not necessarily see the value in accurately fillingin acomplex form - this patient, after all, is dead, and is the time

not better spent on the living? Further, some doctors havecomplained about any degreeof government obligation: "thedoctors cling to their independence and ... for the vast majorityare imposed to any any intervention on the partof the

government or of some bureaucrat".(30) When it comes to useof the tables produced with the list as a basis, in general :"practicing specialists want more categories and urban

statisticians want less". (31) For the specialists it is of particularinterest to know the breakdown of each disease strain, whereasfor the urban statistician suggesting public health policy broadercategories like nutrition and sanitation are more relevant. This

has at times led to a double bind: "So-called administrativestatistics have no value in the eyesof practitioners, who as aresult are completely uninterested init; whereas unless these

76

practitioners provide exact data, then the scientific valueofadministrative statistics has to be called into question".(32)

An example of this tension is to be found in the specialtuberculosis sulrcomrnittee of the League of Nations trying todevelop a new four digit code for use in sanitaria. "Public healthworkers wa nt to know: Doe s a person really have tuberculosis,and, i f so, is it infectious? The worker in vital statistics wantsdetails c oncerning tuberculosis and its relationship to populationgroups. The clinicians' principal interest in classification is is tohave a guide which will help him be defmitive in diagnosis andtreatment of his patients." They noted that it was "exceedinglydifficult for a classification to m eet all these requirements".Current codes ha d been rendered more inadequate by the adventof mass X -rays du ring W orIdWar Two, when millions ofapparently healthy people were found to have early formsofTB:"for which there were not sufficiently exact designations". (33)

. Pressures from the different groups spoke to issues at the coreof the design of the ICD. Statisticians, for example, wanted thefirst ICD to have only 200 categories, since a statistical 'table' asused in censuses could only be a pproximately 200 lines long.For them, lists had to be stable over time and space, forcomparability. This led to a choiceof a particular kind of list,not specifically tied to ca uses of death: "This is why diseasesmust be c lassed according to their seat and not their nature ortheir cause. Because the seat [location] is much more easy todetermine than the nature". [34] They stressed that the role of thelistmakers was not to produce 'philosophy' (which we can read

as referring to the divination of essential causes) but a 'truthful'and 'comparable' list. Spanish authorities wanted the listofgeneral diseases to follow public policy, breaking them down asfollows: general and sporadic; epidemic; imported; common topeople and animals; professional intoxications.[35] Thus, too,one set of statisticians wanted to rearrange the list so that socialbiological factors would be grouped first, and later categorieswould only be used if these could not be filled. Thus 'puerpuralstate', 'malformations' and 'early infancy' would go to the top;not in order to give the m preference: "The point to stressissimply that in statistics the social-biological viewpoint comesfirst and the medico-aetiological second, while the anatomicfunctional is only third in orderof mportance." (36) 'Violent

death' should move up the list too, since this would: "settlevarious doubts ... as to whether consequences due to visible

external cause s are to be classified here, or, for example, underinfectious diseases (a case in point is infectious diseasesofwounds)". Further, there should be a subdivision for diseasesfor which statistics were required under international

conventions (e.g. lead poisoning). (37)

This pluralism brings us uncomfortably close to Foucault'sfamous list. He speaks of an ancient Chinese taxonomy dividinganimals into such fanciful-seeming categories as "those thatbelong to the Emperor, those that are trained, mermaids andfabulous beasts, those that are included in this classification, andthose that 'resemble flies from a distance"'. [38) There is nohomogeneity to the ICD. It is not so much a list of causes ofdeath as a series of dynamic compromises between a wide rangeof players in a number of different dimensions. Or, as oneobserver noted: "In short, the nomenclatureof diseases and ofcauses of death established for the needsof statisticalorganization constitutes a sortof contract between the twoorganizations who are charged with statistical works - that is tosay the service who makes the observations and that whichproduces statistics with the help of these data".(39)

Industrial Actors . We have already said enough to indicate thata vast range of people from a series of different social worldshad a stake in how the ICD was compiled and used. Three other



groups that we will not look at in detail in this paper are:- insurance companies These groups wanted a breakdown of

the ICD statistics in such a way as would be useful for them:"For example, there should be groups corresponding to the ageat which direct compulsory sickness insurance begins, and theage at which compulsory old-age insurance starts". [40]

