copy Blackwell Science Ltd 2002

Health Information and Libraries Journal

19

pp189ndash200

189

Blackwell Science Ltd

HealthCyberMap a semantic visual browser of medical Internet resources based on clinical codes and the human body metaphor

Maged N

Kamel Boulos

Abdul V

Roudsari

amp

Ewart R

Carson Centre for Measurement and Information in Medicine School of Informatics City University London UK

Abstract

Objectives

HealthCyberMap (HCMmdashhttphealthcybermapsemanticweborg)is a web-based service for healthcare professionals and librarians patients andthe public in general that aims at mapping parts of the health informationresources in cyberspace in novel ways to improve their retrieval and navigation

Methods and service description

HCM adopts a clinical metadata frameworkbuilt upon a clinical coding ontology for the semantic indexing classificationand browsing of Internet health information resources A resource metadatabase holds information about selected resources HCM then uses GIS(Geographic Information Systems) spatialization methods to generate inter-active navigational cybermaps from the metadata base These visual cybermapsare based on familiar medical metaphors

Conclusions

HCM cybermaps can be considered as semantically spatializedontology-based browsing views of the underlying resource metadata base Usinga clinical coding scheme as a metric for spatialization (lsquosemantic distancersquo) isunique to HCM and is very much suited for the semantic categorization andnavigation of Internet health information resources Clinical codes ensurereliable and unambiguous topical indexing of these resources HCM alsointroduces a useful form of cyberspatial analysis for the detection of topicalcoverage gaps in the resource metadata base using choropleth (shaded) mapsof human body systems

Introduction and background

HealthCyberMap (HCMmdashhttphealthcybermapsemanticweborg) aims at mapping selected partsof the health information resources in cyberspace(mainly the Web) in unique and novel semantic waysto improve their retrieval and navigation This isachieved through intelligent categorization andinteractive hypermedia visualization of the health

information cyberspace using metadata clinicalcodes (a kind of medical ontology) and GIS(Geographic Information Systems) technologies

Cyberspace and cybergeographymdashthe need for maps

The term lsquocyberspacersquo first used by WilliamGibson

1

literally means lsquonavigable spacersquo (fromthe Greek word

kyber

mdashto steer or navigate) Thelsquosoftrsquo (information) part of the Internet consists ofmany interrelated spaces or services the World

Correspondence M N Kamel Boulos Centre for Measurement andInformation in Medicine City University Northampton SquareLondon EC1V 0HB UK E-mail MNabih-Kamel-Bouloscityacuk

HealthCyberMap

Maged N Kamel Boulos

et al

copy Blackwell Science Ltd 2002

Health Information and Libraries Journal

19

pp189ndash200

190

Wide Web FTP (File Transfer Protocol) e-mailmailing lists bulletin boards newsgroups onlineshopping (usually hosted on the Web) etc

2

Considerable amounts of health-related activitiesand information are exchanged in these spacesevery day

Humans by virtue of their spatio-cognitiveabilities are able to navigate through geographicalspace These cognitive skills also have similar valuein the exploration and understanding of non-geographical information and spaces

3

Maps are powerful graphic tools that classifyrepresent and communicate spatial andor spatial-ized relations (relations projected into someconceptual space for mapping) Maps are also amethod to visualize and navigate a world that istoo large and complex to be seen directly A mapwill show us more in less space and consequentlywe will be able to plan our route in advance

4

There are three distinct but interrelated aspectsof cyberspace that can be mapped

5

1

Networks of multimedia resources and links eghealth information resources as mapped by HCMbased on their subject topics The resultantmaps can be classified as conceptual informationspace maps and can be used as navigationalaids for browsing mapped resources Thegeographical aspects of information resources(coverage and provenance) can also be mappedand are sometimes very useful as an index toinformation resources

6

They are not necessarilythe same as the physical location of hostingservers eg a clinic in London UK might haveits website hosted on a server in CaliforniaUSA however the site remains more relevantto people living in London UK

2

Networks of machines and communicationlinks (the lsquohardrsquo part of the Internet)

3

Demographics of Internet users and cyber-societies Such information can help us tailorweb-based information services like HCM tousersrsquo needsIt has been said the Web is its own map

7

How-ever as we surf the Web we can only appreciate avery small part of it at any given time We cannotfor example figure out the relations of the site weare visiting to the rest of the Web how it measurescompared to the rest of the Web or what the restof the Web looks like We may have difficulties

finding our way back to a resource we have visiteda few hours ago (lsquolost in cyberspacersquo) It is difficultto locate medical web resources that covernarrower broader or similar topics as the one weare currently looking at and except when we haveno particular goal(s) in mind we cannot plan ourweb journeys ahead and maximize their usefulnessWe need a map or set of maps for this purpose

Research literature on the bibliographiccybergraphic uses of GIS to map semantic (informa-tion) spaces is scarce and includes the work doneby Fabrikant

