Evidence-based usability design principles for medication alerting … · 2018. 7. 24. · (MCBZ...

RESEARCH ARTICLE Open Access

Evidence-based usability design principlesfor medication alerting systemsRomaric Marcilly1* , Elske Ammenwerth2, Erin Roehrer3, Julie Niès4 and Marie-Catherine Beuscart-Zéphir1

Abstract

Background: Usability flaws in medication alerting systems may have a negative impact on clinical use and patientsafety. In order to prevent the release of alerting systems that contain such flaws, it is necessary to provide designersand evaluators with evidence-based usability design principles. The objective of the present study was to develop acomprehensive, structured list of evidence-based usability design principles for medication alerting systems.

Methods: Nine sets of design principles for medication alerting systems were analyzed, summarized, andstructured. We then matched the summarized principles with a list of usability flaws in order to determinethe level of underlying evidence.

Results: Fifty-eight principles were summarized from the literature and two additional principles were defined,so that each flaw was matched with a principle. We organized the 60 summarized usability design principlesinto 6 meta-principles, 38 principles, and 16 sub-principles. Only 15 principles were not matched with a usability flaw.The 6 meta-principles respectively covered the improvement of the signal-to-noise ratio, the support for collaborativeworking, the fit with a clinician’s workflow, the data display, the transparency of the alerting system, and the actionabletools to be provided within an alert.

Conclusions: It is possible to develop an evidence-based, structured, comprehensive list of usability design principlesthat are specific to medication alerting systems and are illustrated by the corresponding usability flaws. This list representsan improvement over the current literature. Each principle is now associated with the best available evidence of itsviolation. This knowledge may help to improve the usability of medication alerting systems and, ultimately, decrease theharmful consequences of the systems’ usability flaws.

Keywords: Human engineering, Usability, Alerting system, Decision support, Design

BackgroundMedication alerting systems “provide real-time notifica-tion of errors, potential hazards or omissions” related tothe prescription of medications, and thus help cliniciansto make informed decisions (nota bene: in the presentreport, a “clinician” is defined as any healthcare profes-sional who interacts with the patient; the term thereforeencompasses physicians, nurses and pharmacists) [1].These promising technologies can change prescribers’behavior by helping them avoid errors [2] and, ulti-mately, can improve the quality of the medication

management process [3]. Nonetheless, the design andthe implementation of these tools may introduce nega-tive, unforeseen side effects: poor integration into theclinical workflow [4], acceptance issues, and decreasedsafety and quality of care, for example [5]. Some of theseissues are related to the usability of the alerting systems[6]; they are caused by defects in the design of the sys-tem, i.e. usability flaws. For instance, alerts may bepoorly integrated into the workflow and may appear toolate in the decision-making process – rendering thealerting system useless [7, 8]. In other cases, the contentof the alert is either incomplete or not visible enough toadequately support a clinician’s decision making – lead-ing to incorrect clinical decisions [9]. This lack of infor-mation also increases the clinician’s cognitive load [10].Alerts may be poorly written or explained - causing

* Correspondence: [email protected]. Lille, INSERM, CHU Lille, CIC-IT / Evalab 1403 - Centre d’Investigationclinique, EA 2694, F-59000 Lille, France, Maison Régionale de la RechercheClinique, 6 rue du professeur Laguesse, 59000 Lille FranceFull list of author information is available at the end of the article

© The Author(s). 2018 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, andreproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link tothe Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Marcilly et al. BMC Medical Informatics and Decision Making (2018) 18:69 https://doi.org/10.1186/s12911-018-0615-9

http://crossmark.crossref.org/dialog/?doi=10.1186/s12911-018-0615-9&domain=pdf

http://orcid.org/0000-0002-7077-7267

mailto:[email protected]

http://creativecommons.org/licenses/by/4.0/

http://creativecommons.org/publicdomain/zero/1.0/

misunderstandings or at least creating difficulties in un-derstanding them. These cognitive issues may also leadto incorrect clinical decisions [11–13]. In summary,these and other usability flaws in the alerting systemmay have severe consequences, such as rejection of thealerting system, and incorrect clinical decisions. There-fore, the usability of an alerting system warrants specialscrutiny, with a view to avoiding usability-induced useerrors at least.To prevent the usability of alerting systems from intro-

ducing errors, usability activities (e.g. design specificationsand prototype evaluation) must be undertaken during thetechnology development process [14]. The implementa-tion of those activities requires a sound knowledge ofgood usability design principles (also known as usabilityheuristics and usability criteria). Violation of thoseprinciples may generate usability flaws in the technology.With a view to helping companies to avoid the release ofmedication alerting systems that contain unintentionalviolations of these principles, it is necessary to providedesigners and evaluators with easy access to relevant, illus-trated usability design principles and to convince them ofthe value of applying these principles to design decisions.In summary, designers and evaluators of medication alert-ing systems need to access evidence-based usability designprinciples, i.e. usability design principles that have proventheir value in practice [15]. As far as we know, the presentstudy is the first to have provided evidence-based usabilitydesign principles for medication alerting systems.Putting together a body of evidence relies on the accu-

mulation of results that demonstrate the positive valueof applying design principles. Unfortunately, publicationsin the field of usability evaluation tend to report onlynegative results, i.e. instances of usability flaws. Thisreporting bias prevents the collection of evidence toshow that applying principles is beneficial. Hence,although it is not yet possible to demonstrate the posi-tive value of applying usability design principles, it is still

possible to demonstrate the negative consequences ofviolating them.In previous research, we started to develop a usability

knowledge framework (Fig. 1; [16]). We have used thisframework to gather evidence-based usability designprinciples for medication alerting systems. In a first step,we performed a systematic review of the literature toidentify the usability flaws in medication alerting systemsused in hospital and/or primary care (active or passivealerts, and use as a standalone system or integrated intoa larger information system) [17]. In a second step, wesearched for the consequences of these flaws on users(usage problems; e.g. alert fatigue and missed informa-tion) and on the work system (negative outcomes; e.g. adecrease in effectiveness, and patient safety issues), andlinked them to their cause [6].The third step involves identifying, summarizing, and

organizing published design principles so as to avoid“reinventing” principles as far as possible. The fourthstep (in line with previous work by Nielsen [18]) seeksto match usability flaws to the usability design principlesthat could fix them and thus obtain empirical illustra-tions of the principles’ violation. The present study tack-led the third and fourth steps. The results will help toestablishing evidence in support of these principles.The present study had two objectives. Firstly, it sought

to identify and organize literature reports of usability de-sign principles for medication alerting systems in hos-pital or primary care settings into a comprehensive,structured list of design principles. Secondly, the studysought to match this list with the set of usability flawsidentified in the systematic review [17], in order to as-sess the fit between known usability flaws and knownexisting design principles and thus illustrate violations ofthese principles.

MethodsA two-step methodology was applied.

Fig. 1 Top: a graphical representation of the evidence-based usability knowledge framework. The numbering refers to the four steps, asdescribed in the text. The question marks refer to the steps tackled in the present study. Bottom: an instance of the cause-consequence chainlinking a usability flaw, a usage problem and a negative outcome (adapted from [27])

Marcilly et al. BMC Medical Informatics and Decision Making (2018) 18:69 of 17

Gathering and structuring usability design principlesWe searched peer-reviewed journals and conference pa-pers for published consensus sets of usability designprinciples for medication alerting systems (i.e. principlesthat experts in the field had agreed on). The “grey” litera-ture was excluded because the quality of the informationmay vary. Hence, we searched PubMed, Scopus, andErgonomics Abstracts databases for articles addressingboth “medication alerting systems” and “usability” topics.With this goal in mind, we used the screening and eligibil-ity assessment steps from our previous systematic review[17] to identify papers purposefully providing at least oneusability design principle dedicated to medication alertingsystems. We excluded system-specific papers providingrecommendations on improving usability because theseprinciples are not applicable to a broad range of systems.This task was updated on March 30th 2016. The literaturesearch was intended to provide an overview of publishedsets of usability design principles for medication alertingsystems, rather than being systematic and reproducible.The database search was completed by examining the in-vestigators’ personal libraries and by screening the refer-ences of the selected publications. Two investigators(MCBZ and RM) decided on the final list of publicationsby consensus.Once relevant publications had been identified, one in-

vestigator (RM) extracted all items referring to usabilitydesign principles from each publication. Next, the inves-tigator grouped together principles with similar purposesand organized them hierarchically. A second investigator(MCBZ) independently crosschecked the hierarchicalorganization of the principles. Disagreements weresolved by discussion until a consensus was reached.Lastly, the two investigators summarized the principlesthat had been grouped together.

