Professional Documents
Culture Documents
Designing Human-Centered Distributed Information S
Designing Human-Centered Distributed Information S
Designing Human-Centered Distributed Information S
net/publication/220628131
CITATIONS READS
104 323
5 authors, including:
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Vimla L Patel on 31 May 2014.
Designing Human-
Centered Distributed
Information Systems
Jiajie Zhang, University of Texas at Houston
Vimla L. Patel, Columbia University
Kathy A. Johnson and Jack W. Smith, University of Texas at Houston
and NASA Johnson Space Center
Jane Malin, NASA Johnson Space Center
principles and are typically difficult to learn and use. The fields of human-
computer interaction, interface design, and human factors have made significant contri-
butions to ease of use and are primarily concerned with the interfaces between systems
Human-centered and users, not with the structures that are often more Distributed cognition considers human and artifi-
fundamental for designing truly human-centered sys- cial agents to be indispensable components of a sin-
5
distributed information tems. The emerging paradigm of human-centered gle distributed system. Human activities in concrete
computing (HCC)—which has taken many forms— situations are guided, constrained, and even deter-
design methodology offers a new look at system design. mined by the physical and social context in which
9,10
HCC requires more than merely designing an arti- they operate. One study has argued that the prop-
incorporates the theory ficial agent to supplement a human agent. The erties of a distributed cognitive system consisting of
dynamic interactions in a distributed system com- a group of human agents interacting with complex
of distributed cognition posed of human and artificial agents—and the con- systems—in an airplane cockpit or the control room
text in which the system is situated—are indispens- of a ship—can differ radically from the properties of
with the need for able factors. While we have successfully applied our the components, and they cannot be inferred from
methodology in designing a prototype of a human- the properties of the components alone, no matter
multiple levels of centered intelligent flight-surgeon console at NASA how detailed the knowledge of the properties of those
Johnson Space Center, this article presents a method- components might be.1,11
analyses in system ology for designing human-centered computing The theory of distributed cognition makes three
systems using electronic medical records (EMR) major claims:
design. This systems.
• The unit of analysis is the system composed of
methodology provides Distributed cognition human and artificial agents.
We base our human-centered computing per- • The pattern of information distribution among
systematic principles, spective on the theory of distributed cognition.1–9 human and artificial agents can radically change
According to this theory, people behave in infor- the behavior of the distributed system.
guidelines, and mation-rich environments filled with natural objects, • The behavior of a distributed system should be
artificial objects, and agents, whether human or non- described by the information flow dynamics.
procedures for designing human. This environment, according to the theory,
is grounded in elaborate social and cultural struc- One research project described a similar system as
human-centered tures. In everyday life, people must process infor- the triples rule, which states that the unit of analysis
mation derived from human agents, artificial agents, for cognitive engineering and computer science is a
computing systems. and groups of agents. It is the interwoven process- triple composed of person, machine, and context.12
ing of such information that generates intelligent The goal of a traditional approach is to make user
behavior. interfaces transparent so users can completely engage
Authorized licensed use limited to: IEEE Xplore. Downloaded on February 9, 2009 at 23:31 from IEEE Xplore. Restrictions apply.
Distributed cognition analysis
Multiple levels of analysis
User analysis: Identify characteristics of users Distributed user analysis: Analyze division of
such as expertise and skills, knowledge base, age, labor, overlap of knowledge and skills,
education, cognitive capacities and limitations, pattern of communication, and social interaction
perceptual variations, and so on.