- industrial firms Some of the fIrst groups to produce lists of

causes of death were from the vas t German chemical companiesof the late nineteenth century. For them, relevant variables werewhether the decea sed had touched/not touched certain

compounds, had worked inside/outside etc. Again, a differentset of variables from those interesting other groups;- pharmaceutical companies The claims that can be made fordifferent drugs is i n part a function of the list of diseases. Aclassic case of this is cited by Law and Bijker: because of

religious restrictions, the Spanish pharmacopeia defines wha t wewould describe a s birth control pills with a side-effect of causinghigh blood pressure as a treatment for hypotension with the side

effect of inhibiting birth. [41] Or it can work in a nother way.One of us had a student who was a representative for a largedrug company. Part of her job was interviewing doctors aboutwhether any of their patients had gotten better from one diseasewhile taking one of the company's medications for another. I fyes, that disease might potentially be added to the listof

indications for the illness. The student said that she wasconstantly pressured by he r supervisors to "broaden herindications. "

We will not attempt here to list all the actors involved incompiling and implementing the lCD, but it should be obviousthat something has to give; the list cannot be homogeneous,neutral and appeal to all parties. This is typically the case for

boundary objects, which inhabit a number of different socialworlds simultaneously.

Coordination, Organizational Decision-Making andthe ICD

A number of good people have been working on the problemsposed by the ICD for many years. What sortsof solutions havethey proposed? In this section we inventory a numberof

approaches. It should be emphasized from the outset that someof these solutions are consciously applied and others haveappeared by default, still others became embedded inbureaucratic "decisions."

Distributed Residual Categories. Indeed, our first solution -garbage categories - might seem to be no solution at all, butrather a studied avoidance of the problem. It does, however,

offer some interesting insights. By 'garbage categories' wemean that array of categories where things get put that you don'tknow what to do with - the ubiquitous 'other'. In midnineteenth century Paris, more than 10% of causes of death were

'other causes'. [42] In Berlin at the turn of the century, it washard to get doctors to complete valuable morbidity information;thus one table gave acute bronchitis1571, chronic bronchitis

225 -- but bronchitis, without any other qualifier, 12844. [43]

There were three general causes for the creationof garbagecategories, or "Undefined Diseases": "either because there wasnot enough information or because the disease was badlycharacterized or fmally because the doctor failed to formulate acomplete diagnosis". [44] It would be extremely difficult toenvisage a time when there is no need for these categories.Their manageme nt has been a constant thread throughout thehistory of the ICD. A major feature of this management hasbeen their distribution throughout the list. Thus at the time of

the first revision of the lCD, the United States representatives

suggested getting rid of the categories "Eclampsia (nonpuerpural)" and "Children's convulsions", since they were illdefmed (pun unavoidable). The committee rejected thesuggestion, since it would lead to the attribution of too many:

77

"unknown c ause s ... and this would discredit the statisticS".[45]Or again, the vague term'haemorrhage' was kept, with a viewto: "not over-inflating the figures conceming badly defmed

diseases".[46] This distribution went to the lengths of

distinguishing between two types of generalized other, whichone representative had suggested conflating - viz categories35and 36, 'Other diseases' and 'Unknown or badly defIneddiseases': "Proposed conclusion: Each of these two rubrics isvery important. The latter in particular indicates what is missing

from the other figures in their approach to truth". [47] The needto distribute was urgent - Jacques Bertillon estimated that overhalf the causes of death would be 'other' in Paris in 1900 if all

the residua l ca tegories were gathered together.[48]

These categories tend to fix the maximum levelof granularity.They can signal uncertainty at the levelof data collection orinterpretation - forcing a more precise designation could give afalse impression of positive data. The major disadvantage is thata l ~ z y or rushed doctor will be tempted to overuse 'other.' By

therr nature, boundary objects are only manageablei f here is a

zone of ambiguity written into them: in this case, precisedefmitions would drive a wedge between doctor, statistician andepidemiologist.

Heterogeneous Lists. Throughout the history of the lCD, therehas been a great deal of debate about whether i t constituted anomenclature or a classification. The difference is that anomenclature is merely a list which does not give any indication

of cause whereas a classification gives causes. The advantageofa nomenclature is that it can remain more stable over time. Forexample, a nomenclature based on the 'seat'of the disease canlist a series of indications which can then be used at a seconddegree of analysis to re-diagnose in line with current theory.Systemic diseases - like AIDS or Systemic Lupus Erythematosis- can be tracke d this way, even though the category might nothave existed at the time of original diagnosis. Classifica tions aremore c onvenient immediately, but change frequently.

