using a network menu and the umls information sources map to

Using a network menu and the UMLS InformationSources Map to facilitate access toonline reference materialsBy Jeffrey L. Clyman, M.D.Assistant Professor, Department of Anesthesiology(Medical Informatics)

Seth M. Powsner, M.D.Assistant Professor, Department of Psychiatry

John A. Paton, Ph.D.Director, Office of Academic Computing

Perry L. Miller, M.D., Ph.D.Associate Professor, Department of AnesthesiologyDirector, Center for Medical Informatics

Yale University School of MedicineNew Haven, Connecticut 06510

As computer technology advances, clinicians and biomedicalresearchers are becoming more dependent upon information fromonline databases and information systems. By using speciallyconfigured computer workstations and high-speed computernetworks, it is now possible to access this information in a rapid andstraightforward manner. To empower users by providing thesecapabilities, the authors are assembling a variety of networkworkstations to be located throughout Yale-New Haven MedicalCenter. At the heart of the workstation is NetMenu, a programdesigned to help users connect to a number of important onlineinformation systems, including a hospital order entry and resultsreporting system, a drug reference, bibliographic retrieval systems,and educational programs. In addition, as part of the National Libraryof Medicine's Unified Medical Language SystemS (UMLS®) project,the authors have developed a local prototype of the UMLSInformation Sources Map (ISM) and a companion query assistantprogram to complement the NetMenu in helping users select andconnect automatically to information services relevant to a particularquestion. The ISM query assistant draws from a listing of manyonline information sources accessible via local and internationalnetworks.

INTRODUCTION

Since the introduction of MEDLINE® in 1971, onlineinformation sources have become increasingly valu-able to biomedical researchers and practitioners. Inmany areas, the availability of these resources hasincreased the quality of work to such an extent thataccess to online information is considered vital. Inclinical medicine, literature review guided by an on-line bibliographic search is becoming part of a com-ponent in the recognized standard of clinical care. In

addition to MEDLINE, hundreds of computer-basedinformation sources and online databases may be ac-cessed from a variety of local and international com-puter networks. Traditionally, health care profes-sionals have accessed these networks using isolatedcomputer terminals and relatively slow telephone-based connections. As a result of technological ad-vances in the 1980s, online information is increas-ingly available to sophisticated users with computerworkstations and high-speed network connections.This paper describes the approach taken at the Yale-

Bull Med Libr Assoc 81(2) April 1993 207

Clyman et al.

New Haven Medical Center to facilitate widespreadaccess to a diversity of online information sources,through the combined use of workstations, high-speedcomputer networks, and a local prototype of the In-formation Sources Map (ISM) being developed inconjunction with the Unified Medical Language Sys-temS (UMLS ®) project of the National Library ofMedicine (NLM). Experimentation with the proto-type ISM at Yale is contributing to the developmentand enhancement of the Information Sources Map,distributed as part of the UMLS Knowledge Sources[1-2].

In the past, access to online information has beenextremely variable within the medical center. Isolatedcomputer terminals often are dedicated to specificinformation sources. For example, it is not uncommonto find two or three computer terminals arranged sideby side in offices, hospital wards, and libraries; eachterminal might provide access to a single informationservice, such as a hospital order entry system, clinicallaboratory database, or a bibliographic retrieval sys-tem. In fact, within the medical center, terminals arededicated to at least six different library-based sys-tems, four clinical systems, and myriad administra-tive systems. In some cases, information may be print-ed out, but there is usually no way to transfer datafrom a system accessed on one terminal to anothersystem accessed on another terminal, other than re-entering the data manually. For some workers, suchas admission clerks, a dedicated terminal is adequate.Other workers, such as clinicians, may need to inter-act with several different systems while caring for apatient. In these situations, dedicated terminals arerestrictive and decrease productivity.

It is possible to use a general-purpose desktop com-puter to access a variety of information sources thatmay be connected to different networks. To do so,however, the user must have some level of sophis-tication with regard to the operation of the computerand the connection to the information services. Ac-cessing a specific service requires that the user knowwhich communications program to execute (e.g.,Crosstalk Mark 4); how to use this communicationsprogram to perform functions such as printing, sav-ing data in files, and exiting; how to navigate throughthe international computer networks to access a spe-cific service; how to set the configuration of the localterminal emulator to match the specifications of theremote computer; and how to initiate a session byentering the requisite sign-on codes.Once the connection with an information service

is established, the user also must be fluent in thequery language and vocabulary used by the remotecomputer. Although some information services areentirely menu driven, others offer greater accessthrough use of a customized query language. In mostcases, the query language is unique to a particular

information service or group of services. The com-plexity of some languages has kept many health careprofessionals- from interacting directly with onlineinformation services; instead, many have relied uponhighly trained information specialists or librarians tofunction as search intermediaries.

