lu xu. a visual analysis of ehr flowsheet to assist
TRANSCRIPT
Lu Xu. A Visual Analysis of EHR Flowsheet to Assist Clinicians’ Interpretation of Health-related Data. A Master’s Paper for the M.S. in I.S degree. April, 2019. 32 pages. Advisor: David Gotz
This project designed and implemented visualization interface of EHR Systems based on the requirements of the doctors. The visualizations combined both clinic-reported data and patient self-reported data to provide a better representation of patient health related information for the doctor to make decisions. The visualizations highlighted the trend of values, the outliers and make it possible to compare across time and measures. The user study of ten participants suggests that the visualization interface helped them find the information in an efficient way.
Headings:
Visual Analysis
EHR
Flowsheet
Health-related Data
Patient Self-reported Data
brought to you by COREView metadata, citation and similar papers at core.ac.uk
provided by Carolina Digital Repository
A VISUAL ANALYSIS OF EHR FLOWSHEET TO ASSIST CLINICIANS’ INTEPRETATION OF HEALTH-RELATED DATA
by Lu Xu
A Master’s paper submitted to the faculty of the School of Information and Library Science of the University of North Carolina at Chapel Hill
in partial fulfillment of the requirements for the degree of Master of Science in
Information Science.
Chapel Hill, North Carolina
April 2019
Approved by
_______________________________________
David Gotz
1
Table of Contents
INTRODUCTION ................................................................................................................... 3
LITERATURE REVIEW............................................................................................................ 6
1.1 Doctor Interaction with EHR Systems ........................................................................ 6
1.2 Data Visualization ...................................................................................................... 7
1.3 Data Analysis ............................................................................................................. 7
1.4 Brief Summary ........................................................................................................... 8
METHODS.......................................................................................................................... 9
1. Data Collection............................................................................................................... 9
2. Research Design ........................................................................................................... 11
2.1 Data Analysis and Visualization ............................................................................... 11
2.2 Evaluation ................................................................................................................ 12
2.2.1 Survey ...................................................................................................................... 12
2.2.2 Semi – structured Interview..................................................................................... 13
3. Limitations ................................................................................................................... 14
SYSTEM DESIGN................................................................................................................ 15
1. User Requirement ........................................................................................................ 15
2. System Functions ......................................................................................................... 15
2.1 Measure Types ........................................................................................................ 16
2.2 Measure Display ...................................................................................................... 16
2.3 Self-reported Measures ........................................................................................... 17
USER STUDY AND ANALYSIS ................................................................................................ 19
1. Participants .................................................................................................................. 19
2. Analysis of Survey Results ............................................................................................ 20
3. Limitations and Future Work ....................................................................................... 21
CONCLUSION .................................................................................................................... 22
2
REFERENCE ...................................................................................................................... 24
APPENDIX ........................................................................................................................ 28
3
INTRODUCTION
Diabetes Mellitus is a chronic disease. Management for Diabetes is concentrated in
keeping blood glucose level close to normal. For this purpose, monitoring and control
blood glucose, weight, pulse and other vital signs to a health level are important.
The current Electronic Health Records (EHR) systems that UNC Hospital are using to
record patients’ vital signs and other measures includes two dimensional tables of
variables and value pairs. The patient’s measure values of each visit to the clinic/hospitals
are recorded in the EHR systems, and the patient self-generated health data are not
collected in the systems. The medicine records and measures are in different tables and
are not well connected with each other, so it is almost impossible for clinicians to check
the medicine records and the corresponding measures after each medicine record during
the patient visit time (around 15 min).
In this project, I am interested in data visual analysis of EHR Systems for diabetes
patients. Two types of EHR are visualized and analyzed: the clinical records and patients
self-generated data. Patient self-reported data is generally more frequent than clinical-
reported data. Patients can track their blood pressure, temperature, pulse, weight and
other measures of their body’s basic functions once or multiple times a day by a simple
wearable or fixed equipment. Clinical data, however, can only be measured/recorded
4
when patients visit clinics. If the equipment can upload data automatically and/ or user
can also type in the values, the information provided by both clinical visit data and
patient-self reported data can be much more informative for doctors to understand the
patient’s status.
The client group of the proposed visualization interface is doctors with M.D. degree, who
work in emergency as well as non-emergency environment in the hospitals or clinics, and
are using current EHR systems (such as Epic) to record patients’ health related
information.