Intuitively it might seem desirable to have a single well-defmedgoverning principle for the ICD. However, as for residualcategories (and for the same reason - the arrayof playersinvolved), the solution that has emerged over time has beenrather to fmd the appropriate levelof ambiguity -- to keep the listas heterogeneous as possible for the different actors to fmd theirown concerns represented. This cashes out in the fact thatalthough the list is in appearance homogeneous, there are at leastfour classificatory principles involved: topographical, the seat

of the disease, which part of the body it manifests in;etiological, the origin of the disease (genetic, viral, bacterialetc.); operational, the responses to certain tests; andethical/political. There is no necessary one-to-onecorrespondence between test results and a given topographicalor

etiological feature (though in general one o r the other isasserted). HTLV vs mv is a case in point. HTLV was definedin terms of a positive reaction to a test searching for antibodies.When what we call HIV initially produced the same reaction,Gallo classified it as an HTLV - even though the virus had notbeen isolated. We have given many examplesof ethical andpolitical dimensions above. The definitionsof stillbirth,abortion, suicide, iatrogenesis and euthanasia, for example, arethe outcome of ethical and political decisions.

Parallel Different Lists. Different groups have found that the listjust did not serve their purposes, and so they have modified it.This could happen in a country with a different rangeof medicalproblems. For example, the frrst ICD was drawn up partlythrough a comparison of the Tables ofMortality of six Europeancountries. Little room was left for tropical diseases, and Mrica ncountries produced their own modifications. Again, specialists



is discussed by Fujimura with her concept of "standardizedpackages." [54)

Conclusion: Coordinating Boundary Objects inDecision-Making

It is unrealistic and counter-productive to try to destroy alluncertainty and ambiguity in boundary objects. By their verynature, boundary objects need appropriate degrees of both inorder to work - only in a totally uniform world 6Yi1hin a givenspeciality) would it be even conceivable to try to impose total

precision. This is discussed by L atour and Serres. [55) Ratherthan root out all instances of ambiguity, decision theoristsdealing with standardized lists should instead seek clearly andconsistently to defme the degree of ambiguity appropriate to theobject.

No boundar y object can be defmed once for all. It is the productof negotiation and change. We have noted three spurs for suchchange. First are changes within one of the social worlds thathas a say in the definition of the boundary object. Thus, medicalspecialists might come up with a new test that causes areclassification of a number of diseases. Second are changes inthe bureaucratic background which increase (or decrease) theapplicability of the object -- convergent (or divergent)bureaucracy. Third are technical changes which allow a bettermatch between the actual uncertainty and that "permitted" by thestandard case. Computerization provides such an example.

In the above ca se study, we can identify four types of processes

for resolving problems of the tension between standardizationand local situation. Contingency wins develops as thestringency of standards and constraints on resources collide -- asin the case of "no list", above, or local certifying contingencies.Standardization wins where it fits the work conditions across avariety of settings, and where the parts of the organization are ingood enoug h agreement about codes and conventions.Delegation wins where the problems cannot be resolved, wherethere is an overload on local resources, or where it is politicallyinfeasible to accept standardization attempts. Finally, translationwins when different actors are able to develop workableboundary objects in the context of stable relationships.

Implications for Designers of Tools for CoordinatedWork and Organizational Decision Support Systems

Despite a growing body of evidence from sociology and historyof science, distributed artificial intelligence and distributedcognitive science, images of coordinated decison-making in

large organizations often involve the attempted imposition ofuniversal standardization schemes. While such standards mayemerge in physical systems or certain sorts of market conditions,for the class of phenomena described here no universal standardis possible. The number of actors, the different ways theystructure information, the "moving target" nature of collectingscientific information over time when the science itself ischanging: all these factors, and more, are true of most importantclasses of problems presenting themselves in both the"coordination theory" and ODSS areas. It is often difficult toimagine building tools whose purpose is to collect precise,uniform, and complete information from a large domain over along time -- and invoke the concepts of ambiguity, fuzziness andplastic meanings for their design. The initial designers of theICD certainly did not intentionally build such features into theirdata collection system; on the contrary, they were devoutpositivists, bent on intellectual and moral recruitment to thetruth. Yet as the capital "T" Truth remained elusive, they diddevelop pragmatic workable compromises, many of which used

those features. It is premature to specify specific tools at thispoint in our understanding of highly distributed problem solving

79

and information systems. But some guidelines emerge at this

point.