In recent years, two approaches have been takentoward reducing the barrier posed by complex anddiverse query languages. In an effort to standardizelanguages, information experts and vendors are de-veloping the National Information Standards Orga-nization (NISO) Z39.50 protocol [3]. Although manyvendors have endorsed this protocol, few have com-mitted to full compliance in the immediate future. Asan alternative to a formal query language, programssuch as GRATEFUL MED® (National Library of Med-icine, Bethesda, Maryland) have been developed tofacilitate end-user searching through the use of searchtemplates and graphical user interfaces. Even so, usersstill must learn the details and idiosyncrasies of in-teracting with the various online information sources.

In an effort to overcome or at least reduce the con-straints of online information access, the authors andothers are pursuing the ideal of a network worksta-tion [4-5]. A network workstation would provide easyaccess to a variety of online information services.Workstations would be available throughout themedical center and would not require users to befamiliar with the details of operating the machine,establishing a connection, or conversing in a querylanguage. Furthermore, this system would be capableof maintaining and displaying multiple simultaneoussessions with diverse information sources. For ex-ample, clinicians would be able to enter drug ordersfor a patient by using an order entry system whilelooking up dosage recommendations using an onlinedrug reference. In addition, information could be ex-changed among these various online systems in ei-ther a user-guided or automatic fashion.

Initial work toward this ideal began a few yearsago with development of a prototype known asNetMenu [6]. Developed on a Macintosh computerwith HyperCard software, the prototype offered theuser a window into a world that emphasized infor-mation services rather than technological hurdles. In-formation services appeared as if they were personalcomputer programs and thus could be integrated withother familiar programs and utilities running on thedesktop computer. The prototype was limited by sev-eral factors, including the performance of HyperCardand the inability to support other platforms such asthe PC. Nevertheless, the prototype was valuable inillustrating the concept of providing naive users withtransparent access to network-based information ser-vices. Surprisingly, computer and networking pro-fessionals were as impressed and inspired by NetMenuas were clinicians and biomedical researchers.

Bull Med Libr Assoc 81(2) April 1993208

The UMLS Information Sources Map

THE NETWORK WORKSTATION

Based upon the success of this prototype, in the fallof 1991 the authors began to develop a network work-station that would be practical for widespread de-ployment throughout the medical center. Like theprototype system, the new workstation was designedto provide clinicians and researchers with easy accessto a growing number of network-based informationservices vital to their daily activities. The followingdiscussion of the workstation highlights some of theproblems encountered in its design and the role ofNetMenu in addressing many of these problems.

In comparison to the initial prototype, the networkworkstation project posed additional challenges re-lated to the medical center's diverse collection of in-formation services, host computers, network com-munications protocols, and workstation hardware. Asat many other medical centers, the School of Medicineand the Yale-New Haven Hospital are distinguishedas separate legal and administrative entities. Themedical school has large numbers of Macintosh com-puters as well as PCs, while the hospital has a stan-dardized system of IBM PS/2 personal computers.Similarly, Ethernet and Local Talk dominate theschool's local area networks, while Token Ring is theprincipal connection between hospital workstationsand departmental computers. In regard to commu-nications protocols, TCP/IP and AppleTalk are usedwidely in the medical school, while IBM NETBIOS,Novell IPX, and 802.2 are used on the hospital net-works. Both institutions manage an assortment of hostcomputers running a variety of operating systems,including IBM MVS, DEC VMS, and UNIX.The information services provided also are highly

diverse, not only in content but in terms of confi-dentiality, access restrictions, user interface, mode ofconnection, and relative importance to specific usergroups. For example, the hospital's new physicianorder entry and results reporting system, referred tolocally as CCSS (TDS Health Care Systems, Atlanta,Georgia), is important to practicing clinicians becauseit provides confidential patient information. CCSSrestricts access on the basis of location and user iden-tity, and it can be accessed only through the use ofproprietary commercial software. A computer-basedversion of Current Contents, on the other hand, maybe particularly valuable to researchers, imposes fewaccess restrictions, and may be accessed using a va-riety of software products on different platforms [7].

