Kluwer Academic publishers, Boston, 321-329. 5. Rubin, K.S., Jones, P.M., and Mitchell, C.M., 1988, âOFMspert: Inference of operator intentions in supervisoy.
Application of Cognitive Systems Engineering to Decision Aiding Design Maranda E. McBride, Kaize A. Adams and Celestine A. Ntuen Department of Industrial & Systems Engineering Greensboro, NC 27411 Natalie MaZaeva Department of Industrial Sunny, Buffalo, NY 14228 Abstract This paper presents sample case study in the use of cognitive systems to design of a decision system for triage nursing station information management. The prototype software known as Triage Nursing Station (TNS) provides solutions to many constraints reminiscent in the current manual operations. Usability evaluations of the TNS indicate consistent likelihood support by the experts with no statistical significant differences in opinion
Keywords Cognitive systems engineering, decis ion aiding, Triage Nursing Station
1. Introduction In the past twenty years, cognitive systems engineering (CSE) has gained a significant momentum as a discipline for studying, designing, and analyzing human-centered socio-technical systems. Within the CSE framework, the assumption is that human cognition of the work domain produces the knowledge required for design. For this reason, CSE is considered an embodiment of human cognition [1] which draws upon the “human dimensions” of various disciplines, notably, cognitive psychology, computation and information sciences, and artificial intelligence, and social sciences. Various definitional constructs have been accorded CSE [1-3]. Among these are: 1. CSE is concerned with the synthesis of human knowledge about a task domain and its application to work analysis and design of cognitive enabling supporting tools. 2. CSE is about designing computer-based tools or artifacts to support human activities in a complex work system. 3. CSE involves the analysis and synthesis of cognitive resources vis -a-vis the environmental factors, tasks, and information about the system. 4. CSE evolved as attempts to crystallize human problem solving and decision- making techniques become automated. As noted by Norman [ 2], the aims of cognitive engineering are first, “to understand the fundamental principles of behind human action and performance that are relevant for the development of engineering principles of design, and second, ‘to devise systems that are pleasant to use (p.32)”. These definitional constructs have engendered many psychological and design studies specifically in human-computer interaction and recently in designing team decision aiding and training systems [4 ].
2.Decision Aiding System Design 2.1 Description of Decision Aidi ng Systems Decision aiding systems (DAS) are computer-based decision models for supporting human decision making processes. It is automated in the sense that the human cognitive models are implemented with computer and are designed to assist or collaborate with humans during task performance. These systems are fondly referred to by many names, including, joint-cognitive systems (Wood & Roth, 1988), decision support systems, expert systems, and decision assistants or associates.
2.2 Cognitive Aspects of Decision Aiding Systems Decision aiding systems are functions of cognitive technology and their designs have been mitigated by cognitive engineering system(CSE) principles, including analytical cognitions CSE is about the integration of human knowledge about task (environment and perception), cognition, and artefact behaviors that can lead to execution and control of tasks at various levels of abstractions. As shown in Figure 1, the human is involved in knowledge production through mind and body connections. Usability feedback
Mind/ Cognition
Body
Knowledge Production
HUMAN
jj
HCI
Hardware
HHH Information Production
Decision Aid software
DAS Cognitive Systems Engineering Testbed: Principles & Models Figure 1: Sample CSE testbed showing human and DAS components The human body and mind are the major sources for knowledge production through interrelated activities of the brain. Sample cognitive tasks include planning, estimating, correlating, diagnosing, deciding and executing and monitoring. DAS software contains symbolic and analytical representation of the human knowledge and is often implemented with some hardware. The main purpose of DAS is information production. Through usability analysis, the DAS software can be incrementally refined and improved to serve domain-specific needs.
2.3 Human Knowledge Production for Design of Cognitive Aiding Systems The human operator provides the main source of knowledge for cognitive aid designs. The knowledge varies along a discrete continuum of the operator’s level of expertise, psychological states and traits, and task dimensions. In CSE, the level of expertise is commonly assessed along dimensions of skill-, rule -, and knowledge-based behaviors known to control the decision-making ability of the operator [1 ]. Implicitly, the levels of expertise allow us to replicate human mental model of a system with a computer. There are many knowledge acquisition processes available for collecting human knowledge for system design. Among these methods are cognitive task analysis (CTA) and cognitive work analysis [1].
3. Sample Application of CSE to Design of Triage Nursing Station Decision Aid 3.1 The Design Domain The domain of discourse is a nursing station in an emergency unit of a large hospital. The triage nursing station represents an example of complexity, dynamic information flow, and variations in emergency cases and incidents. It also typifies decision- making processes involving planning, diagnosis, monitoring, and execution. These conditions lead to frequent cognitive workload responsible for decision errors in nursing diagnosis. This scenario fits the domain for CSE application.
3.3 Method of Knowledge Acquisition
Several methods were employed to acquire information regarding triage nursing operations. First, an unstructured interview with the personnel involved was conducted. The nurse director determined the flow of the interview by introducing departmental procedures and tasks. The second method included motor protocols analysis where group members observed how triage nurses performed their tasks (e.g., walking, writing, moving paperwork, etc.). The final method was the analysis of an expert’s decision making in assigning patients to treatment rooms based on nursing diagnosis. Direct interviews were also used to determine the decision processes used at each level of patient incident processing (PIP).
