Mapping the Electronic Health Record: A Method to Study Display Fragmentation

Yalini Senathirajaha, Jinglu Wangb, Elizabeth Boryckic, Andrek Kushnirukc

a Department of Medical Informatics, SUNY Downstate Medical Center, Brooklyn, NY, USA, b Weill Cornell Medical School, Cornell University, New York, NY, USA

c University of Victoria, Victoria, Canada

Abstract

Electronic health records have often been criticized for poor

interaction design. One major problem is the ‘display

fragmentation problem’ i.e. the fact that conventional EHRs

require the user to view many screens and retain information

in memory or external tools rather than being able to view all

relevant information together, increasing cognitive load and

the possibility of errors and inefficiency. We describe a

method for evaluating and depicting the extent of display

fragmentation and discuss its potential uses in comparing

systems, identifying navigation pathways and information

juxtaposition, and improving EHR interaction design.

Keywords:

Electronic Health Records; User-Computer Interface;

Computer Graphics.

Introduction

Healthcare information technology (‘health IT’) and electronic

health records (EHRs) have great promise to improve care,

reduce costs, and create a ‘learning healthcare system’ in

which continuous improvement is possible by using data to

analyze which treatments are most effective. However,

optimal interaction design of such software has proven

difficult, with the potential for health IT to introduce safety

concerns.

The Institute of Medicine 2011 report [1] identifies several

concerns related to fragmented displays and the conventional

interaction approach in which information location is fixed by

the programmer and users navigate through menus. These

concerns include the mismatch between programmer

assumptions and actual work environment and the mismatch

between developer and clinician backgrounds, resulting in

unmet needs. Current displays may not reflect clinical

associations, presenting related data separately. Activities are

treated as belonging to individual clinicians, instead of to a

sociotechnical system with many intercommunicating

components and unpredictable ways [1]. Inflexible order

sequences may require providers to hold orders in mind while

navigating, and time spent on cumbersome data retrieval and

remodeling is time taken from other clinical demands [1].

Display fragmentation occurs in conventional systems when

the end user must click through and view many different

screens or parts of screens in order to view all the relevant

clinical information. This necessitates sequential viewing and

requires retaining information in memory while other parts of

the information are sought. The end user must often then serve

as the ‘de facto integrating agent’ [1], integrating information

in mind rather than using external tools. This is a known cause

of excessive cognitive load and may impair optimal clinical

reasoning performance and other aspects of clinical cognition.

Addressing the display fragmentation problem is therefore an

important task in redesigning clinical information systems for

better efficiency and cognitive support.

Prior work reveals conventional EHRs have an approximate

six-fold greater number of clicks and screen transitions

required to obtain complete information [2] than systems in

which the relevant information for a task can be included on

one screen. Conversely, appropriate information juxtaposition

can foster insight, creativity, sensemaking, and problem-

solving [3,4].

Modular composable systems

Because the display fragmentation problem arises partly from

the fact that information location is typically fixed by the

programmer, one means of addressing it is to by making

systems a) modular (i.e. pieces can be rearranged and

reassembled flexibly according to different needs, like a Lego

set), and b) composable by the end user clinician. Removing

the requirement for a programmer to reconfigure displays and

giving the clinician the ability to assemble any desired

elements together on the same screen can allow patient-

specific display of all relevant information on the same page,

reducing cognitive load and respecting clinicians’ deep

medical expertise in their choice of elements. Ease of use can

mean that drag/drop assembly of ‘objects’ is a simple means

of providing this functionality. Some other domains in which

user expertise and security/reliability are important have also

used this approach; NASA mission control tools are an

example [5].

MedWISER system

Our experimental system, MedWISER, implements a modular

composable architecture, in which elements are backed by a

controlled vocabulary [6]. Separation of the display elements

and back-end data queries allows for a javascript framework

which permits the end user to specify display of chosen

elements (such as a lab value, note or note fragment, mashup

of lab plots, RSS feed) in movable rectangular widgets. These

can also be opened to large-screen size, collapsed with only a

header showing, colored, retitled, and have other

modifications which affect display. This gives the end user

considerable power to select and arrange information elements

[7]. More advanced features are not discussed in this paper.