8

Old

9

and Terpstra

10

Web maps (cybermaps)

Maps published in cyberspace can be either mapscovering topics related to cyberspace itself egmaps of web resources as outlined above or mapsusing cyberspace usually the Web as a publishingmedium eg MARAARMA maps of malaria inAfrica

11

The same principles and opportunities ofweb cartography apply to both types

Non-animated interactive web maps alsocalled clickable maps imagemaps or hypermapscan be served either as client-side imagemaps (as inHCM) or as server-side imagemaps depending onwhere mouse-click co-ordinates are resolved

4

Interacting with a map can stimulate a userrsquos(visual) thinking and encourage explorationClicking an object on such maps can lead to otherweb resources including other web maps and caneven trigger a query against an underlying data-base and display the results It is possible to put allkinds of additional information behind the mapimage thus reducing map clutter and size Mouseevents such as mouse-over (eg ToolTips for mapfeature labelling) and mouse clicking of mapobjects can be associated with this extra informa-tion The map interface itself can be made inter-active by providing the user with control optionslike panning zooming in and out and a smallerinteractive overview map to highlightselect thearea covered by the currently displayed map tile inrelation to a bigger map

4

Figure 1 provides a good example of many ofthese interactivity options and features Notethe country name ToolTip (lsquoUnited Kingdomrsquo) thedifferent map interface buttons on the left and theclickable overview map with a red positional square

HealthCyberMap

Maged N Kamel Boulos

et al

copy Blackwell Science Ltd 2002

Health Information and Libraries Journal

19

pp189ndash200

191

on the right In this particular cybermap healthinformation resources are mapped to countries (ofauthorspublishers) rather than cities and listed ina separate pop-up text window to avoid map clutterNote the lsquoFind resources having the same primarysubject as this one from all over the worldrsquo link atthe end of each resource bibliographic card in theresource list pop-up window to the right

Spatialization and metaphors

Spatialization is the process by which informationwith no inherent spatial attributes (no geogra-phical referent) is mapped onto a defined spatialframework using a variety of spatial metaphorsInformation attributes are transformed into aspatial structure (conceptual space) that can bevisualized and navigated through the applicationof concepts like hierarchy proximity and simil-arity eg a topical hierarchy based on the subjects

of a collection of web resources The goals ofspatialization are to increase the spatial legibilityand comprehension of information spaces improvenavigation through them and enable people tofind the information they are searching for in thesespaces more easily

2

The primary function of a metaphor is toprovide a partial understanding of one kind ofexperience in terms of another kind of experienceSpatial metaphors act as fundamental sensemakers for abstract domains Familiar metaphorstaken from usersrsquo everyday life are usually mucheasier to understand

8

In cybermaps metaphorcomprehension can be enhanced with appropriateuse of visual variables (eg colour) and applicationof sound cartographic design principles

12

It is noteworthy that a graphical browseralready exists for visualizing Read Codes usingthe familiar human body metaphor

13

However noone has yet used such visual interfaces that are

Figure 1 Screenshot of HCM World Map Web interface (httphealthcybermapsemanticweborgworld_map)

HealthCyberMap

Maged N Kamel Boulos

et al

copy Blackwell Science Ltd 2002

Health Information and Libraries Journal

19

pp189ndash200

191

on the right In this particular cybermap healthinformation resources are mapped to countries (ofauthorspublishers) rather than cities and listed ina separate pop-up text window to avoid map clutterNote the lsquoFind resources having the same primarysubject as this one from all over the worldrsquo link atthe end of each resource bibliographic card in theresource list pop-up window to the right

Spatialization and metaphors

Spatialization is the process by which informationwith no inherent spatial attributes (no geogra-phical referent) is mapped onto a defined spatialframework using a variety of spatial metaphorsInformation attributes are transformed into aspatial structure (conceptual space) that can bevisualized and navigated through the applicationof concepts like hierarchy proximity and simil-arity eg a topical hierarchy based on the subjects

of a collection of web resources The goals ofspatialization are to increase the spatial legibilityand comprehension of information spaces improvenavigation through them and enable people tofind the information they are searching for in thesespaces more easily

2

The primary function of a metaphor is toprovide a partial understanding of one kind ofexperience in terms of another kind of experienceSpatial metaphors act as fundamental sensemakers for abstract domains Familiar metaphorstaken from usersrsquo everyday life are usually mucheasier to understand

8

In cybermaps metaphorcomprehension can be enhanced with appropriateuse of visual variables (eg colour) and applicationof sound cartographic design principles

12

It is noteworthy that a graphical browseralready exists for visualizing Read Codes usingthe familiar human body metaphor

13

However noone has yet used such visual interfaces that are

Figure 1 Screenshot of HCM World Map Web interface (httphealthcybermapsemanticweborgworld_map)

