Volume 2, Issue 2A
DECISION AND INFORMATION TECHNOLOGIES
Keeping Up with the Jones’: Influence
from Peer Hospitals on EMR Diffusion
Corey M. Angst, Research Assistant Professor and
Associate Director of CHIDS
Ritu Agarwal, Professor and Robert H. Smith Dean’s Chair
of Information Systems and Director of CHIDS
Evidence suggests that in spite of the value potential of
information technology (IT), institutions in the health
care sector have been slow to adopt it. To the extent
that significant problems in health care such as patient
safety, medical errors, and escalating costs can be
addressed through IT, accelerating its diffusion
throughout the system is an important public policy
issue. This research examines the drivers of adoption
and diffusion of electronic medical records (EMR) in
hospitals across the United States.
Using a survival analysis technique, we model the
likelihood of EMR adoption as a function of three key
drivers – HIT Concentration, Temporal HIT Adoption, and
Spatial Contagion Effect. The first two represent internal
firm characteristics that capture the availability of
complementary technologies and a temporal
component reflecting the cumulative learning and
experience that the firm has with information
technologies. In order to isolate the effects of these
variables, we control for other structural characteristics
such as size and facility type that have been shown in
prior research to affect the likelihood of adoption. Prior
research has also suggested that firms tend to mimic the
behaviors of “similar” others, where similarity is typically
assessed along dimensions such as size and industry.
The diffusion of innovations literature argues that
innovations spread through a social system via a
contagion effect, where “infected” facilities affect others.
We model contagion using a spatial construct reflecting
the effect of localized knowledge spillovers that can
occur among firms that are geographically proximate.
The Nature of Innovations
Studies of diffusion of innovations are of central concern
to disciplines ranging from organizational sciences,
information systems, to marketing, and others.
Researchers seek to classify innovations in various
categories such as: administrative and technical; radical
and incremental; stage of adoption; or product and
Summer 2007
process. More recent work suggests that some
innovations can be highly disruptive. Although scholars
have used various definitions to describe disruptive
innovations, central tenets of these types of innovations
are that they introduce new performance dimensions to
an industry and are inferior – at least in the short term –
to the traditional technology on measures that are
important to the consumer, and they can transform
markets through the disruptions they create.
Descriptions of EMR systems in extant literature and
studies of these systems in practice suggest that as an
innovation, they are radical in that they cause major
shifts in the work practices of clinicians and other
personnel in health care facilities. They are both
technical innovations, because they alter the underlying
technology used in the delivery of patient care, as well
as administrative innovations, because of their effects on
the billing, insurance, and other administrative
practices. EMRs incorporate new products such as an
electronic repository of patient information that is
portable, as well as new processes around patientclinician interaction.
Adoption and Diffusion of Innovations
Firms innovate because innovation has been shown to
lead to competitive advantage. The literature in the
organizational sciences offers significant evidence that
the adoption of innovations by firms is a function of not
just intra-firm characteristics, but also external
influences. In broad terms, organizational factors such
as strategic orientation, IT capabilities, organizational
resources, and structure have been shown to influence
the adoption of innovations. Prior research has also
identified key environmental factors such as
competition, industry, urbanization, etc. as drivers of
innovation.
While research on the adoption of innovations attempts
to model the factors that drive the behavior of a single
entity such as a firm, diffusion models are typically used
to describe the adoption of an innovation by a target
population over a specific period of time. The diffusion
process incorporates an innovation, communication
channels, time, and a social system. Diffusion studies
are useful because they can be used to predict increases
in number of adopters and project future trends.
CHIDS • Van Munching Hall • College Park, MD • University of Maryland • www.smith.umd.edu/chids • chids@rhsmith.umd.edu • 301.405.0702
© 2007 Center for Health Information and Decision Systems. CHIDS Research Briefings are published on a quarterly basis to update CHIDS members.