By permitting choice and assembly of any elements by the

clinician, these can be gathered on the same page before or

during clinical review of the patient case. This then avoids the

fragmentation problem, as repeated navigation to and viewing

of other screens is no longer necessary (or minimally

necessary). This should avoid the consequent cogntive load,

forgetting, navigation ineffiencies and other problems inherent

in our current conventional fixed systems. These typically

MEDINFO 2017: Precision Healthcare through InformaticsA.V. Gundlapalli et al. (Eds.)

© 2017 International Medical Informatics Association (IMIA) and IOS Press.This article is published online with Open Access by IOS Press and distributed under the terms

of the Creative Commons Attribution Non-Commercial License 4.0 (CC BY-NC 4.0).doi:10.3233/978-1-61499-830-3-1138

1138

present the user with a series of tabs, left-hand menus, drop-

down lists or other affordances allowing the user to find and

view different pieces of information configured as a

hierarchical tree, with higher level menus providing access to

more granular information as the tree is traversed.

One way to study and compare systems is to map the locations

of the major information elements as they occur according to

the navigation structure of the EHRs. This can help us

understand the fragmentation problem, suggest design

solutions, and compare systems. Here we describe a method of

doing this, with a few examples.

Methods

Modified Cognitive Walkthrough

Cognitive walkthrough is an expert-based, usability evaluation

method for identifying usability issues. The expert steps

through a typical task, noting all navigation actions, system

responses, and potential problems, according to well-

established heuristics and expert knowledge. We make use of

this basic technique to map information locations, creating the

navigation tree by walking through each step from the top of

the navigation tree (typically the selection of the specific

patient from a patient list) [8]. An example of a map

subsection appears below.

Sunburst Visualization

Many different visualizations of hierarchical structures exist

with advantages and disadvantages. In this technique we have

leveraged advantages of the sunburst visualization. This

visualization dispays tree structures in a circular fashion (as if

the spine of the tree were wrapped around in a

circle),facilitating display of entire trees in a single page, with

drill down (in interactive versions) to details of leaves. This

facilitates apprehension of the relationships between different

regions, and makes it easier to show navigation pathways in

relation to the overall navigation structures. It also addresses

the problem that as the number of leaves expands more space

is needed, since the deeper levels furthest from the trunk

appear in the outer concentric circles, using the greater

available space.

Microsoft Excel 2016 and other tools generate such displays

from spreadsheets or hierarchical text documents. See Figure

1. Figure 1 (b) shows the sunburst visualization for the tree

section shown in Figure 1 (a).

We use as an example the general information review that

clinicians undertake when reviewing a comprehensive patient

case, in one of the major commercial inpatient systems

(Figures 2 and 3). This generally involves the following types

of information: admission notes, laboratory results, orders,

study reports (e.g. of imaging or other studies), images (if

available), discharge summaries, primary care provider notes,

medications, demographic and insurance data, and nursing

notes (if available) as well as automated data from devices, if

applicable.

In the sunburst, sections at the same navigation level appear in

the same concentric circle; with leaves below toward the outer

rim. Items viewable together on the same screen are included

in the same block (with the names all written together in the

same sector). Items not viewable together on the same screen

Figure 1 - (a) Example tree structure subset for clinical data. The user must go to an orders tab, then select laboratory

tests, then chemistry or microbiology, then view the various test panels and their subelements. (b) Sunburst visualization

for the tree subsection shown in Figure 1(a) Numeric values may be assigned to sectors to show them at various sizes

corressponding to desired parameters. Here, sectors at the same level have equal weight.

(a) (b)

Y. Senathirajah et al. / Mapping the Electronic Health Record: A Method to Study Display Fragmentation 1139

Figure 2 - Sunburst diagram of commercial outpatient system navigation with clinical element leaves shown in black. Each concentric

circle represents a level of navigation from the top of the tree. In this case the majority of clinical element values are at least two

steps (e.g. clicks) from the origin, and are displayed separately on different screens from other clinical elements, requiring the user to

view multiple screens to get a comprehensive clinical overview.

but only sequentially as parts of separate screens accessed

from the same menu level appear as separate sectors adjacent

to each other, in the same colour as the higher menu level.