The research question for this project is how to visualize health related measures of
diabetes patients, combined both patient self-reported and clinical reported information,
to assist with clinicians’ understanding of these data and help them to make decisions.
The following are the questions about patient self-report data and clinic-reported data
need to be figured out before the design of visualizations:
1. What measures are important for analysis and are visualized in the chart for the
clinicians to view?
2. Both patient self-reported data and clinical data have different frequency,
different categories, different ranges and different measures. Can they be
displayed in the same chart?
3. If they can be displayed in the same chart, how is the time axis divided? (use the
time point of the patient visit clinic?)
5
4. Are there some good examples to visualize them?
5. How to make unusual values obvious for clinicians to find out?
After meetings with Dr. Arlene Chung and Dr. David Gotz, we selected the measures that
most important for the diabetes patients. Dr. Chung is a UNC Lineberger Comprehensive
Cancer Center member and an Assistant Professor of Medicine and Pediatrics and
Associate Director of Health and Clinical Informatics for the UNC School of Medicine.
The variables we decied to visualize are Blood Glucose (including Glucose POC
(fingerstick) and Glucose Bld (blood draw)), Vital Signs (Weight, Blood Pressure, Pulse,
Pulse Oxygen), and Medicine. After the design and implement of the visualizations, I
recruited ten persons with M.D. degree or with B.S / B.A degrees to participate in user
experiment on clinicians and other participants to evaluate the visualization interface of
information systems that I designed and implemented, in order to find out whether the
visualizations are able to help doctors understand the patient’s health-related information.
6
LITERATURE REVIEW
There are several fields which have relevant studies, such as Electronic Health Records
(EHR), how to visualize health-related data, doctor interaction with EHR, and data
visualization. Following are the theories and opinions which are helpful and necessary to
look at.
1.1 Doctor Interaction with EHR Systems Montague (2013) mentions that EHR improves patient’s satisfaction. However, the
results show that technology-centered physicians spent the most time inputting and
extracting information, which may limit their time to interact with the technology at high
levels or recall information about the patient. For this situation, find an efficient way to
help improve the interactions between the doctors and EHR Systems are necessary.
Benjamin et al. (2018) mentions that EHRs are often perceived by users as difficult to
use. There are issues with difficult-to-read interfaces, confusing displays and iconography
that lacks consistency and intuitive meaning. These usability issues can often lead to
heavy cognitive load.
The difficulties that EHR currently have are hard-to-read interfaces, confusing displays
that lacks of intuitive meaning, and requiring more time to extract the data required.
7
Correspondingly, the information needs of the clients are simple, efficient access of the
patient health related information to help them to have a better understanding of the
patients and distinguish some outliers and abnormal patterns of the patients’ vital signs.
1.2 Data Visualization Data visualization is a very broad concept, and according to the definition by Azzam et
al. (2013), it is a process which is based on qualitative and quantitative data and results in
an image that represents the raw data, and the image is readable by viewers and supports
exploration, examination and communication of the data, while the theory of this
definition is originally from Kosara (2007).
While the technique of data visualization gets much developed and advanced, there are
still concerns and challenges of this field. Kovalerchuk (2017) studies the lossless
visualization of n-D data and considers it better than 2-D in perceptual and cognitive
abilities for discovering visual pattern, but human are limited in their ability to discover
patterns in n-D data using a naked eye, which therefore motivates the development of 2-
D visual representations of n-D data.
1.3 Evaluation of Visualization The evaluation of information visualization is necessary because the visualization
interface is designed and developed to meet the users’ requirements.
8
Carpendale (2008) mentioned that usability questions, perceptual and comprehensibility
questions are important in assessing the appropriateness of a representational encoding
and the readability of visuals.
Studies on data visualization and visual analytics in health informatics applications
commonly use task-based measurements, user feedback, and surveys with both
qualitative responses and scoring scales in a lab setting (Gotz et al, 2019). The PIOS
framework (Participants, Interventions, Comparators, Outcomes, and Study Design)
proposed by Gotz et al. also provide a framework for this project.
The evaluation methods that I applied are task-based questions, surveys about how user
feel about the visualization interface and semi-structured interview to effectiveness,
efficiency and satisfaction.
1.4 Brief Summary Current research has already discussed about the difficulties of current EHR Systems,
advantages of 2-D data visualization over n-D data visualization for human, and
evaluation of visualization approaches for health informatics, which is very informative
for the basic context of this project. But there are very few papers for visualizations of
EHR systems, especially how to visualize both clinical data and patient self-reported
data.