Matching usability design principles to known usabilityflawsOne investigator (RM) checked the list of usability flawspublished in the on-line appendices of Marcilly et al. [17]against the structured list of usability design principles. Asecond investigator (MCBZ) crosschecked the results. Dis-agreements were discussed until a consensus was reached.The items referring to usability flaws were either descrip-tions of the technology’s defects observed during fieldstudies or usability tests, answers to interviews/question-naires, or users’ positive or negative comments about thecharacteristics of the technology collected during theirinteraction with the system. A usability flaw was matchedto a given usability design principle if it was an instance ofa violation of the said principle. Reciprocally, a usabilitydesign principle matched a usability flaw if the applicationof the principle stopped the flaw from occurring. If a flawdid not match any of the usability design principles, then

we considered the possible extension (broadening) of anexisting principle to other contexts so that it covered awider range of flaws. If no principles could be extended tocover the flaw, we defined a new principle.The matching process was intended to be as unequivocal

as possible, i.e. one flaw matched one principle. However, ifa given usability flaw violated several principles (e.g. atdifferent levels of granularity), we matched that flaw to themost significantly violated principles (based on our experi-ence). It should be noted that a given principle could bematched with several instances of the same flaw.Figure 2 illustrates the matching process. Both investi-

gators performed the descriptive analysis of the matches.

ResultsGathering and structuring usability design principlesWe identified 9 publications on design principles dedi-cated to medication alerting systems (Table 1).Figure 3 describes the sets of publications analyzed.

One publication (Zachariah et al. [22]) was included inboth sets; although this publication was an extension ofanother set of design principles described by Phansalkaret al. [20], it contained a few usability design principlesnot found in the original publication [20]. The publica-tion also gave a list of usability flaws detected usingheuristics. Despite the potential for self-matching bias,this publication was included because our objective wasto obtain the most comprehensive possible list of designprinciples. Moreover, it was found that virtually all firstauthors of the set of usability design principles wereco-authors of one or more studies included in the reviewof usability flaws (e.g.[1, 4, 20, 23]).A total of 345 items referring to usability design princi-

ples were extracted from the 9 publications (see Add-itional file 1: Appendix 1) and then organized. The level ofagreement between the two investigators regarding theorganization of the items was very high, with full agree-ment for 92.6% of the combinations, discussion neededfor 6%, and disagreement for 1.3%. After a consensusmeeting, the items were summarized into 58 principles.No significant inconsistencies between principles from dif-ferent publications were noticed. The summarized princi-ples displayed different granularity levels, and some weremore tangible and precise than others; they could there-fore be organized hierarchically into 6 meta-principles, 36principles, and 16 sub-principles (Fig. 4, Table 2).Overall, the 9 publications contributed to different ex-

tents to the set of summarized principles: the contribu-tions ranged from 12% [22] to 69% [24] (see Additionalfile 2: Appendix 2).The level of support for each of the summarized prin-

ciples (in terms of number of publications in which theywere found) varied: one summarized principle was sup-ported by all 9 papers (#49 “Include actionable tools


within the alert”), “Suggest, do not impose” (#42) wassupported by 8 publications, and 6 principles were foundin 1 publication. When considering the overallmeta-principles and their components (i.e. related prin-ciples and sub-principles), the level of support rangedfrom 5 publications for meta-principle E (“Make the sys-tem transparent for the user”) to 9 publications formeta-principle D (“Display relevant data within thealert”) and meta-principle F (“Include actionable toolswithin the alert”).

Matching usability design principles with known usabilityflawsThe two investigators agreed well on the matchingbetween usability flaws and usability design principles,with full agreement for 54.6% of matches, partial

agreement for 31% (agreement on the main corre-sponding principle but a need to match the flaw witha second principle), discussion needed for 13.4%, anddisagreement for 1%. Of the 58 principles, 34 directlymatched at least one instance of a usability flaw, ninewere broadened to cover a flaw, and 15 were notmatched at all (see Additional file 3: Appendix 3).Two new principles were defined so that all flawsmatched a principle (#46, provide “a description ofthe characteristics of the tools included in the alert”to users, and #57, include a “send the alert into theclinical note template” function in the alert).After the addition of the 2 new principles, the final set

comprised 60 summarized usability design principles: 6meta-principles, 38 principles, and 16 sub-principles.The 6 meta-principles were as follows:

Fig. 2 Illustration of the matching process, using meta-principle E (#44) and one of its sub-principles (#48). The usability design principles foundin the literature were summarized and organized hierarchically (left). The usability flaws identified in the systematic review were collated by topic(right). Next, the correspondence between a given type of flaw and a given summarized principle was established based on the principle’s ability(if applied) to fix the usability flaw. This correspondence is represented by a double arrow. When a usability flaw could not be fixed by any of thedesign principles in the literature, we either extended an existing principle or created a new one (single arrow). The illustration presents anextension of principle #48 (in italics)

Table 1 Main characteristics of the publications on usability design principles

First author Year Focus Method used to provide the principles

[2] Bates DW 2003 Design, implementation, monitoring Lessons learned

[19] Kuperman GJ 2007 Design, implementation Expert consensus

[4] Sittig DF 2008 Design, implementation, research Lessons learned / expert consensus

[20] Phansalkar S 2010 Usability Targeted review

[21] Pelayo S 2011 Usability Targeted review & analysis of cognitive andcollaborative tasks

[22] Zachariah M 2011 Development of a usability evaluationinstrument

Phansalkar et al.s’ review and feedback froma preliminary evaluation

[1] Horsky J 2012 Usability Targeted review

[23] Horsky J 2013 Usability Targeted review

[24] Payne T 2015 Usability Expert consensus


A. Improve the system’s signal-to-noise ratio, inorder to decrease the frequency of over-alerting. Inaddition to the drugs ordered, the alert strategyshould take into account parameters such as thepatient’s clinical context or the clinician’s specialty.Moreover, the system must provide tools tocustomize the knowledge implemented within itand to monitor alert overrides.

B. Support collaborative work, advocate a teamapproach, and make the system a team player.The alerting system must encourage collaborationbetween the healthcare professional managingmedications (e.g. physicians, pharmacists andnurses). Overall, alerts must deliver the sameinformation to all clinicians, even if additionalsupplementary data can be presented as a functionof the healthcare professional’s role. The alertingsystem must help clinicians to understand howother healthcare professionals have alreadymanaged the alert.

C. Fit with clinicians’ workflow and their mentalmodel. The alerting system must comply withclinicians’ needs and tasks. Alerts must bepresented at the right moment in the decision-making process. Only the most severe alerts mustinterrupt the users; other alerts must be displayedmore discreetly. Alerts must be concise, under-standable and consistently structured so that userscan easily find the relevant data. Once the alert hasbeen satisfied, the clinicians must be able to resumetheir tasks easily.

D. Display relevant data within the alert. Thesystem must provide clinicians with the

information needed to make informed decisions.This includes the cause of the unsafe event (themedications involved), the description of theunsafe event, the severity/priority of the event,the mechanism of the interaction, the patient’sclinical context, and evidence supporting thealert. Lastly, the system must suggest – but notimpose – a means of remedying or monitoringthe unsafe event.

E. Make the system transparent for the user. Thealerting system must help clinicians to understandwhat the system can and cannot do and how itworks, in order to prevent erroneous interpretationof its behavior. The user must have access to (i) thetypes of data that are checked, (ii) the formulas andrules applied, (iii) the list of the unsafe events thatare targeted, and (iv) a description of the alerts’levels of severity.

F. Include actionable tools within the alert. Thealert must provide several tools that help cliniciansto easily and quickly translate their alert-informedclinical decision into actions: for example, buttonsto modify/cancel/discontinue an order or overridethe alert, to order actions for monitoring an event,and to provide patient education. Other tools arerecommended for managing the alert: pulling upthe alert at a later time, sending the alert into aclinical note, removing the alert for a patient, andgaining access to the patient’s medical records.

The final list of summarized usability design principlesis given in Fig. 4. Table 2 provides a detailed version ofthe principles and corresponding flaws.

DiscussionAnswers to study questionsThe present study sought primarily to provide a specific,comprehensive, structured list of usability design princi-ples for the medication alerting systems implemented inhospital or primary care settings. The secondary object-ive was to pair this list with the set of documented us-ability flaws, assess the match between the usabilityflaws that are known and the existing design principles,obtain illustrations of the existing violations of the prin-ciples, and present evidence that not applying usabilitydesign principles may be detrimental.A total of 60 specific usability design principles for

medication alerting systems were identified andorganized hierarchically around 6 meta-principles: (A)improve the signal-to-noise ratio, (B) support collabora-tive work, (C) fit the clinicians’ workflow and theirmental model, (D) display relevant data within thealert, (E) make the system transparent for the user,and (F) include actionable tools within the alert. The

Fig. 3 Sets of papers analyzed. Left: the set of papers analyzed toestablish a structured list of usability design principles formedication alerting systems. Right: the set of papers analyzed toestablish the list of usability flaws in medication alerting systems [17]


Fig.4

Hierarchicalo

rganizationof

themeta-principles,p

rinciples,and

sub-principles

specifically

relatedto

med

icationalertin

gsystem

s.Meta-principles

aredisplayedat

thetopin

coloredbo

xes.