THE NETMENU PROGRAM

To support users in this heterogeneous and ever-changing environment, the network workstation re-lies heavily on commercial networking and com-munications products. A new version of the NetMenu

Figure 1NetMenu, top-level menu of information services and categories ofservices

_ Yale_

s |i O miniMEDLINEr f % Library

5 J °sO Search assistance_~~~ X Clinical assistance

s & * Education_I _ Wed Apr 15 1 1:16:33 1992

program coordinates the use of these commercialproducts to help the user choose pertinent informa-tion sources and connect to these sources expeditious-ly. To enable workers to interact easily with networkworkstations located throughout the hospital anduniversity, the user interface has a similar look andfeel, regardless of whether the platform is a Macin-tosh or PC. By using XVT (XVT Software, Boulder,Colorado), a cross-platform programming toolkit, theauthors have built NetMenu as a graphical user in-terface application that runs on both the Macintoshand PC with Microsoft Windows [8]. In addition, withfew modifications, the same source code can executein character mode on a UNIX machine and a PC with-out Windows. The UNIX platform, operating in char-acter mode, eventually may support dial-in access tothe campus network from remote sites, providing thefunctionality of workstations connected directly tothe network. Using XVT, one programmer completedthe first phase of the NetMenu project in less thansix months, a relatively short time for software de-velopment.When not in active use, a network workstation may

display the familiar network logo and a menu of spe-cific services or categories of services (Figure 1). Thistop-level menu may list the information services usedmost frequently from that particular workstation (e.g.,miniMEDLINE: Georgetown University, Washing-ton, D.C.). The menu also lists the categories of ser-vices, such as "Library" and "Clinical Assistance." Tosee the specific services within each category, the userclicks the mouse over the category's name or icon,and a second-level menu appears on the screen. Inthis way, the NetMenu can provide access to a largenumber of services that are grouped in a logical fash-ion. From either the top-level or second-level menus,


Clyman et al.

Figure 2NetMenu, descriptive information about Current Contents-aninformation service listed in the library category

abstracts of articles appearing in themost recent issues of leading scientificjournals. From this workstation you mayaccess a computer based version ofCurrent Contents to search for specificccitations or browse issues for the pest

iRncel tConnect

the user may proceed in two ways once the desiredinformation service is found.

If the service appears in the top-level menu (e.g.,miniMEDLINE), the user may click the mouse onceor twice over the name or icon representing the ser-vice. A single click will call up a scrolling text boxcontaining descriptive information about the partic-ular service (Figure 2). To establish a connection andbegin an interactive session, the user may click onthe "Connect" button; pressing the "Cancel" buttonreturns the user to the top-level menu. Alternatively,double clicking on the name of a service in the top-level menu immediately begins automatic connectionto the service display of the descriptive information.

If the service appears in a second-level menu, theconnection process can be activated by double click-ing on the service name. Alternatively, a single clickover the name will highlight that service within thelist. The user then may press the "Info" or "Connect"buttons, either to view descriptive information or toactivate the automatic connection process respective-ly.The content of a particular NetMenu display is de-

termined when the program is executed, based upona configuration file maintained by the workstationadministrator. In this way, a workstation can offer adifferent selection of services depending on its lo-cation, the specific services accessible from that lo-cation, and the most likely users. For example, inpatient care areas of the hospital, a workstation ini-tially displays a menu listing that includes the hos-pital information system (CCSS), miniMEDLINE, adrug reference program, and a category of serviceslabeled consultation programs.At present the NetMenu configuration file exists

as an ASCII text file, which is maintained manually

using a word processing program. Using a standardformat and keywords, a workstation administratorspecifies the arrangement of service or category namesand icons in the top-level menu; the names of infor-mation services listed in the second-level menu; thenames of script files to activate the automatic con-nection process; and descriptive information abouteach service. To simplify maintenance of the config-uration file, the authors are developing a HyperCardprogram that will enable administrators to add, de-lete, and re-arrange the names of services based upona master catalog of information services, such as theISM. The "configurator" program then will automat-ically generate a new configuration file.

CONNECTING TO AN INFORMATION SERVICE

To initiate a connection to an information service, theNetMenu program either launches or transfers con-trol to a commercial communications application (e.g.,Dynacomm Asynch, Future Soft Engineering, Hous-ton, Texas), typically stored on the workstation as partof the overall system. This application executes a pre-defined script to log on to the requested informationservice. In some cases, the script may prompt the userto enter specific sign-on codes required by the infor-mation service. Each script is designed to operate witha specific communications application and may beprecompiled for storage and retrieval from a networkfile server.

