Diagnose Source

AN ABSTRACT OF THE THESIS OF
Kiumars Zolfaghari for the degree of Master of Science in Industrial Engineering

presented on November 21, 2014
Title: Medical Diagnosis: A Functional Model and Diagnostic Aid
Abstract approved:
______________________________________________________
Kenneth H. Funk II
Improving the quality of the medical system has always been of great
importance but has not always made the largest advances. A significant percentage of
medical errors are due to wrong, missed, or delayed medical diagnosis. Diagnostic
errors are among the most preventable categories of medical errors, but when an error
is made it is harder to overcome the harmful effects on a patient, in contrast to other
types of medical errors. Through this research the flaws of the clinical diagnostic
process were explored and addressed by designing a means to present medical
information. Cognitive biases are a major cue in diagnostic errors, and the challenge
of finding their potential countermeasures were also addressed. A systems
engineering approach consisting of problem analysis, modeling, requirements
development, design, and testing was followed to find a solution. Being aware of the
changing state of healthcare, one understands that little effort is currently devoted to
designing appropriate visualization and navigation techniques for presenting medical
history records, in contrast to developing standards for gathering medical records,

which receives substantial attention. This lack of effort worked as a motivation to
develop a solution to the problem.
The Diagnostic Aid was developed according to a formal IDEF0 model of the
diagnostic process, requirements derived from the model, and information
visualization principals. The effectiveness of the Diagnostic Aid in presenting
medical information, preventing cognitive errors, and matching physicians’ need was
tested. A quantitative questionnaire and a qualitative interview were used to evaluate
the subjects’ perceptions about the system and to verify requirements. The users’
overall perceptions of the system were positive and they found the system potentially
helpful. The system did not interfere with their decision making process, at least in
the context that was tested. The Diagnostic Aid was found to be as efficient as the
paper format in presenting medical information, yet it could be improved. Most of the
requirements were evaluated positively and new requirements developed for the next
generation of the Diagnostic Aid.

©Copyright by Kiumars Zolfaghari
November 21, 2014
All Rights Reserved
Medical Diagnosis: A Functional Model and Diagnostic Aid
by
Kiumars Zolfaghari
A THESIS
submitted to
Oregon State University
in partial fulfillment of
the requirements for the
degree of
Master of Science
Presented November 21, 2014

Commencement June 2015
Master of Science thesis of Kiumars Zolfaghari presented on November 21, 2014
APPROVED:
Major Professor, representing Industrial Engineering
Head of the School of Mechanical, Industrial and Manufacturing Engineering
Dean of the Graduate School
I understand that my thesis will become part of the permanent collection of Oregon
State University libraries. My signature below authorizes release of my thesis to any
reader upon request.
Kiumars Zolfaghari, Author

ACKNOWLEDGEMENTS
I would like to express my special appreciation to my advisor, Dr. Funk, for his intellectual
support and patience in guiding me over the years. I would like to thank my graduate
committee members, Dr. Haapala, Dr. Eseonu and, Dr. Pavol for serving as my committee
members and for reviewing this work. Many thanks go to Dr. Lisa Madsen from the
Department of Statistics, who is a great teacher with a wonderful personality.

TABLE OF CONTENTS
Page
1 Introduction ....................................................................................................................... 1
1.1 Background ............................................................................................................... 1
1.2 Motivation ................................................................................................................. 2
1.3 Objectives ................................................................................................................. 3
1.4 Methodology ............................................................................................................. 4
2 Literature Review.............................................................................................................. 6
2.1 Medical Diagnosis .................................................................................................... 6
2.2 Diagnostic Reasoning and its Challenges ................................................................. 8
2.3 Medical Errors .......................................................................................................... 9
2.3.1 Diagnostic Errors ................................................................................................ 10
2.4 Systems Engineering............................................................................................... 19
2.5 Modeling ................................................................................................................. 20
2.5.1 IDEF0 Modeling ................................................................................................. 21
2.6 Requirements .......................................................................................................... 22
2.6.1 Validation and Verification................................................................................. 23
2.7 Information Visualization ....................................................................................... 24
2.7.1 Mapping Data to Visual Form ............................................................................ 26
2.7.2 General Design Principles................................................................................... 27
2.7.3 Using Visualization to Navigate Patient Records ............................................... 28
3 Development of the Diagnostic Aid ............................................................................... 30
3.1 Model Description .................................................................................................. 30
3.1.1 The A-0 and A0 Diagrams .................................................................................. 31
3.1.2 The A2 Diagram ................................................................................................. 33
3.1.3 A21 node- Evaluate Patient’s History, Medical Sign and Symptoms ................ 34
3.1.4 A22 node- Generate Hypotheses ........................................................................ 36
3.1.5 A23 node- Compare, Reject, and Finalize Hypothesis ....................................... 37
TABLE OF CONTENTS (Continued)
Page
3.2 Developing Requirements....................................................................................... 39
3.2.1 Functional Requirements .................................................................................... 41
3.2.2 User Interface Requirements............................................................................... 43
3.2.3 Debiasing Tools .................................................................................................. 46
3.3 Design ..................................................................................................................... 50
3.3.1 Diagnostic Aid Development .............................................................................. 50
4 Test and Results .............................................................................................................. 59
4.1 Participants.............................................................................................................. 59
4.2 Apparatus ................................................................................................................ 60
4.3 Procedure ................................................................................................................ 63
4.4 Results ..................................................................................................................... 65
4.4.1 Technical Questions Results ............................................................................... 65
4.4.2 Usability Questions Results ................................................................................ 67
4.4.3 System Attributes Questions Results .................................................................. 69
4.4.4 Qualitative Interview .......................................................................................... 71
5 Discussion ....................................................................................................................... 77
5.1 System Usability Experience .................................................................................. 77
5.2 Randomized Experiment......................................................................................... 80
5.3 Requirements Verification ...................................................................................... 82
5.4 Positives and Negatives .......................................................................................... 87
5.5 Recommendations ................................................................................................... 88
5.6 Limitations .............................................................................................................. 90
6 Conclusion ...................................................................................................................... 93
6.1 Summary ................................................................................................................. 94
6.2 Future research ........................................................................................................ 94
Bibliography ........................................................................................................................... 98
Appendices ............................................................................................................................ 101
LIST OF FIGURES
Figure Page
Figure 2.1. IDEF0 Semantics……………………………………………………………..…22
Figure 2.2. Reference model for visualization……………………………………………....27
Figure 3.1. IDEF0 Model – A-0 Diagram: Conduct medical encounter.………….......…….32
Figure 3.2. IDEF0 Model – A0 Diagram: Conduct medical encounter sub processes...…....32
Figure 3.3. IDEF0 Model – A2 Diagram: Diagnose the patient problem…………………...33
Figure 3.4. IDEF0 Model – A21 Diagram: Evaluate patient’s history, medical sign and
symptoms.……………………………………………………...…….….……..….…….......35
Figure 3.5. IDEF0 Model – A22 Diagram: Generate hypotheses..……………..…….……..37
Figure 3.6. IDEF0 Model – A22 Diagram: Compare, reject, and finalize hypotheses……...39
Figure 3.7. Patient’s medical history screen..……………………………………..…………53
Figure 3.8. Hypothesis generation screen.………………………………….……....………..55
Figure 3.9. Final diagnosis screen.……………..……...……………………….……………57
Figure 5.1. Usability question results for each participant.…………....……….……………78
LIST OF TABLES
Table Page
Table 4.1. Demographic characteristics of the participant……….……...………………......60
Table 4.2. Results of technical questions for each question………….…..………….……....66
Table 4.3. Results of technical questions for each participant…...……….……..………..…67
Table 4.4. Results of usability questions 1-6 and 12……………..…………………….…....68
Table 4.5. Results of usability questions 7-11 and 13…...……….……..…………….….….69
Table 4.6. Results of system requirements verification……...……….……..…….…………70
LIST OF APPENDICES
Appendix Page
Appendix 1. Entire IDEF0 Diagram ……………...……………………………...……......102
Appendix 2. Requirements….……………………………..…………………………..…..105
Appendix 3. Diagnostic Aid Screens …...………...…………………………………...…..111
Appendix 4. Research Protocol Submitted to IRB.……………..……………….…...........120
Appendix 5. Informed Consent Form……………………………………………………...125
Appendix 6. Clinical Scenarios …...…………..…………….….…………………………128
Appendix 7. Questionnaire …...…………..…….……………………………….……..….132
Appendix 8. Frequency table of Usability Questions 7-11 and 13…...………….……..….142
LIST OF APPENDIX FIGURES
Appendix Page
Figure A.1. Conduct medical encounter ............................................................................... 102
Figure A.2. Conduct medical encounter sub processes ........................................................ 102
Figure A.3. Diagnose the patient problem ............................................................................ 103
Figure A.4. Evaluate patient's history, medical signs and symptoms ................................... 103
Figure A.5. Generate hypotheses .......................................................................................... 104
Figure A.6. Compare, reject and finalize hypotheses ........................................................... 104
Figure A.7. Refute hypotheses not consistent with context of the problem ......................... 105
Figure A.8. Final Diagnostic Aid screen example 1 ............................................................. 112
Figure A.9. Final Diagnostic Aid screen example 2 ............................................................. 113
Figure A.10. Final Diagnostic Aid screen example 3 ........................................................... 114
Figure A.17. Participants’ results for usability Questions 7-11 and 13 ................................ 143
DEDICATION
Lovingly dedicated to my Mom.

1 Introduction
A great deal of mortality in the health care system is the result of medical error.
According to a report by US Institute of Medicine in 2000, medical error was the 8th
leading cause of death in the United States (Kohn, Corrigan, & Donaldson, 2000).
However, an updated estimate that was developed form more recent studies indicates that
the severity of medical errors is more than twofold what was approximated before. James
(2013) concludes that preventable medical errors contribute to the death of at least 210,000
patients annually. This study captured only a fraction of all medical errors and the actual
number is estimated to be more than 400,000 per year, he reasoned, due to the limitations
of the study such as incompleteness of medical records. The severity of medical errors’
outcomes requires immediate and effective action toward patient safety.
1.1 Background
It is important to deliver the most efficient form of treatment to reduce the costs
while increasing patient satisfaction. Various factors of different natures contribute to
medical errors such as poor performance of the clinician, wrong drug treatment, failure in
the process of diagnosis, system errors, and even characteristics of the patient (Kohn et al.,
2000). Neither all medical errors lead to adverse events, nor are all those medical errors
that result in adverse events preventable. According to Leape, Lawthers, Brennan, and
Johnson, (1993), more than sixty six percent of adverse events in healthcare are preventable
and diagnostic errors are contributing factors in approximately thirty percent of

2
these adverse events. However, the absence of considerations of diagnostic errors from
patient safety activities is notable.
There are three different categories of diagnostic errors: no-fault errors, system
errors, and cognitive errors. The advance of medical knowledge over long-term and
system-level improvements is the potential solution for no-fault errors and system errors
respectively. However, the greatest challenge is eliminating or even reducing cognitive
errors, specifically diagnostic errors resulting from cognitive biases (Graber, Gordon, &
Franklin, 2002). Croskerry (2003) provides a reasonably complete list of cognitive biases
that may lead to diagnostic errors and suggests debiasing strategies to reduce them.
Nevertheless, he clarifies that practical solutions to the problem of diagnostic errors are still
unknown. Graber, et al. (2002), explain potential methods for reducing diagnostic errors in
a more generalized way than that of Croskerry, although Graber mentions that the
possibility of decreasing diagnostic errors by the means of heuristics and debiasing
methods should be further explored.
1.2 Motivation
Decision-support systems are usually computerized systems that support humans in
their decision making process. According to Schiff and Bates (2010), ways of improving
diagnostic accuracy should be found other than decision-support systems. They believe that
presenting clinical information, facilitating provider-to-provider communication, and
implementing diagnostic checklists would be more efficient ways to reduce diagnostic
errors.
3
Decision-support systems as an additional backup is welcomed by most physicians
provided that the system does not make the physician slower, does not interfere with their
practice, be customizable to fit the physician’s need, and preferably takes advantage of new
technologies. Therefore, a system that considers all the above factors could be of great help
in increasing the quality of clinical diagnosis.
1.3 Objectives
The primary purpose of this research was to reduce cognitive errors in the
diagnostic process and therefore improve patient safety. To achieve this objective a
Diagnostic Aid for a tablet computer was developed that represents key clinical information
visually and implements cognitive and systematic solutions to diagnostic errors. In this
regard, among the features of the Diagnostic Aid is representing medical history
graphically, tracking medical tests, providing personalized feedback, and supporting the
clinician with a list of diagnoses not considered initially. It was deemed essential
beforehand to identify the important steps and characteristics of the clinical diagnostic
process by developing a functional model and establishing requirements for developing a
Diagnostic Aid. What happens in the clinical encounter process and especially in the
diagnostic process were the key points to learn about diagnostic errors and to find ways to
reduce these errors. The final step to achieve this goal was to validate and test the
effectiveness of the Diagnostic Aid through a usability study and a randomized experiment.
So the main objectives of the study were to:
 Understand the clinical diagnostic process, develop a model of the process, and
establish the requirements for the Diagnostic Aid.

4
 Reduce the chance of cognitive errors and facilitate the clinical diagnostic process
by developing a prototype of the Diagnostic Aid for a tablet computer.
 Investigate the effectiveness of the Diagnostic Aid through user experience
evaluation, verify the requirements, and generate new requirements for the next
generations of the Diagnostic Aid.
1.4 Methodology
A functional model of the diagnostic process was developed employing knowledge
from experts, medical informatics literature, and the medical diagnosis literature. This
model represents clinician-patient and clinician-system interactions in the process of
diagnosis so that common diagnostic errors can be explained and anticipated and
countermeasures could be developed. This model demonstrates a clear picture of what is
happening during the process of diagnosis, and helps to explain diagnostic errors. Next, a
set of requirements for designing and developing a Diagnostic Aid was established based
mainly on the model of the diagnostic process, information visualization literature, and
talking to subject matter experts. The set of requirements considered development of the
Diagnostic Aid from human factors and cognitive science viewpoint and that the
Diagnostic Aid provided potential countermeasures to diagnostic errors in the diagnostic
process. Then, the effectiveness of the Diagnostic Aid and its advantage over the traditional
presentation of clinical scenarios in paper were tested through a randomized experiment
using a prototype. The participants’ general opinion about the Diagnostic Aid and their
specific viewpoint regarding future development of the Diagnostic Aid were collected.
5
The following chapters cover the following topics in order: literature review, design
methodology, results of the experiment, analysis of the results and discussion, and
conclusion.
6
2 Literature Review
This study focuses on facilitating the diagnostic process by reducing diagnostic
errors in a clinical process. This chapter presents a comprehensive literature review of the
disciplines consulted for this research. First, medical diagnosis, diagnostic errors,
cognitive errors, and solutions to these errors are presented. Second, a modified systems
engineering approach for this study and its subsystems such as modeling and
requirements development are introduced. Finally, a review of information visualization
as well as its functions and attributes are presented.
2.1 Medical Diagnosis
The clinician-patient encounter, the interaction between the clinician and the
patient—provider and receiver of care respectively—could be considered as the most
important aspect of the healthcare system. The more accurately and respectfully the
clinician does his1 job, as evaluated by its outcome, the more satisfied the patient would
be. A successful encounter is likely if different steps of the clinician-patient encounter
work integrally. The clinician-patient encounter consists of four steps: interview,
examination, diagnosis, and treatment. This research has its focus on differential
diagnosis which has examination as an integrated segment as well. The terms diagnosis
and differential diagnosis are used interchangeably in this research.
1
Masculine pronouns are used in this thesis to refer to the physician for the purpose of
avoiding ungrammatical prose, knowing that many physicians and patients are females.
7
Differential diagnosis is a systematic method of identifying the existence of an
entity among multiple alternatives. This method is a process of elimination or at least
reducing the candidates to small numbers. In a clinical context, this procedure is used by
clinicians and other medical professionals to consider all possible causes of a problem
and diagnose the specific disease, or, at least, eliminate the potentially life-threatening
conditions (Sox, Blatt, Higgins, & Marton, 2006). Different definitions and models for
diagnosis have been introduced up to the present; however, they all agree that diagnosis
involves both logical reasoning and pattern recognition. A step by step thinking model of
differential diagnosis that has three repeating steps lead to a fourth step, that of a
treatment plan. These steps are listening and generating hypotheses, gathering data to test
hypotheses, evaluating hypotheses, and selecting a course of action. Defining diagnostic
possibilities and estimating their relative probability is the core of the diagnostic process.
Many differences exist between novice and experienced clinicians regarding the
reasoning phase of diagnosis. Experts are capable of grouping their findings into different
clusters or biological, psychological, or sociological clinical problems which are objects
of differential diagnosis. The clinician starts working on these clusters and the patient’s
narrative to reach the potential diagnosis Sox et al. (2006) .
Having identified the context of the problem, the clinician checks his assumptions
with specific medical tests. Collecting more information results in enhancing the
probability and accuracy of the potential diagnosis, eliminating other possibilities, and
finally choosing the most likely diagnosis. Each of these steps of differential diagnosis
has its own clinical reasoning strategies (Baerheim, 2001; Richardson, Wilson, & Guyatt,
2002).
8
2.2 Diagnostic Reasoning and its Challenges
To understand the difficulties in the process of diagnosis, the process is explained
through an example. A patient visits his doctor because of the appearance of a sign. She
wants his problem to be solved fast and with minimum discomfort, thereafter she wants
to know about the reason. In responding to this, the clinician might have difficulty
especially when there is more than one reason for a single sign or symptom. In a
diagnostic hypothesis, the physician has a hard job to recall typical features of the disease
and also might find that these features are common in other diseases. Moreover, the cost
of gathering all necessary data is often unacceptable. The overall encounter might not be
successful enough due to poor management and communication problems. These
numerous issues bother clinicians, especially the newcomers with less practical
experience (Sox et al., 2006).
So why are diagnostic reasoning and clinical decision making so complicated,
ambiguous, and time consuming? The answer is that uncertainty is inherent in medical
knowledge and, therefore, medical decision making. Mamede, Schmidt, and Rikers
(2007) discussed some facts about this uncertainty. First, medical knowledge is
incomplete and will change and progress with new findings as time passes. Second,
although medical knowledge has grown significantly, complexity, insufficiency,
individual differences, and patients’ preferences do not allow it to be definite knowledge.
Third, a doctor as a body of knowledge is the result of several disciplines, beliefs,
experiences, and other sociological and psychological factors that bring subjectivity to
the diagnostic process. Therefore, clinical decision making is not merely an unbiased,
rational application of scientific knowledge and rules, but it is a complex process

9
encompassing multiple dimensions characterized by uncertainty and ambiguity. As a
result, medical errors are more common than what is desired in the practice of medicine.
2.3 Medical Errors
Medical errors cover a broad range of definitions; however, they can be defined as
any human error in the healthcare system. Medical errors include misdiagnosis or an
inaccurate treatment of a disease or other ailment. For the purpose of this research a few
terms are explicitly defined (Kohn et al., 2000).
An error is the failure of a planned action to be completed as intended (i.e., error of

execution) or the use of the wrong plan to achieve an aim (i.e., error of planning).
An adverse event is an injury [to a patient] caused by medical management rather
than the underlying condition of the patient. An adverse event attributable to error is
a “preventable adverse event.”
Negligent adverse events represent a subset of preventable adverse events that
satisfy legal criteria used in determining negligence (i.e., whether the care provided
failed to meet the standard of care reasonably expected of an average physician
qualified to take care of the patient in question). (p. 28).
“How frequently do medical errors occur?” and “What are the contributing
factors?” are two critical questions that should be discussed. James (2013) predicted that
400,000-plus deaths per year are associated with preventable adverse events (PAEs). He
extended the definition of the PAEs to errors of commission, omission, communication,
context, and diagnostic errors. This would make the medical errors the third-leading
cause of death in the US, right after heart disease and cancer. These results are
significantly more dreadful than the results published by US Institute of Medicine in
2000, which estimated the death of 44,000 to 98,000 patients associated with preventable
medical errors each year. These statistics clarify the noteworthiness of errors in the US
healthcare system and state the importance of enhancing patient safety. This research has
10
its focus on preventable adverse events caused by medical diagnostic errors.
2.3.1 Diagnostic Errors
Diagnostic errors account for a substantial portion of all medical errors.
According to Leape et al. (1991) who investigated the state of New York hospitals’
records in 1984, 69.6% of adverse events were found to be preventable (both negligent
and non-negligent errors). More importantly, the second most common type of
preventable errors was diagnostic errors (21.9%), and about 70% of errors in diagnosis
were deemed negligent. Due to the fact that preventable adverse events cost patients in
New York hospitals alone $739 million per year, 71% of the total compensable cost, it
can be estimated that the expenditure of preventable adverse events for the nation was
over $10 billion in 1984. According to a similar study, physicians were more susceptible
to diagnostic errors than drug related errors, and misdiagnoses were more likely to be
considered negligent (Leape et al., 1993).
In order to find strategies to prevent diagnostic errors, the origin of these types of
errors should be clarified. Graber, Gordon, and Franklin (2002) proposed probably the
most comprehensive framework for diagnostic errors and categorized diagnostic errors in
three major groups. First, ‘no-fault errors’ occur when the illness is silent or masked,
resembles another disease, or appears in such an atypical fashion that finding the right
diagnosis is hardly expected. Patient-related errors such as a patient’s non-compliance are
also categorized in this group. Second, ‘system errors’ are flaws in the healthcare system
such as weak policies, inadequate supervision, defective communication, or any other
system factor that can affect the diagnostician in the healthcare system. Finally,
11
‘cognitive errors’, the focus area of this research, can be directly attributed to the doctor
and his failings. Cognitive errors include inaccurate clinical reasoning, inadequate
knowledge, and faulty data gathering and verification. Failing to perceive a lung nodule
on a chest X-ray is an instance of cognitive error.
2.3.1.1 Inattention to Diagnostic Errors
Despite the fact that diagnostic errors cause notable suffering, the subject of
diagnostic errors has been almost absent from patient safety studies. Graber (2005) put
forward three factors for the ignorance of diagnostic errors. First, diagnostic errors are
fundamentally obscure; in other words, diagnostic errors are complicated in nature, hard
to understand, and consequently difficult to identify. Making a diagnosis is a cognitive
process and is influenced by the feelings and attitudes of the clinician. Additionally,
compared to other medical errors, diagnostic errors are subtle and difficult to identify due
to a lack of clear standards for detecting unacceptable failings. Second, healthcare
organizations view diagnostic errors as physician failure rather than as a system problem.
The available data, however, suggest that system-related errors contribute to a substantial
proportion of diagnostic errors. Wachter (2010) argued that we cannot hold a hospital
responsible for diagnostic errors of its medical staff when no convincingly effective
solution has yet been demonstrated. Third, even though clinicians are keenly aware that
diagnostic errors occur, they hardly ever correctly perceive their rate of problematic
diagnoses. Clinicians mostly believe that their diagnoses are correct and errors are made
by others. Lack of feedback and overconfidence are factors that mislead clinicians about
the accuracy of their decisions.