Principles

arepresen

tedin

thelinkedcoloredbo

xes.Sub-principles

arepresen

tedin

thebo

rder-free

areasbe

low

theprinciples


Table

2Summarized

usability

design

principles,and

descrip

tions

ofthemaincorrespo

ndingflaws.Principles

andsub-principles

arepresen

tedrespectivelyas

firstandsecond

inde

nts.Principles

that

have

been

adde

dor

extend

edto

completethematchingprocessaregivenin

italics.Theob

lique

barsindicate

theabsenceof

correspo

ndingflaws

Usabilityde

sign

principles

Summaryof

correspo

ndingflaws

Meta-princ

iple

A:

#1Im

prove

thesigna

l-to-no

iseratioby

improvingthesensitivity

andthespecificity

ofthealertin

gsystem

inorde

rto

decrease

thenu

mbe

rof

irrelevantalerts[1,4,19,20,23]

(e.g.,system

(non

-med

ical)alerts,alertswith

little

eviden

ce,low

clinicalrelevanceor

redu

ndantalerts,alertsthat

require

noaction).

Therearetoomanyalerts[8,25–29]someareredu

ndant[10,25,27,30,31]

orirrelevant[11,31],

othe

rdo

notne

edanyactio

n[29].Poten

tialeventsareover-or

unde

r-de

tected

[10]

dueto

sensitivity/spe

cificity

issues

[32]

orinapprop

riate

trigge

ringthresholds

[11].

#2Che

cktheaccu

racy

oftheinform

ationretrieve

dfro

mtheCPO

E(Com

puterized

PhysicianOrder

Entry)/EHR(ElectronicHealth

Record)[19],che

ckwhe

ther

they

are

outdated

and/or

reconciliated

[1,23].

#3Use

adeq

uate

eviden

ce-based

alertkn

owledg

ebase[19].Itshou

ldbe

regu

larly

up-dated

/maintaine

d[1,2](see

#13).

Alertsareinconsistent

with

EHRdata

[10]

espe

ciallywith

listsof

patient’sactualmed

ications

[10,

12,32,33]or

with

patient’sdiagno

ses[34].

Med

ications

interactions

high

lighted

bythealertsareun

know

nin

pharmaceuticalreferencebo

oks

[27].Kno

wledg

esupp

ortin

gthealertsisno

tup

dated[34].

#4Increa

sethevarietyof

thesourcesof

inform

ationused

inthetrigge

ringmod

el(e.g.severalallergybases[19])a

ndreconcile

multip

lesen

tries[1];whe

ndata

are

missing

(deg

rade

dcond

ition

s),the

system

mustcontinue

tofunctio

n[1].

/

#5Con

sider

temporal

dim

ension

s:intervalbe

tweendrug

s’administration[23]:

distingu

ish“now

”,“stand

ing”,and

“future”orde

rs,evolutio

nof

theun

safe

even

t:increase

theseverityof

theun

safe

even

tifitge

tsworse

[21],tim

elabtestsare

overdu

e[1,19],intervalbetweentheappearance

oftheunsafeeventandthe

administrationofdrugs.

Thealertisirrelevantbe

causetheadverseeffect

itpresen

tshapp

enstoofastto

bemanageable

[29].The

system

does

notdistingu

ishorde

rsspecified

as“now

”andthosespecified

as“fu

ture”

or“stand

ing”

[31].

#6Con

sider

patient

clinical

context[1,19,23,24]:b

esides

thespecificdrug

regimen

(s)(e.g.

dose,rou

te,d

urationof

therapy,sequ

ence

ofinitiatingco-the

rapy,

timingof

co-adm

inistration),add

patient

andlabo

ratory

data

into

theexpe

cted

interaction(e.g.age

,gen

der,bo

dyweigh

t,mitigatin

gcircum

stances,pred

ispo

sing

risks

factors,drug

serum

level,renalfun

ction,co-m

orbidity,and

previous

expe

riences).Co

nsider

thepointduringpatient’sstay

atwhich

thealertispresented.

Med

icationorde

rchecking

isno

tpatient

tailored[26,35]:alertsmay

bevalid

butno

tapplicable

topatient

clinicalcontext[10,31,36]:e.g.pregn

ancy

alertsformalepatientsandwom

enof

non-child-bearin

gage[32],nodistinctionbe

tweentrue

allergiesandside

effects[10,27].An

alertthat

issupp

osed

toappe

arthelastdayof

thestay

(which

isun

foreseeable)

appe

arsevery

day[34].

#7Con

sider

action

salread

ytake

nbytheprovide

r(e.g.d

oseadjustmen

t)[21]

andprovider-specificdata[1,23]

(e.g.clinicalspecialty:som

edrugsmay

beused

off-label,o

thersmay

bevoluntarily

duplicated

trigge

ring“dup

licateorde

rs”[19]).

Alertsappe

arwhilethecorrectiveor

mon

itorin

gactio

nshave

alreadybe

entakenby

the

physicians

[29].Som

ecorrectiveactio

nsthat

areclinicallyrelevant

areno

taccepted

bythe

system

[36].Insomespecialties,adverse

even

tsareintentional[29,32]

(e.g.p

sychiatry[32]);in

othe

r,drug

sareused

off-label(e.g.p

ediatrics[32]).Clinicians

alreadyknow

thealerts[25,29].

#8Includ

efuzzylogic-based

algorithm

s,multi-attributeutility

mod

elandfilters

into

thetrigge

ringmod

elto

change

alerts’activationwhe

ncertaincond

ition

sapply

[1,19,20].Defineapprop

riatelythresholds

totrigge

rthealertsandto

prioritizethe

alertsaccordingto

thepatient’sclinicalcontextandtheseverityof

theun

safe

even

t[1,4,20,23](see

#28).

Alertstrigge

ringthresholds

aretoolow

[26].A

lertsareno

torde

redby

severitylevel[22].

Non

-significant,or

low

incide

nce,alertsarepresen

ted[11,12,29].

#9Com

binealerts

[1]:

/

#10Re

commen

dations

mustbeco

mbined

inaco

nsistent

way

forpatients

withco

morbidities[4].

/

#11Aggreg

atelow

seve

rity

alerts

inasing

ledisplayto

bereview

edallat

once

ataconven

ient

pointin

theworkflow

[1,20,23].

Alertsareno

tgrou

pedaccordingto

theirseverity[22].

#12Su

pportmon

itoring

oftheusag

eof

alerts

andtheircustom

ization

[1,2,19,23,24]

/


Table

2Summarized

usability

design

principles,and

descrip

tions

ofthemaincorrespo


andsub-principles

arepresen

tedrespectivelyas

firstandsecond

inde

nts.Principles

that

have

been

adde

dor

extend

edto


italics.Theob

lique

barsindicate

theabsenceof

correspo

ndingflaws

(Con

tinued)

Usabilityde

sign

principles

Summaryof

correspo

ndingflaws

#13Allowexpe

rtcommitteesin

each

orga

nizatio

nto

adap

tthekn

owledg

e-ba

sean

dsugg

estio

nsforactio

nto

localp

ractices

andgu

idelines

andto

removepo

tential

errorsfro

mthebase.Customizations

shou

ldpersist

acrossversionup

grades

[19,23,24].

Alertsarein

conflictswith

localand

common

practices

[10,27,36].

#14Mon

itor

alerts’ov

erridereason

s:alertsfre

quen

tlyoverrid

denandof

little

valueshou

ldbe

considered

forremovalor

forachange

oftheirpresen

tatio

nform

at(e.g.intrusivene

ss)[1,2,19,23,24].H

owever,d

ono

telim

inateor

turn

off

relevant

alertseven

forspecialists[24].

/

#15Allo

winstitutiona

lflexibility

indetermininginterrup

tive

vs.

non-interrup

tive

alerts

[23,24].

/

Meta-princ

iple

B:

#16Su

pportthecollaborativework.

Advocateateam

approach

[24],m

akethealertin

gsystem

ateam

player

[21].

Thesystem

does

notinform

physicians

whe

ther

pharmacistsreview

theirjustificatio

nof

overrid

eand/or

findthem

useful

[10].

#17Prov

idefunc

tion

sto

supporttheteam

awaren

essof

thealertman

agem

ent

[21,24]:(see

#56).

/

#18Sign

altheavailabilityof

inform

ationto

allu

sers[21]

(evento

non-prescribing

clinicians

[24]).

/

#19Displayov

erride

reason

sen

teredby

aph

ysicianto

nurses

andph

armacistsin

orde

rto

allow

them

toun

derstand

theratio

naleforoverrid

ing[1,19,21,23,24].

Justificatio

nsandcommen

tsaredisplayedto

noon

e[37]

#20Display

alerts

toconcerne

dclinicians

andthen

tono

n-prescribingclinicians

asa

second

check[24].Redirectalertsthatdo

notconcernph

ysicians

tosupp

ortstaff[1].

Pharmacistsreceivealertsthat

concernph

ysicians

(e.g.d

ruginteraction[38]).Ph

ysicians

receive

alertsrelatedto

drug

sadministrationthat

concernnu

rses

[11,29].Ph

ysiotherapistsandnu

rses

receivealertsrelatedto

theorde

ringof

drug

swhilethey

dono

tprescribe[26].

#21Displayco

nsistently

thebasicalertco

nten

t,i.e.the

mainelem

ents

ofthe

alert,am

ongst

allclin

icians

[21,24].Non

ethe

less,the

detailedpresen

tatio

nmay

differb

ased

onclinicians’expertise(e.g.pharm

acistsmay

need

moreph

armacolog

ical

data)[1,23],on

theirrole(privileges,respon

sibilities)andon

thecontextof

use[24].

Detailsmay

bepresentedup

onrequ

est[21].