9
METHODS
This project includes both quantitative and qualitative methods for design and
constructing the visualization interface and user study for visualization interface
evaluation.
1. User Interview To collect the user requirements for this project, Dr. Chung who is experienced with
diabetes patients are interviewed. The questions answered include what data is needed,
where the data is from, what kinds of insights are needed by a doctor.
2. Data Data used for the visualization is mostly artificial dataset. The real patient clinical data
are not considered because it is very sensitive. The patient-self reported data is not
currently available in EHR systems, and it is the future features adding to the current
EHR systems to help inform doctors about what happens between infrequent doctor
visits, so that they are not available for this study. The examples online and Dr. Chung
who has a good knowledge of the range of the normal values helped me to improve the
artificial data quality and made it more realistic for visualization development and user
evaluation.
10
The periodic clinical data with doctor’s office was also captured in today’s modern
flowsheets, and user-generated data captured with high frequency in between visits.
The normal range of measures that included in the visualizations are as follows:
• Weight
Weight has no upper or lower limits because the weight does not change dramatically.
Thus, there are no boundaries and outliers for weight in the visualizations.
• Pulse
Normal pulse rates at rest, in beats per minute (BPM):
newborn
(0–3 months
old)
infants
(3 – 6
months)
infants
(6 – 12
months)
children
(1 – 10
years)
children over 10
years
& adults,
including seniors
well-
trained
adult
athletes
99-149 89–119 79-119 69–129 59–99 39–59
This project selected 59 and 99 (children over 10 years & adults, including seniors) as
lower and upper limit to find out outliers.
• Blood Pressure
Category systolic, mmHg diastolic, mmHg
Normal 90–119
90–129
60–79
60–84
11
• Pulse Oxygen
Oxygen saturation (SO2): Normal range is 75-100 mmHg. Below 60 mmHg indicates
needs for supplemental oxygen. Values under 90 are considered low.
• A1C
Normal Range are between 5.7 - 6.4 percent.
• Glucose
A fasting blood glucose is in between 100-125 mg/dL. Any value between 140 mg/dL
and 199 mg/dL during a two-hour 75g oral glucose tolerance test. Glucose is the same
in terms of units/scale for both POC and BLD. The only difference would be if
showing the hgA1C which is a percentage with an associated average rage of glucose
with it.
• INR
This measure is weekly to a month if done at home; done after change in medication
usually. INR below 1.1 is normal. Often INR intentionally kept higher (e.g., 2.0-3.0
for people with afib). High INR means blood will clot more slowly.
• Medication
The visualization interface added medication of each doctor visit.
3. Research Design
3.1 Data Analysis and Visualization The main contribution of this project will be a visualization interface to show both the
values of clinic-reported health-related data and patient self-reported data, so that the data
analysis will be mainly about extracting and filtering data by different measures, finding
12
out the abnormal values and trend, then visualize them in the interface. The detailed
description of system design is in the next section.
Other than cleaning the data and finding out the attributes needed for visualization, the
design of visualization/interface is also one of the essential parts, which is followed by
actually working on the code.
What’s more, after building the visualization, some analysis will also be included as
illustrating findings and trends of this visualization, mainly for the purpose of identifying
abnormal values and outliers.
3.2 Evaluation As for constructing a data visualization project, it is necessary to have a final part as
evaluation in order to test the efficiency of using this visualization. For this project I
conducted questionnaire and semi-structured interview to evaluate the design of the
interface and collect people’s opinions and suggestions while using the visualization
interface. The evaluation of visualization will be in the User Study and Analysis section.
3.2.1 Participants Subjects of the evaluation are the target users of the project – clinicians in UNC Hospital.
The size of user study group includes ten subjects. But it is difficult to recruit ten
clinicians or people with M.D. degrees, we decided to recruit at least two participants
with M.D. degree, and the rest should have B.S. or B.A. degrees or even higher education
background. The sampling method is convenience sampling. I recruit participants by
13
sending requests to the people I am able to reach and have background in medicine study
or related fields, then ask them to recommend people they can reach out to me.