1. In the face of incompatible information or data structuresamong users or among those s p ~ i f y i n g th : system, attempts tocreate unitary knowledge categones are f u t l ~ e . Rather, parallelor multiple-representational forms are r e q ~ u r e d . So, forexample, instead of trying to represent a d i ~ o r d e r ~ f energydiagnosed with acupuncture as a nervous disease m Westernmedical terms, a parallel representational scheme will avoid

imposition of inappropriate categories.2. Pragmatically, the "Occam's razor" the.codinginformation means that too few categones will result minformation that is not useful ("alive" or "dead," while havingthe virtues of simplicity and [near] exhaustiveness, don't ~ l l usmuch about disease in the world), while too many categoneswill result in increased bias, or randomness, on the part of thosefilling out the form. An ICD with 5 million numbers may bemore scientifically accurate, but most doctors would not e v ~ n look at such a death certificate. Thus, at the level of encodzngtools need to be sensitive to the working conditions of those

encoding the data.3. Imposed standards will produce workarounds. B ~ a u s e imposed standards c a n n o ~ account f o ~ every l?cal contmgency,users will tailo r standard ized forms, mformatlOn systems:schedules, etc. to fit their needs. A good s u ~ a r y ,?f thIS.appeared recently on a feminist button proclaImillg, One SIze

does NOT fit all!" Gasser identified three major classes ofinformal responses to systems which did not fit local

contingencies: fitting, augmenting, and w o r k i n ~ ar?und. [56)

In terms of designing tools for distributed, o r g a n ~ z a t l O n a l decision making, a detailed catalogue a n ~ y s i s o.f suchresDQngs CQuld become part of he deSIgners toolki t;incorporated in the system, could point out styles ofworkarounds at the level of coding.4. Identifying granularity of the problem, then encodi ng it in thesystem where appropriate, would complement existingorganizational information processing. For example, in naturalhistory work, biologists are often classed as "lumpers" vs."splitters." Lumpers tend to identify fewer species, lumpingtogether specimens with fme-grain distinctions, and converselywith splitters. Such individual-level habits or tendencies havealso been documented among those filling out certificates ofdeath. At this level of individual encoding, it is possible to trackdecision making and signal bias in one direction or another (andin fact such capacities exist in several domains, bothcomput erized and manual). However, the monitoring of

relatively simple habits and creating mnemonic tools to correctfor them become impossible at the level of occupationalspecialties or large governmental bodies. Collective memoriesand practices have a different structure, and require much morecomplex representations. Thus, the rule of thumb for designershere would be to tailor the complexity of the representationaround this issue of organizational scale.5. Match the structure of the boundary object informationsystem in the "middle" of the different participants with themismatch of their information needs. For example, in the caseof the lCD, we have a repository maintained by one group of

people, "fed" by forms coming in from a widely distributedconstituency. There is a good match between the types ofinformation being collected (heterogeneous, non-matchinginformation structures) and the repository; similarly between theuse of forms and the far-flung, disparate encoders ofinformation. Another sort of object or system inserted in themiddle of this process could be disastrous -- an abstractanalytical schema with tightly controlled coding requirements,

for example, could severely hamper data collection efforts.



SummaryWe have analyzed the caseof the ICD as an example of thedevelopment of an ODSS over time and space through· he use oftwo boundary objects, the repository and the form. R a ~ ~ r t h ~ strict standardization, there was a balance between pll;fticipantsdifferent needs and capabilities, evolved in concertWIth a setorganizational practices to support.it: There.are manx s u g g e ~ t I v e implications for organizational decISIon making and mformatIon

systems.

The work in this area has just begun -- with t h ~ advent.of v e ~ large scale information systems and t ~ c h ~ o l o g I e s , a n ~ mcreasmgconcern with collaboration and coordinatIon across hIghly .distributed groups, the issues presented h e ~ become p r e ~ s m g . We see our contribution to this set of q u e s t I o n ~ as analy'zmg theways human organizations have r e a c h ~ solutIons to thIS classof problems with and without computIng t e c h ~ ~ l o g y , toreflect back in to the technology the angleof VISIon hIStoryand sociology. On a more p r a c t i c a l l e v ~ l , we ~ o u l d like todefine as precise ly as possible the c ~ a t I ~ n , m a m t e ~ a n c e , perhaps destruction of b o u n ~ a r y . objects J? human m f o r m a t I o ~ processing, especially organIzatIOns. ThIS m ~ a n s understandingthe structure of the information needs for partIcular tasks, andexamining solutions, failed and successful.