Theway

data

aredisplayedisno

tadeq

uate

forallclinicians’types

[10].

#22Sh

arepatient

clinical

inform

ationin

thealertsummaryscreen

withall

clinicians

(e.g.w

ithph

armacists)[1,4].

/

Meta-princ

iple

C:

#23Fittheclinicians’w

orkflowan

dtheirmen

talm

odel

[1,2,4,20].

Men

talm

odelim

plem

entedin

thesystem

does

notfit

users'on

e[12].

#24Displaythealertat

theap

propriatetimedu

ringthede

cision

making[1,2,24]

orlaterdu

ringthemed

icationmanagem

entprocess[19].

Alertsappearou

tof

thelogicalw

orkflow[30],eith

ertoolate

intheorderingprocess[7,8,25,31,

38,39]or

tooearly

[7,31].

#25Alertsmustbedisplaye

dan

dfully

accessible

quickly:screentransitio

ntim

emustbe

wellu

nder

asecond

[1,2],avoidscrolling

andtabs

[24].

Alertsappearance

isdelayed[12]:lags/do

wn-tim

esof

8sec.[28]up

to15

sec.[10].Clinicians

must

exploreseveralpartsof

thealertsto

getallrelevantd

ata[9]:they

mustscroll[10,11,22],explore

severaltabs[9,40],orfindinform

ationintooltips[26]becauseshortversio

nsof

alertsareno

tsufficiently

inform

ative[31].

#26Displayalerts

over

theCPO

E/EH

Rscreen

incloseproxim

ityto

therelevant

controlsanddisplays

[20,23].

Alertsareou

tsidetheregion

ofthescreen

whe

reclinicians

arelooking[26,34].


Table

2Summarized

usability

design

principles,and

descrip

tions

ofthemaincorrespo


andsub-principles

arepresen

tedrespectivelyas

firstandsecond

inde

nts.Principles

that

have

been

adde

dor

extend

edto


italics.Theob

lique

barsindicate

theabsenceof

correspo

ndingflaws

(Con

tinued)

Usabilityde

sign

principles

Summaryof

correspo

ndingflaws

#27Stream

lineusers’workflow

inrespon

seto

alerts

[22].M

aketheresolutio

nof

alertsqu

ickandeasy

(fewsteps)throug

hscreen

operations

[1,22,24]:cancelor

reset

alertsinrespon

seto

theapprop

riate

correctiveactio

n,do

notrequire

acknow

ledg

ment

beforeacorrectiveaction[20].Afterthe

alertisresolved,resum

etheworkflow[19](see

#49).

/

#28Adap

tthedisplayof

thealertto

itstype

(med

icationalerts,system

alerts)and

itsseverity[19,20,23,24].

Alerts

ofdifferent

severitylevelsandof

different

typesareno

tdistingu

ished

[22].

#29Adap

talert’sform

at(e.g.color,sym

bol)andlocatio

non

thescreen

[20,24].

Moresevere

alertsmustbe

placed

with

inthefocalreg

ionof

theuser’svisualfield

inorde

rof

impo

rtance

whileno

nsevere

alertsmustbe

placed

inside

region

s[1,20,23].Distin

guishsystem

vs.m

edicationalertmessage

s[20].

Alertsareno

tdistingu

ishe

dby

severityno

rby

type

[10,22,32].A

llalertslook

thesame[22]

#30Adap

talert’sintrusiven

ess[1,4,19,21,2324].Interrup

tivealertsshou

ldbe

reserved

forh

ighseveritywarning

andused

judiciou

sly:theyshou

ldrequ

irean

explicit

respon

se.Lessimpo

rtantalertsmustbe

displayedlessintrusively(e.g.on-demand)

asmessagesno

trequ

iring

anyactio

ns.D

onotu

sepop-up

alertsforsystem

messages.

Highriskalertsareno

tseen

whenno

tintrusive[31].O

nthecontrary,low

riskalertsthatpo

p-up

anno

yusers.

Moreover,no

n-medicationalertsaretoointrusive[30]andcontrib

uteto

desensitize

theusers[10,27].

#31Use

aco

nsistent

structured

alerttemplate

across

thevariou

ssystem

sused

bytheclinicians

[23,24].

Alertscombine

dbu

tno

tstructured

causevisualizationdifficulties

[10,12,27,34]:usersface

difficulties

tofindspecificdata

[11].The

lack

ofgu

idance

bother

usersandtheirun

derstand

ing

[30,35,41].U

sefulo

ndem

andinform

ationavailablein

EHRisno

tavailablein

thealerts[22].

#32Makethealertconcisean

dactio

nable;thede

scrip

tionof

theprob

lem

shou

ldbe

shorterthan10

words

[1,19,24,24];labelsof

button

mustb

econcise

too[1](see#49).Alertscontaintoomuchtext

orextraneo

usinform

ation[10,25,30,41].

#33Presen

tthemostcriticalinform

ationon

thetop-levelo

fthealert:theun

safe

even

t,its

causes

andits

severity[1,23,24].Th

endisplayon

-dem

and(link

ed)

inform

ationon

backgrou

ndandsecond

aryconsiderations

(con

textualinformation,

mechanism

ofinteractionandeviden

ce[19,24].Thesugg

estio

nof

actio

ncouldbe

presen

tedeither

atthetoplevelo

ron

-dem

and[20,24](see

#36).

Nodataarehigh

lighted

with

inthealert[9,10],the

alertison

aparagraphform

[41].

#34Use

consistent

term

s,ph

rases,classifications,colorsandde

finition

s(e.g.for

theseverity)[1,21,22].

/

#35Te

rminolog

yan

dmessages

shou

ldsuitlocalcon

ventions

[23]

andbe

unde

rstand

ableandno

n-am

bigu

ous[1,23].

Themessage

conveyed

bythealertisno

tun

derstand

able[10,12,40]:the

text

isam

bigu

ous

[41].Icons

used

aremisinterpreted[28,41]as

wellasbu

tton

slabe

ls[35].

Meta-princ

iple

D:

#36Displayrelevant

datawithinthealert[1,20,23,24]

(see

#33).For

therelevant

toolsto

prop

ose,seemeta-principles

F./

#37Th

ecauseof

theun

safe

even

tan

dits

characteristics(e.g.dose)[1,19,20,23].U

sethemed

icationnameas

orde

redas

wellasge

neric

drug

names

whe

niden

tifying

the

interaction[24],d

ono

tfocuson

pharmacolog

ical/therapeutic

classes[24].

Alertdo

esno

tprovideinform

ationon

why

itistrigge

red[10,12,27].

#38Th

eun

safe

even

t(potentialorcurrentlyhapp

ening)

[19,20,23,24].Dono

tuse

generic

term

(e.g.risk),preferconcrete

description[24].Present

thefrequ

ency

orincidence

oftheun

safeevent[24].

Alertdo

esno

tinform

ontheun

safe

even

t[9,10,12,27].

#39Th

eseve

rity

/priorityof

theun

safe

even

t[1,20,23,24]:use

colorcode

anda

sign

alwordto

inform

ontheseverity[20].

Alertdo

esno

tinform

ontheseverityof

theun

safe

even

t[9,10,27].

#40Th

emecha

nism

oftheinteraction(possiblyby

embe

dded

links)[23,24].

Alertdo

esno

tprovideinform

ationon

how

thecauses

oftheun

safe

even

tcond

uctto

the

unsafe

even

t[9].


Table

2Summarized

usability

design

principles,and

descrip

tions

ofthemaincorrespo


andsub-principles

arepresen

tedrespectivelyas

firstandsecond

inde

nts.Principles

that

have

been

adde

dor

extend

edto


italics.Theob

lique

barsindicate

theabsenceof

correspo

ndingflaws

(Con

tinued)

Usabilityde

sign

principles

Summaryof

correspo

ndingflaws

#41Re

levant

datasupporting

thedecision-makingprocess

andthesugg

estio

nsof

actio

n[1,4,23,24]:e.g.

contextualinform

ation,mod

ifyingandpred

ispo

sing

factors

(e.g.co-morbidityor

lab-values).Providealinkto

asummaryof

patient

clinicaldata[23].

Disp

laydata

necessaryto

interpretvalues

provided

(e.g.thresholds).

Thealertdo

esno

tprovideessentialp

atient

inform

ationfortheprescriber

[10].U

sercanlinkto

outsidesourcesof

inform

ationfro

melsewhe

rein

thesystem

,but

thereisno

linkwith

inthealert

[22].Evenwhe

nalertsprovidepatient

biolog

icalresults,the

thresholds

tointerpretthem

are

notpresen

ted[9].

#42Su

ggest,dono

tim

pose.