3.2.2 Questionnaire
I send out a questionnaire with the visualization webpage link
(https://opal.ils.unc.edu/~luxu/ehr/interface.html) to ask participants about the general
view of the visualization interface and to see how the users feel about it and how efficient
it is when using the visualization results to make decisions. I inserted some signals in the
dataset to see whether the participants can find out them.
There are two parts in the questionnaire. The first half is for all participants, and the
second half is for participants with M.D. degree only. For each part, there are 2 to 4
comparatively easy task-based questions tasks (for example: which value do you think is
abnormal?), followed by survey questions about how participants think about this
visualization.
3.2.3 Semi – structured Interview The questionnaire about users’ understanding of the visualizations is not sufficient for
this user study, thus a semi-structured interview follows up to get more detailed feedback
for the visualizations. According to participant’s answers of questionnaire, I asked
questions such as why did they choose this answer, and what do they think about this
visualization interface and why.
14
4. Limitations To engage more real users, the visualization interface should be tested by the real user
and be tested with complex, large and real dataset.
The first limitation of this project is that it used small-size simulated dataset. There are
potentials that some important signals and patterns of the health-related information are
missing in this dataset, so that we cannot evaluate whether the visualization interface
works for such signals and patterns. My approach to use the normal range of measures to
generate artificial data, then add outliers and trend according to the samples can make the
dataset more realistic, but still it cannot be compared with the real, large-size and
complex dataset.
The second limitation is using convenient sampling method to recruit participants for user
study. The participants for the visualization interface tasks, surveys and user feedback are
mostly made up with the Master Students and the Ph.D. candidates who are real users and
are not fully representative for the entire user group.
15
SYSTEM DESIGN
The visualization of the system interface is finished in JavaScript, HTML, CSS
programming languages. The JavaScript Libraries that applied to the project include D3,
Simple Statistics, jQuery and Bootstrap.
1. User Requirement
a. The visualization interface should be easy to understand and require minimum
efforts to learn how to use and interpret it.
b. The visualizations of this project are clear and simple, without very fancy
animations.
c. The visualization interface should display all the measures mentioned above in
the data collection part.
d. It should be easy to find the outliners and abnormal trend.
2. System Functions The table of the interface has the measures taken from each clinic visit and are displayed
in the column corresponding to the date of visit. In between the two clinical visit columns
are the scatter plots of self-reported data measured between these two doctor visits.
16
Figure 1 Visualization Interface
2.1 Measure Types The measures in the table can be selected by the doctors based on the patient type. For
this visualization project particularly for diabetes patients, the measures selected are
mentioned in the data collection part. Among them are the measures of vital signs such as
weight, Blood Pressure, Pulse and measures, and for the diabetes patients specifically,
Blood Glucose. The measures displayed in the interface meet requirement c.
2.2 Measure Display The frequency of user-reported data is more frequent than clinical reported data. Thus,
between each two measures which recorded in the clinic/ hospital, there are a scatter plot
17
of the data measured by the patient themselves or the equipment the they wear. This
function applied two-dimensional visualization, which is easy to interpret for users and
satisfies the requirements a and b.
Figure 2 self-reported data and clinical-reported data
2.3 Self-reported Measures The scatter plots start and end with the clinic-reported data before and after that period.
Each circle is a data point that the patient record.
For the measures with upper and lower limits, such as Blood Pressure, Pulse, Pulse
Oxygen, AIC and Glucose, there are blue dashed lines in the chart for the limits. The
outliers beyond the upper or lower limits are in red color.
Figure 3 Self-reported data
18
The line in gray is a smoothing line connecting all the circles to emphasize how the trend
of the measure changes.
Some measures are single value, but Blood Pressure has compounded values (two
values), which include systolic blood pressure and diastolic blood pressure. The measure
with two values is displayed as scatter plots with same y-axis scales.
Figure 4 Measure of two values
Users can mouse over the circle to read more details, containing the exact value and date
time.
Figure 5 Mouseover function
The functions mentioned above meet the requirements a and d. Thus, all aspects of the
function design meet the four user requirements.
19
USER STUDY AND ANALYSIS
The user study is the evaluation part for the visualization interface of the information
system. The user study includes questionnaire of task-based questions and survey, and
semi-structured interview for each participant. I take around five minutes to walk through
the visualization interface, introducing the participant how each of the function works.