Naturally, there are significant tradeoffs involvedin the decision to use commercial communicationsapplications versus building specialized communi-cations programs. In making this decision, the au-thors considered a number of important issues con-cerning flexibility to customize the user interface;capability to control or limit the user's actions; theexchange of information between NetMenu, ISM ap-plications, and networked information sources; theneed to support many diverse network protocols; thecomplexity of network communications program-ming; and the costs of short-term implementation andlong-term support.The second issue, listed above, concerns maintain-

ing rigid control over what actions users may performon a network workstation. On many public machines,the author would like to prevent users from alteringthe workstation's configuration by changing com-munications parameters, or by resizing, closing, ormoving windows around the desktop. Although suchactions may increase the versatility of a workstationin the hands of sophisticated users, these actions mayrender a workstation entirely useless to unsophisti-cated users. It was therefore important for the net-work workstation to control or limit the user's actions.

Custom-built software is one approach that couldbe taken in addressing many of these important is-



sues. Although a customized application might beexpensive to build initially, the long-term costs mightbe lower if the program was distributed widely andthe project was financed centrally.On the other hand, communications programming

is complex-even programming for a single platformand single protocol. To support a heterogeneous en-vironment, communications applications had to fea-ture multiplatform and multiprotocol support. For-tunately, an increasing number of commercialcommunications products are available for the Mac-intosh and PC that meet many of Yale's specifications.These products are extremely powerful in terms ofoffering multiprotocol support, rich scripting lan-guages, facilities for exchanging data with other ap-plications, and facilities to customize and limit theuser interface.On the PC/Windows platform, the authors chose

Dynacomm Asynch (Future Soft Engineering, Hous-ton, Texas) as the communications application. Inaddition to featuring an extremely rich scripting lan-guage, Dynacomm Asynch supports a variety of dial-up and network connections using network modems,NETBIOS, and TCP/IP, to mention just a few. A Dy-nacomm product featuring the same scripting lan-guage is also available for the Macintosh. However,the current version does not support direct networkcommunications, so PacerTerm (Pacer Software,La Jolla, California) had been selected as the principlecommunications application for the Macintosh. A ma-jor upgrade to the Macintosh Dynacomm product isreportedly in progress. When this upgrade becomesavailable, PacerTerm may be replaced with Dyna-comm, thereby simplifying the creation and main-tenance of scripts, and standardizing the user inter-face for both platforms.

In addition to automating the sign-on process toan information service, scripts are being used to en-hance and even to replace the traditional character-based user interface offered by many host systems.For example, by adding a menu bar to the user in-terface of the Micromedex drug reference (Micro-medex, Denver, Colorado) mounted on a main-framecomputer, the user's task of navigating through thiscomplex and multifaceted reference is simplified. Thescripting language also is used to add "buttons" tothe user interface, allowing easy execution of com-mon functions, such as printing a few screens. Toenhance sessions with miniMEDLINE, a button maybe added to allow users to order reprints of referencedarticles. The script could send an electronic mail mes-sage to the library, requesting that a reprint be pre-pared and mailed to the user. A similar feature, en-abling users to order reprints, is available throughother bibliographic retrieval systems, includingLOANSOME DOCE service provided throughGRATEFUL MED.

For a variety of practical reasons, the system alsomust support PC network workstations that do notrun Windows. For this DOS platform, Crosstalk Mark4 (DCA, Alpharetta, Georgia) was chosen based onits scripting language and broad multiprotocol sup-port. Obviously, this platform limits capabilities forenhancing the user interface and facilitating data ex-change between information services. To limit theuser's actions, memory resident programs have beenbuilt that remap keys and key combinations, effec-tively disabling access to specific functions of thecommunications program. One possible enhance-ment would be the use of a task-switching or mul-titasking program. For a variety of technical and prac-tical reasons, development activities are restricted tofunctions required to simplify workstation operationby the most inexperienced users.

In the present environment, workstations use twoprotocols to manage communications with informa-tion services. In the School of Medicine, computersgenerally communicate with TCP/IP. Macintoshworkstations use MacTCP (Apple Computer, Cuper-tino, California), while PCs running DOS use PC/TCP (FTP Software, Wakefield, Massachusetts). With-in Yale-New Haven Hospital, most workstations areconnected to an IBM Token Ring network. Worksta-tions establish connections using either 802.2 or NET-BIOS, depending upon the requested information ser-vice. To minimize the software requirements of eachworkstation, NETBIOS is used first to establish a con-nection to an IBM Asynchronous CommunicationsServer (LANACS), which in turn connects to infor-mation services through a network modem or directlyusing the TCP/IP protocol.