Figure 1. The methodology of human-centered distributed information design. Each of the four distributed analyses contributes
to its corresponding analysis on the left. The component at the bottom represents the product of functional, task, and
representational analyses. For each level of analysis, our general methodology lets us employ several alternative, specific methods.
in their primary tasks. With transparent inter- ing HCC systems. We selected electronic The components on the left are multiple lev-
faces, users directly interact with the system. medical records systems to demonstrate our els of analyses for single-user, human-cen-
However, user-friendly interfaces are far from methodology because HCC is almost nonex- tered design. The functional, task, and repre-
sufficient as a design goal. In our methodol- istent in EMR systems. sentational analysis levels have degrees of
ogy, user-friendly interfaces are only one of EMR systems are highly complex, dis- abstraction—with the level of functional
four levels of analysis. The three other lev- tributed information systems that, with analysis most abstract and the level of repre-
els—functions, tasks, and users—are just as proper human-centered design, have the sentational analysis most concrete. The user-
important. For example, an excellent user- potential to improve the quality of health care analysis level contributes to each of the lev-
friendly interface might be irrelevant if the dramatically. The lack of even minimum con- els of functional, task, and representational
task the interface supports is not the task the siderations of HCC design principles in most analysis. The components on the right repre-
user performs. current EMR systems makes them very dif- sent the additional analysis needed for design-
ficult to learn and use. In turn, this difficulty ing distributed human-centered systems.
Design methodology leads to strong resistance by physicians, and
Our methodology, called human-centered in some cases it leads to abandoning EMR User analysis
distributed information design (HCDID), systems altogether. The EMR system we dis- User analysis provides user information to
incorporates the theory of distributed cogni- cuss here is not a specific system. Rather, it the functional, task, and representational
tion with the need for multiple levels of is a collection of EMR systems that are cur- analyses. User analysis is the process of iden-
analyses in system design. We developed this rently on the market. tifying user characteristics, such as expertise,
methodology to provide systematic princi- Figure 1 shows the HCDID methodology, skill, knowledge, educational background,
ples, guidelines, and procedures for design- which consists of three major components. cognitive capacity, and so forth. User analy-
Authorized licensed use limited to: IEEE Xplore. Downloaded on February 9, 2009 at 23:31 from IEEE Xplore. Restrictions apply.
H C C a t N A S A
sis is necessary for designing systems that unnecessary and redundant interactions. of detailed domain knowledge and a deep
have an information structure that must Distributed responsibilities let the team understanding of domain structures.
match user knowledge levels. process massive amounts of patient infor- Consider one of the products of doing a
For an EMR system, different users oper- mation, which reduces individual cognitive functional analysis for an EMR system. An
ate different system components. These users loads. Individual expertise contributes to ideal EMR should be able to support the fol-
include installers, maintainers, administra- accomplishing team goals, although team lowing functions: data, alerts, reminders,
tors, nurses, physicians, registration person- performance shows a completely different schedules, clinical decision support, medical
nel, laboratory technicians, billing staff, and structure. knowledge, communication, and other aids.
patients. Different users will have different You can obtain user information with indi- These functions should be complete, accu-
levels of understanding when it comes to the rect or direct methods. With direct methods, rate, and timely. These functions should also
same component of the system. you would collect the information from the be available for all types of health-care pro-
Researchers have written about the impor- users with surveys, questionnaires, field vis- fessionals. Furthermore, these functions
tance of technology in shaping cognitive its, focus groups, interviews, or any of several should be available at all times and at all
activity and the impact of cognition on tech- ethnographic observation techniques. With points of care. The ideal EMR should over-
13
nology design. These studies show that indirect methods, you collect the information come the known problems associated with
using EMR systems requires several changes from other sources, such as textbooks, hand- paper-based records but still provide that
in physicians’and trainees’information-gath- books, protocols, procedures, job descrip- kind of functionality. Finally, and most
ering and reasoning strategies. For example, importantly, an ideal EMR should be able to
there are differences in the content and orga- improve the quality of health care.
nization of information, with paper records A distributed functional analysis of EMR
organized according to a narrative structure One important function of task can show the variety of artificial and human
and computer-based records organized agents and their interrelations. Examples of
according to discrete items of information. analysis is ensuring that the system artificial agents include various modules
This difference in knowledge organization within EMR, such as patient record module,
results in different data-gathering strategies, implementation includes only the alert module, decision-support module, and
where the doctor-patient dialogue is cogni- communication module. Examples of human
tively influenced by the structure of the EMR
system.