Then they are asked to answer the questionnaire made up of single-answer-selection
questions and open-ended questions. The process of finding answer for each question is
similar to the process that the doctors look into the patient’s data and find out more
insights. Afterwards I interview them about how did they answer the questionnaire and
what are their opinions for the visualizations. In general, it takes around 20 min to finish
all the user study parts. The recruit letter, questionnaire and semi-structured interview
questions are in the appendix.
1. Participants Ten participants were recruited for the questionnaires and semi-structured interviews.
Among them, two are M.D. or M.D. Candidates, four are Computer Science Ph.D.
Candidates, one is Library and Information Science Ph.D. Candidate, one is Ph.D.
Candidate in Phycology, one is Ph.D. Candidate in Education, and one is Master of the
School of Government.
20
2. Analysis of Evaluation Results Q1. Find out the Glucose value of this patient on the day of November 28, 2017.
All participants found the correct answer, which indicates that it is not difficult to learn
and find values from the scatter plot.
Q2. Is this Glucose value similar to the Glucose value in about three months, February
10, 2018?
Only 30% of participants selected the correct answer for this question. In the follow-up
interviews, I found it was mostly because the participants have different understanding of
“similar to”. This question should be more specific and provide explanation of what is the
meaning of “similar to”.
Q3. Is the trend of Glucose values during the last two clinical visits (March 4, 2018 and
May 17, 2018) similar to the trend of pulse values during the same time period?
All participants found the correct answer, which indicates that trend of the scatter plots is
not difficult to figure out.
Q4. Does the patient self-reported Glucose data match the readings from doctor’s office
during December 30, 2017 to February 2, 2018?
All participants found the correct answer, which indicates that comparison of clinic
reported data and patient self-reported data are obvious.
Q5. Was it easy to find values for a specific date?
21
Figure 6 Difficulty of finding values
Most participants (90%) think it is not difficult to find specific values.
3. Limitations and Future Work According to the question 7 (What do you find most difficult or limited about this
visualization?) and the user study part, the limitations are as follows:
• The scatter plots are overlapping each other and hard to read if the time interval in
between is very small.
• Users are not able to read all the measures at one time. Instead, they have to hover
each plot to read the value.
• The scale of y-axis is based on the observed maximum and minimum value, thus
the variance of different measures cannot be compared.
0% 10%0%
50%
40%
very hard somewhat hard neutral somewhat easy very easy
22
Future work can be focused on how to solve the overlapping problems if there are too
many values during a short period time. Also, the moved lines can be added so that it will
be easier to align the values and compare with each other.
23
CONCLUSION
The research question for this project is how to visualize health related measures of
diabetes patients, combined both patient self-reported and clinical-reported information,
to assist with clinicians’ understanding of these data and help them to make decisions.
The doctor was interviewed to collect user requirements for the visualization design. The
dataset for visualization design and user study is artificial, because the sensitivity of
patient clinical data and the absence of user self-reported data in current EHR systems.
After design and development of visualization interface, the user study including task-
based questions, surveys and semi-structured interview was conducted to test the
efficiency, effectiveness and satisfaction of the visualization interface.
Overall, the visualization interface of EHR System satisfies the requirements of the
doctors. The way it combines the clinic-reported data and patient self-reported data is
clear and easy to learn. But there are still some functions need to be added to make it
work better.
24
REFERENCE
1. Ashish K. Jha, Catherine M. DesRoches, Eric G. Campbell, Karen Donelan, Sowmya
R. Rao, Timothy G. Ferris, Alexandra Shields, Sara Rosenbaum, and David Blumenthal.
“Use of Electronic Health Records in U.S. Hospitals”, April 16, 2009, N Engl J Med
2009; 360:1628-1638, DOI: 10.1056/NEJMsa0900592
2. A. Hoerbst, E. Ammenwerth. “Electronic Health Records: A Systematic Review on
Quality Requirements”. Methods Inf Med 2010; 49(04): 320-336, DOI: 10.3414/ME10-
01-0038.
3. David Nau, “Effects of a Community Pharmacist-Based Diabetes Patient-Management
Program on Intermediate Clinical Outcome Measures”. J Manag Care Spec Pharm, 2002
Jan; 8(1):48-53.
4. Yixuan Zhang, Kartik Chanana, Cody Dunne. “IDMVis: Temporal Event Sequence
Visualization for Type 1 Diabetes Treatment Decision Support”. IEEE Transactions on
Visualization and Computer Graphics (Volume: 25, Issue: 1, Jan. 2019).