HealthCyberMap

Maged N Kamel Boulos

et al

copy Blackwell Science Ltd 2002

Health Information and Libraries Journal

19

pp189ndash200

192

based on the human body metaphor as theauthors did in HCM to categorize and browseInternet information resources indexed using aclinical coding scheme

Methods and service description

HCM semantic distance metric

In cyberspace conventional map metrics likedistance map projection scale and grid assumenew meanings and new metrics also arise InHCM we used a lsquodistancersquo metric based on thelsquosemantic locationsrsquo of resource topics within aclinical coding scheme projected on a human bodyorganssystems map The clinical coding schemeacts as a semantic conceptual space with resourcesoccupying different locations in this space basedon their meaning (semantics or subject topics)The lsquosemantic distancersquo between two resourceswill then depend on how close (or related) the tworesources are from a semantic perspective (basedon their subjects and their semantic locations as

determined by the clinical coding ontology theyare mapped to) For example a resource onlsquomyocardial infarctionrsquo should be much closer to aresource on lsquoangina pectorisrsquo than to anotherresource on lsquopsoriasisrsquo

We used ICD-9-CM (International Classifica-tion of Diseases ninth revision US ClinicalModification

14

) as the clinical coding schemeontology in HCM pilot Table 1 shows the top-level grouping or classification of resources inHCM based on ICD-9-CM code ranges

Ideally resources on multi-organ-system dis-eases should be spatialized to all relevant humanbody locations not just a single location ie theyshould be listed under all pertinent categoriesThis will ensure that users will always find theinformation they are looking for in the placeswhere they expect it to be present

HCM hierarchical human body topical maps

HCM uses a hierarchical set of human body topicalmaps to navigate resources by body location

Table 1 Top-level grouping or classification of resources in HCM based on ICD-9-CM

Code assigned resource (semantic location) Corresponding body organsystem location (projected spatial location on human body maps)

(001ndash139) Infectious and parasitic diseases(140ndash239) Neoplasms(240ndash259) Endocrine diseases(260ndash279) Nutritional and metabolic diseases and immunity disorders(280ndash289) Diseases of the blood and blood-forming organs(290ndash319) Mental disorders(320ndash389) Diseases of the nervous system and sense organs(390ndash459) Diseases of the circulatory system(460ndash519) Diseases of the respiratory system(520ndash579) Diseases of the digestive system(580ndash599) Diseases of the urinary system(600ndash629) Diseases of the reproductive system(630ndash676) Complications of pregnancy child birth and the puerperium(680ndash709) Diseases of the skin and subcutaneous tissue(710ndash739) Diseases of the musculoskeletal system and connective tissue(740ndash759) Congenital abnormalities(760ndash779) Certain conditions originating in the perinatal period(780ndash799) Symptoms signs and ill-defined conditions(800ndash999) Injury and poisoning(V01-V829) V codes (factors influencing health status and contact with health services eg vaccinations)(E800-E999) E codes (external causes of injury and poisoning)

Sub-ranges should be spatialized to appropriate locations on human body maps

HealthCyberMap

Maged N Kamel Boulos

et al

copy Blackwell Science Ltd 2002

Health Information and Libraries Journal

19

pp189ndash200

193

system according to ICD-9-CM All maps areused to locate (by querying an underlying database)launch health resources on the Web and displaytheir bibliographic metadata records

Semantic zooming

HCM human body maps adopt a semanticzooming approach With a conventionalgeometric zoom all objects change only their sizewith semantic zoom they can additionally changeshape details (not merely size of existing details)or indeed their very presence in the display withobjects appearingdisappearing according to thecontext of the map at hand

15

A good example ofsemantic zoom in HCM is illustrated in Fig 2

the lung digestive system and other body organsdisappear and new (not just enlarged) details ofthe cardiovascular system appear in consecutivemaps

Tools used and service implementation

HCM has been developed as a GIS project andfeatures GIS-driven spatialization GIS takessimple cartography one step further by providingcontextual links between maps and underlyingdatabases (where attributes of features on themaps are stored) On the Web these links can beimplemented as sensitive clickable maps

We used ESRI ArcView GIS v31 for Windows(httpwwwesricom) with BodyViewer v21

Figure 2 Three human body hypermaps from HCM (httphealthcybermapsemanticweborgbodyviewer)

HealthCyberMap

Maged N Kamel Boulos

et al

copy Blackwell Science Ltd 2002

Health Information and Libraries Journal

19

pp189ndash200

193

system according to ICD-9-CM All maps areused to locate (by querying an underlying database)launch health resources on the Web and displaytheir bibliographic metadata records

Semantic zooming

HCM human body maps adopt a semanticzooming approach With a conventionalgeometric zoom all objects change only their sizewith semantic zoom they can additionally changeshape details (not merely size of existing details)or indeed their very presence in the display withobjects appearingdisappearing according to thecontext of the map at hand