In some cases, an information service might be anapplication that executes on the network workstationitself and is stored locally or on a network file server.The user begins operation of the application by se-lecting its name and pressing the "Connect" buttonin NetMenu. Examples of local applications that maybe launched by NetMenu include Hypercard-basedteaching programs for the Macintosh and CD-ROMretrieval programs that could access a network CD-ROM server.

THE INFORMATION SOURCES MAP

Although the NetMenu program is capable of listinga large number of information services on second-level menus, the authors favor restricting these liststo services that are especially important to users andare accessed frequently. Other than displaying de-scriptive information about sources, NetMenu doesnot assist users in selecting sources pertinent to an-swering a specific question. In an effort to providethis kind of assistance to biomedical professionals,the authors are implementing and refining a local


Clyman et al.

prototype of an information sources map to assistNLM in determining the appropriate structure andcontent of the UMLS Information Sources Map.The ISM is a directory of information services. Ap-

plication programs will read the ISM to help the userdetermine what sources are available and which onesmight be most relevant to a particular question. Inaddition, once an information source is selected, theapplication program may help establish an automaticconnection to the source. To provide these functions,the ISM will contain a variety of information abouteach information source, including the name, the pro-vider or owner, a text description, content type (e.g.,bibliographic citations), targeted audience, networkcommunication scripts to allow automatic connec-tions, data input templates to facilitate the exchangeof data between sources, and information to be usedin determining the relevance of the source to a userquery.One major focus ofUMLS has been the construction

of the Metathesaurus to facilitate the translation ofmedical terms from one vocabulary to another [9].Ultimately, the Metathesaurus may be very useful inautomating the transition between informationsources that use different vocabularies, such as ICD-9and MeSH®9. To complement the Metathesaurus, aSemantic Network has been constructed to representthe "relationships among the broad categories or se-mantic types to which all concepts in the Metathe-saurus are assigned" [10]. A third UMLS knowledgebase being developed is the ISM, or a directory ofinformation services [11]. Application programs willread the ISM to help the user determine what sourcesare available, and which might be most relevant to aparticular question. In addition, once an informationsource is selected, the application program may helpmake an automatic connection to the source. To pro-vide this functionality, the ISM will contain a varietyof different information about each informationsource, including the name; the provider or owner;a text description; content type (e.g., bibliographiccitations); a variety of descriptions of its subject cov-erage; targeted audience; network communicationscripts to allow automatic connections; data inputtemplates to facilitate the exchange of data betweendifferent sources; and information to be used in de-termining the relevance of the source to a user query.To explore what type of information would be most

useful in the UMLS ISM, the authors initially built aprototype containing 112 information sources cur-rently available through local and national networks.Most sources were chosen as representative examplesfrom a directory of online databases and can be di-vided into five broad categories: clinical medicine,drug information, biomedical research, environment,and agriculture [12]. The actual entries in the proto-type ISM are quite diverse in scope, depth, and degree

of relevance to biomedicine. The primary focus wason commercially accessible online databases, but anumber of other types of information sources wereincluded, such as clinical consultation and medicalteaching programs.To assess the relevance of an information source to

a user query, the various sources were coded fromthe perspectives of subject content and expected use.Initially, the subject content of each source was de-scribed with a series of high-level Medical SubjectHeading (MeSH) terms and UMLS semantic types.Based on preliminary testing, however, it was deter-mined that MeSH term indexing affords greater dis-crimination among information sources possibly rel-evant to a query [13]. To enhance further the processof selecting sources, several characteristics of thesources were defined, including the likelihood of us-ing the source in particular settings (e.g., patient care,clinical research, basic biomedical research, and healthservices research); the degree of systematic coverage(e.g., comprehensive or sporadic); the degree of con-sensus or review to be expected in material retrievedfrom the source (e.g., NIH expert panel recommen-dations or letter to the editor); and the type of infor-mation offered by the source (e.g., bibliographic ci-tations, patient records, clinical advice, and medicalteaching).

For preliminary testing of the prototype ISM, aninteractive query assistant program was constructedto run on a Macintosh. The function of the queryassistant program was to take, as input, one or twoquery terms and generate a list of information sourceswhich might be relevant to the question expressedby the query terms. First, the query terms are mappedto corresponding MeSH and semantic types using theMetathesaurus. Second, the user specifies additionalcharacteristics of the sources that might be relevant.The query assistant program then selects sources bycomparing the ISM entries for each source with thequery, using the subject content index terms and oth-er coded characteristics. A source might be selectedif, for example, one of its MeSH or semantic typeindex terms matches or is the parent of a query'sMeSH or semantic type.