necessary and sufficient task agents include physicians, nurses, laboratory
technicians, and registration staff. An exam-
For distributed human–computer systems,
it is critical to analyze users’ group proper-
features that match user capacity ple of the interrelations between artificial and
human agents is a mapping that shows which
ties, such as division of cognition and activ- artificial agent supports which human agent,
ity, overlap of knowledge and skills, com- and are required by the task. and to what degree.
munication channels, social status, and so on.
For example, whether two human agents are Task analysis
better than one human agent often depends tions, operation rules, and other training and Task analysis is more concrete than func-
on the overlap of knowledge and skills. The education materials. tional analysis because it involves specific task
ways in which effort is distributed across a structures and procedures. However, it is still
group of human agents can significantly Functional analysis more abstract than representational analysis
affect the outcome of a task. Functional analysis is the process of iden- because it does not involve the details of how
A recent study shows the importance of tifying critical top-level domain structures— information is represented. Task analysis is a
proper user analysis for EMR systems.14 goals that can be largely independent of critical component in cognitive systems engi-
Nurses and physicians—who have partially implementations. Functional analysis is more neering and usability engineering. It consists of
overlapping knowledge bases, skills, expe- abstract than task and representational analy- the process of identifying system functions,
riences, and job responsibilities—rely on ses because it does not involve details of task task procedures, input and output formats, con-
different patient strategies that can lead them processes and representation. In other words, straints, communication needs, organization
to different understandings, diagnoses, and functional analysis is similar to traditional structures, information categories, and task
subsequent activities. Another study char- requirements analysis. Work domain analy- information flow.
acterized the qualitative nature of team inter- sis (which focuses on the structures of a work One important function of task analysis is
action and its relationship to training health domain) and cognitive work analysis (which ensuring that the system implementation
professionals in a primary care unit in a focuses on cognitive activities in the work includes only the necessary and sufficient task
Boston hospital.15 This study showed that domain) are two of the methodologies that features that match user capacity and are
the demarcation of responsibilities and roles can help in conducting functional analysis. required by the task. Fancy features—and fea-
of personnel within the team can become For a distributed system, functional analysis tures that do not match user capacity or are not
fuzzy. also identifies a system’s artificial and human required by the task—might only generate
The nature of the individual expertise agents, their interrelations and constraints, additional processing demands for the user,
required was a function of the patient prob- and their essential roles. For a knowledge- and thus make the system harder to use. But
lem and the interaction goal. Team charac- rich domain such as medicine or aviation, avoiding unnecessary features does not mean
teristics often contribute to reduction of functional analysis requires the acquisition excluding adaptation mechanisms that dynam-
Authorized licensed use limited to: IEEE Xplore. Downloaded on February 9, 2009 at 23:31 from IEEE Xplore. Restrictions apply.
ically adjust the interactions between users for any individual or even a small research of these interactions for designing according
and tasks in changing contexts. group. However, it is essential for designing to human-centered principles. To our knowl-
Hierarchical task analysis, which is the a new EMR system or redesigning an exist- edge, very little analysis of this type has been
basic method for revealing the structure of ing EMR system to make it human-centered. done for EMR systems.