15

A good example ofsemantic zoom in HCM is illustrated in Fig 2

the lung digestive system and other body organsdisappear and new (not just enlarged) details ofthe cardiovascular system appear in consecutivemaps

Tools used and service implementation

HCM has been developed as a GIS project andfeatures GIS-driven spatialization GIS takessimple cartography one step further by providingcontextual links between maps and underlyingdatabases (where attributes of features on themaps are stored) On the Web these links can beimplemented as sensitive clickable maps

We used ESRI ArcView GIS v31 for Windows(httpwwwesricom) with BodyViewer v21

Figure 2 Three human body hypermaps from HCM (httphealthcybermapsemanticweborgbodyviewer)

HealthCyberMap

Maged N Kamel Boulos

et al

copy Blackwell Science Ltd 2002

Health Information and Libraries Journal

19

pp189ndash200

194

for ICD-9 codes an ArcView extension fromGeoHealth Inc (httpwwwgeohealthcombodyviewerhtml) to generate HCM human bodymaps The process is based on an underlyingresource metadata base where ICD codes describ-ing the topics of mapped resources are storedalongside other useful information about theseresources including their web addresses We thenused Zebris WebView 11 (httpwwwzebriscom)the Internet extension to ArcView to translateHCM maps from ArcView to the Web in the formof client-side imagemaps

HCM approach to serving imagemaps with dynamic database links

WebView is much cheaper compared to a dedicatedESRI Internet Map Server (IMS) though not aspowerful as the latter Although it saves users thetrouble of setting-up and running more complexIMS software the basic WebView set-up does notsupport any real GIS database drill-down function-ality (the generated maps cannot communicatewith the corresponding underlying databases)Moreover projects published by WebView onthe Web are uncoupled or disconnected from theoriginal corresponding projects in ArcView

In HCM we developed our own (partial)workarounds for these limitations We made useof WebView HotLink functionality to implementdynamic (ie running in real-time) database drill-down links that will always reflect the latestupdates to this database By clicking differenthotspots on the generated client-side imagemapsin HCM Web interface users are actually trigger-ing server-side preformulated SQL (StructuredQuery Language) queries against the underlyingdatabase of resource metadata

Selecting resources and building the metadata base

Candidate Internet resources are hand-selectedTheir attributes including web address ICD-9-CM codes representing their subjects and anyrecognized qualitycode of ethics rating they bear(eg a Health On the Net Foundation HON sealmdashhttpwwwhonch) are manually compiled inHCM metadata base based on the Dublin Core

(DCmdashhttpwwwdublincoreorg) metadata setscheme with HCM own extensions for resourcequality and geographical provenance

Manual resource indexing ensures the quality

16

of selected resources and the precision of theirtopic indexing Automatic free-text resource index-ing (using conventional web spiders) althoughpossibly providing much wider coverage in lesstime cannot ensure the quality or precision oftopic indexing of spidered resources and cannotindex non-textual multimedia web resources

HCM allows for three DC subject fields perresource record permitting up to three ICD-9-CM codes to be used to describe the topic(s) ofeach selected resource We used two online ICD-9-CM code locators (httpwwweicdcom andhttpwwwe-mdscomicd9) to locate codes thatbest describe topics covered by a given resource

The resource metadata base was implementedin Microsoftreg Access We used ArcView lsquoSQLConnectrsquo feature to connect to HCM metadatabase and import all fields and records from it intoan ArcView table that will refresh each time theproject is opened in ArcView This is the samedatabase running on HCM server which usersquery by clicking the hypermaps

Using BodyViewer to generate multi-level human body maps of ICD-9-coded resources

BodyViewer combines the power of GIS withcomputerized body organ system diagrams It letsusers see where their ICD-coded healthcare data(medical Internet resources in our case) map ontothe human body based on the body region(s) theycover Multiple levels of analysis with multi-levelhuman body maps are provided Each organsystem is broken down into its major componentsfor a finer level of detail eg the Digestive System isbroken into Mouth Oesophagus Stomach LiverGallbladder Pancreas Small Intestine LargeIntestine Rectum Other-Digestive and MetabolicDisorder Public Issues are a special categoryInstead of body diagrams communicable diseasesare categorized according to their mode of trans-mission using meaningful symbols

Our bibliographic use of this extension to mapICD-coded medical Internet resources is the firstof its kind We used BodyViewer to generate HCM

HealthCyberMap

Maged N Kamel Boulos

et al

copy Blackwell Science Ltd 2002

Health Information and Libraries Journal

19

pp189ndash200

195

human body topical maps in ArcView Body-Viewer can aggregate more than one ICD-9 codefield at a time and so was able to use all three DCsubject fields in HCM metadata table combinedto compute resource counts by subject categorybody region

BodyViewer human body map symbols areminiature simplified drawings or icons of thedifferent body organs and systems that observe allapplicable cartographic rules for good map symboldesign

