Visualization of data in health care: Visualizing
uncertainty and visualization for comprehension
This research will investigate how to visually represent the
uncertainty that is inherent in many types of data and processes, and analyze
how these interactive representations are utilized. Often data, or the manner in
which it is acquired, has some type of uncertainty associated with it. This may
be due to how it is collected in that instruments have limitations, or how it is
generated in that simulations are often based on probabilities and so provide us
with stochastic data. Even data free of uncertainty will often acquire
uncertainty due to processing and viewing transformations that modify it. The
process of interpretation or generalization from specific data also has inherent
uncertainty as these processes usually contain non-deterministic mappings.
The process of clinical diagnosis in medicine is a prototypical
scenario for studying the phenomenon of uncertainty. Research has revealed that
physicians are often unaware of the uncertainty that is inherently present when
they make clinical diagnoses. This uncertainty exists largely because medical
practice involves the use of many non-invasive, but also non-definitive,
diagnostic tests that carry a risk of false positive and false negative
diagnoses. An initial planned study in this domain of visualizing diagnostic
uncertainty will focus on pulmonary embolism -- a challenging medical condition
to diagnose, because its detection is typically accomplished through the use of
non-invasive tests that have imperfect sensitivity and specificity. Furthermore,
test results are interpreted in concert with clinical estimates of probability
of disease that clinicians implicitly or explicitly combine with test results to
judge whether pulmonary embolism is present or absent. Inherent in this process is the consideration of
uncertainty in final diagnostic decisions. There are currently existing
computer-based tools in use in the Calgary Health Region (on the regional
hospital order entry system) that are designed to assist clinicians in the
difficult process of accurately diagnosing pulmonary embolism. Anecdotally,
however, those tools have several limitations that limit their use in clinical
settings.
In this planned research, W21C
team members Carpendale, McLaughlin, Ghali, Baylis, and White will lead a formal
user and task analysis of the existing computer-based tool, to assess providers'
views of the existing tool. This will then be followed by the iterative
development of an improved computer-based diagnostic tool, with the goal being
to develop a new diagnostic aid that will draw on fundamental theories of data
visualization to represent the uncertainty present in the diagnostic process.
The resulting tool will then be subjected to a usability analysis and evaluation
of the working system in simulated case settings, followed by a final phase of
formal implementation in live clinical care settings with formal evaluation of
safety and diagnostic efficacy endpoints. The ultimate objective of the planned
work in this area will be to assist clinicians in making more appropriate
clinical decisions for their patients with pulmonary embolism, recognizing that
this ‘research and development template’ applied to pulmonary embolism will also
be applicable to other diagnostically challenging conditions such as deep vein
thrombosis, endocarditis, and coronary artery disease.
In related work, Dr. Carpendale
and colleagues will undertake research into ways of visually presenting health
data for enhanced comprehension. Examples of applications for which prototype
visualizations will be tested included modified presentations of diagnostic
images (e.g. presentation of two-dimensional body images in three dimensions to
enhance comprehension), and modified visual presentations of hospital processes
such as bed planning/projections and patient flow (i.e. creating a form of
hospital ‘mission control room’ visualizing patient placement and flow from
emergency rooms, to in-patient wards, and then to discharge).
The research uniquely bridges expertise in data visualization
and cognitive processes (Carpendale, McLaughlin) and evidence-based diagnosis
(Ghali, Baylis).