ReferencesWe would like to acknowledge the help of our colleagues on t h ~ ICD project, Joan Fujimura and Alberto Cambrosio. Partof thIS

research was conducted with a Faculty Development A ward

from the University of California, Irvine. Annemarie Mol madeextensive and ve ry helpful comments on an earlier draftof thispaper. We also acknowledge the helpful commentsof severalanonymous referees.

[ll John Leslie King and Susan Leigh Star, "Conceptualfoundations for the development of organizational decisionsupport systems," Pa per presented to the Hawaiian InternationalConference on Systems Science, January, 1990.

[2] M. Turoff and S. R. Hiltz, "Computer systems for group vs.individual decisons," IEEE Transactions on Communications,30 (1982), 82-91.

[3] Lucy Suchman. Plans and Situated Action. CambrdigeUniversity Press, 1987; Jean Lave. Cognition and Practice.Cambridge University Press, 1988.

[4] Elihu Gerson and Susan Leigh Star, "Analyzing due processin the workplace," ACM Transactions on Office Informationsystems,4 (1986), 257-70; King and Star, op. cit., note 1; Carl

Hewitt, "Offices are open systems," ACM Transactions onOffice Information systems, 4 (1986), 271-87, and "Thechallenge of open systems," BYTE, 10 (1985), 23-42.

[5] A. Fagot-Largeault, Les Causes de la Mort; HistoireNaturelle et Facteurs de Risque, Paris: Vrin, 1989.

[6] R. H. Sprague, "A framework for the development of

decision support systems," MIS Quarterly, 4 (1980),1-26.

[7] F r a n ~ o i s Sainfort, David Gustafson, Kris Bosworth andRobert Hawkins, "Decision support systems effectiveness:Conceptual framework and empirical evaluation,"OrganizationalBehavior and Human Decision Processes, 45 (1990),232-52.

[8] Jay Nunamaker, Doug Vogel and Benn Konsynski,"Interaction of task and technology to support large groups,"Decision Support Systems, 5 (1989), 139-52.

[9] Goody, Jack, The interface between the written and the oral,Cambridge [Cambridgeshire]: Cambridge University Press,1987; The Domestication of he Savage Mind, Cambridge:Cambridge University Press, 1971.

[10] Foucault, Michel, The orderof hings: an archaeology ofthe human sciences, London: Tavistock Publications, 1970;Patrick Tort, La raison classjficatoire, Paris: Grasset, 1989.

[11] Latour, Bruno, Science in Action, Milton Keynes: Open

University Press, 1988.

[12] See Carl Hewitt, op. cit. note [**] and Susan Leigh Star,Regions of he Mind: Brain Research and the Quest forScientific Certainty (Stanford: Stanford U. Press, 1989).

[13] Ewald, Francois, L'Etat providence, Paris: Grasset, 1986.

[14] Simon Szreter, Th e Importance of Social Intervention inBritain's Mortality Decline c. 1850-1914: a Re-interpretation of

the Role of Public Health', Social History o/Medicine, 1988, 1-37.

[15] Alain Pinnsoneault and Ken Kraemer, "The impactof

technology support on groups: An essessmentof the empiricalresearch," Decision Support Systemes, 5 (1989), 197-216.

[16] M. Maruyama, "Communication between mindscape .types," in J. Van Gigch, ed. Decision Making about DecisionMaking. Cambridge, MA: Abacus Press, 1987.

[17] Susan Leigh Star and James R. Griesemer, "Institutionalecology, 'translations,' and boundary objects: Amateurs andprofessionals in Berkeley's Museum of Vertebrate Zoology,1907-1939," Social Studies 0/ Science, 19 (1989), 387420;Susan Leigh Star, "The structure of ill-structured solutions:Boundary objec ts and heterogeneous distributed problemsolving," Distributed Artificial Intelligence 2, ed. Les Gasserand Michael Huhns. Menlo Park: Morgan Kauffmann, 1988.

[18] King and Star, cited in note 1.

[19] La Reunion du Conseil de la Societe des Nations, SaintSebastian et l'Organisation Internationale d'Hygiene Publique,1923.