p. 2
CHIDS Research Briefing
Volume 2, Issue 2A
Contagion Effect in Innovation Adoption
Prior research has argued that firms tend to model their
practices, business processes, and procedures after
successful peer firms. The innovation adoption
literature suggests that such mimetic behavior is most
likely to be observed between firms that are similar in
characteristics such as size, industry, product, culture,
etc. We argue that the presence of a specific mimetic
phenomenon known as localized knowledge spillover
(LKS) is one reason firms in geographic proximity begin
to resemble each other from a technology infrastructure
standpoint. LKS has been defined as “knowledge
externalities bounded in space, which allow companies
operating nearby the knowledge sources to introduce
innovations at a faster rate than rival firms located
elsewhere.” The LKS literature suggests that the
adoption of highly complex information systems
requires specific knowledge that is fundamentally tacit
in nature. Because this knowledge is not codified, the
geographic proximity to other adopters is very
important. The tacit knowledge can be spread through
face-to-face interaction, personal relationships, or simply
because there is knowledge transfer when knowledge
workers leave one firm for another: all of which tend to
occur with greater frequency when distance between
entities decreases.
Drawing upon the theoretical foundations described
above, the overall research model underlying the study
is shown in Figure 1. The focal dependent variable of
interest is the likelihood of EMR adoption in the next
time period, given the choices and decisions made by
the hospital in previous time periods.
HIT
Concentration
Temporal HIT
Concentration
Spatial
Contagion
Likelihood of
EMR Adoption
Control
Variables
Figure 1. Research Model
Research Hypotheses
EMR systems are often used as the conduit that links and
aggregates the data within isolated legacy systems
within hospitals. Recent work suggests that successful
implementation of a technology requires an already
existing robust IT infrastructure. Other research has
shown that the presence of complementary information
technologies can enhance the adoption of the focal
technology. In all of these examples, the common
theme is that successful implementations of a focal
technology are reliant upon the presence of
complementary IT infrastructure. Therefore, we
propose that the likelihood of EMR adoption will be
positively associated with HIT Concentration. However,
it is not merely the presence of the complementary
technologies that influences the propensity to adopt an
innovation; what is also important is the point in time at
which the complementary HIT’s are adopted. Extensive
bodies of literature ranging from learning-by-doing to
organizational learning to absorptive capacity suggest
that the amount of experience with an IT affects the
success and ultimately, the likelihood of future IT
adoption. This suggests that the likelihood of EMR
adoption will be positively associated with the timeweighted adoption of HIT Concentration.
In an attempt to innovate, firms will actively seek to
capture knowledge beyond their boundaries by luring
key knowledge workers, conducting informal information gathering, and through personal relationships –
particularly within close spatial proximity. We therefore
hypothesize that hospitals within a tightly coupled
geographic cluster of other EMR-adopting hospitals will
be more likely to adopt an EMR than hospitals in a
loosely coupled geographic cluster. More specifically,
we suggest that the influence imparted by other
hospitals to adopt an EMR decreases as the distance
between the focal hospital and other adopters increases.
Research Methods
We use a discrete event hazard model to predict the
likelihood of EMR adoption. The empirical analysis and
hypothesis tests are based on secondary data set
collected via a survey. The data come from a nationwide, annual survey of Care Delivery Organizations in
the USA, conducted by HIMSS AnalyticsTM. The 20042005 HIMSS Analytics Database (derived from the
Dorenfest IHDS+ DatabaseTM) provides information for
3,989 hospitals.
A literature review identified a list of 52 HIT applications
that are commonly used in hospitals. These technologies
were chosen because they are well-known amongst
health care providers, they are often discussed in
conjunction with EMRs, and for the most part, they have
diffused to a greater extent than EMRs. HIT Concentration
was operationalized simply as a count of how many of the
52 applications were adopted. Temporal HIT Concentration
is an assessment of both the number of HIT applications
adopted and the experience (length of time) the hospital
has had with the technology in use.
Summer 2007 p. 3
x miles
a
Focal
Hospital
b
Hospital with EMR
Hospital without EMR
Figure 2. Description of Spatial Calculation
Finally, we control for size of the hospital,
operationalized as the number of beds that are staffed;
type of hospital, operationalized as a teaching/research
hospital or not; and resource characteristics such as
profit versus not-for-profit and age of the hospital.
Analysis and Results
There are 3,989 hospitals included in our analysis. Our
first finding supports the hypothesis that HIT
Concentration (HITC) is related to the likelihood of EMR
adoption. The beta coefficient of HITC was calculated to
be 0.190 (p=.000), which when antilogged (e0.190=1.21),
suggests that in every year measured, the likelihood of
EMR adoption is 21% greater for hospitals whose HIT
Concentration is one standard deviation greater (Mean
HITC = 34.43, Standard Deviation = 6.56). In Figure 3
we provide a visual representation of the effect that HIT
Concentration has on the likelihood of adoption.