4

The icons act as easy-to-understandvisual labels (familiar metaphors) to the differentresource categories that have been classified andmapped according to the ICD-9 codes in theirDC subject fields On the corresponding HCMweb hypermaps these icons are linked to respectivequery pages that are executed on HCM web serverto retrieve the appropriate resources based on theICD-9 codes represented by the clicked icon

A choropleth rendition for spotting topical coverage gaps

BodyViewer classifies resource counts per bodyregion into ranges and associates each rangewith a colour shade or tint ie a choroplethrendition (organs with darker red tints have moreresources associated with them than organs withlighter red shades a grey colour denotes no

resourcesmdashFig 3) This allows us to visually spotlsquoinfogapsrsquo and lsquoinfoclustersrsquo a useful form oflsquocyberspatial analysisrsquo Infogaps represent bodyareas (topics) where resources are deficient andshould be addressed by information providers(topical coverage gaps) They can be also due toinsufficient indexing by HCM

Linking BodyViewer maps to resources

BodyViewerlinking of its maps to the underlying resourcemetadata table within ArcView can only be doneusing one DC subject field at a time In thisregard the corresponding HCM human bodymaps on the Web are superior as the linking SQLquery looks in all three DC subject fields in theunderlying metadata base We inserted a HotLinkfield in BodyViewer map tables to store the webaddresses of corresponding query pages that willrun on HCM web server this field is associatedwith the HotLink mouse event feature of WebView(Fig 3)

HCM BodyViewer maps are available on theWeb at the following address httphealthcybermapsemanticweborgbodyviewer (Fig 4) These humanbody topical web maps can be used to visuallybrowse selected health resources by clinical subject

Maintenance of HCM human body web maps

AsWebView does not allow the dynamic generation

Figure 3 The HotLink field that has been added to the underlying table of a BodyViewer view in ArcView GIS

HealthCyberMap

Maged N Kamel Boulos

et al

copy Blackwell Science Ltd 2002

Health Information and Libraries Journal

19

pp189ndash200

196

of web maps from ArcView some of HCMweb maps will ultimately need to be manuallyregenerated using WebView when the underlyingdata change (if this change has implications on themapsrsquo appearance) For example all BodyViewerchoropleth maps should be regenerated wheneverthe underlying database is updated (resourcesadded andor deleted) as the colour shades of thevarious body organ systems in these maps reflectthe number of resources associated with themThe corresponding web maps must be recreatedin ArcView using WebView then uploaded to theWeb server to replace older ones Associated querypages need not be changed

A broken link checker is used regularly to detectany dead resource links in HCM database (httphealthcybermapsemanticweborg linkcheckerhtm)

Other HCM interfaces

Besides the human body maps described aboveHCM features other forms of spatialization andmaps The different ways of partitioning topicsin the resource metadata base represent differentuseful views of the same resource pool HCM

uses conventional geographical maps to mapInternet health resources to the country of theircorresponding providers (Fig 1) Anothertype of HCM hypermaps categorizes resourcesby type based on DC type field (httphealthcybermapsemanticweborgtypehtm)

There are also alternative ways to browse HCMresource metadata base in case users find it diffi-cult to visually locate what they want on the mapsThese alternative interfaces include a textualResource Index using ICD-9-CM top-level cat-egories and an Advanced Resource SearchEngine by Subject based on user-typed text (httphealthcybermapsemanticweborgicdhtm) Thissemantic search engine goes beyond conventionalfree-text search engines and supports synonymsdisease variants subtypes as well as some semanticrelationships between terms

Discussion

On the use of clinical codes in HCM

HCM cybermaps can be considered assemantically spatialized browsing views of the

Figure 4 Screenshot of HCM BodyViewer map interface on the Web (httphealthcybermapsemanticweborgbodyviewer)

HealthCyberMap

Maged N Kamel Boulos

et al

copy Blackwell Science Ltd 2002

Health Information and Libraries Journal

19

pp189ndash200

197

underlying resource metadata base Mappingconceptual information spaces of web resourcesbased on their semantics has been demonstrated inseveral other systems eg StarWalker

17

(httpwwwbrunelacuk

sim

cssrccc2vrml2starwalker)However HCM adopts a unique clinical metadataframework that builds upon a clinical codingscheme This is very much suited for the semanticcategorization navigation and retrieval of medicalinformation resources on the Internet

A clinical coding scheme can fulfil the followingtasks in relation to digital libraries

18

bull

navigating and browsing through information

bull

indexing knowledgendashboth general medicalknowledge and information about individualpatients (this can form the basis of clinicalproblem-to-knowledge linking

19

mdashsee httphealthcybermapsemanticweborgpkhtm)Reuters Health (httpwwwreutershealthcom)

currently uses

(Systematized Nomen-clature of Medicine) a clinical coding scheme tocategorize medical stories and provide informationspecific to clientsrsquo interests Compared withMedical Subject Headings clinical coding schemeslike

