IEEE Technology and Society Magazine - September 2023 - 82

Feature
article. Final rounds refined concepts with further
comparison with the literature-generated theoretical
coding for further examination. Three theoretical
conclusions were drawn to guide further
analysis of data.
Results and discussion
As results, the large number of coded segments
demonstrate clear patterns of appearance frequency
for subprinciples. Three types of visualization methods
were chosen to display the patterns: code cloud,
code intersection analysis, and analysis of the modal
verb system discussed above.
Code cloud
The frequency of occurrence or the number of
coded segments of each of the subprinciples is presented
by the visualization tool " code cloud " embedded
in MAXQDA. Larger sizes demonstrate higher
frequencies of appearance throughout all documents.
The subprinciples were colored differently based on
the categories they belonged to. The frequency of
occurrence has been impacted greatly throughout different
stages of coding, modifications of search terms,
and expansion of data pool. Figure 4a and b shows
the change in code cloud graphs from the initial stage
to the final stage of coding.
Out of all the 21 subprinciples, " Information on
Design and Deployment " demonstrates the most
occurrence
in
all documents, while
" Sampling
Bias " and " Quality Control " also possess a relatively
high frequency of occurrence. The exact number
of occurrences of each subprinciple is displayed
in Supplemental Materials (Appendix) Table B.
Design and deployment of AI (software and medical
devices) need to comply with regulatory requirements
of safety, efficacy, and accuracy and should
never harm people [6]. Contents that were coded
for this subprinciple primarily consist of broad statements
describing AI applications in healthcare and
the impacts, in general. This explains why this subprinciple
is the most " popular " one-possessing the
highest frequency of occurrence based on the code
map analysis. Interestingly, another subprinciple
" Safety and efficacy, " belonging to the same parent
principle, is also displayed with a fair amount of
weight on the graph. The significance underlying the
design and deployment of AI is the safety of its users,
and ultimately the transparency, explainability, and
intelligibility of the data. Sufficient information about
an AI application should be published and always
remain accessible to all the stakeholders involved in
its design and deployment. In this way, the quality of
operation and safety of users can be ensured. Thus,
transparency of design and deployment of AI is the
baseline of applications of intelligent healthcare systems
explaining why the subprinciple is also correlated
with many other code labels.
On the other hand, a significant increase in the
frequency of occurrence has been observed for
another two subprinciples " quality control " and
" sampling bias, " as seen in Figure 4b.
Code intersection analysis
The code model function of MAXQDA is used to
Figure 4. (a) Code cloud of all documents
(initial coding). (b) Code cloud of all
documents (final results).
82
analyze the frequency of co-occurrences of different
subprinciples. Co-occurrence, or intersection, of two
subprinciples in the MAXQDA code model, stands
for times when two coded segments belonging to different
subprinciples overlap or partially overlap with
one another. While not all subprinciples demonstrate
strong co-occurrence behavior, Figure 5 grasps
a significant portion of the intersection analysis to
best display code labels with strong intersections
and are meaningful for further analysis. Specific
numbers of co-occurrences between each pair of
IEEE Technology and Society Magazine
IEEE Technology and Society Magazine - September 2023

Table of Contents for the Digital Edition of IEEE Technology and Society Magazine - September 2023