[20] CH/Experts Stat!78, Dr E. Roesle, 'The InternationalRecommendations for Determining the causesof Death drawnup by the Health Section of the League of Nations in 1925, andtheir applicability as regards the Reform of the German Statisticsof Causes of Death', 5.

80

[21] Jacques Bertillon, Rapport sur les Travaux de I1nstitut deStatistique (Session de Rome, avril 1887) et sur ['organisationde la direction generale de statistique en Italie, Paris, 1887.

[22] Ibid., 10.

[23] 'Malaria as a problem for the WHO[World HealthOrganization],WHO Archives, 453-1-4, E.J. Pampana, 'Malariaas a problem for the WHO'.

[24] CH/E. STATS/34 2 December 1927 'Reportof the

Committee of the Vital Statistics Section of the American PublicHealth Association on the Accuracyof Certified Causes of Deathand its Relation to Mort'!lity Statistics and the International List',10-11.

[25] Ibid., 11.

[26] Societe des Nations, Box R822, File 12458 Se ssion of theOIHP (Paris 1921).

[27] WHO Archives, 455-3-3, Collaborating with InternationalInstitute of Statistics, 29/1/47, Huber.

[28] Commission Internationale Nomenclature Internationale desMaladies, Proces Verbaux, 1910, Paris, 118.

[29] Ibid., 116-117.

[30] CH/Experts Stat!78 Dr E. Roesle, 'The InternationalRecommendations for Determining the causesof Death drawnup by the Health Section of the League of Nations in 1925, andtheir applicability as regards the Reformof the German Statisticsof Causes of Death', 10.



[31] Societe des Nations, Organisation d'Hygiene, Commissiond'Experts Statisticiens, CH/Experts Stat./l-43, 1927,Communication du Chef de Service de la Statistique Medicale auMinistere Polonnais de l'Interieure, 1-2.

[32] Ibid.

[33] WHO Archives, 1-455-3-1 to 455-9-44, Sixth DecennialRevision of ICD458-1-9, Classification of Pulmonary TB.


[35] Ibid, 17.

[36] CH/Experts Stat/80 'Government Commission onMorbidity and Mortality Statistics in Austria, with reference tothe 4th revision', 4.

[37] Ibid.

[38] Foucault, op.cit., note 2, after Borges.

[39] CH/Expert Stat/43 20 December 1927 Dr P.1. Kurkin,'Note sur la nomenclature des maladieset des causes de deces enRussie', 3.

[40] CH/Experts Stat/80 'Government Commission onMorbidity and Mortality Statistics in Austria, with reference tothe 4th revision', 3.

[41] Law and W. Bijker, Introduction, Proceedings o/theSecond International Conference on the History o/Technology,Cambridge, MA: MIT Press, forthcoming.

[42] Jacques Bertillon, De la Frequence des Principales Causesde Deces aParis pendant la seconde moitie duXIXeme siecle etnotamment pendant la periode /886-1905, Paris, 1906.

81

[43] CH/Experts Stat/88 Dr Teleky, 'La Statistique de Morbiditedes Caisses d'Assurance-Maladie en Allemagne' 19 March,1929, 8.


[45] Ibid, 62.

[46] Ibid, 73.

[47] Ibid, 138.

[48] Ibid, 5.

[49] CH/Experts Stat/87 'Registrar General' of England andWales, 'Observations upon Dr Roesle's Memorandum upon theComparative Study of Morbidity'.

[50] Ibid.

[51] Desrosieres, A., Les categogies socio-pro/essionnelles,Paris: La Decouverte, 1988.

[52] Mohan Tanniru and Hemant Jain, "Knowledge-basedGDSS to support reciprocally interdependent decisions, "Decision Support Systems, 5 (9189), 287-301.

[53] WHO Archives, 455-3-4, 31/3/48, Expert Committee forthe Preparation of the 6th Decennial Revision of the InternationalList of Diseases and Causes of Death, 'Assignment of Causes ofDeath'.

[54] "Constructing 'do-able' problems in cancer research:Articulating alignment," Social Studies 0/Science, Vol. 17

(1987), 257-93.

[55] B. Latouret

aI., 'The Hume Machine', from authors;Michel Serres, Le Passage du Nord-Ouest, Paris: PUF, 1983

[56] Les Gasser, "The integration of c o m p u t i n ~ and routinework," ACM Transactions on Office InformatIOn Systems 4(1986).

{bowker and star} situations vs standsards in long term whide scale decision manking the case of the...

Documents