Ca
C ron
to os
JE MED
M HAEF
e an i
comm
to rea
mail l
ts to s
ere 21
s betw
pted 1
reveal
97/ja
Int
W
ma
ea
ha
settings.1,2 Major advantages are its simplicity of use, the
multiple advantages of asynchronous communication,
and growing options of exchanging visual information,
ren
(an
log
ne
ac
na
ces
W
pe
en
inf
e-m
iss
ve
scr
rea
ho
wa
ma
inf
its
Over recent years the appearance of e-mail messages has
changed from a plain text style with unattractive fonts
(courier) to more flexible hyper text markup language
Af
Cli
He
Co
Me
ver
Ge
Re
Journal of the American Medical Informatics Association Volume 15 Number 2 Mar / Apr 2008 235dering communication less dependent on semantics
d therefore less dependent on education of the dia-
ue partner). In business communities with electronic
tworks, e-mail may thus be valuable for communication
ross social strata while offering advantages to dissemi-
te information without interruption of working pro-
ses.
ith the ease of sending messages and copies to multiple
rsons (one-to-many-messages), e-mail users are experi-
cing information overload and are receiving too much
ormation of low interest (like unsolicited commercial
ail: spamming).3 These problems have become a major
ue, wasting time and resources of e-mail recipients. Con-
rsely, e-mail users have developed strategies to instantly
een incoming messages and to dispose of them without
ding. In large and complex business companies like
(HTML) options that can be defined by the sender and
adjusted to the characteristics of the recipient. Hence, send-
ers have fine-tuned the message design to make it more
appealing to the target population and increase the odds of
the message being read. Indeed, web surveys suggest that
HTML messages with their inherent option of including
graphics are preferred over plain text despite security con-
cerns and their larger size.5
In a randomized controlled trial, we tested the impact of
five different message layouts with e-mails containing the
identical short information on the willingness of the
hospital personnel to retrieve additional information from
a site linked to it. In addition, we analyzed the timing of
collected responses with respect to the e-mail type deliv-
ered.
Methods
We developed a web-based drug information system (AiD-
Klinik) which is accessible in the intranet of our hospital, a
1,680-bed university hospital providing primary and tertiary
care to an urban population. AiDKlinik provides information
on the whole German drug market and can be accessed from
all 4,500 workstations shared by the 6,500 accredited com-
puter users of the hospital. The system is equipped with
filiations of the authors: Department of Internal Medicine VI,
nical Pharmacology and Pharmacoepidemiology, University of
idelberg, Heidelberg, Germany.
rrespondence: Walter E. Haefeli, MD, Department of Internal
dicine VI, Clinical Pharmacology and Pharmacoepidemiology, Uni-
sity of Heidelberg, Im Neuenheimer Feld 410, 69120 Heidelberg,
rmany; e-mail: walter.emil.haefeli@med.uni-heidelberg.de.se Report


haracteristics of Effective Elect
Healthcare Professionals in a H
NS KALTSCHMIDT, MENG, SIMON P.W. SCHMITT, M
ARKUS G. PRUSZYDLO, MMED INFORM, WALTER E.
A b s t r a c t Electronic mailing systems (e-mail) ar
within electronic networks. However, in large business
hospitals it may be difficult to induce account holders
large university hospital we evaluated the impact of e-
with graphics) on the willingness of its
6500 recipien
an integrated link. Overall access rates after 90 days w
respondents opening the link within 3 days. Difference
text, HTML text, animated GIF, and static image prom
the courier plain text message (p
0.001). This study
success of an information campaign.

J Am Med Inform Assoc. 2008;15:235–239. DOI 10.11
roduction
eb-based systems and electronic mailing systems (e-
il) have revolutionized communication and, as an
sily available, informal way to share information, e-mail
s become a cornerstone of communication in health careco
suceived for review: 05/08/07; accepted for publication: 10/21/07.ic Mail Messages Distributed
pital Setting
INFORM,
ELI, MD
mportant means to disseminate information
unities including the hectic environment of
d the e-mail. In two mailings disseminated in a
ayout (three e-mail text versions, two e-mails
eek additional electronic information and open
.1 and 23.5% with more than 70% of the
een different layouts were large and artwork
.2, 1.7, 1.8, and 2.3 times more often access than
ed that layout is a major determinant of the
mia.M2503.