Makethesystem

aclinician’spartne

r[21]:p

rovide

clinicallyapprop

riate

sugg

estio

nsof

actio

nformitigatin

gthepo

tentialh

arm,d

ono

tim

pose

[1,19,20,23,24].Presen

tpo

ssibleancillary

orde

rssuch

asmon

itorin

g/surveillanceactio

ns,d

rugalternative(incl.D

oseandfre

quen

cy)and/or

orde

rmod

ificatio

nor

cancellatio

n[1,2,23,24].Makethesugg

estio

nsactio

nable[24]

(see

meta-principleF).Incase

ofmultip

lesugg

estio

ns,p

rioritizethem

andpresen

ttheircon

ditio

nsof

application[24].Justifysugg

estio

ns[21,24]:checklocally

sugg

estio

ns[24]andinclud

elinkto

institutio

n-specificgu

idelines

[19],m

akeconsensualsugg

estio

ns[1,19].M

onito

rwhether

orno

tusersfollowed

throug

hwith

thesugg

estedactio

nthey

started;ifusersfailno

tifythem

they

didno

tfinish

theactio

nthey

started[22].

Alertsdo

notprovidesugg

estio

nsof

actio

n[10,22,29,41]no

ralternativetreatm

entop

tions

[9].

Alertsprovideerrone

oussugg

estio

nsof

actio

n[35].

#43Ev

iden

cesupporting

thealert(incl.Stren

gthandsource)usingsymbo

ls/le

tter/

numbers[24].Include

alinkto

amorecompletedo

cumentatio

n(m

onog

raph

,evidence,

extend

edinform

ation,context)[1,2,19,23,24].

Alertsdo

notprovideexistin

geviden

ceor

theeviden

cethat

supp

ortsthealertispo

orand

contradict

clinicians’kno

wledg

e[10,27,29].The

alertdo

esno

tpresen

teviden

cereferences

[9].

Meta-princ

iple

E:#4

4Makethesystem

tran

sparen

tfortheuser.The

system

mustno

tbe

ablackbo

xandits

coverage

mustbe

accessibleto

itsuser

[19].Inform

userswhe

nne

cessarydata

aremissing

(e.g.severity

oftheun

safe

even

t)[24].M

akeaccessible:

Thede

cision

supp

ortisinvisibleto

users[42]:capabilitiesandlim

itatio

nsof

thesystem

alon

gwith

type

sof

data

that

arecheckedareno

tshow

nto

theusers[10,27,42].

#45Th

ealerting

algorithm

/logic/form

ulas

implemen

tedwithinthesystem

[1,20,23].

Noalertsareappe

aringafterorde

ringmed

ications

althou

ghclinicians

expe

cton

eto

comeup

forapatient

[12].The

calculationform

ulas

that

thesystem

appliesareno

tun

derstood

byclinicians

[7].

#46Ade

scriptionof

thecharacteristicsof

thetoolsinclud

edin

thealert

(e.g.d

urationof

activ

ation).

Thesystem

isno

texplicitabou

tho

wto

useandmanagealertseffectively[31]:itdo

esno

tmakeitclearthat

onecanturn

offsomealerts[10]

andforho

wmuchtim

e[36].

#47Explana

tion

son

thegradingsystem

s:levelsof

severityused

bythealertin

gsystem

(and

theirnu

mbe

rby

unsafe

even

t)[20,24],andexplanations

oftheir

classificationas

unsafe

even

ts[20]).

Thesystem

does

notexplainthelevelsof

severitythat

areused

[22].

#48Th

eev

ents

that

arech

ecke

dbythealerting

system

[20]

andthetype

and

form

atof

data

(e.g.,free-text,origin

ofthedata

(otherho

spitaldata),nam

eof

drugsvs.

Anatom

icalTherapeutic

Chem

ical-ATC

codes)

Thesystem

does

notexplainwhich

interactions

areactuallychecked[10],w

hich

patients’data

areanalyzed

[31],w

hether

orde

rsin

freetext

arechecked[10]

andwhe

ther

checking

isbased

ondrug

s’names

oron

ATC

code

s[13].

Meta-princ

iple

F:#4

9Includ

eaction

able

toolswithinthealertto

allow

clinicians

totake

actio

nsintuitively,easily

andqu

ickly[1,2,4,19–24];displaythosetoolscloseto

thesugg

estio

nsthey

arerelatedto

[23]

(see

#27and#32).The

listof

actio

nabletoolsshou

ldinclud

e:

Therearede

aden

dsin

which

clinicians

face

noreason

ableop

tions

toproceed[28];the

reare

nouseful

actio

nableop

tions

[22].

#50Mod

ifytheorde

rbe

ingen

tered(orits

dose)[1,22–24].Forinstance,p

ropo

seform

ularydrug

slists[19]forformularydrug

alerts.Alloworderingadrug

sugg

ested[23]

oranewdrug

:inthiscase,clearlystatethattheexistingdrug

willbe

discon

tinuedifthe

newon

eisfinalized

[23],openapre-po

pulatedorderin

gscreen

forthe

newdrug

[23].

Thesystem

does

notgu

ideusersforsw

itching

amed

ication[41].

#51Disco

ntinue

thepreexistingactivedrug

[22–24].

Thesystem

does

notgu

ideuser

fordiscon

tinuing

amed

ication[41].

#52Can

celthe

existin

g/ne

worde

r[1,22–24].

/


Table

2Summarized

usability

design

principles,and

descrip

tions

ofthemaincorrespo


andsub-principles

arepresen

tedrespectivelyas

firstandsecond

inde

nts.Principles

that

have

been

adde

dor

extend

edto


italics.Theob

lique

barsindicate

theabsenceof

correspo

ndingflaws

(Con

tinued)

Usabilityde

sign

principles

Summaryof

correspo

ndingflaws

#53Order

labtestsor

action

sformon

itorin

gas

justified

bythealert[1,24].

/

#54Prov

idepatient

educ

ation[24].

/

#55Delay

thealertforapred

etermined

amou

ntof

time(“sno

oze”

functio

n)[24],

allowusersto

getthealertagain.

Alertscann

otbe

pulledup

later,hind

eringalertresolutio

n[10,36].Moreo

ver,itisno

tpo

ssible

toge

tthealertagain[11].

#56Se

ndthealertto

anothe

rclinician[24].

Thesystem

does

notsupp

ortthetransm

issionof

alertsto

othersclinicians

[28].

#57Send

thealertinto

theclinical

note

template

Thesystem

does

notallow

clinicians

tosend

thealertsinto

atemplateforpatient’srecord

[28,41].

#58Ove

rridethealert(m

eaning

continue

orde

ring,

igno

ringthealert)[22–24].Most

severe

unsafe

even

tsmustbe

moredifficultto

overrid

e(e.g.req

uire

asecond

confirm

ation,or

even

nopo

ssibilityof

overrid

ing[23])thanlesssevere

ones.Alertsmust

requ

irethereason

foro

verride[24](especially

themostcriticalones,op

tionalotherwise

[23].Avoidtextentries,prop

osealistof

3–4(m

ax5items)selectablecodedreason

s;reason

smustbe

1–2wordlong

[19,23,24].

Thesystem

does

notprovideapprop

riate

optio

nsforjustifyingoverrid

es[28].The

system

isno

texplicitabou

tthene

cessity

toen

terajustificatio

n[37];m

oreo

ver,fre

e-text

entriesareno

teffectivein

theoverrid

ejustificatio

nlogic[10]

anden

terin

gdata

tojustify

overrid

esisseen

astim

ebu

rden

[10,36].

#59Allo

wprovidersto

remov

eredun

dant

alerts

forapatient

who

haspreviously

toleratedadrug

scombinatio

n(after

morethan

oneoverrid

e)or

whe

nprovidersfeel

they

have

sufficien

tpracticeandknow

ledg

eabou

tthisalertor

whe

nthealertis

outdated

foraspecificpatient

[1,23].

Thesystem

does

notprovideusersthepo

ssibility

toremoveirrelevantalertsthat

therefore

continue

toappe

ar[28].

#60Allo

wusersaccess

easily

patient’sreco

rdfrom

thealertscreen

tochange

errone

ousdata

(e.g.allergy)or

toaddne

wdata.D

ono

trequ

ireen

terin

gadditio

nal

data

inthealert[1,19].

Thesystem

asks

theusersto

enterdata

inthealertandthen

inthepatient

record,leading

towastin

gtim

eanddo

ubledo

cumen

tatio

n[28].


9 analyzed publications contributed to this list to dif-ferent extents; we consider that the collation of sev-eral sets of usability design principles found in theliterature expands the variety of topics represented ineach individual set.The match between the summarized usability design

principles and the list of documented usability flaws wasquite good: 34 principles were directly matched, and thecontext of application was extended for 9 principles.Nonetheless, 15 principles did not match any of the doc-umented usability flaws. In view of the hierarchicalorganization of the principles, some principles are alsonot matched because their meta-principle or one ormore of their sub-principles are matched - thus artifi-cially reducing the quality of the match. We also identi-fied limited gaps in the principles found in the literature;two new principles had to be created.From a qualitative point of view, a few instances of

usability flaws appear to contradict the correspondingusability design principles. For instance, some princi-ples recommend including non-prescribers (e.g. phar-macists and nurses) in the alert management process,in order to promote collaboration between healthcareprofessionals (e.g. #20). However, it has been reportedthat nurses are annoyed by medication alerts thatinterrupt their work [26]. The balance between pro-moting collaboration between healthcare professionalsand not disrupting non-prescribers’ tasks is delicate.Overall, instances of usability flaws must be used sothat the corresponding design principles are not takentoo literally.