Using thirty representative queries compiled froman NLM test set, the authors tested the ability of theISM coding to discriminate among sources possiblyrelevant to a query. The codings were useful in par-titioning the information sources, allowing a subsetof sources to be retrieved for a specific query. Codingof subject content by MeSH terms afforded better se-lectivity than coding by semantic types, the differ-ence being most pronounced with health services re-search queries. Finally, the additional characteristicscoded in the ISM enabled even further discriminationbetween sources. The main purpose of this study wasto explore issues in coding information sources; other



Figure 3Beginning operation of query assistant

workers have conducted experiments to assess theaccuracy of the selection process itself, i.e., the actualrelevance of selected sources [14].Based on this work with a prototype ISM, a new

version of the query assistant program was construct-ed that would be practical for wider deployment andtesting within the institution. Unlike the prototypeapplication, which is six megabytes in size, the newquery assistant application was developed to functionin a client-server model. The client portion of theprogram simply manages the graphical user interfaceand operates on a Macintosh or Microsoft Windowsplatform. In addition, the client program allows au-tomatic connections to sources selected by the user.The client interacts with the server, which resides ona networked UNIX machine. The server program con-

tains the engine to search the ISM and Metathesaurus,in helping users to select relevant information sources.The client portion of the query assistant program wasbuilt using the Dynacomm scripting language, whichis ported easily between the Macintosh and MicrosoftWindows.To begin operation of the query assistant, the user

fills in text entry and check boxes to describe thequestion at hand and the kinds of information sourcesthe user feels might be helpful (Figure 3). For ex-ample, to identify sources that may provide infor-mation about both hypertriglyceridemia and heartdisease, the user first types "hypertriglyceridemia"in the box labeled "Term 1." When the return key ispressed, the term is transmitted to the ISM serveralong with a request for the names of sources that


Clyman et al.

Figure 4List of alphabetically close terms

might offer information about hypertriglyceridemia.The ISM server then attempts to locate the term inthe Metathesaurus. If an exact match is found, thenthe server uses the term to generate a list of pertinentinformation sources. The list is displayed in a box inthe lower portion of the display.

If there is no exact match to a term in the Meta-thesaurus (e.g., because of a misspelling or the use ofa variant term), the server returns a list of alphabet-ically close terms, which are displayed in a dialog box(Figure 4). The user may choose from this list themost appropriate substitute term to be used by theserver in selecting sources.To narrow the list of possible information sources,

the user may type a second term, e.g., "heart disease,"in the text box labeled "Term 2." The default "AND"check box is selected to request the names of sourcesoffering information about both terms. When the re-turn key is pressed this time, both terms are sent tothe server. The returned list of information sourcesmay be refined further based upon specific criteriaand interests indicated by check boxes in the middleof the screen.

From the list of suggested information sources, theuser may select a specific source and initiate an au-tomatic connection by pressing the "Connect" but-ton. To accomplish this, the query assistant requestsinformation from the server on how the source isaccessed. Often, a specific source is available throughseveral networks or information vendors (e.g., fromthe NLM, BRS, or DIALOG), each requiring pass-words and the use of different query languages. Inthese instances, the user is asked to choose from thelist of vendors or networks. Finally, the query assis-tant program requests from the server informationfor use by a communications script, which is executedto establish a connection.The authors are currently in the early stages of

deploying network workstations and NetMenu with-in the Yale Medical Library. These versions ofNetMenu allow access to a variety of bibliographicsystems, including ORBIS (the Yale University Li-brary catalog implemented with NOTIS), mini-MEDLINE, and Current Contents; a search assistanttool, which performs automated Metathesaurus termlook-up and expansion to potentially related MeSH



terms; two clinical advice programs (COAG-ADVI-SOR and HEPATITIS-ADVISOR), running at the Cen-ter for Medical Informatics [15]; and several teachingprograms, which run locally on the Macintosh work-station. In the near future, the ISM query assistantprogram will be added to the list of services accessiblefrom NetMenu. In addition, a Microsoft Windowsversion of NetMenu will be installed soon at the Li-brary of the Department of Epidemiology and PublicHealth, and a DOS character version as a clinicalworkstation at a number of locations within Yale-New Haven Hospital. Users who dial into the net-work from outside the medical center will be able toaccess network information services by interactingwith the DOS character version of NetMenu runningon a Chaterbox dial-in system (J&L Information Sys-tems, Chatsworth, California).