spitals, continued information exchange is an essential
y to promote excellence in clinical care.4 Therefore,
king an addressee turn his attention to a given piece of
ormation and peruse it is as essential as the message
elf.mputerized physician order entry (CPOE) and decision
pport systems, e.g., for drug-drug interactions or dosage

ad
cu
of
cat
we
ma
co
sys
e-m
Ea
lay
ele
(3)
tio
we
an
Fig
sta
for
(dr
(Ta
pr
ha
ele
en
ser
Us
the
pr
Me
pix
sm
ful
W
da
int
ind
ser
do
dif
wa
be
lay
the
dif
e-m
mo
aft
the
en
we
sid
Ta
Ac
Cla
HT
Ar
Sta
An
236 KALTSCHMIDT et al., Effect of Layout on E-mail Attentionjustment in renal failure. It is updated twice a month and
rrently supports about 2,000 searches per day. Extensions
the functionality of the system are routinely communi-
ed in an e-mail to all
6,500 e-mail account holders who
re all equipped with MS Outlook 2000 as the standard
iling program. Separated by an interval of six months we
mmunicated (a) the introduction of a drug interaction
tem and (b) the addition of pregnancy information by
ail sent to all accounts within the intranet of our hospital.
ch mailing consisted of one message in five different
outs (Table 1). Each layout consisted of five different
ments: (1) a subject line, (2) an informative short title with
a two-sentence statement and (4) a link to access addi-
nal information, or (5) an image. Title, text, and image
re variable parts: three e-mail layouts only contained title
d text with the same wording but variable fonts (Table 1,
ure 1A). Two layouts only contained an image which was
tic in one e-mail and an animated graphics interchange
mat (GIF) file of 5.5 (pregnancy message) or 7.7 seconds
ug interaction message) duration in the other e-mail
ble 1, Figure 1B). Graphics were used because the visual
esentation and communication of information in hospitals
s been shown to be more effective.6 The relevant text
ments were set in a large font (at least 14 point) to
sure good readability.7 For lettering two fonts with
ifs and one font without were used.
ing identifying parameters each message was linked to
same uniform resource locator (URL) i.e., a webpage
oviding illustrations and details on the new knowledge
ble 1 y Characteristics of E-mail Messages Randomly
counts in a University Hospital
E-mail Layout Title Font Statement
ssic (black and white) Courier Courier
ML text Times New Roman Times New
twork text Arial with shadow Arial with s
tic image None None
imated image None Noneses. While this webpage illustrated the new feature,
ing a pertinent example, it neither contained data nor
ve access to the respective knowledge bases. It was
refore not suitable to answer specific questions, but
ed the advantages of not being limited by the narrow-
ss of the e-mail viewer’s preview pane.
ssages were designed to be the same size (600  300
; equals 22  12 cm at 1024  768 display resolution),
all and readable, even if the message was not opened to
l screen or viewed only in the Outlook preview pane.
e then imported all hospital e-mail accounts into a
tabase, sorted them alphabetically, and divided them
o five groups of roughly equal size to be mailed as
ependent packages. This was done to reduce mail
ver load. Within each package, addressees were ran-
mized to one of the five messages to assure that all the
ferent types of messages were included in each mailing
ve. Within each package, accounts were serially num-
red from 1 to 5 and were assigned to the five e-mail
outs. E-mails were only sent once to each receiver and
receivers were neither informed about the existence of
ferent e-mails nor who else was receiving the mail. The
ails were sent on a Monday (1st mailing) and six
nths later on a Wednesday (2nd mailing). For 90 days
er each mailing, we anonymously logged every visit to
linked pages. This information was the primary
dpoint of the study and the frequency distributions
re analyzed by 2-analysis. A p-value 0.05 was con-
ered significant.
buted in Five Different Layouts to All E-mail
Link Graphics
Underscored blue text as hyperlink None
HTML-Button None
Whole E-mail None
HTML-Button Static image
HTML-Button Animated GIF
F i g u r e 1A. E-mail with a text
message (artwork text) sent to
inform addressees about a new
feature in their web-based drug
information system (AiDKlinik)
(1st mailing). The message states:
On average hospitalized patients
are exposed to eleven different
drugs. The computerized physi-
cian order entry system (Medibox)
supports the evaluation of the re-
sulting 55 combination pairs for
drug interactions.ba
giv
ga
the
us
ne
Distri
Font
Roman
hadow

Re
Bo
clu
an
sag
sag
Ov
1,5
eac
(1s
be
an
aft
we
bo
low
oth
tha
lin
en
art
we
co
Di
Ra
be
to
a c
po
the
att
tio
no
In
wh
co
su
rat
co
rea
de
bo
no
ou
ex
lik
sp
ex
wh
no
Ou
fea
tio
sy
he
fun
su
cei
res
me
tan
cei
ma
im
wi
an
ha
F i
me
wi
ale
int
for
ma
Journal of the American Medical Informatics Association Volume 15 Number 2 Mar / Apr 2008 237sults
th mailings were successfully and uneventfully con-
ded. Mails not deliverable were excluded from further
alysis, leaving 6309 (1st mailing, drug interaction mes-
e) and 6767 (2nd mailing, pregnancy information mes-
e) e-mail accounts (set to 100%) in the data set.