Study strengthsThe results of the present study represent an improve-ment with respect to the current literature. We did notchange the principles extracted from the literature. Bycombining and summarizing the extracted principles,they are now clearly identified, listed, and organizedhierarchically into a comprehensive, consistent, andstructured hierarchy. Furthermore, the process ofmatching the principles to the usability flaws allows oneto identifying evidence to show that not applying theseprinciples has negative consequences. Each principle isnow associated with the best available evidence of itsviolation. As far as we know, the present study is thefirst to have drawn up this type of list.In addition to providing evidence, the matching

process also provided concrete illustrations of violationsof usability design principles. The illustrations may helppeople designing and evaluating alerting systems toidentify the “usability mistakes” that should not be madeor to catch these mistakes during the evaluation phases.In fact, the illustrations provide a clearer understandingof the design principles to be applied.

Study limitationsThe retrieval of the usability design principles mighthave biased the representativeness of the principlesand the flaws. We considered only publicationsreporting general sets of design principles, rather thanevaluations giving system-specific usability recommen-dations. Grey literature was excluded. Moreover, mostof the analysis was performed by one investigator,with a second investigator independently crosscheck-ing the results. Together, these biases might havecaused us to miss a few relevant principles. Conse-quently, the principles that we extended or created inthe present study may have already been described inother publications (e.g. as system-specific recommen-dations on usability). Likewise, some usability flawsmight have been missed during the systematic review[17] due to publication and reporting biases: it mighthave been possible to match principles not matchedin the present study with usability flaws documentedoutside our review [17]. Despite these limitations, thematch between the usability design principles and theusability flaws was quite good and ensured that theprinciples and flaws retrieved were representative.This good level of matching might be due (at least inpart) to the inclusion of Zachariah’s publication [22]and reports written by closely linked authors in bothsets of publications (i.e. the set used to establish thelist of principles and the set used to establish the listof flaws, e.g. [4, 20, 23]). In the present study, therisk of self-matching bias was considered to be ac-ceptable because our objective was to obtain the mostcomprehensive possible list of design principles andcorresponding flaws. On the contrary, not including apublication in one set because its authors had alsoworked on a publication included in the other setcould have led us to ignore relevant usability designprinciples and/or usability flaws.The frequency of appearance of the design principles

was analyzed in order to establish the level of supportfor the design principles (i.e. the number of publicationsthey were found in). However, we do not interpret thisnumber as an indicator of which principles should beprioritized. Firstly, reporting and publishing biases anddifferences in the focus of the publications analyzed mayhave biased the frequency of appearance. Even withoutthese biases, prioritizing the principles would imply thatwe are able to predict the severity of the consequencesof the related usability flaws. However, the severity de-pends on many other factors, such as the system’s otherfeatures and other flaws, and the context of use. This isone reason why most sets of design principles - whetherdeveloped for interactive systems (e.g. Nielsen’s [43] andScapin’s [44] sets) or for a specific type of technology(e.g. the ones included in our analysis) - do not prioritize


Fig.5

Norman’s“seven

stages

ofactio

n”mod

el,asappliedto

theinteractionwith

amed

icationalertin

gsystem

.Top

:Loo

pa(ontheleft;“display/read”thealert)represen

tsthecore

interaction.

Loop

b(“ackno

wledg

emen

t”)represen

tsasecond

orde

rinteraction:no

tallalertingsystem

srequ

irealertsto

beacknow

ledg

ed.U

sabilityde

sign

principles

that

could(if

applied)

improvethequ

ality

oftheinteractionarelinkedto

thecorrespo

ndingstageof

themod

el,w

henpo

ssible


principles. Design principles can be prioritized by a per-son who is aware of the alerting system’s characteristicsand context of use.In the present study, we addressed the evidence in

favor of usability design principles by examining theviolation of these principles. Evidence to suggest thatapplying design principles is beneficial has not yetbeen considered, due to reporting bias in the litera-ture. Even though our present evidence is not basedon instances of successful design, it may be convin-cing enough to persuade designers to apply usabilitydesign principles. Once researchers have begun toreport on the positive usability characteristics ofmedication alerting system, the present analysis willhave to be updated.

The significance of the present results for a userinteracting with a medication alerting systemUsability design principles are related to various compo-nents of the alerting system: the triggering model, theknowledge implemented, the cognitive model implementedin the system, the information displayed, and the tools pro-posed within the alert. Applying these usability design prin-ciples might improve the clinician-alerting systeminteraction and the collaboration between clinicians. Ac-cording to Norman’s “seven stages of action” model [45],the user’s interaction with a system encompasses twostages: the action stage translates a goal into an actionsequence, and the evaluation stage compares the changesperceived in the world with the initial goal of the action(see Fig. 5). A clinician interacts with an alerting system inorder to check the appropriateness of the prescriptions(step 1). Two “action and evaluation loops” may then bedescribed. The main loop is “display/read” the alert. Thesecond “acknowledgement” loop depends on the alertingsystem model; in some models, acknowledgment is notrequired.For the “display/read” loop (loop a, Fig. 5, left), improving

the overall usability of the alerting system by applying thewhole set of design principles may facilitate the interactionand increase the clinician’s intention to use the alerting sys-tem (step 2). More specifically, the whole “improving thesignal-to-noise ratio” meta-principle may help to improvethe relevance of the alerts, decrease alert fatigue, and thusincrease the clinician’s will to use the alerting system. Instep 3, principles such as “signal the availability of informa-tion to all users” (#18) and “display the alert at the appro-priate time” (#24) could make it easier to notice andretrieve alerts. In step 4, applying the “fit the clinician’sworkflow” meta-principle may help the clinician to displaythe alerts. Once alerts are displayed, clinicians have to readand interpret them (in steps 5 and 6). Applying the “use aconsistent structured alert template”, “display relevantdata”, and “make the system transparent” principles (#31,

#36, and #44, respectively) may make the alerts morereadable and help the clinicians to interpret them. Lastly,“extend[ing] the sources of information used in the trigger-ing model” (#4) may make it easier for clinicians to assessthe alerts’ relevance (step 7).Once alerts are interpreted, physicians may have to

acknowledge them (loop b, Fig. 5, right). Applying the“suggest - do not impose” principle (#42) may increasethe probability with which a clinician acknowledges thealert and perform corrective actions (step 2). Next,“includ[ing] actionable tools within the alerts” (#49) maymake it easier and quicker to specify and executecorrective actions (e.g. modify the order; step 3). If aphysician overrides an alert and enters the reason why,“display[ing] override reason” (#19) might help otherclinicians to interpret the alert’s acknowledgement status(step 6) and decide whether or not the alert has beenproperly assessed.In summary, applying this set of usability design princi-

ples might improve both the action and evaluation stagesof a user’s interaction with the alerting system - mainly inthe “display/read” loop but also in the “acknowledgement”loop. Some principles go beyond Norman’s model, whichrelates to an individual’s interaction with the alertingsystem and not interactions between clinicians or theclinicians’ workflow. Adhering to the “fit the clinicians’workflow” meta-principle might decrease the risk of rejec-tion. Moreover, if the “support collaborative work”meta-principle were to be applied, the alerting systemcould truly help clinicians to gain the same mental repre-sentation of the prescription being checked; this wouldhelp them to coordinate their actions and improve patientsafety.

Generalizability of the studyThe list of usability flaws used in the matching processmight increase over time, depending on whether newpublications report usability flaws. Moreover, technologyevolves rapidly, and the related principles might changeaccordingly. For instance, the principles presented hereare formulated for medication alerting systems imple-mented on laptop and/or desktop computers. However,as mobile health technologies are refined and expanded,alerting systems will be progressively installed on mobiledevices. This might modify the applicability of the us-ability design principles listed here. It will therefore beessential to update this work regularly and take accountof the latest trends and developments. However, themaintenance of this knowledge may be time-consuming,and represents a challenge for human factors specialistsin the field of medical informatics. Manufacturers shouldbe associated with this process.Some design principles insist on the need for promot-

ing collaboration between clinicians (#16) but ignore the


key person in the medication management process - thepatient. Only principle #54 mentions the patient (beingable to “provide patient education”). However, as forother information technologies, the implementation ofan alerting system changes the nature of thepatient-clinician interaction [46]. It is important to en-sure that poor usability has not damaged thepatient-clinician interaction. On the contrary, increasedusability should underpin patient-clinician discussion,empower the patient [47], and ensure that care remainspatient-centered. The current literature on the usabilityof medication alerting systems does not consider the pa-tient as a stakeholder in medication management. Futureresearch on the usability of medication alerting systemshould integrate patients as stakeholders in medicationmanagement, so as to adapt or extend usability designprinciples to their specific features.Although the structured design principles target

only medication alerting systems implemented inhospital or primary care settings, some principles maybe applied to other kinds of alerting systems. For in-stance, part of the “fit the clinicians’ workflow”meta-principle could also be applied to laboratory re-sult alerting systems. Nonetheless, the evidence thatunderpins the principles presented here is valid formedication alerting systems only.In addition to the results, the method used to build

this set of evidence-based usability design principlescould also be applied to evidence-based usability designprinciples for other kinds of technology. However, thismethod is very time-consuming, and requires in-depthknowledge of the usability of the technology in questionif the data are to be analyzed correctly.