5. S. E. Inzucchi, R. M. Bergenstal, J. B. Buse, M. Diamant, E. Ferrannini, M. Nauck,
A. L. Peters, A. Tsapas, R. Wender, D. R. Matthews. “Management of hyperglycaemia in
type 2 diabetes: a patient-centered approach. Position statement of the American Diabetes
Association (ADA) and the European Association for the Study of Diabetes (EASD)”.
2012.
6. Nancy DBrener, John O.GBilly, William R, GradyM.A.b. “Assessment of factors
affecting the validity of self-reported health-risk behavior among adolescents: evidence
from the scientific literature”. Journal of Adolescent Health, Volume 33, Issue 6,
December 2003, Pages 436-457
7. Seppo Miilunpalo, Correspondence information about the author Seppo Miilunpalo,
Ilkka Vuori, Pekka Oja, Matti Pasanen, Helka Urponen. “Self-rated health status as a
health measure: The predictive value of self-reported health status on the use of physician
25
services and on mortality in the working-age population”. UKK Institute for Health
Promotion Research, Tampere, Finland. May 1997Volume 50, Issue 5, Pages 517–528.
8. Benedetto, Simone; Caldato, Christian; Bazzan, Elia; Greenwood, Darren
C; Pensabene, Virginia; et al. “Assessment of the Fitbit Charge 2 for monitoring heart
rate”. PLoS One; San Francisco Vol. 13, Iss. 2, (Feb 2018): e0192691.
DOI:10.1371/journal.pone.0192691.
9. Azzam, T., Evergreen, S., Germuth, A. A. and Kistler, S. J. (2013), Data Visualization
and Evaluation. New Directions for Evaluation, 2013: 7–32. doi:10.1002/ev.20065
10. Kosara, R. (2007, July). Visualization criticism—The missing link between
information visualization and art. Information Visualization, IV’07. 11th International
Conference (pp. 631–636). doi: 10.1109/IV.2007.2
11. Melissa Mallon (2015) Data Visualization, Public Services Quarterly, 11:3, 183-192,
DOI: 10.1080/15228959.2015.1060147
12. Olsen, Kai A. "Data Visualization." Computer Sciences, edited by K. Lee Lerner and
Brenda Wilmoth Lerner, 2nd ed., Macmillan Reference USA, 2013. Science in Context,
link.galegroup.com/apps/doc/CV2642250148/SCIC?u=unc_main&xid=495a6d14.
Accessed 6 Sept. 2017.
13. T. Munzner (2015). Visualization analysis and design. CRC Press
14. Kovalerchuk, B. (2017). Visual Cognitive Algorithms for High-Dimensional Data
and Super-intelligence Challenges. Cognitive Systems Research.
15. Nicoletta Di Blas , Mirjana Mazuran , Paolo Paolini , Elisa Quintarelli , Letizia
Tanca, Exploratory computing: a challenge for visual interaction, Proceedings of the
2014 International Working Conference on Advanced Visual Interfaces, May 27-29,
2014, Como, Italy [doi>10.1145/2598153.2600037]
16. Pomarede, D; Courtois, HM; Hoffman, Y (2017). Cosmography and Data
Visualization, PUBLICATIONS OF THE ASTRONOMICAL SOCIETY OF THE
PACIFIC, Volume 129, Issue 975
17. Wang Y, White W, Andersen E (2017). PathViewer: Visualizing Pathways through
Student Data. Proceedings of the 2017 CHI Conference on Human Factors in Computing
Systems. pp: 960-964
26
18. Logre I, Mosser S, Riveill M (2015). Composition challenges for sensor data
visualization. Companion Proceedings of the 14th International Conference on
Modularity, pp: 25-26
19. Klomklao T, Ratanarungrong P, Phithakkitnukoon S (2016). Tweets of the nation:
tool for visualizing and analyzing global tweets. Proceedings of the 2016 ACM
International Joint Conference on Pervasive and Ubiquitous Computing Adjunct, pp:
1349-1357
20. Devitt, P.(2015). Research essentials: Understanding quantitative data analysis.
Nursing Children and Young People, Volume 27, Issue 7
21. Lucko, Gunnar; Mitchell, Zane W (2010). Quantitative Research: Preparation of
Incongruous Economic Data Sets for Archival Data Analysis. Journal of Construction
Engineering and Management. 01/2010, Volume 136, Issue 1
22. Marron, J. S. and Alonso, A. M. (2014), Overview of object oriented data analysis.
Biom. J., 56: 732–753. doi:10.1002/bimj.201300072
23. Shruti Kohli and Himani Singal. 2014. Data Analysis with R. In Proceedings of the
2014 IEEE/ACM 7th International Conference on Utility and Cloud Computing (UCC
'14). IEEE Computer Society, Washington, DC, USA, 537-538.