(and to a lesser extent ICD-9-CM)offer more precise coding more specificity ofmedical conditions (narrower terms) and moresophisticated relationships

20

Meaningful maps without clutter

Using a clinical coding ontology as a metric forspatialization (lsquosemantic distancersquo) to generatemeaningful navigational cybermaps is unique toHCM Ontology-based information visualizationis a rapidly growing research field

21

to which HCMproudly belongs by adopting an ontology-basedframework (ICD-9-CM) for the classification andvisualization (browsing and navigation) ofInternet health resources

The authors believe that the use of familiarmedical metaphors for visualizing these resourcesis far superior to using abstract map symbols torepresent these resources on a map (like the starsand dots in StarWalker

17

and Visual Net PubMedinterface

22

mdashhttppubmedantarcticastart)In HCM map query results (resources) are

listed in a separate text window (Fig 1) to avoidmap clutter The latter would have been unavoid-

able had we opted to represent each resourceusing a distinct point symbol on the map (cfVisual Net PubMed interface

22

) Query resultswill always reflect the latest updates carried onHCM metadata base without the need to changeany code

Complementary interfaces

The different forms of spatialization andcorresponding hypermaps in HCM complementeach other rather than being mutually exclusiveAlthough no one who is interested in informationfor example about lsquoangina pectorisrsquo would tryto search and call up this information by lookingfor and clicking on a map with the geographicallocation of the servers carrying that information(they would go instead to the human body mapsfor this kind of query) the geographical worldmaps can still prove useful when browsing forno specific reason (exploring) or doing someanalytical research on the provenance of differentresources or looking for location-specific healthservices disease rates or guidelines

HCM hypermaps should also be perceived asa complementary improvement over rather thantotal replacement of HCM textual interfacesDepending on userrsquos prior knowledge and queryhypermaps could be more intuitive and faster thantextual category lists and keyword searches inlocating and selecting topicsresources

HCM intended audience

The intended audience for HCM includeshealthcare professionals and librarians patientsand the public in general Meeting all the needsof such a widely varied audience is not an easytask and was the main reason for experimentingwith the different (but complementary) HCMinterfaces described above

There is a growing trend to see medical know-ledge as a single corpus or pool of knowledgerelevant to both doctors and patients and thusshould be made accessible to both groups withoutany distinction Lay persons sometimes showmore knowledge and understanding of their owncondition than their doctors do Supporters of thistrend think that patients should be empowered

HealthCyberMap

Maged N Kamel Boulos

et al

copy Blackwell Science Ltd 2002

Health Information and Libraries Journal

19

pp189ndash200

198

and given more information and control of theirconditions

23

HCM pilot does not currently organize informa-tion resources by their intended primary audienceWe are asking users in our online formative evalua-tion questionnaire (see below) about the usefulnessof doing so in a future implementation

HCM versus topic maps

HCM is clearly sharing most of ISO Topic Mapsrsquopivotal concepts

24

Thanks to its resource meta-data base HCM can automatically and dynamic-ally categorize (classify) the resources in its indexin many different ways to generate different setsof visual and textual lsquotopic mapsrsquo Although theacquisition of metadata in HCM depends on ahuman cataloguer the automated categorizationof these resources based on clinical codes andother metadata fields should save the cataloguertime and effort The underlying clinical codingscheme could also help the automatic generationof a list of topicsresources related to a given resourcesubject code ISO Topic Maps on the other handlargely depend on manual categorization

The DC metadata set can be easily mappedto ISO Topic Maps In HCM the actual topics(concepts) are the clinical codes which are them-selves extracted from a separate ontology ICD-9-CM (to populate the DC subject field) Theoccurrences are the web resources themselves(DC identifier field) Occurrence roles correspondto DC type field eg image of lsquopalmoplantarpsoriasisrsquo versus fact-sheet on the same subject

HCM evaluation

The authors believe that evaluation of any serviceshould run throughout its lifetime and not onlyfor a limited time This ensures that the servicecontinues to deliver what was promised and helpsprevent designersrsquo blindness (deficiencies overlookedby designers and only seen by users) For thisreason we have launched a formative evaluationof HCM pilot service using an online user ques-tionnaire (httphealthcybermapsemanticweborgquestionnaireasp) and server logs (By formative wemean initial evaluation of concepts in their infancyrather than evaluation of a full-blown service)

Crawford provides a short practical guide on theevaluation of library and information services Hesees evaluation as an internal control mechanismthat ensures resources dedicated to the evaluatedservice are used to the best interests of usersEvaluation can help justifying a service plan andplanning for future improvements Differing needsof different user categories (eg healthcare profes-sionals and lay persons) might be also highlightedduring evaluation

25

Technical developments usually precede usabil-ity questions so it is not surprising that there is nolandmark web map usabilityevaluation researchpublished yet