Turning the results into a usable, practical tool fordesigners and evaluatorsThe present set of evidence-based usability design princi-ples for medication alerting systems must be made access-ible to and usable by designers and evaluators. At present,the principles are presented as a printable table (Table 2)that might not be ideal for optimal use. We intend to usethe table to develop tools that present the evidence-basedknowledge in a way that suits the needs of the various sys-tem designers and evaluators (usability experts, computerscientists, etc.) in various contexts of use (design,evaluation, procurement processes, etc.). With that aim inmind, we have started to identify the needs of medicationalerting system designers and evaluators [48]. Accordingly,we developed (i) a checklist that measure the appropriateuse of evidence-based principles in the design of medica-tion alerting systems, and (ii) a set of interactive designinstructions illustrated by visual representations of goodand bad usability practices, in order to help designers makeinformed design decisions.

This list of usability design principles should help de-signers to make evidence-based usability design decisions.Nonetheless, and even though we believe that the list ishelpful, it is not intended to be used as a stand-alone sys-tem or to replace the requirement for expertise in usabilityand design. Firstly, the present list does not include gen-eral design principles for unspecified interactive systems;it must therefore be used in combination with sets ofgeneral usability design principles for interactive systems(e.g. [43, 44]). Secondly, several principles require insightsinto the users’ cognitive tasks and their decision-makingprocesses in order to adjust (for instance) an alert’s formatand the moment at which it appears (e.g. #24). Hence,work system and cognitive work analyses [49] must beperformed so the principles are applied in an optimal way.Thirdly, principles moderate each other; they must not beapplied alone or in an unquestioning manner. Humanfactors specialists and designers must use their expertiseto determine which principles must be applied and howthey must be applied, given the characteristics of the aler-ting system and the setting in which it is implemented in(hospital vs. primary care, for example). In summary, thisstructured list of usability design principles must be usedas a support for expertise and not as a substitute for it.Applying some of the principles listed here may

present specific technical and organizational challengeswhen seeking to tailor alerts. For instance, the “prioritizethe alerts according to patient’s clinical context and theseverity of the unsafe event” (#8) principle requiresaccess to valid data on the patient’s clinical context, stay,and treatment. However, these data are often not stan-dardized or structured enough to be used in the alertingsystem’s set of rules [50]. Further research is needed toovercome these challenges.Ultimately, presenting designers and evaluators with

evidence-based knowledge may help to decrease theoccurrence of unforeseen and potentially harmfulusability-induced use errors. Nonetheless, one must beaware that improving the usability of an existing system orensuring that the usability of a system under developmentis optimal is no guarantee of success. Other issues arisingduring the development of a medication alerting system(e.g. an error-ridden knowledge base, a poor implementa-tion process, unsuitable settings, etc.) can ruin even opti-mal levels of usability. Even though it is necessary toconsider usability during the design, development, andevaluation of medication alerting systems, one must neverneglect the relevant technical, social, and managerial fac-tors that also contribute to the system’s success or failure.

ConclusionsIn the present study, we developed an evidence-based,structured, specific, comprehensive list of usability designprinciples for medication alerting systems, and then


illustrated them with the corresponding usability flaws. Thislist should help designers and usability experts to gain abetter understanding of usability design principles. We ex-pect that the list can be used during the design and evalu-ation processes of medication alerting systems, in order toprevent usability issues that could have a counterproductiveimpact on clinicians (e.g. alert fatigue) and potentiallyharmful outcomes for patients (e.g. errors in medicationdosing). Although operational barriers may complicate thedeployment and maintenance of the evidence-based usabil-ity design principles presented in the present study, our re-sults show that the approach is feasible. Indeed, ourapproach could be transferred to other health informationtechnologies for the generation of specific lists ofevidence-based usability design principles. In this way, de-signers and evaluators could be provided with tools to helpthem avoid usability design issues in health informationtechnology and thus decrease the likelihood of unforeseenand potentially harmful usability-induced use errors.

Additional files

Additional file 1: Appendix 1. List of usability design principlesidentified in the 9 papers and the corresponding usability design principlessummarized (for definitions, please refer to Table 2). (DOCX 248 kb)

Additional file 2: Appendix 2. The 9 papers’ contributions to thesummarized principles. Crosses show that a given principle is mentioned in apaper. The right-hand-most column gives the number of papers mentioning agiven principle. The bottom two rows present the number of principles men-tioned by each paper and the proportion of the full list of principles mentionedby each paper. It should be noted that the percentages are based on the 58principles summarized in step 1. Principles #46 and #57 were created after thematching process and therefore were not included here. (DOCX 56 kb)

Additional file 3 Appendix 3. Results of the matching betweeninstances of usability flaws (from Marcilly et al. [17]) and the usabilitydesign principles summarized in the present study. (DOCX 82 kb)

AbbreviationsATC: Anatomical Therapeutic Chemical; CPOE: Computerized Physician OrderEntry; EHR: Electronic Health Record

AcknowledgementsThe authors would like to thank the staff at the University of Lille 2 library fortheir very efficient work in retrieving the required publication. The authorwould like to thank Melissa Baysari for her feedback on the wording of Table2, Emmanuel Castets and Pierre-François Gautier for the figures’ design. Fi-nally, the authors would like to thank the reviewers and editors for their con-structive comments.

Availability of data and materialsThe dataset(s) supporting the conclusions of this article is(are) includedwithin the article (and its additional file(s)).

Authors’ contributionsRM designed the study, retrieved the data, performed the analysis and wrotethe paper. EA helped to design the study, provided methodological supportand supported the writing of the paper by reading it several times andproviding advice to improve the report of the study. ER supported thewriting of the paper by reading it several times and validating it.Additionally, ER checked English spelling and grammar. JN provided amethodological support and supported the writing of the paper by readingit several times and providing advice to improve the report of the study.MCBZ retrieved the data, performed the analysis and supported the writing

of the paper by reading it several times and providing advice to improve thereport of the study.All authors approved the present version of the paper.

Ethics approval and consent to participateNot applicable.

Competing interestsThe authors declare that they have no competing interests.

Publisher’s NoteSpringer Nature remains neutral with regard to jurisdictional claims inpublished maps and institutional affiliations.

Author details1Univ. Lille, INSERM, CHU Lille, CIC-IT / Evalab 1403 - Centre d’Investigationclinique, EA 2694, F-59000 Lille, France, Maison Régionale de la RechercheClinique, 6 rue du professeur Laguesse, 59000 Lille France. 2Institute ofMedical Informatics, UMIT – University for Health Sciences, MedicalInformatics and Technology, 6060 Hall in Tirol, Austria. 3eHealth ServicesResearch Group, School of Engineering and ICT, University of Tasmania,Private Bag 87, Hobart, Tasmania 7001, Australia. 4General Electric HealthcarePartners, 92772, Boulogne Billancourt cedex, France.

Received: 22 June 2017 Accepted: 23 May 2018

References1. Horsky J, Schiff GD, Johnston D, Mercincavage L, Bell D, Middleton B.

Interface design principles for usable decision support: a targeted review ofbest practices for clinical prescribing interventions. J Biomed Inform. 2012;45:1202–16.

2. Bates DW, Kuperman GJ, Wang S, Gandhi T, Kittler A, Volk L, Spurr C,Khorasani R, Tanasijevic M, Middleton B. Ten commendments for effectiveclinical decision support: making the practice of evidence-based medicine areality. J Am Med Inform Assoc. 2003;10:523–30.

3. Jaspers MW, Smeulers M, Vermeulen H, Peute LW. Effects of clinicaldecision-support systems on practitioner performance and patientoutcomes: a synthesis of high-quality systematic review findings. J Am MedInform Assoc. 2011;18:327–34.

4. Sittig DF, Wright A, Osheroff JA, Middleton B, Teich JM, Ash JS, Campbell E,Bates DW. Grand challenges in clinical decision support. J Biomed Inform.2008;41:387–92.

5. Campbell EM, Guappone KP, Sittig DF, Dykstra RH, Ash JS. Computerizedprovider order entry adoption: implications for clinical workflow. J GenIntern Med. 2009;24:21–6.

6. Marcilly R, Ammenwerth E, Roehrer E, Pelayo S, Vasseur F, Beuscart-ZephirMC. Usability flaws in medication alerting systems: impact on usage andwork system. Yearb Med Inform. 2015;10:55–67.

7. Horsky J, Kaufman DR, Patel VL. Computer-based drug ordering: evaluationof interaction with a decision-support system. Stud Health Technol Inform.2004;107:1063–7.

8. Ash JS, Sittig DF, Campbell EM, Guappone KP, Dykstra RH. Someunintended consequences of clinical decision support systems. AMIA AnnuSymp Proc. 2007:26–30.

9. Duke JD, Bolchini DA. Successful model and visual design for creatingcontext-aware drug-drug interaction alerts. AMIA Annu Symp Proc. 2011:339–48.

10. Russ AL, Zillich AJ, McManus MS, Doebbeling BN, Saleem JJ. Prescribers'interactions with medication alerts at the point of prescribing: a multi-method, in situ investigation of the human-computer interaction. Int J MedInform. 2012;81:232–43.