12
2.3.1.2 Cognitive Errors
Humans have limitations in their ability to process information and therefore use
heuristics. These limitations guarantee the persistence of cognitive errors that reflect
misdiagnosis from inadequate knowledge or faulty data gathering, inaccurate clinical
reasoning, or faulty verification. Graber et al. (2002) explain how cognitive errors happen
during different stages of the diagnostic process, such as perception, hypothesis
generation, data interpretation, and verification. The majority of diagnostic errors result
from the individual’s cognitive process. According to a study of diagnostic errors in three
medical centers, cognitive errors occurred in 74% of error cases, making them more
common than system-related errors and no-fault errors (Graber, Franklin, & Gordon,
2005). The same study concludes that diagnostic error is multifactorial in origin, due to
the fact that cognitive and system-related factors tend to co-occur. The root causes of
diagnostic errors lie in the three areas of attitudes, skills, and knowledge; cognitive skill
errors (cognitive biases) are considered to be a lack of skill and are more common than
errors caused by knowledge gaps. Graber et al. (2005) stated that faulty information
processing, faulty verification, faulty data gathering, and inadequate knowledge are
cognitive contributors to diagnostic errors, in descending order.
2.3.1.3 Cognitive Biases
Heuristics, self educating and problem solving by experience and trial and error
usually work well, facilitate clinical decision making, and are considered economical, and
effective; however, when they fail, they are described as cognitive biases. “Bias” simply
is a person’s propensity to behave and respond in a particular way, but it carries a

13
negative meaning of prejudice (Croskerry, 2002). Croskerry (2003) provides a list of
about thirty failed heuristics and biases involved in diagnostic errors and proposes
strategies to reduce them. Several studies provide and explain dominant and frequent
cognitive biases in the diagnostic process (Croskerry, 2000; J. P. Kassirer & Kopelman,
1989; Norman & Eva, 2010). According to these studies the most prevalent cognitive
biases in clinical decision making are as follows.
 Availability bias: This heuristic involves assessing events or outcomes based on
similar events that are available and come to mind easily. An impressive or
striking event is easy to recall because a combination of findings or certain causal
connections bring it readily to mind. Recent experience with a disease may
increase the chance of its being diagnosed; conversely, if a disease has not been
detected for a while, it may be underdiagnosed. For example, if an emergency
room physician sees a patient with a headache that proved to be a subarachnoid
hemorrhage, it would be more likely to bring subarachnoid hemorrhage to mind
when the next headache comes along (Tversky & Kahneman, 1973; Reason,
1990).
 Anchoring bias: This heuristic involves assessing the likelihood of events or
outcomes based on information available at the outset (i.e., first impression gained
on first exposure). Adhering to salient features of the patient’s initial presentation
might lead to a premature closure of thinking and prevent the physician from
adjusting the initial hypothesis based on further information gained. Thus, the
patient may be labeled very early with an incorrect diagnosis, which is difficult to
14
remove once attached. This cognitive bias usually compounds with confirmation
bias (Kassirer & Kopelman, 1991; Tversky & Kahneman, 1974).
 Representativeness bias: This heuristic is a technique used in probability
assessment and results from finding an event similar to other well-defined events.
In clinical language, it refers to estimating the probability of a disease by judging
how similar it is to a disease category or physician mental templates.
Representativeness is a powerful heuristic for emergency room physicians since it
accentuates the pattern recognition strategy (Croskerry, 2002; Tversky &
Kahneman, 1974).
 Confirmation bias: This heuristic, effective in information processing, is the
tendency to search for or interpret information in a way that verifies one’s beliefs.
In a diagnostic context, it is looking for confirming evidence to support an
existing hypothesis instead of trying to find disconfirming evidence to refute it,
despite the fact that the latter is often more definitive (Reason, 1990; Yates,
1990).
 Premature Closure: This is the leading cognitive bias and has a broader definition
than other biases. The tendency to stop considering other possibilities after
reaching a diagnosis is counted as premature closure. Not taking the medical
history of the patient completely, failing to do a full physical examination if
necessary, refraining from ordering critical tests, and failing to consider the
correct diagnosis are all examples of premature closure. Premature closure could
be the result of failing to use other heuristics appropriately (Kassirer &
Kopelman, 1991; Voytovich, Rippey, & Suffredini, 1985).

15
These biases may mislead diagnostic reasoning throughout the process. Some of
them, including availability, representativeness, and anchoring, have their effects mostly
in hypothesis generation. Others, such as confirmation bias, anchoring, and premature
closure, affect particularly hypothesis refinement (Mamede et al., 2007). These are the
biases that many sources support as the most prevalent biases in clinical diagnosis,
however, there are other biases such as base rate neglect or overconfidence bias that are
also common in clinical decision making process.
Norman and Eva (2010) provided a clinical scenario to illustrate these biases in a
clinical context for further clarification. In their scenario the physician diagnoses a
migraine in a patient with a headache that is actually caused by a tumor. The physician
might be affected by the bias of representativeness if the patient shows some symptoms
consistent with migraine presentation, or by that of availability because he may have seen
a patient with a migraine problem recently. The physician might be accused of
confirmation bias if he only looks for information consistent with migraines, and forgets
to collect the data that match other diagnoses, or he might be accused of premature
closure and anchoring because he did not order a CT scan to look for a potential tumor.
However, if the patient does in fact have a migraine, these heuristic approaches will lead
to a correct diagnosis, and whether or not the physician was affected by cognitive biases
will never be revealed. This example brings out three serious issues about the notion of
cognitive bias. Norman and Eva (2010) stated three important facts about heuristics and
biases in practice. First, heuristics and biases are mental abilities that deal efficiently with
uncertainties in the real world; they are basically good, although they occasionally fail.
Second, it is hard to observe biases in situations where the biases lead to the correct
16
diagnosis, even though cognitive biases have been established experimentally under
carefully manipulated conditions. Third, not considering the correct diagnosis and hence
not gathering the relevant data are indications of premature closure that could be mainly
because of a knowledge deficit rather than a reasoning error.
2.3.1.4 Reducing Cognitive Errors
The greatest challenge in reducing diagnostic errors is the minimization of
cognitive errors, biases, and failed heuristics that underlie them. The unconquerable
obstacle in getting rid of cognitive errors is the impossibility of knowing all of the
relevant facts, and also the inability to process the known facts quickly. Graber et al.
(2002) proposed two approaches to enhance the quality of thinking, perception, and
reasoning, and therefore to reduce cognitive errors: cognitive solutions and system-
related solutions.
Cognitive Solutions: The idea is to reduce cognitive errors and improve clinical
decision making skills by improving cognition. Training is one of the verified methods
that have direct and positive impact on the outcome of the clinical reasoning; however,
effective ways of teaching clinicians to avoid biased judgment is yet to be determined
(Regehr & Norman, 1996). Nevertheless, medical trainees instructed about the pitfalls of
heuristics had better performance confronting different cases than those who were not
similarly trained.
Beside knowledge and skills, attaining accurate diagnoses requires awareness of
one’s thinking, which is achievable through metacognitive training. Metacognition is a
process that involves stepping back from the immediate problem to examine and reflect
17
on the thinking process. One of the applications of metacognitive training is to maintain
open-mindedness regarding cases that require searching for more possibilities and
evidence. Individuals who practice this process tend to fall into the trap of cognitive
biases less and produce better decisions. Briefly speaking, cognitive solutions reduce the
rate of diagnostic errors by directly improving cognition (Baron, 2008; Graber, 2003).
Croskerry (2003), an expert in the domain of cognitive biases, proposed a list of
debiasing strategies for reducing cognitive diagnostic errors. He believed that
metacognition is the mainstay of this approach, which works more efficiently if the
clinician becomes aware of various cognitive pitfalls. Those of his strategies that are
different from what Graber et al. (2002) suggested, and fit to the scope of this research
are as follows:
 Feedback: Provide rapid and dependable feedback to decision makers for
recognition and correction of errors.
 Cognitive forcing strategies: Develop strategies to stay away from expected biases
in particular problems.
 Decrease reliance on memory: Improve the quality of judgment through cognitive
aids such as mnemonics, checklists, guidelines, and hand-held computers.
 Consider alternatives: Initiate forced consideration of other possibilities such as
starting and working through a different diagnosis.
 Minimize time pressure: Provide adequate time for decision making.
On the other hand, Graber (2003) acknowledged that diagnostic accuracy can be
improved if clinicians can be “debiased” by metacognitive training, but he doubted
whether this improvement can be accomplished in real practice. He believed that a few
18
laboratory setting training exercises that were successful in debiasing the learners do not
guarantee the same results in the unique setting of medical decision making. Moreover,
he mentioned that biases are integral components of human information processing, and
awareness does not prevent clinicians from being susceptible to the effects of their biases.
System Solutions: The idea here is to compensate for errors caused by a biased
mind through changes at the system level. The quality of data presentation has a
significant influence on enhancing perception that leads to detecting abnormalities more
easily. Perception can be enhanced by graphic presentations, presentations that facilitate
trend recognition, computer-aided diagnosis implementation, or any other technology that
embraces human factors engineering principles for supplementing perception (Graber et
al., 2003).
According to the same study, another beneficial strategy for reducing cognitive
errors is the use of a second opinion that the clinician obtains either through double-
checking the medical treatment or by asking for help from another expert. Graber and his
team suggested clinical guidelines and clinical decision-support systems for improving
diagnostic accuracy. Guidelines are themselves heuristics, and so they cannot provide a
full solution; however, they can improve the accuracy of diagnosis by standardizing the
approaches to clinical problems and minimizing the variability in response patterns
(Graber et al., 2002). According to a study by (Hunt, Haynes, Hanna, & Smith, 1998),
clinical decision-support systems improved the clinical performance for drug dosing,
preventive care, and the general process of care, but the results were not positive for
clinical diagnosis. The most potentially valuable approach is probably to integrate
decision aids directly with the active practice of medicine.

19
The first step in solving the problem of clinical diagnostic errors involves careful
scrutiny of the problem, its facets as well as the framework in which the problem is
defined. This scrutiny often suggests a well-ordered cognitive and systematic approach
for considering all relevant factors to explain determinants of a diagnostic problem. After
understanding the problem, potential solutions could be considered in respect to the
system’s limitations and strengths. Proposing the idea of the Diagnostic Aid and its
characteristics would be the next stage. In accordance, a systems engineering method was
selected for the methodology part of this research.
2.4 Systems Engineering
According to (Chapanis, 1996), systems engineering is the process by which
systems are analyzed, designed, and constructed. Budurka (1984) defined this concept in
more detail:
Systems engineering is the iterative but controlled process in which user needs are
understood and evolved, through incremental development of requirements
specifications and system design, to an operational system. Systems engineering
includes the control and integration of all disciplines throughout the system life
cycle in a manner so as to assure that all user requirements are satisfied. (p. 41).
To contribute effectively to the diagnostic process from an engineering viewpoint,
a systems engineering approach was used in this research. Budurka’s definition calls
attention to serving human purposes or fulfilling human needs. The need to improve
patient safety through facilitating the diagnostic process defines the reason to develop a
new system. Budurka’s definition first emphasizes the iterative nature of system
development. In the IDEF0 modeling phase of system development, ideas are tried,
modified, and abandoned in favor of other alternatives. Next, the definition takes notice
20
of specifications. Engineers and human factors professionals are required to prepare
specifications. Therefore, requirements, in the current project, have been derived from the
medical diagnosis literature with the benefit of human factors inputs. Finally, Budurka’s
definition indicates that system development is a multidisciplinary activity and requires a
team with members from different disciplines with various points of view. Human factors
engineering, cognitive psychology, medicine, and information visualization are among
the specialties that played a role in establishing the requirements for the Diagnostic Aid
in the current research project. This Diagnostic Aid aims to improve the physician
performance in the diagnostic process through visualizing medical information and
providing features to prevent or reduce the effect of cognitive biases. With a little change
in systems engineering approach developed by (Chapanis, 1996), the current research
involved analysis, modeling, requirements development, design and test as its main steps.
These concepts are explained in more detail in the following sections.
2.5 Modeling
A systematic approach is required to understand the diagnostic process as the
major component of a healthcare system and from that understanding derive requirements
for a means to facilitate and improve diagnosis. Such an approach raises the necessity of
a comprehensive modeling language to model a very complex process in detail. It has
been shown that the IDEF0 language is capable of representing a wide variety of
healthcare services, is very easy to learn, is efficient in precise expression, and is flexible
enough to facilitate interaction between humans and processes (Mutic, Brame, Oddiraju,
Michalski, & Wu, 2010).

21
IDEF0 is based on a graphical language known as Structured Analysis and Design
Technique (SADT), developed for the US Department of Defense. IDEF0, as a function
modeling tool, was first designed to model manufacturing and logistics processes. IDEF0
is structured to represent complex processes by means of a set of rules and techniques for
creating graphical representation and enhances communication between systems analysts,
developers and users (Gaithersburg, 1993; Kim & Jang, 2002).
2.5.1 IDEF0 Modeling

An IDEF0 model provides shared understanding among different parties working
on the system. IDEF0 is a top-down graphical modeling language that describes the scope
of the system at the highest level; at lower levels the model can be decomposed into
details gradually. In an IDEF0 diagram, the flow of information cascades from the upper
left to lower right from the most dominating tasks to lesser ones. As shown in Figure 2.1,
IDEF0 has a simple syntax of boxes and arrows with five elements. A box contains a
verb or verb phrase and represents an activity, process, or function. Inputs shown as
arrows entering the left side of the activity box are anything that is transformed by the
activity such as matter, energy, information, or the state of the system or some
subsystems; outputs shown as arrows exiting from the right side of the activity box are
the result of the process; controls or constraints are represented by arrows that flow into
the box from the top and define how the process works; and the mechanisms that carry
out the activity are represented by arrows that flow into the box from the bottom.
There is a parent and child relationship between different diagrams at subsequent
levels of an IDEF0 model that continues as far as needed to show the information in
22
appropriate detail. Reading an IDEF0 diagram starts with the highest level that contains
the general purpose of the model. Then, at the next level, all activities decomposed from
the main process and the arrows define their relationships. Diagrams should be read from
the upper left to lower right to understand how arrows interact with activities. Usually a
critical path of important information that consists of the main inputs, outputs, and
controls is identifiable in the diagram (Whitman, Huff, & Presley, 1997; Kim & Jang,
2002; Mutic et al., 2010).
Control
Input Function Name Output
Mechanism
Figure 2.1. IDEF0 Semantics
2.6 Requirements
A requirement describes something that is needed or desired in a system (or
product or process) that is to be designed and implemented. One of the primary focuses
of this research was to determine the requirements for developing an aid to facilitate
correct diagnosis. Developing requirements helps in understanding the scope of the
system and identifying its features and constraints. A good requirement states something
23
that is necessary, verifiable, and attainable and these attributes could be verified in the
prototype by examination, analysis, test, or demonstration. Requirements should be in the
form of simple sentences to be unambiguous and not become complicated by unnecessary
explanations of operations, design, or other related but unnecessary information. A well-
written requirement only states what the system needs to do and be. Requirements can be
elaborated by using the IDEF0 model or a similar flow chart of the process (Hooks,
1994).
2.6.1 Validation and Verification
Validation is the process of confirming the completeness and correctness of
requirements. Validation ensures that requirements meet certain quality characteristics
and are clear and understood by the developers and that the requirements truly address
the customers’ or users’ needs.
Verification is the process of confirming that the designed and built product fully
meets documented requirements. Among the reasons that might prevent the requirements
to be verifiable are using ambiguous terms and over-specifying. Ambiguous terms are
subjective and could be interpreted differently by different individuals. For example, the
terms “rapid”, “user-friendly”, “sufficient”, and “easy” are examples of ambiguous,
unverifiable terms. Over-specifying is caused because of either stating unnecessary or
overly stringent requirements (Debbabi, Hassaïne, Jarraya, Soeanu, & Alawneh, 2010).
Some of the requirements for developing the Diagnostic Aid must address the
way the information is displayed, and one way of doing that is through information
visualization.
24
2.7 Information Visualization
The systems engineering approach consists of analysis, modeling, requirements
development, design and refinement. Since the requirements are developed for a tablet
computer they use visualization concepts inherently. However, the importance of
information visualization and why they are crucial part of developing the Diagnostic Aid
are explained below.
Visualization, meant as the “The act or process of interpreting in visual terms or
of putting into visible form” (Merriam-Webster’s online dictionary, n.d.), expanded
greatly in recent years due to the advancement of inexpensive technologies. Ruddle,
Brodlie, and Dimitrova (2002) believe that visualization, facilitated by graphical external
representations, is a cognitive activity from which people construct internal mental
representations.
The discipline of Information Visualization (IV) as a branch of the Human-
Computer Interaction field, was introduced in the late 1980’s with basic principal of
deploying computers to generate interactive, visual representation of data and to use
human perceptual abilities for their interpretation. Information Visualization literally is:
“The use of computer-supported, interactive, visual representations of abstract data to
amplify cognition” (Card, Mackinlay, & Shneiderman, 1999). To understand why this
field can be considered helpful in developing a Diagnostic Aid, some terms should be
analyzed.
Visual Representation: There are many situations in the real world where
phenomena are understood better through graphics rather than text. Graphical
representation of data, compared to textual representation, exploits human visual

25
perception to convey large amounts of data to our mind and allow us to make inferences.
Diagrams can be effective in reducing the search required by grouping related
information together, avoiding the need to match symbolic labels by using location to
group information about a single element, and supporting a large number of perceptual
inferences (Larkin & Simon, 1987).
Information Graphics: Information graphics is the use of graphics to present
information in such a way as to make the retrieval of specific information quick and
efficient. Information graphics can present data in a compact, schematic, and efficient
way. It is also useful since it can facilitate the finding of relationships and trends among
data entities (Card et al. 1999).
Cognitive Amplification: Graphics, in this context, uses visual representation to
amplify cognition. Implementation of graphics in certain ways increases the memory and
processing resources available to users, so the user will be able to detect patterns, and
perform inferences more easily.
The amount and complexity of information available to clinical staff are
constantly increasing as a result of information technology advancements. Regrettably,
the amount of information that a user can process at a given instant is very limited due to
cognitive limitations and, therefore, users will not be able to take advantage of these large
amounts of data and might even be overwhelmed by them. The answer to this problem is
through IV that presents information effectively and interactively (Chittaro, 2001). IV has
become a rapidly growing discipline since it is a useful tool in facilitating the way we
present and understand complex datasets. In summary, Card et al. (1999, p16) state six
principles that show how visualization can amplify cognition:

26
1. Increase the memory and processing resources available to the users;
2. Reduce the cost of searching information;
3. Enhance data pattern detection;
4. Enable perceptual inference operations;
5. Use perceptual attention mechanisms for monitoring; and
6. Encode information in a customizable medium.
IV aims at making the user an active element of pattern recognition and can play
an important role in developing most kinds of medical systems. From this viewpoint, IV
components and data mining components can have synergistic roles inside the same
application to achieve several goals in the development of a medical system. These goals
can be organized in three major groups (Chittaro, 2001):
1. Present medical data visually in easy to understand, learn, navigate, and manage
formats.
2. Visually intensify subtle facets of the diagnostic, therapeutic, and healing process
which otherwise could be difficult to notice.
3. Prevent clinical staff to be overwhelmed by huge quantities of information.
The challenge is how to convert abstract data into a graphical representation that
keeps the meaning of the data and also improves the human perception.
2.7.1 Mapping Data to Visual Form
Card et al. (1999, p17) propose a reference model for visualization. Using this
model simplifies the understanding of transformation of data through IV. Figure 2.2
shows the model of mapping data to visual form to the human perceiver. In brief, arrows
27
from left to the right indicate a series of data transformation while arrows from the user to
the left indicate the adjustments to the transformations. This process starts with
transforming raw data into a well-organized data format followed by mapping the data set
into visual structures. The next step embeds this visual form into views by specifying
graphical parameters. Finally, views will be presented to the user through the human
visual system. The user interprets the view to reconstruct the underlying information and
can interact with all previous steps to make changes to the resulting visualization, and
make further interpretations.
Figure 2.2. Reference model for visualization
2.7.2 General Design Principles
A good display, dependent on the creativity of the designer, should consider the
nature of the data, the information to be represented, and its use. Graphics should follow
some basic principles to build effective representations of data. A graphic should show
the data, avoid distorting what the data says, show large data sets coherently in a small
28
space, encourage inferential processes, and give different perspectives on the data – from
broad overview to the fine structure. Graphics facilitate IV, but a number of issues must
be considered (Card et al., 1999; Tufte, 2001):
 Data is nearly always multidimensional, while graphics represented on a
computer screen or on a paper are presented on a 2D dimensional surface;
 Sometimes we need to represent a huge dataset, while the amount of data
representable on a computer screen or on a paper is limited;
 Data may vary over time, while graphics are static;
 Humans, in general, have remarkable abilities to select, manipulate and rearrange
data, so the graphical representations should provide users with these features.
2.7.3 Using Visualization to Navigate Patient Records
Plaisant et al. (1998) provided a visualization environment for personal medical
history named LifeLines that can be applied to medical records. The medical version of
LifeLines is capable of representing patient records, problems, diagnoses, test results, and
medications. A one-screen overview showing multiple facets of the records and filtering
tools allows users to focus on part of the record. In LifeLines, the medical record is
summarized as a set of lines and events on a scalable timeline. Line color and thickness
illustrate relationships or significance of events. LifeLines uses graphical time series that
facilitates making comparisons and relationships between quantities displayed. Highly
interactive interfaces enable the graphed data to show time evolution of quantitative data
and access to whatever data is needed and hence is a great example of cognitive
amplification. Lifelines can also reduce the chances of missing information, facilitate the
29
recognition of abnormalities, and streamline access to details. This paper mentions that
the data encoding is very subjective and a single robust design would be hard to be
agreed on. Overall, LineLines concentrated on the visualization aspects of the record and
not the data entry issues.