any task that has goals and subgoals, Commitment from a large institution is
describes a task in terms of a hierarchy of essential for a comprehensive task analysis. Representational analysis
operations by using a graphical representa- We have carried out preliminary task analy- Functional analysis and task analysis deal
tion. Hierarchical task analysis decomposes sis of several small components of EMR sys- with system functions, structures, and pro-
a high-level task into its constituent subtasks tems. For example, one analysis showed that cesses, whereas representational analysis
and operations, and is essential for any sys- writing a prescription using an EMR system deals with the interface between systems and
tem consisting of both human and artificial requires more steps and much more time than users. Representational analysis is based on
agents. This type of analysis typically shows writing a paper prescription, although the a robust phenomenon called representational
the big picture of the task components and mental effort required is less for EMR than effect,5 in which different representations of
the relations among them. It also takes into for paper prescription writing. a common abstract structure or process can
account both physical and mental activities. Additionally, performing a hierarchical task generate dramatically different representa-
The theory of distributed representations is analysis to identify an EMR system’s under- tional efficiencies, task difficulties, and
the foundation for the analysis of the distri- lying data structure can help pinpoint some behavioral outcomes.
bution of information patterns among human The form of a representation can influence
and artificial agents.5,6 Human agents inter- and sometimes determine what information
act with each other and with artificial agents can be easily perceived, what processes can
synchronously and asynchronously—at one Distributed task analysis can show be activated, and what information can be
place and at different places. The analysis of derived from the representation. For a com-
locations and activities can reveal how tasks crucial interactions among human plex or novel task, some portion of the task
are distributed across space and over time. space might never be explored, and some
One study demonstrated this type of analy- and artificial agents and can structures of the task might never be discov-
sis on more than 20 medical personnel, who ered without a change in representation. Rep-
interacted over the Internet and through con-
ference calls over a period of two years. The
provide an understanding of these resentational analysis can be performed on
system properties that are identified through
purpose of the study was to develop software
that could be used in any hospital system.16
interactions for designing according functional and task analysis. With direct-
interaction interfaces, users can efficiently
Information flow analysis can help analyze engage in the primary tasks they intend to
how the information gets propagated and to human-centered principles. perform. And they can avoid the interface
transformed among human and artificial housekeeping tasks that typically act as bar-
agents. Distributed task analysis, by the same riers between users and systems.
token, can reveal critical task structures that fundamental problems. For example, one study Representational analysis can help sys-
cannot be identified by conventional task showed that current EMR systems use one of tematically generate innovative ideas for data
analysis, which focuses on a single individ- two data structures, neither of which is driven displays and information for exploration,
ual’s interaction with a system. by human-centering.13 One EMR data struc- evaluation, and selection. This type of analy-
Task analysis can help identify task struc- ture, for example, uses a hierarchical data sis works through developing a taxonomy
tures, interaction among procedures, and model to capture information used by specific of displays and tasks, and through map-
information flow. For example, task analysis applications. It is primarily a patient-record ping principles between displays and tasks.
can identify overlooked tasks, relative task system added onto a billing system. The other Researchers have applied representational
importance, overlapping task information, data structure makes extensive use of an event- analysis to relational information displays
function groups, relation to user analysis, and based approach by recording information and even cockpit instrumentation. A repre-
so on. It can also help pinpoint task bottle- according to a time-oriented view to facilitate sentational taxonomy of relational informa-
necks where special design requirements its reuse by multiple applications. Unfortu- tion displays, when combined with a task
come into play. Another product of task nately, these two data structures do not support taxonomy from task analysis, can systemat-
analysis is a taxonomy based on information- the daily tasks of health-care professionals, so ically determine the best display format for
processing needs. For example, there are they are not human-centered. A typical daily a specific task.
information tasks for retrieval, gathering, task, such as making a diagnosis, is better sup- Representational analysis can help identify
seeking, encoding, transformation, calcula- ported by an EMR system that organizes infor- alternative displays for each task in the taxon-
tion, manipulation, comparison, organization, mation around problems. omy of information-processing needs and can
navigation, and so on. Identifying different Because EMR systems are used by many also help determine the best match between a
information-processing needs is essential for types of people—synchronously and asyn- display and a task for each unique event under
the creation of task-specific, context-sensi- chronously, and at the same or different loca- each unique situation. This type of analysis then
tive, and event-related information displays. tions—a distributed task analysis can show determines the mapping between a display and
A systematic and comprehensive task crucial interactions among human and artifi- a task with a mapping principle: the display
analysis of an EMR system is out of reach cial agents and can provide an understanding should carry exact information for the task, no
Authorized licensed use limited to: IEEE Xplore. Downloaded on February 9, 2009 at 23:31 from IEEE Xplore. Restrictions apply.