3.2 Application of CSE Models 3.2.1 Decision ladder model: The decision ladder model is a framework based on information structure abstraction, primarily used to portray the map of information flow from abstract to concrete dimension [1]. Figure 2 shows example of decision ladder model for the triage nursing station information flow. 3.2.2 Operator function model (OFM): The operator function model (OFM) is a framework for modeling cognitive processes in a dynamic system [5]. The OFM allows for mathematical representation of human-system behaviors based of state transitions mitigated by task allocation and function requirements. Figure 3 shows the representation of the triage nursing information flow with the OFM model. 3.2.3 Model Implication: A computerized triage nursing station was designed using Microsoft Access (Version 2000) and Visual Basic. The software programs were designed to assume responsibility for automatic nursing functions. Figure 4 shows an example knowledge acquisition screen. The screen allows the nurse to record patient’s information on personal record as well as vital signs. Once the information is completed, the nurse will activate the decision aid model to support in cognitive tasks of room assignment, and prioritizing patient status.
Knowledge-based Domain
Ambiguity
Restore patient to pre-injury health status Injury could be serious, possibly deadly
Interpret consequences for current task - safety, efficiency, etc.
n o w
Target state
ID present state of system
Define task
Send patient to the appropriate department for care
Patient has undergone some type of injury Set of observations
Task
Observe information and data
Formulate procedure; plan sequence of actions
s s i l y n a da s e b a e d g l e
System state
w n o
K
Ultimate goal
K
l e d g e b a s e dp l a n n i n g
Evaluate performance criteria Patient can go to Acute Care or Fast Track for care
Administer treatment according to procedural guidelines
Patient arrives at ER Alert
Procedure
Rule-based Domain Activation; detection of need for action
Release of preset response
Skill-based Domain
Execute; coordinate manipulations
Figure 2 Decision ladder model of information flow in the triage nursing station
Operator Function Model Legend:
T
4
Activity 1 Triage Patient
2
Monitor System (Activity 4)
Top-Level Activities
1 2 3 4 5 6 7 8 9 10
1
Patient arrives at ED (I) Patient has to be registered (I) Patient has to be treated (I) Patient has been triaged (T) Patient has been registered (T) Patient has been treated (T) Patient has been examined (I&T) Patient has been categorized (I&T) Patient has to be examined (I) - Patient is sent to location
3
Seq
Activity 3 Treat Patient
T Activity 2 Register Patient
5
(not within the scope of this project)
6
T
And
Activity 1a Perform initial exam
8
1
Level 2 Decomposition
Seq
Activity 1b Categorize patient
0
Activity 1e Respond to requests in person
7
9 8
T T
Lower Level Activities (Actions)
Activity 1c Send patient to appropritate location
T
7
9
T
Action 1 - Acquire all information regarding injury
T
Activity 1d Give telephone advice
Action 2 - Take vital signs Action 3 - Get medical history
Action 1 - Acquire all pertinent information (personal, insurance, health, employer, etc.)
Action 2 - Complete the appropriate forms
Figure 3 Operator function model for the triage nursing station
4. Evaluation The nurse experts were used to conduct heuristic usability evaluation on the system Questions used to evaluate the efficacy of the system was to ensure that TNS implements specifications and determines how well the design conforms or simulates the mental model of the system. The TNS was evaluated primarily on the likelihood that the nurses use the system as an associate in triage nursing tasks. The likelihood ratio or diagnostic support given the decision aid is defined by L(e/TNS) = P(e/TNS)/ P(e/~TNS), where e is the evidence of performance improvement or acceptability, P(e/TNS) is the observed probability of performance when TNS is used, and P(e/~TNS) is the initial performance without TNS. Given a task space, e can be observed as a vector of task performances, e = (e 1 , e2 , e3 , …,en ).
5 Conclusion The development of TNS demonstrates the use of cognitive systems engineering principles and models for design of a human-centered system. The system was designed on the using human knowledge on the nursing tasks The TNS provides solutions to many constraints reminiscent in the current manual operations. Usability evaluation of the TNS indicates consistent likelihood support by the experts with no statistical significant differences in opinion (student tstatistics = 3.067 > t (6,0.05) = 1.943).
Figure 4 Sample screen for knowledge acquisition in the TNS domain.
References 1. 2. 3. 4.
5.
Rasmussen, J., 1986, Information Processing and Human-Machine Interaction: An Approach to Cognitive Engineering, North Holland: New York. Norman, D., 1986, “Cognitive engineering:, appears in User-Centered System Design, Norman, D. and. Draper, S. (eds.), Lawrence Erlbaum Associates, New Jersey, 31-61. Woods, D., and Roth, E., 1988, “Cognitive engineering: Human problem solving tools”, Human Factors, 30(4), 415-430. Ntuen, C.A., and Pioro, B.T., 1996, “ Theoretical issues and design principles for human interaction with complex systems”, appears in Human Interaction with Complex Systems: Conceptual Principles and Design Practice, Ntuen, C.A., and Park, E.H. (eds.), Kluwer Academic publishers, Boston, 321-329. Rubin, K.S., Jones, P.M., and Mitchell, C.M., 1988, “OFMspert: Inference of operator intentions in supervisoy control using a blackboard architecture”, IEEE Transactions on Systems, Man, and Cybernetics, 8, 618-637.