4 especially in relation to naviga-tional cybermaps

User questionnaires complement and patchmany of the deficiencies of server logs With serverlogs alone we cannot know whether or not userswere satisfied or whether they have found whatthey were looking for (usersrsquo perceived qualityutility) Another problem with anonymous userlogs is that we cannot reliably know how often thesame user comes back to the site

Testing usability and user acceptance is a criticalpart of any web-based information service26

Usability evaluation could include query scenariosbased on representative information seeking tasksand real-world data8 For example asking the userhow easy is it to find resources on say lsquodiabetesmellitusrsquo using HCM maps Ideally effectivenessusability studies need to care for different userprofiles It would be very useful to know more aboutthe background and characteristics of users as thiscould affect their ability to perceive andor to com-prehend a map or visual metaphor eg age previ-ous education existing knowledge and experienceand browserdevice4 For example the authorsdonrsquot expect every user to know in advance thatresources on lsquodiabetes mellitusrsquo are classifiedunder lsquoendocrine disordersrsquo although the intuitiveexploratory nature of the maps can help usersdiscover and learn new things

Some possible future directions

Future possibilities includebull The use of a more comprehensive clinical coding

scheme like combined with betterhuman body maps to care for different user

HealthCyberMap Maged N Kamel Boulos et al

copy Blackwell Science Ltd 2002 Health Information and Libraries Journal 19 pp189ndash200

199

types and needs and a true terminologyserver27 The latter would allow us to reasonwith clinical codes (resource subjects) in moresophisticated semantic ways when retrievingresources

bull The introduction of additional resource group-ings and visual metaphors based on the sameunderlying resource metadata eg small imagesof the different blood cells linked to resourceson blood diseases classified according to themajor blood cell type affected in each diseaseFor skin conditions a regional and morpholo-gical grouping of resources could prove veryuseful

bull Supporting customization based on userrsquosgeographical location to deal with languageas well as any specific health needsonlineresources related to userrsquos location (see httphealthcybermapsemanticweborgiphtm)

Conclusions

This paper describes a novel and unconventionaluse of GIS to map conceptual spaces occupiedby collections of health information resourcesBesides mapping the semantic and non-geographical aspects of these resources usingsuitable spatial metaphors HCM also collects andmaps some geographical aspects of these resourceslike provenance

Metadata-driven information classification andretrieval is usually associated with better precisionand recall rates compared to automated spiderindexing Using clinical codes to describe thesubjects of medical web resources can furtherenhance metadata quality and hence offer sup-erior topical categorization and retrieval of theseresources

The web hypermaps in HCM are client-sideimagemaps with dynamic metadata base linksHCM human body maps with their lsquosemanticzoomingrsquo feature allow the navigation of Internethealth resources by body locationsystem accord-ing to ICD-9-CM codes which act as HCMmedical ontology and are used to describe resourcesubjects in the metadata base The lsquosemanticdistancersquo between two resources on these mapsdepends on how close (or related) the tworesources are from a semantic perspective based

on the lsquosemantic locationsrsquo of their topics withinICD-9-CM The maps are used to locate launchhealth resources on the Web and display theirbibliographic metadata records

HCM addresses many cyber-knowledge needsof Internet health information providers andconsumers The authors believe that the visualcategorization of Internet health resources usingfamiliar spatial metaphors for imagendashword asso-ciation could give users a broad overview andunderstanding of what is available in this complexconceptual space and help them navigate it moreefficiently and effectively Topical coverage gapscan be also easily identified (using the humanbody choropleth maps of resource counts) andaddressed by information providers

Acknowledgements

The authors would like to thank Dr ChristopherAustin president of GeoHealth Inc USA whosupplied BodyViewer v21 Extension Version(ICD-9) free of charge for this research We alsoextend our thanks to Dr David Hunt president ofYaki Technologies USA for providing us withtheir proprietary ICD-9-CM search technologyas a research grant to build HealthCyberMaprsquosAdvanced Resource Search Tool

References

1 Gibson W Neuromancer London Harper Collins 1984 672 Dodge M amp Kitchin R Mapping Cyberspace London

Routledge 20013 Skupin A From metaphor to method cartographic

perspectives on information visualization Proceedings of IEEE Symposium on Information Vizualization (Infovis 2000) Salt Lake City Utah October 2000 Available from httpwwwgeogucsbedusim sarateachinggeo234papersskupinpdf and httpwwwgeogunoedusim askupinresearchinfovis2000figures

4 Kraak M J amp Brown A Web Cartography Developments and Prospects London Taylor amp Francis 2001

5 Dodge M An atlas of cyberspaces Available from httpwwwcybergeographyorgatlasatlashtml (accessed 27 February 2002)

6 Ding J Gravano L amp Shivakumar N Computing geographical scopes of web resources Proceedings of the 26th Very-Large Database (VLDB) Conference Cairo Egypt September 2000 Available from httpwwwcscolumbiaedu7EgravanoPapers2000vldb00pdf