30
3 Development of the Diagnostic Aid
The purpose of this study was to design a Diagnostic Aid for the iPad or other
mobile computing platforms that can be helpful in reducing diagnostic errors by reducing
the mental workload of the physician during the diagnosis, by visualizing the patient’s
medical information, and by providing the physician with debiasing tools. This chapter
explains the process of creating the IDEF0 model for the diagnosis process, establishing
requirements for the design of the Diagnostic Aid, and also developing a prototype of the
Diagnostic Aid for the iPad.
The IDEF0 model helps to understand the process of diagnosis in more detail and
also to clarify at which stage each of the influential cognitive biases might interfere
within this process. Requirements worked as a guideline and standard for creating the
Diagnostic Aid. The design of the Diagnostic Aid was based on LifeLines, the IDEF0
model, standards required by the established requirements, design guidelines from
information visualization, and also the suggestions of subject matter experts.
3.1 Model Description

The modeling phase of this research was started by gathering essential
information regarding the diagnostic process to gain a deep understanding of it. Various
sources of information were used for generating the desired model of the diagnostic
process. As discussed in the previous chapter, the literature on approaching medical cases
and medical decision making was reviewed to obtain a comprehensive view of the
31
diagnostic process. Then, subject matter experts evaluated the model and gave their
feedback on the model. The model changed continuously during an approximately
yearlong period of meetings and revisions; however, it still might not reflect the full
insight of the reviewers individually. Physicians take different approaches in medical
decision making based on their backgrounds and experience levels. This variation could
be in the form of decision making and inferential methods or the sequence and repetition
of steps undertaken by the physician. The IDEF0 language easily resolves this potential
problem since the flow of information and box positions are not necessarily in
chronological order.
3.1.1 The A-0 and A0 Diagrams
Tong (2011) developed an IDEF0 model of the whole process of a patient-
physician encounter that was used to outline the activities in this process. Her model has
the top-level A-0 node of “Conduct medical encounter” that has five sub processes
(Figure 3.1). The IDEF0 model of this research is integrated in her model and expands
the A2 node “Diagnose the patient problem”. The A0 diagram (Figure 3.2) demonstrates
the relationship between the diagnostic activity and other activities of the encounter
model, as well as the effective elements. By conducting a single medical encounter (A-0),
the physician diagnoses the patient problem and reaches the final diagnosis or diagnoses
(output O7) based on the information provided by the chief complaint, history of present
illness, signs and symptoms, and physical examination.

32
Medical guidelines
Clinician's factors
Environment/system factors
Patient medical condition
I_Patient's medical history
I_Medical test results
I_Chronological account of illness
I_Patient's medical signs & symptoms
Patient perceived problems
Existing patient-clinician relationship Healing patient

Clinician's initial understanding of patient's problems Updated medical records
New/returning patient Follow-up schedule
Clinician's updated understanding of patient's problem aft
Patient's enhanced understanding of problems
Conduct medical encounter
Ongoing patient-clinician relationship
Final diagnosis(es)
Final treatment plan
Updated patient's medical history
Medical tests ordered
A0
Information System
Medical equipment
Patient
Facilities
Clinician
Figure 3.1. IDEF0 Model – A-0 Diagram: Conduct medical encounter
C4 C9 C3 C2 C1 C6 C7 C8 C5
Patient medical condition Medical guidelines
Patient perceived problems Hypothetical-Deductive strategy
Environment/system factors Pattern Recognition
Clinician's factors I_Medical test results

Existing patient-clinician relationship Ongoing patient-clinician relationship
I1 Establish & O6
Clinician's initial understanding of patient's problems Patient's
Clinician's updated understanding of patient-clinician factors
relationship Clinician's updated understanding of patient's problem after the encounter
I2 maintain O4
patient-clinician
relationship
A1

O10
Physical examination result Updated patient's medical history

O9
Diagnose the patient problem Final diagnosis(es)

O7
Clinician's understanding of final diagnosis
Discussion about the final diagnosis

A2
Patient in treatment Healing patient

O1
New/returning patient Updated medical records
I3 O2
Treat patient Final treatment plan
O8
Clinician's understanding after implementing the treatment.
Discussion about treatment plan
A3
Clinician's understanding with discussed follow-up plan
Plan follow-up Follow-up schedule

O3
O5
Discussion of follow-up plan
A4
Patient Medical equipment

Clinician Facilities Information System
M1 M3 M2 M4 M5
Figure 3.2. IDEF0 Model – A0 Diagram: Conduct medical encounter sub processes
33
3.1.2 The A2 Diagram
The A2 diagram (Figure 3.3) defines the nodes and elements of the diagnostic
process and represents the scope of this research.
C12C10C9 C2 C4 C11C3 C1 C5 C8 C6 C7
I_Patient's medical history The rule of parsimony
I_Patient's medical signs & symptoms Hypothetical-Deductive strategy
I_Chronological account of illness Pattern Recognition
Patient's factors
Clinician's factors
Medical guidelines
Clinician's updated understanding of patient-clinician relationship Evaluate patient's history, medical Context of the problem
I1 signs and symptoms
A21
Diagnostic information need
Physical examination result Generate Updated context of the problem

I2 hypotheses
Initial set of hypotheses
A22

O2
O5
Compare, reject and finalize Clinician's understanding of final diagnosis
hypotheses O4
Final diagnosis(es)
O3
O1
A23
Clinician
Patient
Facilities Medical equipment
M3 M2 M1 M4
Figure 3.3. IDEF0 Model – A2 Diagram: Diagnose the patient problem
The IDEF0 model of the diagnostic process is mostly based on the medical encounter
explained by Sox et al. (2006) and modified by the findings of Toy and Patlan (2012),
other clinical diagnostic literature, and feedback from subject matter experts (Hunink et
al., 2014; Mutic et al., 2010). Differential diagnosis has three steps in this model:
 A21 “Evaluate patient’s history, medical signs, and symptoms”.
 A22 “Generate hypothesis”.
 A23 “Compare, reject and finalize hypothesis”.

34
The elements attributed to each of the activities change progressively according to
the model flow. For example, the state of the input “Clinician’s updated understanding of
patient-clinician relationship”, input 1, changes through the A21, A22, and A23 nodes to
the “Clinician’s understanding of final diagnoses” (output O4). Also, mechanisms and
controls usually affect the activities and inputs, for example, the clinicians, as a
mechanism, and medical guidelines, as a control, have an effect on evaluating patient
medical information.
The mechanisms that play a role in the process of diagnosis are the clinician,
patient, medical equipment, and facilities. The corresponding controls to the patient are
the “Patient’s medical signs and symptoms,” the “Patient’s medical history,” “Medical
test results,” and “Chronological account of illness.” The “Clinician’s factors” that can
affect the process are the clinician’s specialty, knowledge, experience, attitude, fatigue,
personality, etc. The “Patient’s factors” that affect the diagnostic process are age, gender,
anatomy, family history, personality, ability to speak, and so on. Influential factors
related to the environment, “Environment/system factors”, include lighting, temperature,
humidity, equipment used, functional limitations, etc. “Medical guidelines,” which
provide supportive information and diagnostic reasoning methods, are also considered as
effective controls.
3.1.3 A21 node- Evaluate Patient’s History, Medical Sign and Symptoms
In the A21 node, “Evaluate patient’s history, medical signs, and symptoms”, the
clinician interacts with the patient and employs the factors that serve as controls to
evaluate the patient’s condition and to develop the “Context of the problem”. Some of the
35
influential controls such as “Medical guidelines” are very general for many medical cases
and some of them, such as “Chronological account of illness” or the “Patient’s medical
test results”, vary from case to case. Evaluating the patient’s medical information is
shown in more detail in Figure 3.4; the clinician starts by reviewing the patient’s medical
history and lab results, listening to the chief complaint of the patient, and analyzing signs
and symptoms.
C5 C1 C10C6 C8 C9 C7 C4 C2 C3
Patient medical condition I_Patient's medical signs & symptoms
I_Chronological account of illness Anchoring cognitive bias
I_Patient's medical history Representativeness cognitive bias
I_Medical test results Availability cognitive bias
Patient's factors
Medical guidelines
Clinician's factors
Clinician's updated understanding of patient-clinician relationship Clinician's underestanding of patient medical test results
I1
Review patient's medical
history and test results
Clinician's understanding of patient's medical history
A211
Analyze medical signs & Clinician's understanding of patient medical signs & symptoms
symptoms
A212
Listen to patient's illness Clinician's understanding of patient account of illness(HPI)

chronology
A213
Context of the problem

O1
Develop a cognitive representation

of the problem
A214
Clinician
Facilities
Patient
M2 M1 M3
Figure 3.4. IDEF0 Model – A21 Diagram: Evaluate patient’s history, medical signs and
symptoms
The clinician collects all key information about past medications, hospitalizations,
diagnoses, family history, allergies, or any other necessary information. Then, according
to this information and other effective controls, the clinician obtains an understanding of
the patient’s problem; however, it is possible that the clinician’s understanding is

36
influenced by cognitive biases shown in the diagram such as the “anchoring cognitive
bias” that works as a control for the A214 node, “Develop a cognitive representation of
the problem.” The more information the clinician collects at this stage, the less
susceptible he will be to error in decision making.
3.1.4 A22 node- Generate Hypotheses
In the A22 node, ”Generate Hypothesis” which is broken down into its sub-
activities in Figure 3.5, the clinician comes to an initial set of hypotheses by integrating
his understanding of the patient’s problem derived from medical information and by
conducting a physical examination. After developing some potential hypotheses, the
physician conducts the physical examination to find more about the patient’s condition
and verify, refute, and prioritize his list of hypotheses. Conducting physical examination,
A222 node, is complex, varied, and influenced by the potential diagnosis in the mind of
the clinician. However, the physical examination is not necessarily performed at this
stage and might be executed sooner or later in the diagnostic process based on the
clinician’s experience and findings. After developing some hypotheses, the clinician, if
not convinced yet by his findings, might request additional information through
laboratory tests or other resources. The decisions made at this level are also affected by
cognitive biases. Anchoring, representativeness, and availability are cognitive biases that
affect the clinician in generating hypothesis, and the clinician might display confirmation
bias in defining additional required information.

37
C3 C5 C4 C2 C1
Anchoring cognitive bias Environment/system factors Confirmation cognitive bias
Availability cognitive bias
Representativeness cognitive bias
Patient's factors
Clinician's factors
Context of the problem Generate hypotheses based Initial generated hypotheses list
I1 on aggregated knowledge
A221
Clinician's understanding of the patient's problem

Conduct physical
examination
Physical examination result Updated context of the problem
I2 O2
A222
Prioritize list of initial Initial set of hypotheses

hypotheses O3
A223
Identify additional
information needed for Diagnostic information need
O1
consistency check
A224
Clinician
Facilities
M2 M1 M3 M4
Figure 3.5. IDEF0 Model – A22 Diagram: Generate hypotheses
3.1.5 A23 node- Compare, Reject, and Finalize Hypothesis
The diagnostic process then continues with node A23, “Compare, reject, and
finalize hypothesis” as the final stage of differential diagnosis (Figure 3.6). It starts with
strategies (hypothetical-deductive and pattern recognition) to refute the hypotheses that
are not consistent with the context of the problem. Node A231 compares the signs and
outcomes of the potential hypothesis with the patient’s medical condition. If the
outcomes are far from the current state of the problem then the hypothesis will probably
be rejected from the list. If more than one hypothesis remains, then the next step is to
38
choose one or more hypotheses that best match the patient’s medical condition. Those
hypotheses that have the same treatment can be categorized in one group at this stage.
After the rejection of unlikely hypotheses, the clinician prioritizes the remaining
hypotheses and gathers additional required information. At this stage, the clinician might
ask for further medical lab tests, or even conduct a physical examination if he cannot
decide confidently from what he has in hand. After collecting all key clinical information
the clinician will probably come to the final diagnosis or diagnoses by doing a pairwise
comparison among the remaining hypotheses and rejecting the less likely one according
to the rule of parsimony. Confirmation bias plays an important role in the final decision
of the clinician at this level. Premature closure is also prevalent at this level if the
clinician is not likely to consider all relevant and necessary information before finalizing
his decision. An “Updated patient medical history,” “Final diagnoses,” a “Discussion
about the final diagnosis,” and “Clinician’s understanding of final diagnoses” are outputs
of the process. The entire IDEF0 model of the diagnosis process is represented in
Appendix 1.
39
C10C6 I1 C1 C5 C4 C2 C9 C3 C7 C8
Pattern Recognition Premature closure error I_Medical test results The rule of parsimony
Medical guidelines
Updated context of the problem

Patient's factors
Confirmation
Clinician's factors cognitive bias
Hypothetical-Deductive strategy
Initial set of hypotheses Selected hypotheses that match patient's condition

I2
Refute hypotheses not consistent with

context of the problem
Physical examination result

I3
A231
Additional information needed
Prioritize hypotheses and identify

additional information needed
Prioritized list of hypotheses
A232

O5
Order required medical tests Updated Context of The Problem
Remained hypotheses
A233
Compare hypotheses one by one Final hypotheses

and refute the unlikely ones
A234
Explain the O1
Final diagnosis(es)
patient's problem O4
with final O2
hypothesis(es) Clinician's understanding of final diagnosis
O3
Plan a follow up visit
A235
Clinician
Patient
Facilities
Medical equipment
M1 M4 M2 M3
Figure 3.6. IDEF0 Model – A22 Diagram: Compare, reject, and finalize hypotheses
3.2 Developing Requirements
For this research, requirements were produced based on the IDEF0 model, clinical
and information visualization literature, user-centered design principles, previous work
on a comprehensive physician patient encounter model, and brainstorming. From the
activities indicated in the IDEF0 model, the user’s needs are addressed and requirements
to better serve the user are defined. Cognitive errors that might happen during the
diagnostic process are also addressed in the IDEF0 model as controls, so analysis of
errors during the requirement development aided in prevention of errors and
enhancement of the Diagnostic Aid performance.

40
Requirements provide guidance with regard to the interface features and how the
Diagnostic Aid’s screens look, features and functionalities of the Diagnostic Aid, and
memory tools that aim to prevent cognitive biases. Requirements should explain all the
functions of the diagnostic process in an easily understood way, describing exactly what
each aspect of the Diagnostic Aid should do.
A few key requirements are described in the following sections: user interface,
functional, and debiasing tools. A full list of the requirements, their categories, their
specific action area, the source from which they were derived, and whether they were met
is presented in Appendix 2.
User interface requirements represent a display or control feature of the
Diagnostic Aid, while functional ones relate to what the Diagnostic Aid does. For
instance, “the Diagnostic Aid shall present patient’s vital signs and allergies” is an
example of a user interface requirement, while “the Diagnostic Aid shall provide means
to share the patient’s problem with remote specialists” is a functional requirement. The
user interface category contains a wide variety of requirements that support the entire
Diagnostic Aid design, for example, “the Diagnostic Aid shall present the patient's
personal information on all pages”. The last category, debiasing tools, is applied to
indicate the most appropriate ways of conveying required information to reduce the
effects of prevalent cognitive biases in the diagnostic process. Requirements were
iteratively refined and verified during the modeling and design process since one of the
methods of requirements verification is through implementing them in the design.

41
3.2.1 Functional Requirements
Functional requirements refer to the capabilities of the Diagnostic Aid regarding
required medical actions. For example, that the clinician should be able to easily move
among different pages of the Diagnostic Aid to reduce information access cost is a
general requirement. Some authors claim that information availability should be
customizable according to the user’s need. Physicians as subject matter experts consulted
in the research also believed that records should be filtered to remove uninteresting
results. Moreover, adding the ability of entering data to improve the overall functionality
of the system was proposed. To streamline documentation, implementing tools such as
voice recognition have also been suggested (Schiff & Bates, 2010). Considering these
findings, the user of the Diagnostic Aid should be able to take notes at all stages of the
diagnosis. This is consistent with A214 node, “Develop a cognitive representation of the
problem”; and A235 node, “Explain the patient’s problem with final hypothesis.” The
clinician should also be capable of highlighting especially significant information and
filtering out negligible information. Conducting the physical examination, node A222, is
essential in nearly every instance of the diagnostic process. The Diagnostic Aid must
allow the clinician to store the information and findings of the physical examination.
Having access to medical journal articles, diagnostic cues, the latest possible
treatments and so on in a short amount of time can dramatically enhance the quality of
health services (Wachter, 2010; Schiff & Bates, 2010). Besides providing access to
information sources, offering a second opinion or consultation that the clinician can take
full advantage of are useful tools that should be considered. Graber (2008) suggested
providing physicians with access to the Internet, electronic medical reference texts, and
42
journals as potential sources to improve clinical decision making. He also encouraged
providing a second opinion and enhanced feedback. Therefore, among the Diagnostic Aid
requirements is to provide means to share the patient’s problem with remote specialists
for a second opinion. The A22 node in the IDEF0 model requires the clinician to generate
a list of hypotheses and prioritize them according to his findings. The clinician should be
able to cover these criteria and also should have access to search engines and electronic
medical references such as PubMed.com and Uptodate.com. The list of functional
requirements includes:
1. The Diagnostic Aid shall provide means to take record of clinician's final
understanding of the problem.
2. The Diagnostic Aid shall provide means to share the patient’s problem with
remote specialists.
3. The Diagnostic Aid should provide feedback of the clinician final diagnosis(es).
4. The Diagnostic Aid shall provide means to access electronic medical references.
5. The Diagnostic Aid shall provide means to list the clinician’s hypotheses.
6. The clinician shall be able to take notes of his physical examination findings.
7. The clinician shall be able to add to and remove from the hypotheses list and
prioritize hypotheses.
8. The Diagnostic Aid shall allow the clinician to take and save notes at all of the
stages of the diagnostic process.
9. The Diagnostic Aid shall provide means for the user to highlight especially
significant information.
43
10. The Diagnostic Aid shall provide means to filter out all but the highlighted
information.
11. The Diagnostic Aid shall provide means to move among different pages easily.
3.2.2 User Interface Requirements
According to (Haines, n.d.) in developing a Tracking Board Prototype, the
patient’s medical information should be available on all pages of a multipage Diagnostic
Aid to reduce information access cost. She also suggested locating related patient medical
information close together.
Schiff and Bates (2010) investigated the effects of electronic medical records
(EMR) on quality of care and believed that electronic documentation is a key to this
effort and suggested a rational approach in designing documentation databases. Some of
their findings and recommendations were used to develop requirements for the
Diagnostic Aid. One of their suggestions that was implemented is some means be
included to document clinicians’ thoughts rather than just using usual point-and-click
methods. Due to human memory limitations, providing a practical checklist that calls up
important medical questions, diagnoses, and diseases probabilities was proposed. From
these considerations and the IDEF0 A21 node, “Evaluate patient’s history, medical signs
and symptoms,” three requirements were developed.
 The patient’s medical information shall be readily accessible from all pages.
 The Diagnostic Aid shall present the history of present illness, history of
hospitalizations and previous illnesses, family history, and medications patient is
taking, in chronological order.

44
 The Diagnostic Aid shall present related pieces of information close together to
reduce information access cost.
Displaying the patient’s vital signs was also a requirement developed according to
Schiff and Bates’ recommendations and the A212 node in the IDEF0 model. In the A231
node, the clinician compares the expected symptoms of each potential hypothesis with
the patient’s condition and rejects the unlikely hypothesis if any discrepancy arises.
Therefore, the Diagnostic Aid must present all symptoms associated with each diagnosis
in the hypotheses list. Coding should be employed to make more important aspects of a
patient’s information salient.
Sawyer et al. (1996) investigated the impact of design upon effective use of
medical devices, and discussed design-induced errors and mistakes in the use of such
devices. According to them, in designing the user interface, accepted symbols, colors,
and icons should be used to facilitate the comprehension of information. In addition,
labels should be unambiguous, corresponding to the conventions of the user population,
so they can easily be read and understood. Some design principles were derived from
rules of thumb and are spread over different pages of the Diagnostic Aid. For example,
“Information shall be represented consistently across the Diagnostic Aid,” or “The
Diagnostic Aid shall help the clinician keep track of the diagnostic process by showing
the current step in the process.” A full list of user interface requirements includes:
1. The Diagnostic Aid design shall be consistent with the way the physician
conducts diagnosis.
2. The Diagnostic Aid shall present the patient's personal information on all pages.
3. The patient’s medical information shall be readily accessible from all pages.
45
4. The Diagnostic Aid shall present the history of present illness, history of
hospitalizations and previous illnesses, family history, and medications patient is
taking, in chronological order.
5. The Diagnostic Aid shall present the patient’s vital signs and allergies.
6. The Diagnostic Aid shall present all symptoms associated with each diagnosis in
the hypothesis list.
7. The Diagnostic Aid shall present a list of suggested diagnoses after the clinician
has reviewed the patient's data.
8. The Diagnostic Aid shall utilize a standard format for every page.
9. The Diagnostic Aid shall use coding corresponding to established conventions.
10. The Diagnostic Aid shall help the clinician keep track of the diagnostic process by
showing the current step in the process.
11. The Diagnostic Aid shall present related pieces of information close together to
reduce information access cost.
12. The Diagnostic Aid shall utilize accepted symbols, icons, colors and
abbreviations.
13. The Diagnostic Aid shall provide unambiguous labels for objects so they can be
easily read (symbol size, contrast, color).
14. The Diagnostic Aid shall present diagnostically significant relations and trends
among data.
15. The Diagnostic Aid shall minimize the cost of retrieving information.
16. Information shall be represented consistently across the Diagnostic Aid.
17. Symbols should be easily recognizable by the operator.