erall, the linked pages were accessed 1,333 times (1st) and
90 times (2nd) by responders during the 90 days after
h mailing, rendering an average response rate of 21.1%
t) and 23.5% (2nd). Response rates varied considerably
tween the five message formats (Figure 2, Table 2). Total
d individual responses showed circadian fluctuations and
er approximately 96 hours, 67–80% of the total responses
re received, irrespective of the mailing type (Figure 2). In
th mailings the plain text (courier font) version had the
est response rate, which was significantly lower than all
ers (Table 2). The highest rate of response observed was
t to the e-mail with the static image, through which the
ked page was accessed by 372 (1st) and 469 (2nd) recipi-
ts. The combined analysis of both mailings revealed that
work text, HTML text, animated GIF, and static image
re accessed 1.2, 1.7, 1.8, and 2.3 times more often than
urier plain text (Table 2).
scussion
pid dissemination of information is critical in any hospital
cause of the need to coordinate complex services and react
the rapidly changing information in health care. E-mail is
onvenient, rapid, and inexpensive way to reach a target
pulation and by providing links instead of attachments
processor load can be kept to a minimum. However, if
achments are not opened or links not accessed, informa-
n may not be spread in the intended way because it will
t reach the addressee.
order to objectively assess the number of employees
o were attracted by the e-mail and scrutinized its
ntent, we calculated a response rate similar to e-mail
rveys.8,9 E-mail surveys have generally lower response
es than mail surveys8,10 and their response rates have
ntinuously declined over the past 20 years, recently
9,11
g u r e 1B. E-mail with a
ssage containing a GIF file
th or without animation to
rt account holders to a new
eraction tool in their drug in-
mation system (AiDKlinik) (1st
iling).ching values between 20 and 30%. Although we
liberately did not use any of the common methods to
tio
waost responses, such as personalized messages, pre-
tice letters before the mailing, or reminders thereafter,8
r average response rate of 22% was in the currently
pected range.9,11 Compared to different other settings
e an e-mail survey in an academic environment (re-
onse rate 27%)12 and considering that we did not
plicitly solicit for a response this is a favorable result
ich may even underestimate response rates because
t all addressees are involved in patient care.
r mailing was intended to inform users about a new
ture implemented in our hospital-wide drug informa-
n and prescription system (AiDKlinik). Because this
stem is designed to be used without formal training, we
avily rely on e-mail communication to announce new
ctionalities to the hospital staff. Similar to earlier
rvey responses, over 70% of the responses were re-
ved within the first 96 hours after mailing.11 However,
ponse rates varied considerably between the different
ssage formats, indicating that design is a rather impor-
t modulator. Indeed, compared to the employees re-
ving the plain text courier message, more than twice as
ny users receiving the HTML version with a static
age were induced to access the corresponding webpage
th extended information. It is more difficult to design
e-mail message with graphics; but even if the e-mail
d been sent as a plain text HTML, message 70% more
ployees would have been reached. These results match
recent internet survey where roughly three times
many respondents (66.8%) favored e-mail messages in
HTML format over those who preferred plain text
ssages (only 22.7%).5 The main reasons given for this
eference were better readability (78%), more attractive
play (68%), ease of scanning (64%), and overall design
%).
henever e-mail is used as a vehicle, the message has
withstand being classified as unwanted spam. Thus, we
re somewhat surprised that layout may have such a
bstantial impact. We also learned that the most laborious
sign (because of the animation) was not the most success-
. E-mail messages with a static figure consistently
ompted the most people to access the additional informa-em
a
as
an
me
pr
dis
(64
W
to
we
su
de
ful
prn page. Hence, using graphics was the most successful
y to reach employees. Whether it is worthwhile spending

ad
wi
pa
Se
be
gu
lin
vie
me
cli
rep
W
ne
rea
de
ha
inf
Pr
co
the
Se
rea
wo
int
sh
he
du
ind
pr
mu
sp
In
tio
sh
su
giv
me
F i
ho
sec
Ta a Uni
Di
onses
Cla
HT
Ar
Sta
An
To
Th
*2
238 KALTSCHMIDT et al., Effect of Layout on E-mail Attentionditional resources to design graphics to a specific message
ll depend on the aims of the information campaign and
rticularly on how critical it is to reach a large audience.
g u r e 2. Cumulative calls of the link page in the first 96
urs after sending the first (interaction; panel A) and
ond mailing (pregnancy; panel B).
ble 2 y Response Rates of E-mail Account Holders in
fferent Layouts
E-mails Sent Responses after 96 Hours
Layout N N (% of Column Total) % of Total Resp
ssic 2616 268 (12.5%) 74.0%
ML text 2616 464 (21.6%) 74.8%
twork text 2615 324 (15.1%) 71.7%
tic image 2615 625 (29.1%) 74.3%
imated image 2614 466 (21.7%) 71.9%
tal 13076 2147 (100%) 73.5%e results of two independent e-mail campaigns are combined.