DISCUSSION

The work just described represents the first of manyphases in a project to deploy network workstationsthroughout the medical center. The NetMenu pro-gram has been constructed to coordinate a variety ofcommercial communications products, enabling in-experienced users to select and access a number ofimportant information services in an easy, intuitivefashion. To supplement NetMenu, an ISM query as-sistant program has been constructed to help usersselect relevant information sources from a large num-ber of diverse sources accessible through local andinternational networks. At present, NetMenu servesas an application launcher (e.g., shell) and help fa-cility. In a later phase of the project, the program willbe enhanced to help simplify access to informationservices that require personal identification and pass-words. For this purpose, authentication systems suchas Kerberos could be incorporated into the NetMenuprogram to interact directly with other network-basedauthentication modules [16].

In addition, NetMenu eventually will play a moreactive role in coordinating the exchange of infor-mation between network services and workstation-based applications designed for decision support andthe presentation of clinical data. By using the emerg-ing technologies for interprocess communications onpopular computer workstations (e.g., publish/sub-scribe, dynamic data exchange [DDE], and object link-age embedding [OLE]), NetMenu and related pro-grams might help facilitate the desktop integrationof medical information. Such integration has beendifficult to achieve, due to the architecture of tradi-tional information systems. This situation shouldchange as workstations become more powerful andmore accessible to health care professionals.Over the past several years, there has been consid-

erable interest in facilitating the retrieval of infor-

mation from an increasing number of online data-bases. In the commercial sector, Telebase (Malvern,Pennsylvania) offers a service that allows a user tosearch a number of diverse online databases using asingle query expression. The Wide Area InformationServer (WAIS) project is another example [17]. De-veloped by a consortium of four corporations (DowJones & Co., Thinking Machines Corp., Apple Com-puter, and KPMG Peat Marwick), the WAIS projecthas established a uniform protocol to enable inex-perienced users to search a number of free text da-tabase servers, accessible via local and national net-works. First, the user must choose specific WAISservers that might contain information or documentsrelevant to a query. For assistance in choosing theseservers, Thinking Machines has created a Directoryof Servers with indexed textual descriptions of allknown servers.Compared to the Telebase and WAIS systems, the

approach described in this paper differs in a numberof respects. Most importantly, NetMenu and the ISMquery assistant do not interact directly with onlineinformation sources other than the ISM itself. As aresult, individual documents or citations are retrievedthrough further interaction with specific informationsources using their respective query languages. TheTelebase system uses a "brute force" technique toselect relevant information sources. To the author'sknowledge, Telebase makes no attempt to narrow theassortment of information sources based upon thespecific query. Every source listed in a domain cate-gory is searched automatically. By contrast, the ISMquery assistant helps the user narrow the selectionof sources.

In effect, the query assistant provides a functionsomewhat analogous to that provided by the WAISDirectory of Servers. Unlike the ISM, the WAIS Di-rectory of Servers does not use standard, consistentinformation to index databases available on WAISservers. The ISM query assistant program does notrestrict the user to only those information sourcesthat comply with the WAIS communications protocol;instead, a user may connect to many different typesof local and remote sources using diverse commu-nications protocols. In addition, NetMenu providesa direct connection to local information sources with-out the need for an intermediate server. The ISM isalso unique in being customized for biomedical in-formation sources by incorporating special codingsfor subject content and additional characteristics. TheWAIS and ISM approaches are complementary, how-ever. The WAIS servers each can be viewed as onlineinformation services that may be cataloged in theISM. By using the WAIS protocol, available in thepublic domain, the authors eventually may augmentthe query assistant program so it can interact directlywith WAIS servers accessible via the Internet.


Clyman et al.

CONCLUSION

Two complementary methods have been developedto facilitate the selection of relevant online infor-mation services and to provide easy access to theseservices. The NetMenu program operates on both per-sonal and public network workstations and may becustomized for the information sources available fromparticular locations as well as for the needs of itsprimary users. On the desktop machine, NetMenuguides users in connecting to common sources of in-formation and could coordinate the exchange of in-formation between these sources.Whereas NetMenu offers access to an important

subset of services, the ISM query assistant helps usersselect from a comprehensive collection of biomedicalinformation sources. The ISM knowledge base andserver reside on a central machine, thus enablingfrequent revisions and Internet access by a large num-ber of clients. In the future, as medical libraries be-come more electronic and less centralized, facilitiessuch as NetMenu and the ISM may become powerfultools to help biomedical professionals navigatethrough vast stores of information and tasks that maybecome indispensable in the performance of theirdaily work.