DOI=http://dx.doi.org/10.1109/UCC.2014.81
24. Danyel Fisher. 2016. Big data exploration requires collaboration between
visualization and data infrastructures. In Proceedings of the Workshop on Human-In-the-
Loop Data Analytics (HILDA '16). ACM, New York, NY, USA, Article 16, 5 pages.
DOI: https://doi.org/10.1145/2939502.2939518
25. Peter Norvig. 2011. Internet scale data analysis. In Proceedings of the 17th ACM
SIGKDD international conference on Knowledge discovery and data mining (KDD '11).
ACM, New York, NY, USA, 2-2. DOI: https://doi.org/10.1145/2020408.2020412
26. Rahul C. Basole, Mark L. Braunstein, and Jimeng Sun. 2015. Data and Analytics
Challenges for a Learning Healthcare System. J. Data and Information Quality 6, 2-3,
Article 10 (July 2015), 4 pages. DOI: https://doi.org/10.1145/2755489
27. Richard A. Becker and John M. Chambers. 1984. Design of the S system for data
analysis. Commun. ACM 27, 5 (May 1984), 486-495.
DOI=http://dx.doi.org/10.1145/358189.358078
27
28. Tova Milo and Amit Somech. 2016. REACT: Context-Sensitive Recommendations
for Data Analysis. In Proceedings of the 2016 International Conference on Management
of Data (SIGMOD '16). ACM, New York, NY, USA, 2137-2140. DOI:
https://doi.org/10.1145/2882903.2899392
29. Shyr, J. and Spisic, D. (2014), Automated data analysis. WIREs Comput Stat, 6: 359–
366. doi:10.1002/wics.1318
30. Carpendale, S. (2008). Evaluating information visualizations. In Information
visualization (pp. 19-45). Springer, Berlin, Heidelberg.
31. Wu, D. T., Chen, A. T., Manning, J. D., Levy-Fix, G., Backonja, U., Borland, D., ...
& Kandaswamy, S. (2019). Evaluating visual analytics for health informatics
applications: a systematic review from the American Medical Informatics Association
Visual Analytics Working Group Task Force on Evaluation. Journal of the American
Medical Informatics Association, 26(4), 314-323.
28
APPENDIX
Questionnaire 1. Find out the Glucose value of this patient on the day of November 28, 2017.
____________
2. Is this Glucose value similar to the Glucose value in about three months, February 10, 2018?
Yes No
3. Is the trend of Glucose values during the last two clinical visits (March 4, 2018 and May 17, 2018) similar to the trend of pulse values during the same time period?
Yes No
4. Does the patient self-reported Glucose data match the readings from doctor’s office during December 30, 2017 to February 2, 2018?
Yes No
5. Was it easy to find values for a specific date? ----------+--------------------+-------------------+-------------------+-------------------+---------- Completely not Probably not Not sure likely Of course 6. What do you like most about this visualization?
_______________________________________________________________ 7. What do you find most difficult or limited about this visualization?
________________________________________________________________
29
The following questions are for participants with M.D. only.
8. Can you find out any outliers and unusual patterns of the values before Jan 4, 2018 which may indicate that the patient’s blood glucose will rise on February 3, 2018?
----------+--------------------+--------------------+--------------------+--------------------+---------- Completely not Probably not Not sure likely Of course
9. After the Medicine started form January 2, 2018, do you think the patient’s glucose
was under control? ----------+--------------------+--------------------+--------------------+--------------------+---------- Completely not Probably not Not sure likely Of course
10. Would you find this visualization useful if it becomes part of the Epic systems? ________________________________________________________________
11. When would you use this visualization? ________________________________________________________________
12. What would you change this visualization?
________________________________________________________________
30
Recruit letter Hello, I am doing a research project on flowsheet visualizations to help clinicians have a better understanding of diabetes’ data. Can you join me walk through the visualizations and then conduct a questionnaire and semi-structured interview, which will in total cost around 20 min? Thank you so much. Your effort will be greatly appreciated! Best regards, Lu