H C C a t N A S A
Authorized licensed use limited to: IEEE Xplore. Downloaded on February 9, 2009 at 23:31 from IEEE Xplore. Restrictions apply.
References References 7. J. Holland, E. Hutchins, and D. Kirsh, “Dis-
tributed Cognition: Toward a New Founda-
lection, Organization, and Reasoning,” J. Am.
Medical Informatics Assoc., vol. 7, no. 3,
1. E. Hutchins, Cognition in the Wild, MIT tion for Human-Computer Interaction July–Sept. 2000, pp. 569–585.
Press, Cambridge, Mass., 1995. Research,” ACM Trans. Computer-Human
Interaction, vol. 7, no. 2, June 2000, pp. 14. C.M. Johnson et al., “Understanding the Men-
174–196. tal Models of Nurses and Physicians Reading
2. D.A. Norman, Things that Make Us Smart, Patient Records,” Proc. AMIA, Am. Medical
Addision-Wesley, Boston, Mass., 1993. 8. D.D. Woods, “Designs Are Hypotheses About Informatics Assoc., Washington, D.C., 2002.
How Artifacts Shape Cognition and Collab-
3. G. Solomon, Distributed Cognitions: Psycho- oration,” Ergonomics, vol. 41, no. 2, Feb. 15. V.L. Patel et al., “The Collaborative Health
logical and Educational Considerations, Cam- 1998, pp. 168–173. Care Team,” Teaching and Learning in Med-
bridge Univ. Press, Cambridge, U.K., 1996. icine, vol. 12, no. 3, Summer 2000, pp.
9. W. J. Clancey, Situated Cognition, Lawrence 117–132.
Erlbaum, Mahwah, N.J., 1997.
4. P.C. Wright, R.E. Fields, and M.D. Harrison,
16. V.L. Patel et al., “Toward a Framework for
“Analyzing Human-Computer Interaction as
10. R.R. Hoffman et al., “The Triples Rule,” IEEE Computer-Mediated Collaborative Design in
Distributed Cognition: The Resources
Intelligent Systems, vol. 17, no. 3, May–June Medical Informatics,” Methods of Informa-
Model,” Human-Computer Interaction, vol.
2002, pp. 62–65. tion in Medicine, vol. 38, no. 3, Sept. 1999,
15, no. 1, 2000, pp. 1–41.
pp. 158–176.
11. L. Suchman, Plans and Situated Actions,
5. J. Zhang and D. A. Norman, “Representations Cambridge Univ. Press, Cambridge, U.K., 17. J.J. Cimino et al., “What is Wrong with
in Distributed Cognitive Tasks,” Cognitive 1987. EMR?,” Proc. Am. Medical Informatics
Science, vol. 18, no. 1, Jan./Mar. 1994, pp. Assoc., Washington, D.C., 1999.
87–122. 12. E. Hutchins, “How a Cockpit Remembers Its
Speed,” Cognitive Science, vol. 19, no. 3,
6. J. Zhang, “A Distributed Representation July–Sept. 1995, pp. 265–288.
Approach to Group Problem Solving,” J. Am. For more information on this or any other com-
Soc. of Information Science, vol. 49, no. 9, 13. V.L. Patel et al., “Impact of a Computerized puting topic, please visit our Digital Library at
July 1998, pp. 801–809. Patient Record System on Medical Data Col- http://computer.org/publications/dlib.
you@computer.org
The email address
of computing professionals
Authorized
View publication stats licensed use limited to: IEEE Xplore. Downloaded on February 9, 2009 at 23:31 from IEEE Xplore. Restrictions apply.