REFERENCES

1. HUMPHREYS BL, LINDBERG DAB. The Unified MedicalLanguage Systems project: a distributed experiment in im-proving access to biomedical information. In: Lun KC, De-goulet P, Piemme TE, Rienhoff 0, eds. MEDINFO 92: pro-ceedings of the Seventh World Congress on MedicalInformatics, September 6-10, 1992, Geneva Palexpo, Swit-zerland. Amsterdam: North-Holland, 1992:1496-500.2. MAsYS DR, HUMPHREYS BL. Structure and function ofthe UMLS Information Sources Map. In: Lun KC, DegouletP, Piemme TE, Rienhoff 0, eds. MEDINFO 92: proceedingsof the Seventh World Congress on Medical Informatics,September 6-10, 1992, Geneva Palexpo, Switzerland. Am-sterdam: North-Holland, 1992:1518-21.3. NATIONAL INFORMATION STANDARDS ORGANIZATION (U.S.).Information retrieval service definition and protocol spec-ifications for library applications. New Brunswick, NJ:Transaction Publishers, 1989:52 (National information stan-dards series; ANSI/NISO Z39.50-1988).4. CLAYTON PD, ANDERSON PK, HILL C, MCCORMACK M. Aninitial assessment of the cost and utilization of the Inte-grated Academic Information Management System (IAIMS)at Columbia-Presbyterian Medical Center. In: Clayton PD,ed. Proceedings of the Fifteenth Annual Symposium onComputer Applications in Medical Care: assessing the value

of medical informatics, November 17-20, 1991, Washing-ton, DC. New York: McGraw-Hill, 1992:109-13.5. STAIR TO, CoRN M, BROERING NC. First year's experienceof the MAClinical computer workstations project. Acad Med1990 Jan;65(1):20-2.6. PATON JA, CLYMAN JI, LYNCH P, MILLER PL ET AL. Strategicplanning for IAIMS: prototyping as a catalyst for change.In: Miller RA, ed. Proceedings of the Fourteenth AnnualSymposium on Computer Applications in Medical Care:standards in medical informatics, November 4-7, 1990,Washington, DC. Los Alamitos, CA: IEEE Computer SocietyPress, 1990:709-13.7. PATON JA, BELANGER A, CHEUNG K, GRAJEK S ET AL. On-line bibliographic information: integration into an emerg-ing IAIMS environment. In: Proceedings of the SixteenthAnnual Symposium on Computer Applications in MedicalCare: supporting collaboration, November 8-11, 1992, Bal-timore, MD. In press.8. LEVITT J. XVT; safe haven in the GUI wars. UNIX Today1990 Oct.9. TUTTLE MS, NELSON SJ, FULLER LF, SHERERTz DD ET AL.The semantic foundations of the UMLS MetathesaurusG.In: Lun KC, Degoulet P, Piemme TE, Rienhoff 0, eds. MED-INFO 92: proceedings of the Seventh World Congress onMedical Informatics, September 6-10, 1992, Geneva Pal-expo, Switzerland. Amsterdam: North-Holland, 1992:1506-11.10. HUMPHREYS, op. cit.1 1. MASYS, op. cit.12. Directory of online databases. New York: Cuadra/El-sevier, 1988.13. MILLER PL, WRIGHT LW, FRAWLEY SJ, CLYMAN JI ET AL.Selecting relevant information resources in a network-basedenvironment: the UMLS Information Sources Map. In: LunKC, Degoulet P, Piemme TE, Rienhoff 0, eds. MEDINFO92: proceedings of the Seventh World Congress on MedicalInformatics, September 6-10, 1992, Geneva Palexpo, Swit-zerland. Amsterdam: North-Holland, 1992:1512-7.14. MASYS DR. An evaluation of the source selection ele-ments of the prototype UMLS Information Sources Map.In: Proceedings of the Sixteenth Annual Symposium onComputer Applications in Medical Care: supporting col-laboration, November 8-11, 1992, Baltimore, MD. In press.15. CLYMAN JI, SHIFMAN M, MILLER PL. The routine use ofclinical advice systems in a university hospital. In: Amer-ican Medical Informatics Association Spring Congress, 1992(transcript available on request).16. STEINER JG, NEUMAN C, SCHILLER JI. Kerberos: an au-thentication service for open network systems. In: ProjectAthena. Cambridge, MA: Massachusetts Institute of Tech-nology, 1988.17. KAHLE B, MEDLAR A. An information system for cor-porate users: wide area information servers. Online 1991Sep;15(5):56-60.

Received September 1992; accepted October 1992


using a network menu and the umls information sources map to

Documents