Figures 2 and 3 show the essential clinical elements coloured

in black. Figure 3 shows the extent of display fragmentation;

with each element in a separate sector if it must be navigated

to separately, permitting analysis of the EHR structure and

where elements are juxtaposed or not.

While figures shown here are relatively simple, visualization

software is capable of rendering hierarchies with thousands of

elements, with drill-down and expansion capabilities to

facilitate examination of relationships between elements,

pathway identification, and the study of subsections.

Quantification of Display Fragmentation

The fact that clinical elements appear in different subsections

of the tree can be quantified, with respect to useful parameters

such as the number of clicks or screen transitions required to

access individual elements (starting from the top of the tree, or

from intermediate points), as well as the likely numbers of

clicks/screen transitions required for typical case review.

Derivation of an expression incorporating these parameters

(including such things as requirements to scroll, or filtering

functions) could allow comparison across commercial

systems, and comparison of whether an interaction design

improvement has successfully reduced the numbers of actions

and fragmentation.

Y. Senathirajah et al. / Mapping the Electronic Health Record: A Method to Study Display Fragmentation1140

Figure 3 - Display fragmentation in a commercial inpatient system. (Clinical review elements only).

A total of 37 clicks is required to access all the elements shown in black.

Results

Initial trials of this method revealed that a majority of end

clinical elements (end leaves) of the navigation tree are

located at least two levels below the top, and that

consequently numbers of clicks and screens may reach dozens

in order to view all relevant information. This echoes

complaints by clinician organizations that ‘too many clicks’

are plaguing physicians in practice and delaying rather than

hastening good use of the EHR and expected efficiencies

[9,10].

Some systems attempt to reduce the problem by having pages

for summaries, with varying degrees of success. Two

problems identified in our preliminary work are the lack of

inclusion of important clinical information in the summary,

and the need for scrolling within elements in the summary,

with uniform presentation of blocks of data regardless of

whether the blocks are populated or not. This has the potential

of wasting space and confusing the user, who may be unaware

that the data exists elsewhere. The display fragmentation

problem still exists where the user must take an action to view

all the data on screen. Roman et al. describe the significance

of within-page and between-page navigation as an important

factor for EHR usability [11]. An example of the first problem

is an outpatient system in which the clinical summary page

includes demographic, billing, encounter and allergy data, but

not other important clinical variables.

Discussion

Comparative studies across commercial and home-grown

EHRs are under way, in order to identify the extent of

fragmentation as a problem in general, and specific design

Y. Senathirajah et al. / Mapping the Electronic Health Record: A Method to Study Display Fragmentation 1141

patterns which may contribute to it or conversely reduce

fragmentation for specific tasks. The advent of mobile EHRs

has also affected the fragmentation in both positive and

negative ways. As mobile screens typically display less

information, designers may have paid more attention to screen

flow in order to ensure correct distribution of clinical values in

a way meaningful and convenient to physicians. They are also

able to make use of design patterns such as sequential tree

menus to focus actions on immediate tasks without extraneous

data. On the other hand, if poorly done, this can increase

fragmentation, given the screen size limits and need to cater to

finger- rather than mouse-driven interaction, which typically

has smaller resolution.

The modified cognitive walkthrough method combined with

sunburst display has several advantages for data collection.

Cognitive walkthrough for the purpose of defining navigation

structures is easy to teach, and easily grasped by non-

researchers, allowing for others such as clinicians with access

to systems to be data collectors. This can facilitate broader

studies as specialized access need not be arranged. The use of

Excel or similar spreadsheets is familiar to most people, and

the branching structure, once explained, is also easy to

understand. Automatic generation of sunburst as one of the

now standard visualizations allows for easy experimentation.

Formatting options allow manual coloring of subsectors. Some

visualization tools generate the path from the top of the tree as

a part of the legend, changing interactively as the user moves

over different elements.