HealthCyberMap Maged N Kamel Boulos et al

copy Blackwell Science Ltd 2002 Health Information and Libraries Journal 19 pp189ndash200

200

7 Staple G C Notes on mapping the net from tribal space to corporate space TeleGeography 1995 Global Telecommunications Traffic Statistics amp Commentary TeleGeography Inc October 1995 Available from httpwwwtelegeographycomPublicationsmappinghtml

8 Fabrikant S I Spatialized browsing in large data archives Transactions in GIS 4(1) 65ndash78 Available from httpwwwgeogucsbedusim sarateachinggeo234paperstig99pdf

9 Old L J Using spatial analysis for non-spatial data Proceedings of the 20th Annual ESRI International User Conference San Diego California June 2000 Available from httpconservationesricomlibraryuserconfproc00professional papersPAP196p196htm

10 Terpstra P Mapping cyberspace with GIS Proceedings of the 18th Annual ESRI International User Conference San Diego California April 1998 Available from httpwwwesricomlibraryuserconfproc98PROCEEDTO650PAP615P615HTM

11 The MARAARMA Collaboration Mapping malaria risk in Africa Available from httpwwwmaraorgza (accessed 27 February 2002)

12 Fabrikant S I Spatial metaphors for browsing large data archives Unpublished PhD Dissertation Boulder CO USA University of Colorado-Boulder 2000 Available from httpwwwgeogucsbedu 7Esarahtmlresearchdiss spatializationhtml and httpwwwgeogucsbedu 7Esarahtmlresearchdisssf_disszip

13 The Visual Read Company UK Graphical Read Codes Browser v60 Example Screenshot Available from httpwwwvisualreadcomvisreadpage04a_graphical_imagehtm (accessed 27 February 2002)

14 Centers for Disease Control and Prevention (CDC)mdashNational Center for Health Statistics USA Classification of Diseases (ICD-9-CM) Available from httpwwwcdcgovnchsicd9htm (accessed 27 Feburary 2002)

15 Spence R Information Visualization Essex UK ACM Press 2001

16 Kamel Boulos M N Roudsari A V Gordon C amp Muir Gray J A The use of quality benchmarking in assessing web resources for the dermatology virtual branch library of the National electronic Library for Health (NeLH) Journal of Medical Internet Research 2001 3(1) e5 Available from httpwwwjmirorg20011e5gt

17 Chen C Thomas L Cole J amp Chennawasin C Representing the semantics of virtual spaces IEEE Multimedia 1999 6(2) 54ndash63 Available from http

wwwbrunelacuksim cssrccc2papersieee_multimediachen99pdf

18 Rector A L Clinical terminology why is it so hard Methods of Information in Medicine 1999 38 239ndash52

19 Kamel Boulos M N Roudsari A V amp Carson E R A dynamic problem-knowledge coupling semantic web service In Della Mea V Beltrami C A Woodall J amp Arvanitis T N (eds) Proceedings of the 6th World Congress on the Internet in Medicine Udine Italy December 2001Technology and Healthcare 2001 9 477ndash479 Amsterdam IOS Press Available from httpmednet2001drmmuniudit proceedingspaperphpid=44

20 McKillen D News Report SNOMED RTreg Enables Reuters Health to Categorize Medical Stories and Provide Information Specific to Clientsrsquo Interests Available from httpwwwsnomedorgreuterspdf and httpwwwsnomedorgprodtepr_reuters00pdf (accessed 27 February 2002)

21 van Harmelen F Broekstra J Fluit C ter Horst H Kampman A van der Meer J amp Sabou M Ontology-based information visualisation Presented at the Workshop on Visualisation of the Semantic Web (VSW rsquo01) September 2001 London in Conjunction with the 5th International Conference on Information Visualisation Available from httpwwwaidministratornl usersdevelopmentfilesVSW01pdf

22 Antarctica Systems Inc Canada Visual Net PubMed Interface Available from httppubmedantarcticastart (accessed 27 February 2002)

23 Muir Gray J A The Resourceful Patient Oxford eRosetta Press 2002 Available from httpwwwresourcefulpatientorg

24 ISOIEC 13250 Topic Maps Available from httpwwwy12doegovsgmlsc34document0129pdf (accessed 3 December 1999)

25 Crawford J Evaluation of Library and Information Services 2nd edn London Aslibimi 2000

26 Mazzi C P amp Kidd M A framework for the evaluation of internet-based diabetes management Journal of Medical Internet Research 2002 4(1) e1 Available from httpwwwjmirorg20021e1gt

27 Bechhofer S K Goble C A Rector A L Solomon W D amp Nowlan W A Terminologies and terminology servers for information environments Proceedings of STEP rsquo97 Software Technology and Engineering Practice 1997 Available from httpciteseernjneccom354766html