46
18. The salience of cues should be consistent with their value.
19. The Diagnostic Aid shall make the user capable of processing different cues
equally.
20. The Diagnostic Aid should use constraints to prevent possible errors.
3.2.3 Debiasing Tools
In the diagnostic process, the clinician generates a list of potential hypotheses
based on the patient’s condition, refutes the unlikely ones by gathering clinical
information, and eventually forms the final diagnosis. Due to the complexity of diagnosis
and the amount of mental workload involved, human errors inevitably occur. The
Diagnostic Aid is to promote a systematic decision process while improving the quality
of decisions by reducing the prevalent cognitive diagnostic biases. As mentioned earlier,
prevalent cognitive errors in clinical diagnosis are anchoring, availability,
representativeness, confirmation, and premature closure. It is noteworthy that these
cognitive biases affect the clinician in different stages of the decision making. As shown
in the IDEF0 A214 node, “Develop a cognitive representation of the problem,” and A221
node, “Generate hypothesis based on aggregated knowledge,” the availability, anchoring,
and representativeness biases are influential in forming an understanding of the problem
and generating hypotheses and are therefore shown as controls. Confirmation bias affects
the clinician at the A224 node, “Identify additional information for consistency check,”
and A233 node, “Order required medical test.” Confirmation bias occurs because there is
a tendency to seek confirming information rather than to find disconfirming data to test
the competing hypothesis. In addition, ordering additional laboratory tests may increase
47
the clinician’s confidence in his decision, even though additional sets of data do not alter
the probability of the diagnosis. Anchoring bias, besides in hypothesis generation, plays a
role in finalizing the hypotheses, so it is shown as a control in the A232 node, “Prioritize
hypotheses and identify additional information needed,” and A234 node, “Compare the
hypotheses one by one and refute the unlikely ones.” Finally, premature closure is a
general heuristic, which mostly prevents the clinician from collecting additional
information in the final stages of the diagnosis. Thus, it acts as a control in the A232
node, which involves identifying additional information needed.
The idea of the Diagnostic Aid is to call cognitive biases to the attention of the
clinician, so that the clinician’s awareness of potential pitfalls in information processing
prevent him from making inaccurate decisions. The Diagnostic Aid should provide
debiasing tools to reduce the likelihood of cognitive biases. Requirements in this part can
be categorized into two general and specific groups. Examples of general requirements
follow:
 The Diagnostic Aid shall provide means (cognitive debiasing tools) to reduce the
likelihood that cognitive biases interfere with diagnosis.
 Debiasing tools shall not interfere with the process of decision making.
 Debiasing tools shall be specifically attributed to different stages of diagnostic
reasoning.
On the other hand, there are requirements that are individually ascribed to specific
cognitive biases considering the debiasing methods mentioned before. For example, for
anchoring bias, the Diagnostic Aid should provide means to encourage the clinician to
review all the patient’s information before making a final decision. For preventing the
48
availability bias, the debiasing tool should encourage the clinician to consider the history
of recent diagnoses. The representativeness bias usually happens when the patient’s
condition is atypical of a diagnosis, so to prevent this tendency, the clinician must search
for inconsistencies between the patient’s symptoms and the potential diagnosis.
Informing the clinician of the probability of a diagnosis might also be helpful in raising
awareness of the representativeness effects. Looking for disconfirming data, being alert
to the absence of supporting evidence, and early switching between hypotheses are
potential countermeasures of confirmation bias that should be supported by the
Diagnostic Aid. In general, premature closure happens when the clinician is in a rush and
avoids collecting all key clinical information. To prevent the harmful effects of this
tendency, debiasing tools should encourage the clinician to gather more data before
making a final decision. A full list of debiasing tools requirements is presented below.
1. The Diagnostic Aid shall provide means (cognitive debiasing tools) to reduce the
likelihood that cognitive biases interfere with diagnosis.
2. Design of debiasing tools shall be consistent with the rest of the Diagnostic Aid
standards.
3. Subtle but clear visible feedback shall be implemented for debiasing tools.
4. Debiasing tools shall not increase the user’s cognitive burden.
5. Appropriate labels shall be used to convey the functionality of the memory tools
quickly.
6. Debiasing tools shall not interfere with the process of decision making.
7. Debiasing tools shall be specifically attributed to different stages of diagnostic
reasoning.
49
8. Debiasing tools shall help the clinician to reach the correct diagnosis more
accurately.
9. Debiasing tools shall help the clinician to reach the correct diagnosis more
quickly.
10. The anchoring debiasing tool shall encourage the clinician to review all the
patient’s information before making a decision.
11. The availability debiasing tool shall encourage the clinician to consider the
history of recent diagnoses.
12. The availability debiasing tool shall encourage the clinician to consider
hypotheses with higher absolute probability rather than those easily brought to
mind.
13. The representativeness debiasing tool shall encourage the clinician to search for
inconsistencies between patient’s symptoms and the potential diagnosis.
14. The Diagnostic Aid shall inform the clinician about the probability of a potential
hypothesis.
15. The confirmation bias debiasing tool shall encourage the clinician to look for
more data that provide disconfirming evidence.
16. The Diagnostic Aid should be able to notify the clinician of absence of crucial
supporting evidence to prevent confirmation bias.
17. The Diagnostic Aid shall make the user capable of switching among hypotheses
with minimal mental effort.
18. The debiasing tool shall reduce the likelihood of physician being subject to
cognitive biases.
50
19. The Diagnostic Aid shall encourage the clinician to search for more evidence to
decrease the chance of overconfidence bias.
20. The Diagnostic Aid shall encourage the clinician to collect more data and
consider more hypotheses before making final decision.
3.3 Design
The objective of this part of the research was to develop a Diagnostic Aid to
facilitate the diagnostic process. The Diagnostic Aid benefits from visualized information
to assist the clinician to find and analyze information more easily due to increasing the
unused memory resources. One strategy to minimize the effect of cognitive biases in
diagnostic reasoning is to call these processes to the attention of clinicians, therefore,
clinicians might be more willing to use tools to help to overcome these tendencies.
However, it is unreasonable to expect every clinician to become a decision analyst, but it
is not unrealistic to ask clinicians to revise a diagnosis by considering the clinical story in
reverse order. Since heuristics and therefore biases are integral components of human
information processing, efforts in this domain have not been very successful so far.
Nevertheless, more thoughtful consideration of the limitation of information processing
will enhance the quality of decision making (Elstein, 1999).
3.3.1 Diagnostic Aid Development
The IDEF0 model defined the general outline of the Diagnostic Aid and
established requirements, and subject matters experts’ thoughts guided the design of it.
The Diagnostic Aid that was designed, for example, has three different main screens
51
based on three sub-processes of the diagnosis, A21, A22, and A23 in the IDEF0 model.
The design process mainly followed the requirements and human factors principals, so
that each functional, interface design, or debiasing feature of the Diagnostic Aid, was
sourced from a requirement. Then, the design was modified several times after being
reviewed by subject matter experts at different levels. It is notable that, besides the
mentioned factors, designing a new thing is always a matter of creativity and innovation.
Therefore, having the same IDEF0 model and requirements, two different designers
would definitely develop two graphically different presentations. The Diagnostic Aid
developed for this project is designed for the iPad, and Keynotopia as a rapid prototyping
application was utilized to develop the Diagnostic Aid prototype.
3.3.1.1 Medical History Screen
To better explain the envisioned, fully functional Diagnostic Aid and its
characteristics, a partly functional prototype was developed based on the diagnosis
process for Sarcoidosis (a chronic disease of unknown cause that affects lungs, lymph
nodes, liver, or other tissues). The three main screens of the Diagnostic Aid are medical
history, hypothesis generation, and final diagnosis. The medical history screen presents
demographic information such as the patient’s name, date of birth, sex, race and
ethnicity. This information is also available from other screens of the prototype. To fulfill
the requirements regarding clinical information, the Diagnostic Aid presents the
following information:
 The reason for admission for care as the chief complaint.
 The patient’s vital signs and other information regarding patient’s background.
52
 The history of prevalent illnesses in the patient’s family.
 Any allergies to food or medications.
 The patient’s initial diagnostic impressions and conditions as symptoms.
 Physical examination results.
 Any diagnoses or consultations reports.
 Results of diagnostic tests.
 Hospitalizations, treatment and services provided to the patient before his arrival.
 Medication administered, including the dose and frequency.
The Patient’s medical history is represented using a graphical time scale, which
has proven to be advantageous to data interpretation (Plaisant et al., 1998). As shown in
Figure 3.7, the medical records are displayed as a set of lines and events on a scalable
data span of 12 months. Lines display varying status conditions such as symptoms, while
icons indicate discrete events, such as laboratory test results. The time scale clarifies the
relation among quantities displayed. Line color and thickness illustrate the severity of
events. The icons’ location record their occurrence in the time frame. A quick glance at
the medical history of this patient reveals that this patient has suffered from a
nonproductive cough for 4 months. She also suffers from dyspnea (shortness of breath),
fatigue and malaise (generalized feeling of discomfort), for a few months, while her
fatigue and malaise has been more intense during the last month. She visited an
ophthalmologist about six months ago; by clicking on the icon of that visit a detailed
information window appears on the right hand side containing the reason of the visit and
the diagnosis. Vital signs measures are demonstrated by making the normal ranges clear,
so if an indication is out of normal range, it would be recognized easily.

53
Figure 3.7. Patient’s medical history screen
As users touch an event on the screen, a window with detailed information pops
up in the right margin of the screen. In this case, it seems that bumps (a swelling on the
skin) appeared on her skin about 2 weeks prior to this visit. One can glance down from
the bumps line to the medication section and see that the appearances of bumps were
soon after she changed her contraceptive pills. So, the user does not need to remember
the duration and starting point of events, making it easier to detect concurrent events.
Connecting to external resources is also possible from this page. The physician can also
54
connect to remote specialists and ask for their opinions through this feature. Finally, the
user can take notes on any findings that he wants and save them. Ideally the physician
will switch to the second page of the Diagnostic Aid, hypothesis generation, after
reviewing all the essential information of the patient in the medical history screen. All
screens for a single scenario are provided in Appendix 3.
3.3.1.2 Hypothesis generation screen
After collecting the patient’s medical information and listening to the illness
chronology, the clinician would be able to provide a list of possible causes while
prioritizing them in his head. The clinician is also able to highlight selected information
that he believes plays an important role in diagnosing the disease. However, he can have
access to full medical information by going back to the first screen. The main difference
between the first two stages of diagnosis, as represented in the Diagnostic Aid, is the
effect of cognitive biases in the hypothesis generation phase. As is seen in the debiasing
tools section, there are three questions that address anchoring, availability, and
representativeness heuristics, respectively (Figure 3.8). The idea behind these questions is
to remind the user of pitfalls related to cognitive biases. Addressing the anchoring bias by
asking the user “Is your first diagnosis your final diagnosis? If so, what is your next most
probable diagnosis?” makes the user think of how considering the first hypothesis as the
final one involves the tendency of relying too heavily on the initial piece of information
and first inferential hypothesis. So, informing the clinicians that anchoring might
contaminate their decisions hopefully makes them adjust their decision making process.
The second question, “Have you considered this diagnosis more than one time
55
recently? If so, reconsider and review other possible diagnosis”, tries to call the attention
to the availability bias. This heuristics occurs when something is recalled very easily then
considered disproportionately important. In this context, if the clinician diagnoses a
problem for the second time in a short period of time, it might be because of recalling it
easily and considering it important. If the clinician were cautious that considering a
diagnosis even more than one time might be due to the availability heuristic, he might
reevaluate his way of thinking and analyzing the patient’s information.
Figure 3.8. Hypothesis generation screen
The third question, “Could this be an atypical manifestation of another condition?
If so, consider another diagnosis”, addresses the potential threat of representativeness

56
bias in generating hypothesis. The Representativeness debiasing tool tries to alert the
clinician that when a patient’s condition reflect the salient characteristics of a group of
illnesses does not necessarily mean that the problem is definitely belongs to that group.
3.3.1.3 Final diagnosis screen
After providing a list of potential hypotheses by the clinician, it is time to finalize
the case by coming to a final diagnosis or diagnoses. The last screen looks very much the
same as the hypothesis generation screen except that the debiasing tools address
confirmation bias and the premature closure error, and the presentation of all symptoms
associated with each diagnosis (Figure 3.9). The Diagnostic Aid also provides a list of
hypotheses that the clinician has not considered yet according to the patient’s medical
information. This, however, requires the Diagnostic Aid to either be connected to clinical
databases or has these types of information in itself. Providing a list of unconsidered
hypotheses and associated symptoms for both considered and unconsidered hypotheses
would be a valuable tool for the clinician in the hypothetical deductive strategy. The
clinician could easily compare the patient condition with each hypothesis outcome and
refute from or keep the hypothesis in the list.
The first question in the debiasing tools part, “Have you looked for disconfirming
information? If no, try a disconfirming test?”, tries to suppress the effects of confirmation
bias. Confirmation bias is a tendency to favor information that confirms one’s hypothesis
and in the clinical context would mostly appear as trying to gather and remember
information selectively. In an attempt to confirm the initial hypothesis, this debiasing tool
aims to persuade the clinician to look for additional information, either through physical
57
examination, laboratory test, or any other form that rejects the most anticipated
hypothesis.
Figure 3.9. Final diagnosis screen
The final debiasing question, “What else can it be? What is your next priority?” is
designed to avoid a premature closure error. Premature closure is very common in
differential diagnosis and happens when the clinician fails to consider reasonable
alternatives after an initial diagnosis is made. Premature closure can also be considered as
a type of anchoring heuristic error. The debiasing question tries to make the clinician take
time to go through differential diagnosis and consider other alternatives.

58
The section that provides the hypothesis list and associated symptoms is intended
to be helpful in preventing all possible sorts of biases that might lead to premature
closure. A symptom for a hypothesis that is not seen in the patient might act as powerful
disconfirming evidence and providing a list of unconsidered hypotheses can be a good
hint for preventing premature closure.
As mentioned before, the Diagnostic Aid prototype was developed with a
prototyping tool for the Apple iPad iOS operating systems. The prototype Diagnostic Aid
functionalities are limited and it does not have all the capabilities that were listed in the
requirements list. However, the Diagnostic Aid is partially operative and adequate for
testing in a controlled experiment for its major features.

59
4 Test and Results
In line with the general purpose of this research to develop a model of the
diagnosis and to introduce a Diagnostic Aid tool to reduce physician-related diagnostic
errors, this chapter describes the process used to test the effectiveness, evaluate the
usability and general fit of the Diagnostic Aid, and verify whether the requirements were
met or not. The terms “system” and “Diagnostic Aid” are used interchangeably in this
chapter.
4.1 Participants
The sample of physicians as participants for this study was selected from among
Oregon clinicians and residents that have been involved in the process of diagnosis in
their careers. Clinicians were contacted after Institutional Review Board (IRB) approval
and finally six clinicians were recruited for this experiment through a chain referral
process. The research protocol submitted to the IRB and the informed consent signed by
the participants are shown in Appendices 4 and 5 respectively. The sample recruited for
this study was relatively small and not a perfect representative of all practicing physicians
nationally. All participants can be categorized in the experienced class of physicians with
at least ten years of being professionally active. The design and findings of the study
might direct future similar studies. The self reported demographics of physicians are
shown in Table 4.1.

60
Table 4.1. Demographic characteristics of the participants
Job Title #
Family Nurse Practitioner 1
Family Medicine Physician 1
Physician (MD) 2
Pediatric Physician 1
Retired Obstetrician/Gynecologist 1
Work Experience
10 to 20 years 2
20 years or more 4
Involved in Differential Diagnosis on a Daily Basis
Yes 6
No 0
Age
46-55 4
56 or older 2
4.2 Apparatus
Equipment used during data collection are explained in this section.
Clinical Scenarios: Three different but equivalent fake medical scenarios were
adapted from standard medical cases for medical exams and developed for this study. All
of these scenarios were about the same length, had the chief complaint of fatigue in
common, covered the same medical background information of the patients, and had
similar lab tests results reported. A short version of one of the clinical scenarios includes:
A 36-year-old woman presents to her physician’s office due to 4 months
of nonproductive cough and fatigue. Her medical record shows that she
was referred to an ophthalmologist because of blurred vision six month ago
which was diagnosed as anterior uveitis. Her father died with the diagnosis
of small cell lung cancer at the age of 67 years, and her mother died at the
61
age of 61 due to myocardial infarction. Vital signs showed T: 99 F, and
blood pressure of 142/87 mmHg. She has tender red papules over her shins.
EKG reveals occasional PR interval prolongation. Lung examination
revealed diffusely decreased lung sound. Laboratory findings are:
WBC count: 5100/mm3
Hemoglobin: 13.7 mg/dL
PPD test showed induration of 3 mm after 72 hours. Culture of
bronchoalveolar lavage fluid was negative.
These scenarios were represented both visually as in the Diagnostic Aid form and
also traditionally as in scripts. One of the scenarios was only used as a training scenario
to familiarize participant with the Diagnostic Aid. The other two scenario scripts used for
testing are presented in Appendix 6.
iPad/visualized scenario: Keynote was used to develop the clinical scenarios in a
visualized format for the iPad. How to work through the scenarios on the iPad was
explained to the participants.
Questionnaire: The questionnaire consisted of an introduction and four different
sections (Appendix 7). The first section, technical questions, consisted of mostly open-
ended ones and asked about the details of the scenario and the participant’s decision
about the problem of the patient. For example, “What is/are the chief complaints of the
patient?” or “What is/are your final diagnosis(es)?” are among the questions in this
section. The next two sections assessed the participant’s level of agreement regarding
usability and the extent to which the requirements were met in the design. The usability
questions, in the 1-7 Likert scale, were designed in order to provide a measure of the
62
system’s ease of use according to the score gained. Some of the usability questions were
modified versions of those of the System Usability Scale (SUS) method and some of
them were developed specifically according to the design of the Diagnostic Aid (Brooke,
1996). Examples of these questions are:
 The application is easy to use generally.
Strongly Disagree 1 2 3 4 5 6 7 Strongly Agree
 The system is too inconsistent.
 The presentation of patient vital signs is effective.
The system attribute questions, also in the Likert scale, were developed to validate
the requirements in the design of the Diagnostic Aid. Participants could also clarify their
opinions by writing comments. Some of these questions were:
 The debiasing tools reduced the likelihood of me being subject to cognitive
biases.
 It is NOT easy to retrieve the desired information by using this application.
 The application effectively encouraged me to collect more data and consider more
hypotheses before making my final decision.

63
A few questions from this part were eliminated from the analysis part since they
were very close to questions in the usability sections.
The last section was a short demographic survey that asked about participants’ title, age,
and working experience.
Interview: A qualitative interview was conducted with each participant. The
interview questions were designed to derive more detailed information regarding the user
experience with the system. The questions addressed what the participants liked or
disliked about the Diagnostic Aid, assessed whether or not the tools and features of the
Diagnostic Aid were helpful, and evaluated the satisfaction with the product as an
everyday tool. The interview guide included the following questions:
 Would you like to use this system every day if it were fully functional
 What do you like most and least about the system?
 What functionalities of the system were useful? What were not? What should be
added to it? What should be removed?
 What is your recommendation regarding future development of this application?
 How did the debiasing tools affect your decision making process?
 Were the debiasing tools helpful in any way? How they could be improved?
4.3 Procedure
First, the objectives of the study were explained to each participant. Then, as a
preparation for data collection, the researcher introduced the Diagnostic Aid and taught
the participant to use the device to review the test clinical scenario and to become
familiar with working with the Diagnostic Aid. Thereafter, the questions regarding the
64
use of the Diagnostic Aid were answered. After the participants reviewed the test
scenario by using the Diagnostic Aid for 10-15 minutes, they worked on one of the
scenarios designed for the actual test (A or B). After reviewing the patient’s information
and going over different pages of the Diagnostic Aid, the subject answered the technical
questions and also the usability and system attributes questions in the questionnaire.
Next, another scenario presented in script format was provided to the participant. The
participant worked on this scenario for about 10 minutes and then answered similar
technical questions designed for the new scenario (excluding the usability and
requirements development questions). Finally, in a semi formal interview the participant
was asked questions about his or her experience with the Diagnostic Aid. The second
group of participants did the script format scenario first, and the visualized one
afterwards, but the rest of the experimental procedure was as described above.
The design of this study was a two-factor cross-over design. To counterbalance
the learning effect, participants were randomly assigned to two treatment groups:
1. First working on the Diagnostic Aid scenario and then working on the script
scenario.
2. First working on the script scenario and then working on the Diagnostic Aid
scenario.
In this study, two different clinical scenarios were required during each setting of
the study to reduce the learning effect. The alternation between the scenarios and their
presentation method was intended to reduce the effect of practice, fatigue, or any other
confounding variables.
65
4.4 Results
The results for different sections of the study including technical questions,
usability questions, and system attribute questions, and also interview analysis is
represented in this subsection. The analysis of the results and corresponding conclusions
are presented in the following chapters.
4.4.1 Technical Questions Results
The technical questions were either open ended or multiple choice questions
about certain facts in the clinical scenarios. This section investigated whether the
visualized format of clinical scenarios helped the participants to better memorize and
recall the patient’s medical information and also to better finalize their decision regarding
the patient’s problem. The responses’ correctness categorized by the paper/Diagnostic
Aid format for each question and each participants are shown in Table 4.2 and 4.3.
66
Table 4.2. Results of technical questions for each question
# of
# of Paper
Diagnostic
Questions Incorrect
Aid Incorrect
Answers
Answers
1 What is/are the chief complaint(s) of the patient? 0 0
Which vital sign(s) is/are not in the normal range, if
2 1 2
any?
Which of these conditions are not among the
3 1 2
patient’s medical symptoms?
How many months did swollen feet /nonproductive
4 1 3
cough last?
What does the physical exam reveal about the
5 0 0
patient?
6 What is the WBC count? 3 3
Are there any abnormalities in the patient’s lab
7 4 2
results?
8 What medication(s) is/are the patient taking? 0 0
9 Which symptom(s) started soonest? 3 0
10 Which symptoms are most severe? 2
11 What is/are your final diagnosis(es)? 1 0
≅ 2.3 (out of ≅ 2.1 (out of
Average
10 questions) 11 questions)
Most of these questions were only testing whether the Diagnostic Aid helped the
participant to memorize and recall the information more accurately while a few others,
such as questions 7 and 11, required some knowledge in a specific medical area as well.
Question 10 was only measurable through the Diagnostic Aid and was not counted
toward paper results.