11. van der Sijs H, van Gelter T, Vulto A, Berg M, Aarts J. Understanding handlingof drug safety alerts: a simulation study. Int J Med Inform. 2010;79:361–9.

12. Russ AL, Saleem JJ, McManus MS, Zillich AJ, Doebbling BN. Computerizedmedication alerts and prescriber mental models: observing routine patientcare. Proc Human Factors Ergon Soc Annu Meet. 2009;53:655–9.

13. Hartmann Hamilton AR, Anhoj J, Hellebek A, Egebart J, Bjorn B, Lilja B.Computerised physician order entry (CPOE). Stud Health Technol Inform.2009;148:159–62.


https://doi.org/10.1186/s12911-018-0615-9

https://doi.org/10.1186/s12911-018-0615-9

https://doi.org/10.1186/s12911-018-0615-9

14. International Standardization Organization Ergonomics of human systeminteraction - part 210: human centered design for interactive systems (rep N°9241-210). Geneva: international standardization Organization; 2010.

15. Marcilly R, Peute L, Beuscart-Zephir MC. From usability engineering toevidence-based usability in health IT. Hat evidence supports the use ofcomputerized alerts and prompts to improve clinicians' prescribingbehavior? Stud Health Technol Inform. 2016;222:126–38.

16. Marcilly R, Beuscart-Zephir MC, Ammenwerth E, Pelayo S. Seeking evidenceto support usability principles for medication-related clinical decisionsupport (CDS) functions. Stud Health Technol Inform. 2013;192:427–31.

17. Marcilly R, Ammenwerth E, Vasseur F, Roehrer E, Beuscart-Zephir MC.Usability flaws of medication-related alerting functions: a systematicqualitative review. J Biomed Inform. 2015;55:260–71.

18. Nielsen J. Enhancing the explanatory power of usability heuristics. CHI '94proceedings of the SIGCHI conference on human factors in. Comput Syst.1994:152–8.

19. Kuperman GJ, Bobb A, Payne TH, Avery AJ, Gandhi TK, Burns G, Classen DC,Bates DW. Medication-related clinical decision support in computerizedprovider order entry systems: a review. J Am Med Inform Assoc. 2007;14:29–40.

20. Phansalkar S, Edworthy J, Hellier E, Seger DL, Schedlbauer A, Avery AJ, BatesDW. A review of human factors principles for the design andimplementation of medication safety alerts in clinical information systems.J Am Med Inform Assoc. 2010;17:493–501.

21. Pelayo S, Marcilly R, Bernonville S, Leroy N, Beuscart-Zephir MC. Humanfactors based recommendations for the design of medication relatedclinical decision support systems (CDSS). Stud Health Technol Inform. 2011;169:412–6.

22. Zachariah M, Phansalkar S, Seidling HM, Neri PM, Cresswell KM, Duke J,Bloomrosen M, Volk LA, Bates DW. Development and preliminary evidencefor the validity of an instrument assessing implementation of human-factorsprinciples in medication-related decision-support systems–I-MeDeSA. J AmMed Inform Assoc. 2011;18(Suppl 1):i62–72.

23. Horsky J, Phansalkar S, Desai A, Bell D, Middleton B. Design of decisionsupport interventions for medication prescribing. Int J Med Inform. 2013;82:492–503.

24. Payne TH, Hines LE, Chan RC, Hartman S, Kapusnik-Uner J, Russ AL, et al.Recommendations to improve the usability of drug-drug interaction clinicaldecision support alerts. J Am Med Inform Assoc. 2015;22:243–1250.

25. Baysari MT, Westbrook JI, Richardson KL, Day RO. The influence ofcomputerized decision support on prescribing during ward-rounds: are thedecision-makers targeted? J Am Med Inform Assoc. 2015;18:754–9.

26. Kortteisto T, Komulainen J, Makela M, Kunnamo I, Kaila M. Clinical decisionsupport must be useful, functional is not enough: a qualitative study ofcomputer-based clinical decision support in primary care. BMC Health ServRes. 2012;12:349–57.

27. Russ AL, Zillich AJ, McManus MS, Doebbeling BN, Saleem JJA. Humanfactors investigation of medication alerts: barriers to prescriber decision-making and clinical workflow. AMIA Annu Symp Proc. 2009:548–52.

28. Saleem JJ, Patterson ES, Militello L, Render ML, Orshansky G, Asch SM.Exploring barriers and facilitators to the use of computerized clinicalreminders. J Am Med Inform Assoc. 2005;12:438–47.

29. van der Sijs H, Aarts J, van Gelter T, Berg M, Vulto A. Turning off frequentlyoverridden drug alerts: limited opportunities for doing it safely. J Am MedInform Assoc. 2008;15:439–48.

30. Feldstein A, Simon SR, Schneider J, Krall M, Laferriere D, Smith DH, Sittig DF,Soumerai SB. How to design computerized alerts to safe prescribingpractices. Jt Comm J Qual Saf. 2004;30:602–13.

31. Krall MA, Sittig DF. Clinician's assessments of outpatient electronic medicalrecord alert and reminder usability and usefulness requirements. Proc AMIASymp. 2002:400–4.

32. Weingart SN, Massagli M, Cyrulik A, Isaac T, Morway L, Sands DZ, WeissmanJS. Assessing the value of electronic prescribing in ambulatory care: a focusgroup study. Int J Med Inform. 2009;78:571–8.

33. Russ AL, Saleem JJ, McManus MS, Frankel RM, Zillich AJ. The workflow ofcomputerized medication ordering in primary care is not prescriptive. ProcHuman Factors Ergon Soc Annu Meet. 2010;54:840–4.

34. Wipfli R, Betrancourt M, Guardia A, Lovis CA. Qualitative analysis ofprescription activity and alert usage in a computerized physician orderentry system. Stud Health Technol Inform. 2011;169:940–4.

35. Khajouei R, Peek N, Wierenga PC, Kersten MJ, Jaspers MW. Effect ofpredefined order sets and usability problems on efficiency of computerizedmedication ordering. Int J Med Inform. 2010;79:690–8.

36. Patterson ES, Nguyen AD, Halloran JP, Asch SM. Human factors barriers tothe effective use of ten HIV clinical reminders. J Am Med Inform Assoc.2004;11:50–9.

37. Chused AE, Kuperman GJ, Stetson PD. Alert override reasons: a failure tocommunicate. AMIA Annu Symp Proc. 2008:111–5.

38. Koppel R, Metlay JP, Cohen A, Abaluck B, Localio AR, Kimmel SE, Strom BL.Role of computerized physician order entry systems in facilitatingmedication errors. JAMA. 2005;293:1197–203.

39. Khajouei R, de Jongh D, Jaspers MW. Usability evaluation of a computerizedphysician order entry for medication ordering. Stud Health Technol Inform.2009;150:532–6.

40. Saleem JJ, Patterson ES, Militello L, Anders S, Falciglia M, Wissman JA, RothEM, Asch SM. Impact of clinical reminder redesign on learnability, efficiency,usability, and workload for ambulatory clinic nurses. J Am Med InformAssoc. 2007;14:632–40.

41. Trafton J, Martins S, Michel M, Lewis E, Wang D, Combs A, Scates N, Tu S,Goldstein MK. Evaluation of the acceptability and usability of a decisionsupport system to encourage safe and effective use of opioid therapy forchronic, noncancer pain by primary care providers. Pain Med. 2010;11:575–85.

42. Chan J, Shojania KG, Easty AC, Etchells EE. Usability evaluation of order sets in acomputerised provider order entry system. BMJ Qual Saf. 2011;20:932–40.

43. Nielsen J. Usability Engineering Boston: Academic Press; 1993.44. Scapin DL, Bastien JMC. Ergonomic criteria for evaluating the ergonomic

quality of interactive systems. Behav Inf Technol. 1997;6:220–31.45. Norman DA. The Design of Everyday Things. New-York: Basic Book; 1988.46. Pearce C. Computers, patients, and doctors—theoretical and practical

perspectives. In: Shachak A, Borycki EM, Reis SP, editors. Health Professionals'Education in the Age of Clinical Information Systems, Mobile Computingand Social Networks: Academic Press, Elsevier; 2017.

47. Náfrádi L, Nakamoto K, Schulz PJ. Is patient empowerment the key to promoteadherence? A systematic review of the relationship between self-efficacy,health locus of control and medication adherence. PLoS One. 2017;12

48. Marcilly R, Monkman H, Villumsen S, Kaufman D, Beuscart-Zéphir M-C. Howto present evidence-based usability design principles dedicated tomedication-related alerting systems to designers and evaluators? Resultsfrom a workshop. Stud Health Technol Inform. 2016;228:609–13.

49. Vicente KJ. Cognitive work analysis. Mahwah, NJ: Lawrence ErlbaumAssociates; 1999.

50. Riedmann D, Jung M, Hackl WO, Stuhlinger W, van der SH, Ammenwerth E.Development of a context model to prioritize drug safety alerts in CPOEsystems. BMC Med Inform Decis Mak. 2011;11:35.


Evidence-based usability design principles for medication alerting … · 2018. 7. 24. · (MCBZ...

Documents

Transcript of Evidence-based usability design principles for medication alerting … · 2018. 7. 24. · (MCBZ...