In preliminary studies, users are found to be mostly unfamiliar

with the sunburst (unless their particular work requires it), but

find it relatively simple to understand once its tree structure

basis is explained. Prior empirical work on comparative

information visualization of hierarchical structures has shown

sunburst visualizations to be more efficient for tasks involving

perception of the hierarchy [12], with a shorter learning curve.

Interaction Design Implications and Other Use Cases

Mapping can support composable approaches’ reduction of

fragmentation, and assist creation of safer design patterns,

such as replacements for extensive dropdowns or other

undesireable interaction features. Radial menus, for example,

leverage humans’ good capacity to distinguish small angles.

Mapping of redesigns can provide a measure of progress.

Multiple workflow paths could be overlaid on the map, as a

way of specifying efficient transitions.

EHR display fragmentation maps have other uses besides

research. Interactive sunburst visualizations can be used to

convey the navigation paths to particular elements, facilitating

instruction. It is conceivable that use of maps for comparison

of different EHRs might be used for purchasing processes, as

part of evaluation of whether the system flow can be made to

match workflow needs. In some systems, changes do not

propagate to all areas automatically, and a map could facilitate

any manual changes necessary. Development uses include to

give initial overviews to developers or researchers seeking to

improve workflow, and assessing the effect of software

modifications on navigation structures, and thus likely effects

on time, efficiency, and fit to task.

Conclusions

Display fragmentation is a critical problem for the usability of

the electronic health record, and mapping EHR navigational

structure with visualization presents useful ways to understand

and address it. In addition such maps can be useful for related

purposes such as instruction and software development. The

method will allow comparison of fragmentation across EHRs,

facilitating our understanding.

Acknowledgements

This work is funded by AHRQ 1R01HS023708-01A2.

References

[1] Committee on Patient Safety and Health IT, Technology, Health IT and

Patient Safety: Building Safer Systems for Better Care, Institute of

Medicine, 2011.

[2] Y. Senathirajah and S. Bakken, Visual clustering analysis of CIS logs to

inform creation of a user-configurable Web CIS interface, Methods Inf

Med 50 (2011):337-348.

[3] A. Kerne, E. Koh, Representing Collections as compositions to support

distributed creative cognition and situated creative learning, New Rev

Hypermedia M 13 (2007):135-162.

[4] A. Kerne, E. Koh, S. Smith, H. Choi, R. Graeber, A. Webb, Promoting

emergence in information discovery by representing collections with

composition, In: ACM SIGCHI conference on Creativity and Cognition

Proceedings, 2007.

[5] J. Trimble, T. Dayton, and M. Quinol, Putting the users in charge: A

collaborative, user composable interface for NASA's Mission Control,

In: Hawaii International Conference on System Sciences, 2012.

[6] Y. Senathirajah and S. Bakken, Architectural and usability

considerations in the development of a Web 2.0-based EHR, Stud

Health Technol Inform 143 (2009):315-321.

[7] Y. Senathirajah, D. Kaufman, and S. Bakken, The clinician in the

Driver's Seat: Part 2: Intelligent Uses of Space in a User-composable

electronic health record, J Biomed Inform 52 (2014):177-188.

[8] A.W. Kushniruk, V.L. Patel, Cognitive and usability engineering

methods for the evaluation of clinical information systems, J Biomed

Inform 37 (2004): 56-76.

[9] TooManyClicks.com, accessed December 2016.

[10] J.S. Ash, C.M. Kilo, M. Shapiro, J. Wasserman, C. McMullen, and W.

Hersh, Roadmap for Provision of Safer Healthcare Information Systems:

Preventing E-Iatrogenesis, Institute of Medicine, Washington, DC,

2010.

[11] L.C. Roman, J.S. Ancker, S.B. Johnson, Y. Senathirajah, Navigation in

the electronic health record: A review of the safety and usability

literature, J Biomed Inform 67 (2017):69-79.

[12] J. Stasko, R. Catrambone, M. Guzdial, K. McDonald, An evaluation of

space-filling information visualizations for depicting hierarchical

structures, Int J Hum-Comp St 53 (2000):663-694.

Address for correspondence

Corresponding author: Yalini Senathirajah yalini@zoho.com

Y. Senathirajah et al. / Mapping the Electronic Health Record: A Method to Study Display Fragmentation1142