67
Table 4.3. Results of technical questions for each participant
Paper Incorrect Diagnostic Aid

Participants
Answers Incorrect Answers
1 4 3
2 1 2
3 2 3
4 2 1
5 2 2
6 3 3
The participants performed equally well on the technical questions, therefore
neither of the presentations showed any advantage over the other. The average number of
wrong answers for paper format was 2.3 questions versus 2.1 for the Diagnostic Aid. The
question — which symptoms are most severe? — was only included in the Diagnostic
Aid question set since it could only be perceived through the Diagnostic Aid’s graphs and
it could not be highlighted in the paper presentation. Since the total number of questions
in the two presentations were different and sometimes the participants did not answer the
questions appropriately, such as choosing more than one answer in a multiple choice
question, the participants’ scores are an estimation of the incorrect answers identified.
4.4.2 Usability Questions Results
The usability questions mainly evaluated the participant’s level of agreement
regarding the usability of the system and the effective presentation of the patient’s
medical information. Questions 1-6 and 12 were related to the user’s general perception
of the system and the rest of the questions in this section focused more on the quality of
data presentation and the ease of navigation through the system. The first set of questions
was combined into a single composite score during the data analysis (Likert scale data),
68
while the rest of the questions were considered as single questions and were analyzed
individually (Likert-type). For the Likert scale data (Questions 1-6 and 12), the mean and
standard deviation for each participant was reported (Table 4.4). On the other hand, for
the other questions the mean and standard deviation for each item over all the participants
were considered for analysis (Table 4.5). The range of the possible responses was from 1
to 7, where 1 indicated strong disagreement while 7 indicated strongly agreement toward
the usability of the Diagnostic Aid.
Table 4.4. Results of usability questions 1-6 and 12
Questions 1-6 ,12 Questions 1-6 ,12

Participant
Mean Standard Deviation
1 6 0.82
2 5 0.82
3 6.7 0.49
4 4.4 0.53
5 6.3 0.49
6 5.9 0.9
Total 5.7 1.02
Almost all participants found the system usable and easy to work with (mean
response >5); however, three participants (mean response >=6) had a higher degree of
agreement than the other two.

69
Table 4.5. Results of usability questions 7-11 and 13
Standard
Questions Mean
deviation
7 It is easy to find specific information 5.8 1.3
8 Navigation through the application is easy 6.2 1
9 The presentation of patient medical history is effective 4.7 1.4
10 The presentation of patient laboratory tests is effective 6.2 1.2
11 The presentation of patient vital signs is effective 6.7 0.5
13 The application would be more beneficial for cases that are
5.3 1.9
more complex than the one that I just did
Participants agreed that finding information and navigating through the
Diagnostic Aid was easy. They also found the presentation of vital signs (mean=6.7) and
laboratory results (mean=6.2) more effective than that of medical history (mean=4.7).
They also believed that in more complicated clinical scenarios the Diagnostic Aid would
be of more help. The frequency table of these questions for different participants is also
presented in Appendix 8.
4.4.3 System Attributes Questions Results
In this section of the questionnaire, participants responded with their level of
agreement that each individual requirement was met. The most important requirements
were addressed through the requirement verification questions. However, one of the
questions was very similar to one of the usability questions and was omitted from the
analysis. The response range varied from 1 to 7 for strong disagreement to strong
agreement toward meeting these requirements. The mean and standard deviation of
participants’ results for each of the questions are presented in the Table 4.6 below.
70
Table 4.6. Results of system requirements verification
Standard
Requirements Mean
deviation
The application design is consistent with the way I think about
1 5.5 0.55
and conduct diagnosis
2 The labels are unambiguous and easy to understand 6.2 1.17
The debiasing tools reduced the likelihood of me being subject
3 5.6 1.2
to cognitive biases
The debiasing tools helped me reach the correct diagnosis more
4 4.8 0.75
accurately
The debiasing tools helped me reach the correct diagnosis more
5 4.5 0.84
quickly
The debiasing tools interfered with the process of decision
6 1.8 0.98
making
It is NOT easy to retrieve the desired information by using this
7 2.5 0.55
application
The application benefited me to balance my speed and accuracy
8 5.5 1.38
in order to come to a timely and correct diagnosis
The application helped me understand relationships and trends
9 4.5 1.51
in the data
The anchoring debiasing tool effectively encouraged me to
10 5.2 1.72
consider other alternatives before making a decision
The availability debiasing tool effectively encouraged me to
11 4.5 1.64
consider other alternatives before making a decision
The representativeness debiasing tool effectively encouraged
12 me to search for inconsistencies between the patient’s 5.2 .6
symptoms and my potential diagnosis
The confirmation bias debiasing tool effectively encouraged me
13 5 1.79
to look for disconfirming data
The application effectively encouraged me to collect more data
14 6 1.26
and consider more hypotheses before making my final decision
Results were categorized by mean response for each requirement:
 High positive degree of agreement (≥ 80%, mean ≥ 5.6)
 Positive degree of agreement (60% - 80%, mean: 4.2 - 5.5 )
 Low degree of agreement (40% - 60%, mean 2.8 - 4.1)

71
 Negative degree of agreement (≤ 40%, mean ≤ 2.8)
Overall, four requirements received a high positive degree of agreement. All the
other requirements received a positive degree of agreement.
4.4.4 Qualitative Interview
As mentioned before, an interview with each participant was conducted after
he/she finished answering the Likert-scale questions. As the most important part of the
evaluation, the purpose of the interview was to gather participants’ feedback regarding
their overall experience with the Diagnostic Aid, their general perception of the system,
and their opinion on its overlooked aspects. In the interview the participants could openly
make their recommendations regarding future developments of the system.
The first interview question asked whether the physician would be willing to use
this system every day if it was fully functional. All participants expressed that they would
use this system if it were fully functional, but they had some concerns about the system
too. Four participants expressed their concern about entering the information in this
system; they would be willing to use the system if information input is easy and efficient.
I might [use the system], but I can’t tell how easy it is to enter the data and who
is entering the data. Technology doesn’t make you faster, it actually probably
makes you a little slower.
The next question asked what participant liked and disliked about the system, and
the third question asked what was useful and what was not, as well as what should be
added to the system and what should be removed. Since these questions were somewhat
similar to each other and the participants’ responses for these two questions overlapped
72
with each other, the results were combined in this section. The elements that participants
liked and the functions they found useful were:
 The ability to go from one part of the system to another easily
 The ability to pull up information such as laboratory data quickly and succinctly
on the margin
 The ability to access external resources
 The ability to go back in time if needed
 The separation of information
 Differential diagnoses and their associated symptoms suggested by the system
 Having medical information on different screens
 The visual graphics of the symptoms plotted against time
 The arrangement of the symptoms, their ranges, and the different colors used to
show the symptoms’ severity
 Using different colors in the overall design
 Having reference ranges for vital signs
The participants lauded two of these aspects more frequently, indicating these
features’ relative importance. One was the organization of patient information that made
it easier to review and navigate. The other was the list of hypotheses and associated
symptoms suggested by the system. Participants found it very beneficial when the system
synthesized the symptoms and suggested diagnoses according to the patient’s condition.
They mentioned that this feature reminded them whether they had considered all of the
differential diagnoses or not. A less experienced participant found the suggested

73
hypotheses very helpful since she realized that she had not considered some of the
diagnoses at all.
A few participants believed that the information was presented in an order that
was not their preferred order and it was aggravating to jump from one part of the system
to the one that they expected to go to next. A participant made a very critical point when
the interviewer asked him whether or not a system based on his model of thinking would
likely be different from another physician’s.
…Not only [would one] physician perhaps have a different order of sequence of
areas that he would think about, that might be different from mine, I would not
use the same sequence of thinking reliably with each patient.
Accordingly, the most satisfactory design as suggested by a participant would be
a customizable layout so that each physician could configure it the way he/she naturally
thinks.
There were some more subtle refinements that participants believed should be
considered:
 The ability to take notes, underline, and highlight the data on the laboratory
results
 The ability to have access to external resources at the final page
 To include the normal range for laboratory results
 To have different reference range of vital signs for different patients
 To have chief complaint under the interview box and allergies in the medication
section
 To have negative symptoms, such as not having fever, as well as symptoms in the
condition section
74
 To have medication and hospitalization precede the physical examination in the
order of presentation
In another question the participants were asked about their suggestions regarding
future developments of the system. They believed that refinements in the design of the
system should be applied according to the subject matter experts’ opinion and then tested
by physicians to come up with the most acceptable design. One of the physicians
suggested that this process could be done in a mock clinic by having people pretend to be
fake patients with fake complaints. Data entry was still one of the concerns, so that two
participants suggested the system be tested in an office setting to see whether it makes the
physician slower or not. However, one participant suggested that it should be embedded
in an EMR system to have access to existing information. The other thing that was
mentioned was that the system should be able to cover all contingencies that might
happen during a clinical encounter so that the physician does not have to jump back and
forth if, for example, the patient is presenting a very odd case or the patient is reluctant to
provide answers to the physician’s questions. A participant suggested that if any of the
patient’s symptoms appeared in the associated symptoms of suggested differential
diagnoses, then they should be highlighted. Another participant proposed that the system
should remind the physician to check the most common symptoms based on the chief
complaint at the reviewing medical information stage. She also added that it would be
very beneficial if the system could suggest essential lab tests at the very beginning steps
of the diagnosis. One participant mentioned that the spelling and terminology should
pertain to how a physician is educated and thinks. The last thing worth mentioning is that
it was suggested that the system by default should present the information in an
75
acceptable format for most physicians, while also allowing the physician to arrange the
information in a more individualized format if he wants.
You don’t want the software program to only make your data entry more
difficult and give you no help at all, and at the other extreme you do not want
the software package to be so dominant that it essentially dictates the patient
and doctor can only think the way the software engineer thinks and no other order
and therefore arrive at a predetermined conclusion that the software engineer
thought of.
The last two questions of the interview part asked the participants whether
debiasing tools were effective and how they could be improved to be more beneficial.
Participants expressed that the debiasing questions had effects on them to some extent
that reminded them not to jump to the first diagnosis. It was also mentioned that
debiasing tools force experienced physicians, who are likely to skip steps of differential
diagnosis, to rethink and go step by step through the process. Although the participants
found the debiasing questions helpful in expanding their thinking, they believed that
debiasing tool were too generic to secure the attention and involvement of the user.
Moreover, a participant mentioned that the debiasing tools could possibly be even more
beneficial for residents that do not have years of experience.
Participants suggested various ways to improve the debiasing tools:
 More interactive tool that offers alternatives in the process of hypothesis
generation
 More visually stimulating
 Better wording for debiasing questions
 Mnemonics can be used as reminders

76
Participants indicated that the debiasing tools might be more effective if the tools
suggest alternative hypotheses while the physician enters his hypothesis list in the
Diagnostic Aid. This feature would remind the physician of certain unconsidered
differential diagnoses while encouraging him to be open-minded and also make the
physician to rethink more carefully about the diagnosis that he/she landed on. Two
participants that self-identified as auditory learner and visual learner believed that the
debiasing tool could be more engaging by using different graphics, brighter colors, or
having it on a separate page that obliges the physician to go through it. Another
participant did not find the second question (availability bias) as helpful as the other
questions, arguing that “common things are common”, meaning a physician must
consider common diagnoses before rare ones. Two other participants mentioned that the
wording of the first question is vague and should be altered. It was also suggested that the
familiar medical mnemonics (such as SIGECAPS – sleep change, interest (loss), guilt,
energy (lack), concentration problem, appetite loss, psychomotor, and suicide – for
depression diagnosis) be used in the embedded debiasing tool so that it forces the
physician to think through all known steps of differential diagnosis for at least particular
problems and therefore for the physicians be able to rule in and rule out other diagnoses.
77
5 Discussion
The results that were presented in the previous chapter are reviewed, analyzed,
and discussed in this chapter. The general outcomes of the experiment conducted are
emphasized and recommendations are also offered. Limitations of the study are covered
at the end of this chapter as well.
5.1 System Usability Experience
Participants’ overall perception of the system was positive and they found the
system usable and functional. An overall average rating of 5.7 out of the 7-point Likert-
scale with the standard deviation of about one for usability questions indicates that
participants found the Diagnostic Aid easy to use on average. Only one participant had an
average score of 4.4, which was considerably less than the total average over all
participants (Figure 5.1). Even though the Diagnostic Aid was not completely functional,
the majority of participants easily learned how to work with it and evaluated it as reliable
in usability questions. Participants were also positive about using this system on a daily
basis if the system works correctly and the data entry does not make them slower;
however, the physician’s inclination to adopt a new system couldn’t be completely
evaluated without a fully functional system.

78
7 6.7
6.3
6 5.9
6
5.7
5
5
4.4
Total Average
3
Figure 5.1. Usability question results for each participant
It was observed that the physicians were very open to the idea of new
technologies in the medical system and were inclined to the idea of having a Diagnostic
Aid as a decision aid tool. However, the fact that the sample size was relatively small and
the participants were not selected randomly favors the hypothesis that more individuals
more accepting of new technology agreed to participate in this study. Nevertheless, the
high ratings are of interest considering the fact that all participants were over forty years
old and had at least ten years of medical experience. If the participants were selected
independently and a bigger sample was obtained, a stronger conclusion could be formed
regarding the tendency of more experienced physicians to use new technologies in their
daily practice. It seemed that data entry in current EMR systems is complex and time
consuming since all participants’ first concern regarding the Diagnostic Aid was about
79
the difficulty of inputting the data. However, improving the quality of diagnosis by
amplifying cognition through reducing mental workload was of more of interest in this
study than the data entry difficulties. Therefore, attention was more on developing the
standards for medical information presentation, and data entry issues were not addressed.
As a result, the Diagnostic Aid should be considered as an isolated system designed only
for facilitating diagnosis that is embedded in an EMR system. A complete
implementation of the Diagnostic Aid would require more thought on mechanisms of
data capture and data entry.
According to the usability questionnaire results, finding information and
navigation through the system was uncomplicated and the presentation of vital signs and
laboratory tests was effectual due to different sections, colors, and graphs. Nevertheless,
participants found the presentation of medical history less efficient while they desired a
more complete and concise history portion. One of the participants mentioned that the
physicians are accustomed to a different order for medical information presentation
which starts by chief complaint followed by history of present illness (HPI), review of
systems (ROS), past medical history (PMH), social history, family history and so on.
Therefore, the inconsistency between the Diagnostic Aid’s presentation of medical
history and what the physicians are using in their everyday practices was the main reason
that they found the medical history section not effective.
In general, positive opinions of the participants regarding the Diagnostic Aid
reveals that this system is potentially a good platform for implementing ideas regarding
improving the diagnosis.

80
5.2 Randomized Experiment
One of the objectives of this research was to find the effectiveness of the
Diagnostic Aid in facilitating the diagnostic process. Besides the usability study, a
randomized experiment was conducted to explore the differences of the Diagnostic Aid
and paper format presentation of the data. As mentioned in the literature review, among
the solutions for reducing cognitive errors are decreasing reliance on memory, and
improving data presentation quality. The Diagnostic Aid implemented these solutions
along with other features that make it a potential solution to the problem since the
technical questions were mostly about measuring the memory and recall abilities of the
participants. Therefore, more accurate answers on the technical questions when using the
Diagnostic Aid could have been interpreted as demonstrating the effectiveness of this
system in decreasing the physician’s memory workload, and therefore increasing the
quality of diagnosis.
The scenarios used for the study were very brief and designed in order to
represent medical information visually. This was probably the major reason that the
participants performed almost equally well in the technical questions section for both
formats. In addition, the multiple-choice questions or even open-ended questions could
not adequately evaluate the effectiveness of information presentation. One reason is that
the participants had a tendency to choose more than one answer in a few of the multiple-
choice questions. Additionally, a careful scrutiny of each participant’s results revealed
that the second test taken was affected by the first test due to the fact that the medical
scenarios’ content were very similar.