-test: p  0.001 vs. classic version.ere were also considerable differences between the three
ferent text messages which are in agreement with general
edictors of successful web design.7 However, even though
th the Courier and Times New Roman formats had
tering with serifs, the Times New Roman messages
ompted the addressee to access the information page 1.7
es more frequently. This suggests that the Courier font is
s aesthetic and should be avoided. On the other hand, the
adowed Arial font was only slightly more successful than
urier, perhaps because its contrast to the blue message
ckground was less distinct, reducing its readability.7
veral limitations of this report merit mentioning. First,
cause logging was done anonymously we did not distin-
ish between individual users. Hence access rates to the
k may exceed the actual number of users if individuals
wed the page, which only advertised the new enhance-
nt, more than once. Because this webpage did not contain
nically useful data and only expanded the information
eated access by the same user was unlikely.
e were not able to track the pattern or rates of use of the
w AiDKlinik knowledge sources due to the following
sons: in our CPOE system drug interaction messages are
livered whenever drug pairs with interaction information
ve been prescribed (immediate feedback). Therefore such
ormation is always delivered and tracking not possible.
egnancy information was delivered as an icon and the
mbination of color and code of the icon already revealed
risk category.
cond, we do not know to what occupational group the
ders belonged and whether professional background
uld modify the preferences in layout. But since drug
eractions are a critical issue in all medical disciplines and
ould therefore be prevented in all sectors, stratification by
althcare provider specialty could have actually intro-
ced another source of bias. As another example, drug
uced risks in pregnancy as exemplified by endocarditis
ophylaxis which has to be considered by dentists just as
ch as by cardiologists, surgeons, or any other prescribing
ecialty.
conclusion, this study showed that a substantial propor-
n of the hospital staff may be reached by e-mail within a
ort period of time and that the layout of the message will
bstantially influence the extent to which the addressees
e appropriate attention to the content of the e-mail
ssage and seek additional information.
versity Hospital to E-mails Sent in Five
Total Responses after 90 Days
N (% of Column Total) % of E-mails Sent ( Response Rate)
362 (12.4%) 13.8%
620* (21.2%) 23.7%
452* (15.5%) 17.3%
841* (28.8%) 32.2%
648* (22.2%) 24.8%
2923 (100%) 22.4%Th
dif
pr
bo
let
pr
tim
les
sh
Co
ba

References y
1. Car J, Sheikh A. Email consultations in health care: 1—scope
and effectiveness. BMJ. 2004;329:435–8.
2. Neville RG, Marsden W, McCowan C, Pagliari C, Mullen H,
Fannin A. A survey of GP attitudes to and experiences of email
consultations. Inform Prim Care. 2004;12:201–5.
3. Olson GM, Olson JS. Human-computer interaction: Psycholog-
ical aspects of the human use of computing. Annu Rev Psychol.
2003;54:491–516.
4. Scally G, Donaldson LJ. Looking forward: Clinical governance
and the drive for quality improvement in the new NHS in
England. BMJ. 1998;317:61–5.
5. Wilson RF. Survey of e-mail format preferences and programs.
Web Marketing Today 2003; Issue 124. Available at URL:
http://www.wilsonweb.com/wmt8/email_format_preferences.
htm. Accessed May 3, 2007.
6. Powsner SM, Tufte ER. Graphical summary of patient status.
Lancet. 1994;344:386–9.
7. Nielsen J. Designing web usability: The practice of simplicity.
Indianapolis: New Riders; 2000.
8. Schaefer DR, Dillman DA. Development of a standard e-mail
methodology. Pub Opin Q. 1998;62:378–97.
9. Sheehan K. E-mail response rates: a review. JCMC. 2001;6(2).
Available at: URL: http://jcmc.indiana.edu/vol6/issue2/
sheehan.html. Accessed May 3, 2007.
10. Kiesler S, Sproull LS. Response effects in the electronic survey.
Pub Opin Q. 1998;50:402–13.
11. Yun GW, Trumbo CW. Comparative response to a survey
executed by post, e-mail, & web form. JCMC. 2000;6(1). Avail-
able at: URL: http://jcmc.indiana.edu/vol6/issue1/yun.html.
Accessed May 2, 2007.
12. Fraze S, Hardin K, Brashears T, Smith JH, Lockaby J. The effects
of delivery mode upon survey response rate and perceived
attitudes of Texas Agri-Science teachers. Available at: URL:
http://www.depts.ttu.edu/aged/research/frazesurvey.pdf.
Accessed May 3, 2007.
Journal of the American Medical Informatics Association Volume 15 Number 2 Mar / Apr 2008 239