81
On a few questions the participants did very close on both formats while on the
other questions there were some difference in the accuracy of responses. Questions 4 —
How many months did swollen feet /nonproductive cough last? — and 6 — What is the
WBC count? — received the most wrong answers among the Diagnostic Aid format
while questions 6 and 7 — Are there any abnormalities in the patient’s lab results? — and
9 — Which symptom started soonest? — had the same situation for the paper format. For
questions 4 and 6 the presentation of information in the paper format was more
straightforward and salient than that of the Diagnostic Aid, and it could be a reason why
the paper format was superior in question 4; however, in question 6, the results were
quite the same and also for question 7 which was similar to the question 6, the
participants did better with the Diagnostic Aid. This defied expectations that the paper
format should be superior regarding the results of these two questions; however, it should
be considered that question 7 required some medical knowledge while question 6 did not.
The similarity of cases and subjectivity of the correct answers were among the reasons
that make the outcome of this part inconclusive. Probably the most interesting result was
for question 9 that the participants did significantly better on Diagnostic Aid since the
answer was easily recognizable in the Diagnostic Aid format due to the graphic bars but
not so evidently in the paper format. This is consistent with the findings of Plaisant et al.
(1998) in efficiency of the timeline for recalling interval comparisons and inter-
categorical connections. Overall, the small sample size had its most negative effect on
this part of the study since for getting a statistically significant result a bigger sample size
and a more randomly selection of subjects would be required. Since the scenarios were
very similar in content, a bigger time interval between executing the two parts of the
82
experiment could have guaranteed more independent results. Even with the same set of
questions and a bigger sample size, a greater distinction between the Diagnostic Aid and
paper format on specific details, such as question 9, would be expected even if not on the
overall average for all questions. It also can be concluded that the Diagnostic Aid
presentation of information is at least as good as the paper presentation and might be
more advantageous in more intricate cases.
With the current result we cannot conclude that the Diagnostic Aid design and its
debiasing features were effective in improving the quality of the diagnostic process.
However, all participants except one agreed that such a system could be even more
beneficial for more complex scenarios. In more complex cases or in real practice the
physicians face cases that require gathering and analyzing information to a great extent
and therefore the physicians would be more vulnerable to heuristic errors.
5.3 Requirements Verification
Requirements were initially developed for a system that is fully functional and
interactive and were partly verified in the development process of the system through
talking to physicians as subject matter experts. Moreover, physicians used the system
attributes questionnaire for evaluation of the requirements. Due to the fact that the
Diagnostic Aid was not fully functional, not all the requirements could be verified.
Jordan (2002) indicates that if at least 80% of participants have a positive opinion
regarding a system attribute, then that requirement could be considered verified. The
nature of requirements used in this study is different from his field of attention but still a
similar approach is justifiable. Since the participants were six in total, at least 80% is
83
achieved with five participants; with a bigger sample size the verification results would
be more valid. After data analysis, the requirements could be grouped into three
categories: verified, not verified, and not evaluated. If responses of equal and greater than
five out of seven are perceived as the positive agreement, then seven requirements could
be considered verified. These requirements, and one other requirement that was verified
through the usability questions, include:
User interface (1): The Diagnostic Aid design shall be consistent with the way
that physician conduct diagnosis.
User interface (10): The Diagnostic Aid shall provide unambiguous labels for
objects so they can be easily read.
User interface (15): The Diagnostic Aid shall minimize the cost of retrieving
information.
Debiasing tools (6): Debiasing tools shall not interfere with the process of
decision making.
Debiasing tools (10): The anchoring debiasing tool shall encourage the
clinician to review all the patient’s information before making a decision.
Debiasing tools (13): The representativeness debiasing tool shall encourage the
clinician to search for inconsistencies between patient’s symptoms and the
potential diagnosis.
Debiasing tools (18): The debiasing tools reduced the likelihood of me being
subject to cognitive biases.
Debiasing tools (20): The Diagnostic Aid shall encourage the clinician to collect
more data and consider more hypotheses before making final decision.
84
Requirements regarding the Diagnostic Aid’s design, information presentation,
navigation, and effective encouragement for more data collection were strongly approved
by participants. Requirements about the anchoring and representativeness biases were
also verified with a lesser degree of confidence. If the mean response for each question is
considered for verification (> 80%, mean= 5.6 out of 7) instead of the number of
participants that approved the requirement, then two requirements regarding the
anchoring and representativeness debiasing tool could not be considered verified.
A few other requirements were verified through the interview section, when
participants found a specific feature helpful or mentioned the necessity of a feature that
was considered in the initial requirements list, but was not implemented in the current
Diagnostic Aid. These potentially verified requirements include:
Functional (2): The Diagnostic Aid shall provide means to share the patient’s
problem with remote specialists.
Functional (3): The Diagnostic Aid should provide feedback of the clinician
final diagnosis(es).
Functional (9): The Diagnostic Aid shall provide means for the user to
highlight especially significant information.
Functional (11): The Diagnostic Aid shall provide means to move among
different pages easily.
User interface (3): The patient’s medical information shall be readily
accessible from all pages.
User interface (5): The Diagnostic Aid shall present the patient’s vital signs
and allergies.
85
User interface (6): The Diagnostic Aid shall present all symptoms associated
with each diagnosis in the hypothesis list.
User interface (7): The Diagnostic Aid shall present a list of suggested
diagnoses after the clinician has reviewed the patient's data.
User interface (20): The Diagnostic Aid should use constraints to prevent
possible errors.
Debiasing tools(3): Subtle but clear visible feedback shall be implemented for
debiasing tools.
A list of requirements that were not verified includes:
Debiasing tools(8): The debiasing tools shall help the clinician to reach the
correct diagnosis more accurately.
Debiasing tools(9): The debiasing tools shall help the clinician to reach the
correct diagnosis more quickly.
Debiasing tools(14): The Diagnostic Aid shall present diagnostically
significant relations and trends among data.
Debiasing tools(12): The availability debiasing tool shall encourage the clinician
to consider other alternatives rather than those easily brought to mind.
Debiasing tools(15): The confirmation bias debiasing tool shall encourage the
clinician to look for more data that provide disconfirming evidence.
Most of the non-verified requirements were related to the debiasing tools. The
participants did not find the debiasing tools helpful to come up with a final diagnosis
more accurately and quickly. The availability and confirmation bias requirements also did
86
not seem to encourage additional data collection. In general, the debiasing tool section
was not a good implementation of the debiasing techniques for diagnostic process. The
physicians ignored it easily during the process since its design and presentation was not
salient enough to demand their attention. However, that the participants mentioned that it
reminded them of the possibility of cognitive biases, so it would not be overoptimistic to
conclude that debiasing questions could be helpful if designed to be more salient and
more specific to the case under consideration. As mentioned in previous chapters,
reminding the physicians of the pitfalls of heuristics is the first step in the process of
debiasing and the debiasing questions were partially effective in this regard. The rest of
the requirements were not evaluated for verification, yet some of them were verified
through the design process. The requirements regarding the design of the Diagnostic Aid
that were implemented during the development of the system were considered verified.
The following requirements are based on the interviews with participants and
were not considered before. The participants’ needs validated the necessity of these
requirements and therefore they should be considered for the future generations of the
Diagnostic Aid.
 The Diagnostic Aid shall make the physician capable of customizing the layout of
the information presented.
 The Diagnostic Aid shall present negative symptoms as well as positive ones.
 The Diagnostic Aid shall present lab results with a reference range.
 The Diagnostic Aid shall implement different vital signs’ reference range for
different patients.
87
 The Diagnostic Aid shall make the physician capable of accessing to external
resources on all pages.
 Debiasing tools shall be presented visually stimulating.
 Debiasing tools shall be interactive and help the physician when heuristics are
likely to fail.
5.4 Positives and Negatives
According to the quantitative and qualitative analysis, there were a few of the
system’s characteristics that were highlighted by the participants as positive. A list of the
most commended positive features and functions include:
 Presentation of essential information on different screens.
 The visual presentation of the symptoms and medication against time using bar
graphs.
 The visual presentation of vital signs with a reference range.
 The utilization of different colors in information presentation.
 Easy navigation through the system to find different information.
 Synthesizing the symptoms and offering differential diagnosis.
 The ability to retrieve medical information quickly on the right hand column of
the display.
 The links to external resources.
 The debiasing questions segment.

88
On the other hand, the most cited drawbacks to the system were identified as:
 Unconventional presentation of medical history information.
 Lack of reference range for lab results.
 Generic and not adequately stimulating debiasing questions.
5.5 Recommendations
The overall results of this study reveal where the most effort should be placed for
future developments of the Diagnostic Aid. The elements that were recognized as helpful
should be kept and improved in the system and the less useful ones should be removed or
significantly altered. Some new features also should be added to the system to improve
system usability, functionality, and effectiveness. For instance, the normal range for
laboratory results is very crucial, especially for uncommon laboratory tests. Mnemonics
or checklists should be used to explore all the serious symptoms, to remind the physician
of all the diagnostic process steps, and to propose all possible differentials. The
presentation of negative symptoms in a different color under the list of all other
symptoms would probably make the list of symptoms more complete and informative.
Regardless of whether the Diagnostic Aid achieved the goals of this research, these
refinement will increase the chance of having a more efficient and reliable system in the
future.
One of the most-mentioned drawbacks of the Diagnostic Aid was the presentation
of medical history. The physicians are accustomed to another format and order of medical
history presentation and were not very comfortable with the information display.
Therefore, for future developments of the Diagnostic Aid, the medical history
89
presentation should be more consistent with current EMR formats. Additionally, the
layout of data presentation should be customizable for each physician since not a single
format would match all physicians’ expectations and mindsets. So, the medical history
should be presented in a modifiable way that is aligned with the current EMR’s format of
presentation so that the information could be retrieved with less effort. On the other hand,
physicians found the presentation of medical information over a timeline very interesting,
while unconventional. Although the idea of timeline for presenting medical records is not
original to this research and was presented by Powsner and Tufte (1994) and other studies
showing its advantages in better recall and faster response time, it has not been tried in
the current EMR systems used by medical professionals (Plaisant et al., 1998). This
feature should be kept and improved so that the physician could search for medical
conditions over time by expanding the timeline.
One of the main objectives of this study was to introduce an aid for facilitating the
diagnosis and reducing common cognitive biases. The debiasing section was designed so
as not to interfere with the decision making process of the physician. The researcher was
cautious to make the debiasing tool, as an unconventional section, as unobtrusive as
possible, while the physicians were more willing to work with a more engaging decision
aid tool. Since the debiasing section was one of the main original ideas presented by this
effort, it would have been more valuable if the participants found it interfering and made
comments about it rather than ignorable. The total experiment manifested that physicians
were open to a more interactive debiasing tool that reminds them of all the possible
pitfalls during the diagnostic process and directs them through that process. Regehr and
Norman, (1996) indicate that just the awareness of the biases is not enough to avoid them
90
and neither is passive training. But recognizing the situations where the heuristics are
likely to fail would be more helpful. Therefore, first, the debiasing section should be
more stimulating and engaging by incorporating brighter colors and/or interactive
buttons. It was suggested to implement debiasing tools in a completely separate page that
the physician has to go through, forcing him to think about all possibilities. Second, the
debiasing tools, embedded in the Diagnostic Aid, should consider the patient’s
information and suggest useful directions to take. For example, it might remind the
physician to consider certain differential diagnoses or suggest doing specific lab tests; a
participant suggested that the system should also ask about the existence of serious
symptoms according to the chief complaint. Since cognitive biases affect the physician
from the early stages of the process, it would be more beneficial that the differential
diagnoses be offered sooner in that process.
5.6 Limitations
This study was a combination of a usability study and a randomized experiment.
The small sample size and the not fully functional Diagnostic Aid caused some
limitations in the study. The non-functionality of the Diagnostic Aid reduced the
participants’ level of perception of the system. For instance, there were a few potential
functionalities that were not fully interactive in the prototype developed; the Diagnostic
Aid was supposed to let the physician take notes, enter an initial list of hypotheses, and
also connect to medical and non-medical external resources. Since these functions were
not fully functional in the Diagnostic Aid prototype, they were not used in the experiment
91
and therefore were not evaluated by the participants, except by means of a few comments
in the interview section.
The sample size of participants was not particularly small for a usability study,
and the participants probably uncovered most of the obvious issues, but it was quite small
for a randomized experiment. To report a significant statistical result and make causal
inference, a larger sample of participants would be required. Difficulties in recruiting
busy physicians2 prevented acquiring of a bigger sample. Although the participants were
randomly assigned to four different groups to minimize the effect of unwanted factors,
they were not a representative enough sample of medical professionals since they were
selected through career connections. In addition, all participants were in the experienced
stratum of their profession and therefore were not representative of a less experienced
population of physicians. Therefore, the results of the study could not be generalized to a
bigger population of physicians.
The medical scenarios developed for the training and for the experiment were
very similar to each other and caused confusion in answering some of the technical
questions. The scenarios and the associated technical questions were not adequate for the
participants to distinguish the Diagnostic Aid from the paper if there was any advantage
of the former. Likely, more complicated scenarios with similar levels of difficulty but
differentiated content would have provided better results for the experiment. The
questionnaire was also not designed in the best way. The analysis of Likert-type
questions without having a reference score is usually a very subjective task. Using
standard systems such as SUS with a reference score would make the interpretation of the
2
It took six months to recruit only six participants.
92
result much easier. Due to the presence of the researcher, the participants might have
been subject to response bias and provided desirable responses rather than choosing
answers that were reflective of their true opinion. Using indirect questioning or methods
that do not require the presence of the interviewer could be helpful in overcoming this
bias (Fisher, 1993). Online surveys for questionnaire parts might also be beneficial since
respondents feel more anonymous in online surveys and less susceptible to social
desirability bias (Joinson, 1999).
The analysis of the survey questions was very subjective due to the lack of a
baseline for inference. The interviews were transcribed and analyzed only by the
researcher and therefore the results might be subject to measurement bias. To remove this
bias, the interview voice records, could be independently transcribed by an additional
person and the results compared to the researcher’s.

93
6 Conclusion
The main purpose of this research was to improve patient safety by reducing
cognitive errors in the clinical diagnostic process. To achieve this goal, understanding the
clinical diagnostic process and developing a model of this process was the initial step.
The literature on clinical diagnosis and conversations with subject matter experts helped
to understand the process up to a level that an IDEF0 model could be developed for the
diagnostic process. The IDEF0 model demonstrated most of the influential factors in this
process. Next, the requirements were established for the development of the Diagnostic
Aid. Overall, the literature review, IDEF0 model of the diagnosis, and requirements
formed the structure for the development of the Diagnostic Aid.
The Diagnostic Aid is the semi functional application for the Apple iPad used for
facilitating the diagnostic process and reducing the chance of cognitive errors. The
Diagnostic Aid visualized information and provided features that tried to amplify
physicians’ cognition through reducing their mental workload. In the randomized
experiment, the Diagnostic Aid was found to be as effective as the paper format
regarding information presentation, yet not significantly better. So it cannot be concluded
that Diagnostic Aid facilitated the diagnosis process. However, the physicians accepted
the overall Diagnostic Aid’s design and evaluated its features positively.
The last objective of this research was to verify the requirements through the
experiment and surveys. The majority of the requirements tested were verified and some
new requirements were generated. The Diagnostic Aid addressed numerous needs of the
94
physicians, while participant’ articulated demands and preferences that had not been
previously considered shed light on the necessity of new requirements.
6.1 Summary
The Diagnostic Aid was not effective enough in improving the quality of care
through reducing cognitive errors. Yet although the system as a whole did not meet
expectations for facilitating diagnosis, its specific features for information presentation,
suggesting alternatives, reminding of potential biases, etc. were evaluated positively by
the physician participants. Therefore, it is still possible that the effectiveness of the
Diagnostic Aid could be established using a more complete implementation of the design
and a better setting for the experiment.
Participated physicians seemed very open to the adoption of new systems with
new technologies as decision-support systems as long as these systems do not just make
them slower with no particularly added advantage. The physicians most desired a system
immune to biases and intelligent enough to guide them through the diagnostic process
while they enter the information in the system.
6.2 Future research
This study provides insight into the physician’s expectations of a decision-support
system that would be beneficial in the process of diagnosis. The results of this study are
limited to the sample of participants while the methods used could be extended to larger
sample sizes for further studies. A larger sample of participants with more diverse
specialties and skills would make the results more generalizable. Relevant studies should
benefit from a fully functional system for increasing the validity of the evaluations.
95
The IDEF0 model of the diagnostic process and the requirements developed could
be used for further studies and references regarding the clinical diagnostic process. The
partially functional Diagnostic Aid would be a robust framework for future studies that
continue with the development of decision aid tools. In the fully functional Diagnostic
Aid the physician should be able to easily enter and edit the essential information. The
debiasing tools require substantial change in function and design as just general strategies
toward caution are unlikely to be helpful in reducing the chance of biases occurring.
Therefore, the debiasing tools should be more specific, more engaging, and better
integrated into the system.
A system that could reduce the mental workload of physicians by providing
essential information and guides at the right time would be valuable. Developing such a
complicated system would likely require the collaboration of very different specialties
such as medicine, psychology, information systems, data mining, artificial intelligence,
and so on. The core of such a system is expected to be an intelligent and powerful
computation system that has the capability of learning from entered data. In addition,
according to what the physicians expected from such a system, a single system could not
be effective enough for all medical areas, indicating that customized systems within each
discipline are required. Therefore, with the current state of the technology and established
medical policies, effectively reminding the physicians of the potential errors seems to be
the most promising way of reducing medical diagnosis errors.

96
Bibliography
Baerheim, A. (2001). The diagnostic process in general practice: has it a two-phase

structure? Family Practice, 18(3), 243–245.
Baron, J. (2008). Thinking and deciding. New York: Cambridge University Press.
Brooke, J. (1996). SUS - A quick and dirty usability scale. In CRC Press.
Budurka, W. J. (1984). Developing strong systems engineering skills. IBM Technical

Directions, 10(4), 40–48.
Card, S. K., Mackinlay, J., & Shneiderman, B. (Eds.). (1999). Readings in Information
Visualization: Using Vision to Think (1st ed.). Morgan Kaufmann.
Chapanis, A. (1996). Human Factors in Systems Engineering. New York, NY, USA:
John Wiley & Sons, Inc.
Chittaro, L. (2001). Information visualization and its application to medicine. Artificial

Intelligence in Medicine, 22(2), 81–88.
Croskerry, P. (2000). The cognitive imperative: thinking about how we think. Academic
Emergency Medicine: Official Journal of the Society for Academic Emergency
Medicine, 7(11), 1223–1231.
Croskerry, P. (2002). Achieving quality in clinical decision making: cognitive strategies

and detection of bias. Academic Emergency Medicine: Official Journal of the
Society for Academic Emergency Medicine, 9(11), 1184–1204.
Croskerry, P. (2003). The importance of cognitive errors in diagnosis and strategies to

minimize them. Academic Medicine: Journal of the Association of American
Medical Colleges, 78(8), 775–780.
Debbabi, M., Hassaïne, F., Jarraya, Y., Soeanu, A., & Alawneh, L. (2010). Verification
and Validation in Systems Engineering: Assessing UML/SysML Design Models
(2010 edition.). Heidelberg ; New York: Springer.
97
Elstein, A. S. (1999). Heuristics and biases: selected errors in clinical reasoning.

Academic Medicine: Journal of the Association of American Medical Colleges,
74(7), 791–794.
Fisher, R. J. (1993). Social Desirability Bias and the Validity of Indirect Questioning.
Journal of Consumer Research, 20(2), 303–315.
Graber, M. L. (2003). Metacognitive training to reduce diagnostic errors: ready for prime
time? Academic Medicine: Journal of the Association of American Medical
Colleges, 78(8), 781.
Graber, M. L. (2005). Diagnostic errors in medicine: a case of neglect. Joint Commission

Journal on Quality and Patient Safety / Joint Commission Resources, 31(2), 106–
113.
Graber, M. L. (2008). Taking Steps Towards a Safer Future: Measures to Promote

Timely and Accurate Medical Diagnosis. The American Journal of Medicine,
121(5, Supplement), S43–S46.
Graber, M. L., Franklin, N., & Gordon, R. (2005). Diagnostic error in internal medicine.
Archives of Internal Medicine, 165(13), 1493–1499.
Graber, M. L., Gordon, R., & Franklin, N. (2002). Reducing diagnostic errors in
medicine: what’s the goal? Academic Medicine: Journal of the Association of
American Medical Colleges, 77(10), 981–992.
Haines, J. (n.d.). Urgent Interactions: Evaluating Usability and Incorporating Information

Visualization in Emergency Medicine Interfaces. Retrieved from
http://www.juliakhaines.com/home/files/JHainesCapstoneFinal.pdf
Hooks, I. (1994). Writing Good Requirements. In proceedings of the 4th Annual

International Symposium of the National Council on Systems Engineering (Vol.
2). San Jose, CA.
Hunt, D. L., Haynes, R. B., Hanna, S. E., & Smith, K. (1998). Effects of computer-based
clinical decision support systems on physician performance and patient outcomes:
a systematic review. JAMA, 280(15), 1339–1346.
James, J. T. (2013). A New, Evidence-based Estimate of Patient Harms Associated with

Hospital Care: Journal of Patient Safety, 9(3), 122–128.
98
Joinson, A. (1999). Social desirability, anonymity, and internet-based questionnaires.

Behavior Research Methods, Instruments, & Computers, 31(3), 433–438.
Jordan, P. W. (2002). Designing Pleasurable Products. London; New York: CRC Press.
Kassirer, J. P., & Kopelman, R. I. (1989). Cognitive errors in diagnosis: instantiation,

classification, and consequences. The American Journal of Medicine, 86(4), 433–
441.
Kassirer, J. P., & Kopelman, R. I. (1991). Learning clinical reasoning. Williams &
Wilkins.
Kim, S., & Jang, K. (2002). Designing performance analysis and IDEF0 for enterprise
modelling in BPR. International Journal of Production Economics, 76(2), 121–
133.
Kohn, L. T., Corrigan, J., & Donaldson, M. S. (2000). To err is human building a safer
health system. Washington, D.C.: National Academy Press.
Larkin, J. H., & Simon, H. A. (1987). Why a Diagram is (Sometimes) Worth Ten
Thousand Words. Cognitive Science, 11(1), 65–100.
Leape, L. L., Brennan, T. A., Laird, N., Lawthers, A. G., Localio, A. R., Barnes, B. A.,
… Hiatt, H. (1991). The nature of adverse events in hospitalized patients. New
England Journal of Medicine, 324(6), 377–384.
Leape, L. L., Lawthers, A. G., Brennan, T. A., & Johnson, W. G. (1993). Preventing
medical injury. QRB. Quality Review Bulletin, 19(5), 144–149.
Mamede, S., Schmidt, H. G., & Rikers, R. (2007). Diagnostic errors and reflective
practice in medicine. Journal of Evaluation in Clinical Practice, 13(1), 138–145.
Mutic, S., Brame, R. S., Oddiraju, S., Michalski, J. M., & Wu, B. (2010). System
mapping of complex healthcare processes using IDEF0: a radiotherapy example.
International Journal of Collaborative Enterprise, 1(3), 316–331.
NIST, National Institute of Standards and Technology (1993). Integration Definition for
Function Modeling (IDEF0), Federal Information Processing Standards
Publication 183. Gaithersburg, MD: Author.
99
Norman, G. R., & Eva, K. W. (2010). Diagnostic error and clinical reasoning. Medical
Education, 44(1), 94–100.
Plaisant, C., Mushlin, R., Snyder, A., Li, J., Heller, D., & Shneiderman, B. (1998).
LifeLines: using visualization to enhance navigation and analysis of patient
records. Proceedings of the AMIA Symposium, 76–80.
Powsner, S. M., & Tufte, E. R. (1994). Graphical summary of patient status. Lancet,
344(8919), 386–389.
Reason, J. (1990). Human Error (1st ed.). Cambridge University Press.
Regehr, G., & Norman, G. R. (1996). Issues in cognitive psychology: implications for
professional education. Academic Medicine: Journal of the Association of
American Medical Colleges, 71(9), 988–1001.
Richardson, W. S., Wilson, M., & Guyatt, G. (2002). The process of diagnosis. Users’
Guide to the Medical Literature: A Manual for Evidence-Based Clinical Practice.
2nd Ed. New York, NY: McGraw Hill Medical, 399–406.
Ruddle, R., Brodlie, K., & Dimitrova, V. (2002). Communication, visualisation and
interaction. University of Leeds, School of Computing.
Sawyer, D., Aziz, K., Backinger, C., Beers, E., Lowery, A., Sykes, S., … Trautman, K.
(1996). Do It By Design - An Introduction to Human Factors in Medical Devices.
Schiff, G. D., & Bates, D. W. (2010). Can Electronic Clinical Documentation Help
Prevent Diagnostic Errors? New England Journal of Medicine, 362(12), 1066–
1069.
Sox, H. C., Blatt, M. A., Higgins, M. C., & Marton, K. I. (2006). Medical Decision
Making (1st ed.). The American College of Physicians.
Tong, X. (2011). Process Modeling the Physician-Patient Encounter (Master’s thesis).

Oregon State University, Corvallis, OR
Toy, E., & Patlan, J. (2012). Case Files Internal Medicine, Fourth Edition (4th ed.).
McGraw-Hill Medical.
100
Tufte, E. R. (2001). The visual display of quantitative informations (2nd ed.). Cheshire,
Conn.: Graphics Press.
Tversky, A., & Kahneman, D. (1973). Availability: A heuristic for judging frequency and
probability. Cognitive Psychology, 5(2), 207–232.
Tversky, A., & Kahneman, D. (1974). Judgment under Uncertainty: Heuristics and
Biases. Science, 185(4157), 1124–1131.
Visualization. (n.d.). Retrieved September 16, 2013, from http://www.merriam-

webster.com/dictionary/visualization
Voytovich, A. E., Rippey, R. M., & Suffredini, A. (1985). Premature conclusions in

diagnostic reasoning. Journal of Medical Education, 60(4), 302–307.
Wachter, R. M. (2010). Why diagnostic errors don’t get any respect--and what can be
done about them. Health Affairs (Project Hope), 29(9), 1605–1610.
Whitman, L., Huff, B., & Presley, A. (1997). Structured models and dynamic systems
analysis: the integration of the IDEF0/IDEF3 modeling methods and discrete
event simulation. In Winter Simulation Conference: Proceedings of the 29 th
conference on Winter simulation (Vol. 7, pp. 518–524).
Wickens, C. D., Lee, J. D., Liu, Y., & Gordon-Becker, S. (2003). Introduction to Human
Factors Engineering (2nd ed.). Prentice Hall.
Yates, J. F. (1990). Judgment and decision making. Prentice Hall.

101
Appendices
102
Appendix 1- Entire IDEF0 Diagram
Medical guidelines
Clinician's factors
Patient perceived problems
Existing patient-clinician relationship Healing patient

Clinician's initial understanding of patient's problems Updated medical records
New/returning patient Follow-up schedule
Clinician's updated understanding of patient's problem after the encounter
Conduct medical encounter
Final diagnosis(es)
Final treatment plan
A0
Information System
Medical equipment
Patient
Facilities
Clinician
Figure A.1. Conduct medical encounter
C4 C9 C3 C2 C1 C6 C7 C8 C5
Patient medical condition Medical guidelines
Patient perceived problems Hypothetical-Deductive strategy
Environment/system factors Pattern Recognition
Clinician's factors I_Medical test results

Existing patient-clinician relationship Ongoing patient-clinician relationship
I1 Establish & O6
Clinician's initial understanding of patient's problems Patient's
Clinician's updated understanding of patient-clinician factors
relationship Clinician's updated understanding of patient's problem after the encounter
I2 maintain O4
patient-clinician
relationship
A1

O10
Physical examination result Updated patient's medical history

O9
Diagnose the patient problem Final diagnosis(es)

O7
Clinician's understanding of final diagnosis

A2
Patient in treatment Healing patient

O1
New/returning patient Updated medical records
I3 O2
Treat patient Final treatment plan
O8
Clinician's understanding after implementing the treatment.
Discussion about treatment plan
A3
Clinician's understanding with discussed follow-up plan
Plan follow-up Follow-up schedule

O3
O5
Discussion of follow-up plan
A4

Clinician Facilities Information System
M1 M3 M2 M4 M5
Figure A.2. Conduct medical encounter encounter sub processes

103
C12C10C9 C2 C4 C11C3 C1 C5 C8 C6 C7
I_Patient's medical history The rule of parsimony
I_Patient's medical signs & symptoms Hypothetical-Deductive strategy
I_Chronological account of illness Pattern Recognition
Patient's factors
Clinician's factors
Medical guidelines
Clinician's updated understanding of patient-clinician relationship Evaluate patient's history, medical Context of the problem
I1 signs and symptoms
A21
Physical examination result Generate Updated context of the problem

I2 hypotheses
Initial set of hypotheses
A22

O2
O5
Compare, reject and finalize Clinician's understanding of final diagnosis
hypotheses O4
Final diagnosis(es)
O3
O1
A23
Clinician
Patient
Facilities Medical equipment
M3 M2 M1 M4
Figure A.3. Diagnose the patient problem
C5 C1 C10C6 C8 C9 C7 C4 C2 C3
Patient medical condition I_Patient's medical signs & symptoms
I_Chronological account of illness Anchoring cognitive bias
I_Patient's medical history Representativeness cognitive bias
I_Medical test results Availability cognitive bias
Patient's factors
Medical guidelines
Clinician's factors
Clinician's updated understanding of patient-clinician relationship Clinician's underestanding of patient medical test results
I1
Review patient's medical
history and test results
Clinician's understanding of patient's medical history
A211
Analyze medical signs & Clinician's understanding of patient medical signs & symptoms
symptoms
A212
Listen to patient's illness Clinician's understanding of patient account of illness(HPI)

chronology
A213
Context of the problem

O1
Develop a cognitive representation

of the problem
A214
Clinician
Facilities
Patient
M2 M1 M3
Figure A.4. Evaluate patient's history, medical signs and symptoms

104
C3 C5 C4 C2 C1
Anchoring cognitive bias Environment/system factors Confirmation cognitive bias
Availability cognitive bias
Representativeness cognitive bias
Patient's factors
Clinician's factors
Context of the problem Generate hypotheses based Initial generated hypotheses list
I1 on aggregated knowledge
A221
Clinician's understanding of the patient's problem

Conduct physical
examination
Physical examination result Updated context of the problem
I2 O2
A222
Prioritize list of initial Initial set of hypotheses

hypotheses O3
A223
Identify additional
information needed for Diagnostic information need
O1
consistency check
A224
Clinician
Facilities
M2 M1 M3 M4
Figure A.5. Generate hypotheses
C10C6 I1 C1 C5 C4 C2 C9 C3 C7 C8
Pattern Recognition Premature closure error I_Medical test results The rule of parsimony
Medical guidelines

Patient's factors
Confirmation
Clinician's factors cognitive bias
Hypothetical-Deductive strategy
Initial set of hypotheses Selected hypotheses that match patient's condition

I2
Refute hypotheses not consistent with

context of the problem

I3
A231
Additional information needed
Prioritize hypotheses and identify

additional information needed
Prioritized list of hypotheses
A232

O5
Order required medical tests Updated Context of The Problem
Remained hypotheses
A233
Compare hypotheses one by one Final hypotheses

and refute the unlikely ones
A234
Explain the O1
Final diagnosis(es)
patient's problem O4
with final O2
hypothesis(es) Clinician's understanding of final diagnosis
O3
Plan a follow up visit
A235
Clinician
Patient
Facilities
Medical equipment
M1 M4 M2 M3
Figure A.6. Compare, reject and finalize hypotheses

105
C4 C5 C9 C8 C6 C2 C1 C7 C10 C3
Diagnostic information need Confirmation cognitive bias
Clinician's factors Patient medical condition
Hypothetical-Deductive strategy Pathognomonic strategy
Patient's factors factors
Environment/system
Medical guidelines
Pattern Recognition
Initial set of hypotheses Identify expected outcomes Expected outcome of each hypothesis
I1 of each hypothesis
A2311
Matched hypothesis(es)
Compare each hypothesis outcome with

patient's condition and reject the
unlikely ones
I2
A2312
Refute hypotheses that don't match Selected hypotheses that match patient's condition
patient condition O1
A2313
Clinician
Facilities
M3 M1 M4 M2
Figure A.7. Refute hypotheses not consistent with context of the problem
106
Appendix 2 – Requirements
Table A.1. Complete list of requirements, their action area, and sources
Functional Action area Source Verification
1 The Diagnostic Aid shall provide means to take Explain the patient’s problem IDEF0 Model – Verified by
record of clinician's final understanding of the with final hypothesis(es) A235 design
problem.
2 The Diagnostic Aid shall provide means to share the General (Wachter, 2010) Verified by
patient’s problem with remote specialists. questionnaire
3 The Diagnostic Aid should provide feedback of the General (Graber, 2008) Verified by
clinician final diagnosis(es). questionnaire
4 The Diagnostic Aid shall provide means to access Generate hypotheses (Graber, 2008) Verified by
electronic medical references. Compare, reject, and finalize design
hypotheses
5 The Diagnostic Aid shall provide means to list the Generate hypotheses bases on IDEF0 Model – Verified by
clinician’s hypotheses. aggregated knowledge A221 design
6 The clinician shall be able to take notes of his Conduct physical examination IDEF0 Model – Verified by
physical examination findings. A222 design
7 The clinician shall be able to add to and remove Prioritize list of initial IDEF0 Model – Verified by
from the hypotheses list and prioritize hypotheses. hypotheses A223 design
Refute Hypotheses that don’t IDEF0 Model –
match patient condition A2313
8 The Diagnostic Aid shall allow the clinician to take Develop a cognitive IDEF0 Model – Verified by
and save notes at all of the stages of the diagnostic representation of the problem A214 design
process.
107
Table A.1. Complete list of requirements, their action area, and sources (Continued)
9 The Diagnostic Aid shall provide means for the user General (Haines, n.d.) Verified by
to highlight especially significant information. interview
10 The Diagnostic Aid shall provide means to filter out Evaluate patient’s history, Subject matter Not evaluated
all but the highlighted information. medical signs and symptoms experts
IDEF0 Model –
A21
11 The Diagnostic Aid shall provide means to move General (Sawyer et al. Verified by
among different pages easily. 1996) interview
User Interface Action area Source Verification
1 The Diagnostic Aid design shall be consistent with General Verified by

the way the physician conducts diagnosis. questionnaire
2 The Diagnostic Aid shall present the patient's Evaluate patient’s history, (Ginsburg, 2007) Not evaluated
personal information on all pages. medical signs and symptoms IDEF0 Model –
A21
3 The patient’s medical information shall be readily General (Haines, n.d.) Verified by
accessible from all pages. interview
4 The Diagnostic Aid shall present the history of General (Schiff & Bates, Not evaluated
present illness, history of hospitalizations and 2010)
previous illnesses, family history, and medications IDEF0 Model –
patient is taking, in chronological order. A21
5 The Diagnostic Aid shall present the patient’s vital Analyze medical signs & (Schiff & Bates, Verified by
signs and allergies. symptoms 2010) interview
IDEF0 Model –
A212
108
6 The Diagnostic Aid shall present all symptoms Identify expected outcomes of IDEF0 Model – Verified by
associated with each diagnosis in the hypothesis list. each hypothesis A2311 interview
7 The Diagnostic Aid shall present a list of suggested Prioritize hypotheses and Subject matter Verified by
diagnoses after the clinician has reviewed the identify additional information experts interview
patient's data. needed IDEF0 Model –
A232
8 The Diagnostic Aid shall utilize a standard format General (Wickens, Lee, Not evaluated
for every page. Liu, & Gordon-
Becker, 2003)
9 The Diagnostic Aid shall use coding corresponding General Sawyer et al., Not evaluated
to established conventions. (1996).
10 The Diagnostic Aid shall help the clinician keep General Sawyer et al., Not evaluated
track of the diagnostic process by showing the (1996).
current step in the process.
11 The Diagnostic Aid shall present related pieces of General Sawyer et al., Not evaluated
information close together to reduce information (1996).
access cost.
12 The Diagnostic Aid shall utilize accepted symbols, General Sawyer et al., Not evaluated
icons, colors and abbreviations. (1996).
13 The Diagnostic Aid shall provide unambiguous General Sawyer et al., Verified by
labels for objects so they can be easily read (symbol (1996). questionnaire
size, contrast, color).
14 The Diagnostic Aid shall present diagnostically General Subject matter Not verified
significant relations and trends among data. experts
15 The Diagnostic Aid shall minimize the cost of General Subject matter Verified by
retrieving information. experts questionnaire
109
16 Information shall be represented consistently across General Sawyer et al., Not evaluated
the Diagnostic Aid. (1996).
17 Symbols should be easily recognizable by the General (Wickens, Lee, Not evaluated
operator. Liu, & Gordon-
Becker, 2003)
18 The salience of cues should be consistent with their General (Wickens, Lee, Not evaluated
value. Liu, & Gordon-
Becker, 2003)
19 The Diagnostic Aid shall make the user capable of General Not evaluated
processing different cues equally.
20 The Diagnostic Aid should use constraints to General (Wickens, Lee, Verified
prevent possible errors. Liu, & Gordon-
Becker, 2003)
Debiasing Tools Action area Source Verification
1 The Diagnostic Aid shall provide means (debiasing General Not evaluated
tools) to reduce the likelihood that cognitive biases
interfere with diagnosis.
2 Design of debiasing tools shall be consistent with General Not evaluated
the rest of the Diagnostic Aid standards.
3 Subtle but clear visible feedback shall be General Verified
implemented for debiasing tools.
4 Debiasing tools shall not increase the user’s General Not evaluated
cognitive burden.
110
5 Appropriate labels shall be used to convey the General Not evaluated

functionality of the memory tools quickly.
6 Debiasing tools shall not interfere with the process General Verified
of decision making.
7 Debiasing tools shall be specifically attributed to General Verified by
different stages of diagnostic reasoning. design
8 Debiasing tools shall help the clinician to reach the General Not verified
correct diagnosis more accurately.
9 Debiasing tools shall help the clinician to reach the General Not verified
correct diagnosis more quickly.
10 The anchoring debiasing tool shall encourage the Anchoring bias IDEF0 Model – Verified with
clinician to review all the patient’s information A212 & A221 less degree of
before making a decision. confidence
11 The availability debiasing tool shall encourage the Availability bias IDEF0 Model – Not evaluated
clinician to consider the history of recent diagnoses. A212 & A221
12 The availability debiasing tool shall encourage the Availability bias IDEF0 Model – Not verified
clinician to consider hypotheses with higher A212 & A221
absolute probability rather than those easily brought
to mind.
13 The representativeness debiasing tool shall Representativeness bias IDEF0 Model – Verified with
encourage the clinician to search for inconsistencies A212 & A221 less degree of
between patient’s symptoms and the potential confidence
diagnosis.
14 The Diagnostic Aid shall inform the clinician about Representativeness bias IDEF0 Model – Not evaluated
the probability of a potential hypothesis. A212 & A221
111
15 The confirmation bias debiasing tool shall encourage Confirmation bias IDEF0 Model – Not verified
the clinician to look for more data that provide A224 & A231-
disconfirming evidence. A234 & A2312
16 The Diagnostic Aid should be able to notify the Confirmation bias IDEF0 Model – Not evaluated
clinician of absence of crucial supporting evidence A224 & A231-
to prevent confirmation bias. A234 & A2312
17 The Diagnostic Aid shall make the user capable of General Not evaluated
switching among hypotheses with minimal mental
effort.
18 The debiasing tool shall reduce the likelihood of General IDEF0 Model – Verified
physician being subject to cognitive biases. A232 & A233 &
A234
19 The Diagnostic Aid shall encourage the clinician to Overconfidence bias Not evaluated
search for more evidence to decrease the chance of
overconfidence bias.
20 The Diagnostic Aid shall encourage the clinician to Premature closure Verified
collect more data and consider more hypotheses
before making final decision.
112
Appendix 3- Diagnostic Aid Screens
Figure A.8. Final Diagnostic Aid screen example 1

113

114

115

116

117

118

119

120

121
Appendix 4 – Research Protocol Submitted to IRB
RESEARCH PROTOCOL
Sep 20, 2013
1. Protocol Title: Designing a medical diagnostic mockup that helps the clinician to improve the
quality of diagnosis.
PERSONNEL
2. Principal Investigator Kenneth H. Funk II
3. Student Researcher(s) Kiumars Zolfaghari
4. Co-investigator(s) N/A
5. Study Staff N/A
6. Investigator Qualifications
Kenneth Funk (PI) is Associate Professor in the School of Mechanical, Industrial, and
Manufacturing Engineering. Dr. Funk specializes in human factors engineering, especially in
the healthcare, military, aviation, and manufacturing domains. His primary research addresses
human performance in complex, high risk systems, like submarines, surface warships,
airplanes, and operating rooms.
Kiumars Zolfaghari (Student Researcher) has a B.Sc. in Industrial Engineering and is a M.Sc.
student in Industrial Engineering at OSU. His research focus is human factors and cognitive
psychology in designing new systems. He completed the NIH online training course
“Protecting Human Research Participants” and has done coursework projects that involved
human subjects at Oregon State University.
7. Training and Oversight

Dr. Funk will supervise the study at all times, including the recruitment of subjects, data
collection and storage, and submissions of IRB documents. He has supervised many students
conducting experiments involving human subjects and will meet with the Student Researcher
on a weekly basis to supervise all aspects of this project. Kiumars Zolfaghari developed the
methodology for this research and the PI has approved it. He has completed the online
training “Responsible Conduct of Research” offered by the Collaborative Institutional
Training Initiative (CITI). Dr. Funk will continue in his role through the completion of the
study.
FUNDING
8. Sources of Support for this project (unfunded, pending, or awarded)
This study is unfunded.
DESCRIPTION OF RESEARCH
9. Description of Research
This research aims to study the effect of medical data visualization on physicians’ diagnostic
performance through a diagnostic mockup (non-functional). The objective of this research is
to develop and validate a mockup for an iPad that represents key clinical information visually
and implements cognitive and system features to reduce cognitive errors in the diagnostic
122
process. In this regard, among the features of the mockup will be representing medical history
graphically, tracking medical tests, providing personalized feedback, and supporting the
clinician with a list of diagnoses not considered initially. The focus of this project will be 1)
to represent patient’s clinical information on a tablet computer, 2) to earmark cognitive biases
that are prevalent in the diagnostic process, and 3) to reduce the mental effort of the clinician
while making decision regarding a patient’s problem.
The products of the research include a formal model of the diagnostic process, a set of
requirements to develop a visual, interactive, and graphical aid for the diagnostic process, and
a mockup of that aid. Clinician feedback on the use of the mockup will provide knowledge
for better mockups and prototypes in the future. The mockup will also be useful for training
and quality studies.
This study is part of Kiumars Zolfaghari’s master’s thesis and it will be published as a thesis
at OSU and as journal articles in medical and human factors journals.
10. Background Justification

According to the US Centers for Disease Control and Prevention, medical error is the 8th
leading cause of death in the United States. Various factors with different natures contribute
to medical errors such as poor performance of the clinician, wrong drug treatment, failure in
the process of diagnosis, system errors, and even characteristics of the patient. More than
66% of adverse events in healthcare are preventable and diagnostic errors are contributing
factors in approximately 30% of these adverse events. However, diagnostic errors have been
largely unaddressed in patient safety activities.
According to Schiff and Bates (2010), ways of improving diagnostic accuracy should be found
other than decision-support (e.g., automated diagnosis) systems. They believe that presenting
clinical information, facilitating provider-to-provider communication, and implementing
diagnostic checklists would be more efficient ways to reduce diagnostic errors.
In this research, the development of a mockup diagnostic aid aims to find a new way to
improve patient safety through implementation of both already discussed methods plus some
new innovative techniques.
11. Multi-center Study

OSU is the only institution conducting this study.
12. External Research or Recruitment Site(s)

N/A
13. Subject Population

The subject population of this study consists of physicians, interns, residents, and other
trained medical professionals experienced enough to perform differential diagnosis. The
subject population is not restricted to any gender or ethnic group. Ideally, participants will
represent different ages, medical experience and levels of expertise in performing differential
diagnosis.
Total target enrollment is 25.
123
Inclusion and exclusion criteria:

Inclusion: Any medical expert who has finished his/her third year of studies in the field
of medicine and who has enough knowledge to perform differential diagnosis.
Exclusion: Anyone who does not have enough medical knowledge and has never been in
charge of making differential diagnosis. Non-English Speakers. Children.
Recruitment: Participants will be recruited by email (attached). Initial contact will be to
personal contacts and to Oregon medical professionals who are listed in public information
sources (telephone book, internet search). Contacts will be expanded through chain referrals;
initial contacts will be asked to forward the emailed invitation to other possible participants.
Those who respond to the invitation to participate will be contacted back to confirm that they
meet the inclusion criteria and to schedule the study.
14. Consent Process

All relevant elements of the consent are included in the Informed Consent document
(attached). The consent form will be reviewed and signed by the participant at the place of the
study before starting any activity related to this study. No audio recording will be taken prior
to this. This consent form informs the participants of the nature of the research activity and
provides them with information about the survey, the potential risks and benefits associated
with participation, and contact information for the principal investigator, and student
researcher. To ensure comprehension of the consent, the researcher will ask the following
open-ended questions to the participants:
o What questions can I answer for you?
o So that I am sure that you understand what the study involves, would you
please tell me what you think we are asking you to do?
o In your own words, can you tell me what the biggest risk to you might be if
you enroll in this study?
Signatures on the consent form. Subjects who consent to the study will sign the consent
form, which indicates the study has been explained to them, all their questions have been
answered, and that they agree to be in the study. The researcher will also sign the consent
form indicating that the study was explained to the subject, comprehension was assessed and
found to be sufficient, and the subject provided consent to participate in the study.
15. Assent Process

N/A
16. Eligibility Screening

The inclusion criteria will be presented in the recruitment email. If participants’
characteristics match the inclusion criteria and they are willing to participate, they will be
invited to schedule a time to conduct the study. Since the inclusion criteria will be explained
before obtaining informed consent, there will be no need for screening process.
17. Methods and Procedures

124
Clinical scenario
Internal medicine clinical scenarios will be used to test the effectiveness and usability of the
mockup. Three case scenarios were developed by clinical experts and will be implemented in
the mockup for test purposes. These scenarios describe different patients who suffer from
fatigue and malaise problems for different reasons. The clinician participant will interact with
the mockup, as much as it is functional, collect information, analyze the patient’s problem
and, after developing a diagnosis, evaluate how the system would fit his/her needs.
Questionnaire
Each participant will answer a questionnaire (attached) after interacting with the mockup.
This questionnaire consists of three different parts. The first part, technical questions, consists
of mostly open-ended ones and asks about the clinical details of the scenario. The next two
parts assess the participants’ level of agreement regarding usability and compliance with
design requirements by Likert-scale questions. Participants can also clarify their opinions by
writing comments.
Interview
A qualitative interview will be conducted with each participant following a written guide.
The interview questions are designed to derive more detailed information regarding the user
experience with the system. The interview will last about 10 minutes and will be audio
recorded.
Procedure
Participants will be divided to two groups randomly for the purpose of this study. After
signing the informed consent, the student researcher will introduce the mockup and train
participants to use the mockup. Questions regarding the use of the mockup will be answered.
This will be a training stage of the experiment, so the participants become familiar with how
the mockup works.
Then the participants will review the first scenario by using the mockup for 10-15 minutes
and reviewing the patient information on it. Then they will answer the questionnaire. The
student researcher will remain close to clarify any possible questions.
Then, another scenario presented in paper format will be provided to the participants.
Participants will work on the scenario in the paper format for about 10 minutes and then
answer the same technical questions with different multiple-choice answers (excluding the
usability and requirements questions). Finally they will be interviewed and asked questions
about their experience with the mockup and how they compare it with the paper presentation.
Questions will address participants’ perception of the usability and satisfaction with the
system, what they like or dislike about it, and whether the tools and features of the mockup
were helpful or not.
The second group of participants will do the paper format scenario first, but the rest of the
experiment will be in the same order and format.
The experiment will take place either at Oregon State University or at the participants' places
of employment. In the latter case, or in case the experiment is conducted during the
participants’ work hours, permission will be obtained from the participants’ employers prior
to conducting the research.
Data Analysis
Data from the Likert scale questions will be analyzed using descriptive statistics. Each
specific question will be analyzed separately, and also summed with other related items to
create a score for usability, pleasureability, and so on. We will also analyze the technical
125
questions for the paper format and the mockup format and compare the quality of results;
looking to see if the mockup format has any advantageous over paper format or not. Another
category for the analysis will be the experience of participants and their familiarity with
differential diagnosis.
18. Compensation
All the participants who complete the experiment will receive a $20 gift card.
19. Costs
There are no costs associated with this study for the participants.
20. Drugs or Biologics
N/A
21. Dietary Supplements or Food
N/A
22. Radiation
N/A
23. Biological Samples
N/A
24. Anonymity or Confidentiality
Documents, including consent forms, questionnaires, audio recordings, and notes will be
stored in a secure location (student researcher’s locked office for paper documents and
password-protected computer for digital audio files ) during the study. Each participant will
be assigned a number randomly, not associated with the participant’s name, to identify the
participant’s group. However, there is a chance that we could accidentally disclose
information that identifies a participant. All the documents from the study will only include
the participant identification number and not the participant’s name. No list linking
participant numbers to names will be kept. After the study is completed, all the paper
documents will be kept in in the PI’s locked office and the digital data will be saved to the
PI’s password-protected computer and will be kept for three years.
25. Risks
The research is considered to be minimal risk. This is not a web-based research and no
information will be collected online during the time of the study.
26. Benefits
We do not believe there are any direct benefits to individuals who participate in the study.
But, the findings of this study could be used in development of diagnostic applications,
ultimately beneficial to both clinicians and patients.
27. Assessment of Risk: Benefit ratio

The contribution of the study to the healthcare research outweighs the minimal risk. In line
with this way of thinking, fewer diagnostic errors in clinical context may be expected in the
future as a result of this work.
126
Appendix 5 – Informed Consent Form
CONSENT FORM
Project Title: Evaluating a medical diagnostic mockup designed to help

improve the quality of diagnosis by reducing the chance of
diagnostic errors.
Principal Investigator: Kenneth H. Funk II, Ph.D.
Student Researcher: Kiumars Zolfaghari
Version Date: 22 October 2013
1. WHAT IS THE PURPOSE OF THIS FORM?

This form contains information you will need to help you decide whether to be in this
study or not. Please read the form carefully and ask the study team member(s) questions
about anything that is not clear.
2. WHY IS THIS RESEARCH STUDY BEING DONE?

This study tries to find ways to reduce diagnostic errors that happen during the
differential diagnosis process, specifically, the cognitive errors that have root in the
common fallibilities of human beings. We designed a mockup application for the iPad
that presents medical information visually and has tools and features that support the
human’s memory and decision-making heuristics. The usability and effectiveness of the
mockup will be evaluated through this experiment.
Kiumars Zolfaghari is conducting this study for his Master’s thesis research.
3. WHY AM I BEING INVITED TO TAKE PART IN THIS STUDY?

You are being asked to take part in a research study. The subject population of this study
consists of physicians, interns, residents, and other trained medical professionals
experienced enough to perform differential diagnosis.
4. WHAT WILL HAPPEN IF I TAKE PART IN THIS RESEARCH STUDY?

The study activities include two phases. In one of them, we will ask you to read a one
page scenario of a patient, analyze his/her condition, make your decision about final
diagnosis(es), and answer a few questions based on your interpretation of patient’s
problem.
In the other phase you will work on another scenario represented by the diagnostic
mockup. This time, you will interact with the mockup and its features on an iPad to
finalize your decision. The same questions plus a group of questions about the
functionality of the mockup will be asked after this part. At the end, a short interview will
be performed regarding your experience with the mock up.
127
Study duration: Your participation in this study will last about one hour. The paper
scenario phase will take about 15-20 minutes and the mockup scenario will take 20-25
minutes. The final short interview will also last about 10 minutes.
Recordings: An audio recording will be made only during your interview part. You still
can enroll to participate in this study if you do not want your opinions to be audio
recorded.
______I agree to be audio recorded.

Initials
______I do not agree to be audio recorded.

Initials
Randomization: This study involves a process called randomization. Randomization

means that you are put into one of the groups by chance. It is like flipping a coin. Neither
you nor the people doing the study will choose what group you will be in. You will have
an equal chance of being placed in any group. The difference between these groups is the
order of doing a paper or a mockup scenario first.
5. WHAT ARE THE RISKS AND POSSIBLE DISCOMFORTS OF THIS STUDY?

We believe there are no foreseeable risks to participants.
6. WHAT ARE THE BENEFITS OF THIS STUDY?

This study is not designed to benefit you directly. However, your participation would be
part of an effort to improve the clinical diagnostic quality in the US medical centers.
7. WILL I BE PAID FOR BEING IN THIS STUDY?

You will receive a $20 gift card.
8. WHO WILL SEE THE INFORMATION I GIVE?

The information you provide during this research study will be kept confidential to the
extent permitted by law. Research records will be stored securely and only researchers
will have access to the records. Federal regulatory agencies and the Oregon State
University Institutional Review Board (a committee that reviews and approves research
studies) may inspect and copy records pertaining to this research. Some of these records
could contain information that personally identifies you.
Every precaution will be taken that the information you share will not be identified with
you personally. However, there is a chance that we could accidentally disclose
information that identifies you.
Only the researchers will have access to your tape recordings and the data will only be
used for data analysis. To help ensure confidentiality, your name will not be associated
with any information that you provide and if the results of this project are published, your
identity will not be made public. All data will be stored in password-protected electronic
files whose access is restricted to the researchers.
128
9. WHAT OTHER CHOICES DO I HAVE IF I DO NOT TAKE PART IN THIS

STUDY?
Participation in this study is voluntary. If you decide to participate, you are free to
withdraw at any time without penalty. You will not be treated differently if you decide to
stop taking part in the study. If you choose to withdraw from this project before it ends,
the researchers may keep information collected about you and this information may be
included in study reports.
Optional Questions: In the questionnaire and in the interview, you are free to skip any
questions that you would prefer not to answer.
10. WHO DO I CONTACT IF I HAVE QUESTIONS?

If you have any questions about this research project, please contact:
Kenneth H. Funk II, Principal Investigator

School of Mechanical, Industrial, and Manufacturing Engineering
204 Rogers Hall
Oregon State University
Corvallis, OR 97331-6001
Email: [email protected]
Phone: 541-737-2357
If you have questions about your rights or welfare as a participant, please contact the
Oregon State University Institutional Review Board (IRB) Office, at (541) 737-8008 or
by email at [email protected]
WHAT DOES MY SIGNATURE ON THIS CONSENT FORM MEAN?

Your signature indicates that this study has been explained to you, that your questions have
been answered, and that you agree to take part in this study. You will receive a copy of this
form.
Do not sign after the expiration date:
Participant's Name (printed): ____________________________________________
_________________________________________ __________________________
(Signature of Participant) (Date)
_________________________________________ __________________________
(Signature of Person Obtaining Consent) (Date)
129
Appendix 6 – Clinical Scenarios
Scenario A
A 36-year-old woman presents to her physician’s office due to 4 months of non-
productive cough and fatigue. Her medical record shows that she was referred to an
ophthalmologist because of blurred vision six month ago which was diagnosed as anterior
uveitis. She was under omeprazole 20 mg daily for suspicious gastrointestinal reflux
disease (GERD) since 1 year ago. She complains that since two month ago she has
developed dyspnea while going up stairs, which she had not before. Simultaneously she
complains of some fatigue and malaise which have been more intense during the last
month. She had weight loss of 4 Lbs. during the last 3 months but denies any fever,
chills, sweats, recent travel, or sick contacts. She was a smoker for two years and quit
smoking 8 years ago. She consumes alcohol occasionally while partying with her friends.
Her father died with the diagnosis of small cell lung cancer at the age of 67 years, and her
mother died at the age of 61 due to myocardial infarction. Her sister has been diagnosed
as diabetic and is on metformin bid. She works on the assembly line of an electronics
plant. She has allergy to peanuts, but no allergy to penicillin. She has complete
vaccination and her vaccination history is up to date. Vital signs showed T: 99 F, blood
pressure of 142/87 mmHg, heart rate: 80/min and respiratory rate of 16/min. Physical
examination reveals a young woman who seems anxious. She has tender red papules over
her shins. The patient said she first noticed the bumps when she changed oral
contraceptive pills, but assumed they would disappear. Heart examination reveals
decreased S2 in the apex, but without murmur. EKG reveals occasional PR interval
prolongation. Lung examination revealed diffusely decreased lung sound. X-ray of the
130
chest shows bilateral hilar lymphadenopathy with pulmonary infiltrates. Laboratory
findings are:
WBC count: 5100/mm3
Platelet count: 276,000/mm3
Na+: 143 mEq/L
K+: 4.3 mEq/L
CL-: 111 mEq/L
Ca2+: 14.5 mg/dL
CO2: 27 mmol/L
Blood urea nitrogen: 9 mg/dL
Creatinine: 1.1 mg/dL
Thyroid function test was normal. Parathyroid hormone was in the normal limit. PPD test
showed induration of 3 mm after 72 hours. Culture of bronchoalveolar lavage fluid was
negative.
Scenario B
A 33-year old woman with sarcoidosis presents to her primary care physician
complaining of progressive fatigue and shortness of breath over the past 2 months. She
had fatigue and some occasionally shortness of breath during last 6 months, but the recent
shortness of breath is somehow different from the old one in terms of intensity. She also
reports that her socks and shoes do not fit the way they used to for the last four months
and that she fainted a few weeks ago for the first time in many years. She had regular
131
annually appointments with an ophthalmologist, and the last visit around 5 months ago
showed normal vision, and a mild anterior uveitis. She denies any recent illness and only
takes medications (prednisolone) to control her sarcoidosis. She states that she is more
comfortable sitting than lying down. Her father died due to stroke at the age of 65, and
her mother died of congestive heart failure at the age of 67. Her sister and brother have
diabetes, and are on insulin. She had a complete vaccination history, and did not report
any serious disease during childhood. She denies any drug or food allergy. The physical
examination reveals a young lady who seems tired. Her blood pressure is 134/87 mm Hg,
respiratory rate is 17/min, plus is 96/min, and temperature is 98.9F. She has jugular
venous distension, which increases with inspiration. Heart auscultation showed a
decreased S1 and S2, but without any audible murmur. She also has decreased breath
sounds bilaterally at the bases. Chest X-ray showed bilateral perihilar lymph node
enlargement, along with the bilateral reticulonodular appearance especially in lung bases.
Laboratory chemistry showed
WBC count: 4500/mm3
Platelet count: 312,000/mm3
Na+ : 141 mEq/L
K+: 3.7 mEq/L
CL-: 111 mEq/L
Ca2+: 12.3 mg/dL
Blood urea nitrogen: 11 mg/dL
Creatinine: 0.9 mg/dL

132
All other chemistries were within the normal ranges. ECG shows decreased QRS voltage
in all the leads. An echocardiogram shows an ejection fraction of 45%, decreased to
normal end diastolic volume, normal right ventricle, and thick left ventricle.
133
Appendix 7 – Questionnaire
Phase One (Paper):
You are supposed to find the problem of a patient with the chief complaint of fatigue
through differential diagnosis. A clinical scenario will be provided for you in one page
which includes essential clinical information. Then you will decide on the most accurate
hypothesis.
Phase Two (Mockup):

During the pilot study, you will work with the iPad application and a clinical scenario for
about 10 minutes. The mockup has some specific features that you will find out about in
the pilot study. Then, you are supposed to find the problem of another patient with chief
complaint of fatigue through differential diagnosis. Essential patient’s clinical
information including medical history, signs and symptoms, laboratory results,
medications, physical examination results, and so on will be provided by the mockup in
different screens.
Please complete the following tasks:
 Start using the mockup on the iPad.
 Go through different screens of the mockup and gather information as needed.
 Compare your potential hypotheses with mockup’s suggested ones
 Decide on the most accurate diagnosis.
134
Phase one (Paper)

Questionnaire
This is a research project and for this project you are asked to complete a short
questionnaire. This questionnaire aims to gather information about your experience with
the paper format of the medical scenario you have used today. Your answers will not be
released to anyone and will remain anonymous. Your name will not be written on the
questionnaire or be kept in any other records. If you have any questions, please ask the
student researcher. Thank you for your time.
Technical Questions
1. What is is/are the chief complaint(s) of the patient?

________________________________________
2. Which vital sign(s) is/are not in the normal range, if any?

A. Temperature
B. Blood Pressure
C. Respiratory rate
D. Pulse
F. None
3. Which of these conditions are not among the patient’s medical symptoms?
A. Cough
B. Night Sweats
C. Dyspnea
D. Malaise
4. How many months did the nonproductive cough last?

_______________________________
5. What does the physical exam reveal about the patient?
A. Tender red papules over her shins

B. Sleep deprivation
135
C. Massive enlargement of liver

D. Tiredness
6. What is the WBC count?

________________________
7. Are there any abnormalities in the patient’s lab results?

________________________
8. What medication(s) is/are the patient taking?

________________________
9. Which symptom(s) started soonest?

________________________
10. What is/are your final differential diagnosis(es)?
1. Berylliosis
2. Fungal infection
3. Lymphoma
4. Sarcoidosis
5. Tuberculosis
6. Other:________________
136
Phase Two (Mockup)

Questionnaire
This is a research project and for this project you are asked to complete a short
questionnaire. This questionnaire aims to gather information about your experience with
the mockup you have used today. Your answers will not be released to anyone and will
remain anonymous. Your name will not be written on the questionnaire or be kept in any
other records. If you have any questions, please ask the student researcher. Thank you for
your time.
Technical Questions
1. What is/are the chief complaint(s) of the patient?
________________________________________
2. Which vital sign(s) is/are not in the normal range, if any?

A. Temperature
B. Blood Pressure
C. Respiratory rate
D. Pulse
F. None
3. Which of these conditions are not among the patient’s medical symptoms?
A. Breast Shortness
B. Fainting
C. Weight loss
D. Fatigue
4. How many months did swollen feet last?

_______________________________
137
5. What does the physical exam reveal about the patient?

A. Massive enlargement of liver
B. Decreased S1 and S2, without any audible murmur
C. Sleep deprivation
D. Anxiousness
6. What is the WBC count?

________________________
7. Are there any abnormalities in the patient’s lab results?
________________________
8. What medication(s) is/are the patient taking?
________________________
9. Which symptom(s) started soonest?
________________________
10. Which symptoms are most severe?
________________________
11. Which, if any, of the application’s recommended diagnoses did you not consider at
all?
12. What is/are your final diagnosis(es)?

1. Aortic stenosis
2. Cardiac tamponade
3. Hypertensive heart disease
4. Pericarditis
5. Restrictive cardiomyopathy
6. Other:________________
138
Usability Questions
The following questions are related to your experience with the system. To answer them
please imagine that the system is fully functional.
1.The application is easy to use generally.

2. I would like to use this system frequently.

3. I would be able to rely on the system.

4. The information presented on screens is easy to comprehend quickly.

5. The system is too inconsistent.

6. I would have confidence in this system.

7. It is easy to find specific information.

8. Navigation through the application is easy.

9. The presentation of patient medical history is effective.

139
10. The presentation of patient laboratory tests is effective.

11. The presentation of patient vital signs is effective.

12. The application performed the way that I expected.

13. The application would be more beneficial for cases that are more complex than the
one that I just did.
140
System Attributes Questions
The following statements are related to characteristics of the system. To answer them,
please imagine that the system is fully functional.
1. The application presents patient medical history effectively.

2. The application design is consistent with the way I think about and conduct diagnosis.
3. The labels are unambiguous and easy to understand.

4. The debiasing tools reduced the likelihood of me being subject to cognitive biases.
5. The debiasing tools helped me reach the correct diagnosis more accurately.
6. The debiasing tools helped me reach the correct diagnosis more quickly.
7. The debiasing tools interfered with the process of decision making.

8. It is NOT easy to retrieve the desired information by using this application.

9. The application benefited me to balance my speed and accuracy in order to come to a

timely and correct diagnosis.
141
10. The application helped me understand relationships and trends in the data.
11. The anchoring debiasing tool effectively encouraged me to consider other alternatives
before making a decision.
12. The availability debiasing tool effectively encouraged me to consider other

alternatives before making a decision.
13. The representativeness debiasing tool effectively encouraged me to search for

inconsistencies between the patient’s symptoms and my potential diagnosis.
14. The confirmation bias debiasing tool effectively encouraged me to look for
disconfirming data.
15. The application effectively encouraged me to collect more data and consider more
hypotheses before making my final decision.
142
Personal Questions
1. What is your job title or profession?
________________________________________________
2. How long have you been in a practicing physician?
 0 to 2 years
 2 to 5 years
 5 to 10 years
 10 to 20 years
 20 years or more
3. Does your work require performing differential diagnosis on a daily basis?
YES NO
4. What is your age?
 22-25 years old

 26-35 years old
 36-45 years old
 46-55 years old
 56 years or older
143
Appendix 8 – Frequency Table of Usability Questions 7-11 and 13
8
7
6 Participant 1
5 Participant 2
4 Participant 3
3 Participant 4
Participant 5
2
Participant 6
1
0
Question 7 Question 8 Question 9 Question 10 Question 11 Question 13
Figure A.17. Participants’ results for usability Questions 7-11 and 13

Diagnose Source

Uploaded by

Copyright:

Available Formats

Diagnose Source

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Diagnose Source

Uploaded by

Copyright:

Available Formats

AN ABSTRACT OF THE THESIS OF

Kiumars Zolfaghari for the degree of Master of Science in Industrial Engineering

Title: Medical Diagnosis: A Functional Model and Diagnostic Aid

is made it is harder to overcome the harmful effects on a patient, in contrast to other

process were explored and addressed by designing a means to present medical

of finding their potential countermeasures were also addressed. A systems

engineering approach consisting of problem analysis, modeling, requirements

designing appropriate visualization and navigation techniques for presenting medical

history records, in contrast to developing standards for gathering medical records,

develop a solution to the problem.

diagnostic process, requirements derived from the model, and information

visualization principals. The effectiveness of the Diagnostic Aid in presenting

tested. A quantitative questionnaire and a qualitative interview were used to evaluate

generation of the Diagnostic Aid.

Oregon State University

Presented November 21, 2014

Major Professor, representing Industrial Engineering

Head of the School of Mechanical, Industrial and Manufacturing Engineering

Dean of the Graduate School

Kiumars Zolfaghari, Author

Department of Statistics, who is a great teacher with a wonderful personality.

Lovingly dedicated to my Mom.

outcomes requires immediate and effective action toward patient safety.

while increasing patient satisfaction. Various factors of different natures contribute to

and diagnostic errors are contributing factors in approximately thirty percent of

patient safety activities is notable.

respectively. However, the greatest challenge is eliminating or even reducing cognitive

possibility of decreasing diagnostic errors by the means of heuristics and debiasing

methods should be further explored.

Decision-support systems are usually computerized systems that support humans in

presenting clinical information, facilitating provider-to-provider communication, and

implementing diagnostic checklists would be more efficient ways to reduce diagnostic

Decision-support systems as an additional backup is welcomed by most physicians

in increasing the quality of clinical diagnosis.

process by developing a functional model and establishing requirements for developing a

So the main objectives of the study were to:

establish the requirements for the Diagnostic Aid.

by developing a prototype of the Diagnostic Aid for a tablet computer.

 Investigate the effectiveness of the Diagnostic Aid through user experience

generations of the Diagnostic Aid.

A functional model of the diagnostic process was developed employing knowledge

model represents clinician-patient and clinician-system interactions in the process of

countermeasures could be developed. This model demonstrates a clear picture of what is

Diagnostic Aid provided potential countermeasures to diagnostic errors in the diagnostic

presentation of clinical scenarios in paper were tested through a randomized experiment

This study focuses on facilitating the diagnostic process by reducing diagnostic

requirements development are introduced. Finally, a review of information visualization

as well as its functions and attributes are presented.

2.1 Medical Diagnosis

patient—provider and receiver of care respectively—could be considered as the most

be. A successful encounter is likely if different steps of the clinician-patient encounter

work integrally. The clinician-patient encounter consists of four steps: interview,

and differential diagnosis are used interchangeably in this research.

Differential diagnosis is a systematic method of identifying the existence of an

entity among multiple alternatives. This method is a process of elimination or at least

hypotheses, evaluating hypotheses, and selecting a course of action. Defining diagnostic

clusters or biological, psychological, or sociological clinical problems which are objects

narrative to reach the potential diagnosis Sox et al. (2006) .

2.2 Diagnostic Reasoning and its Challenges

To understand the difficulties in the process of diagnosis, the process is explained