.10
.08
.06
HIT CONC
41 or more
.04
38-40
35-37
.02
Hazard
We use an Excel® add-in program to calculate the
Euclidian distance between two zip codes. We use this
information to compute the total number of hospitals
within three radii – 5 miles, 50 miles, and 100 miles. The
HIMSS Analytics database is used to cross-reference zip
codes with those hospitals that have an EMR and also
the total number of hospitals in a given zip code radius.
We modified the Excel program to first input a database
of all hospitals and, second, input a database with all
hospitals with an EMR. From this, we extract all
hospitals that meet the criteria and perform a count. In
the particular case highlighted in Figure 2, the EMR
Count for the focal hospital ‘a’ is 3 (the focal hospital is
not counted even if it is an adopter), and the hospital
count is 6. Assuming all hospitals within the light
yellow area are in the same city, it is apparent that these
hospitals can have markedly different ratios of adopter
to total number of hospitals. For example, as noted,
hospital ‘a’ has an EMR count=3 and hospital count=6,
and hospital ‘b’ has values of 3 and 5, respectively.
30-34
29 or less
0.00
0
•1975
5
1980
10
1985
15
1990
20
1995
25
2000
30
2004
Figure 3. Effect of HIT Concentration on the
Likelihood of Adoption
Our second finding suggests that there is a temporal
component associated with the likelihood of EMR
adoption; however, the explanatory power of temporal
HITC (THITC) is quite low (0.005, p<.000), suggesting
that a one standard deviation increase in THITC will
result in only a 0.5% increase in the likelihood of EMR
adoption (Mean THITC = 224.16, Standard Deviation =
124.33).
With respect to the spatial influence, we standardize the
data by using characteristics of the entire distribution and
control for the number of hospitals within each radius so
as to eliminate bias present in regions with a relatively
high number of hospitals. From this analysis, we find
some cities emerge as very highly concentrated in EMR
adoption (see Table 1, page 4). We then test the influence
imparted at increasing radii and find that the strongest
influence emerges from those hospitals within a 5-mile
radius, relative to those in a 50 or 100-mile radii (β5mi=.323,
p=.000; β50mi=.091, p=.000 β100mi=-.025, p=.000). What
this suggests is that a high ratio of EMR-adopting hospitals
to total number of hospitals (i.e. EMR-adopting/total
number hospitals) within a 5-mile radius is a much
stronger influence on the likelihood of adoption, than the
same ratio in a 50- or 100-mile radius. Thus, our assertion
holds that the influence imparted by other hospitals to
adopt an EMR increases when the distance between the
focal hospital and other adopters is smaller. Therefore
adoption is more likely in tightly clustered regions of high
EMR use.
CHIDS Research Briefing
Table 1. City Rankings by Standardized Spatial EMR
Adoption (three or more hospitals)
Rank
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
City, State
Jamaica, NY
Loma Linda, CA
Baltimore, MD
Honolulu, HI
Akron, OH
Albuquerque, NM
Winston-Salem,
Davenport, IA
Salt Lake City, UT
Boston, MA
Lexington, KY
Portland, OR
Grand Rapids, MI
Brooklyn, NY
Augusta, GA
# Hospitals in Standardized Mean # Hosp
5-mi Radius
w/ EMR in 5-mi Radius1
3
2.41
3
2.21
15
1.81
7
1.74
4
1.70
11
1.62
4
1.53
3
1.53
9
1.53
12
1.41
8
1.36
8
1.31
5
1.27
18
1.24
5
1.23
Lowest rank by Mean
157 Wichita Falls, TX
158 Miami, FL
159 Dayton, OH
160 Hialeah, FL
161 Irving, TX
162 San Angelo, TX
163 Wichita, KS
164 Milwaukee, WI
165 Lancaster, PA
166 Madison, WI
167 Rochester, NY
168 St. Louis, MO
169 Newark, NJ
170 Detroit, MI
171
Coral Gables, FL
3
12
4
3
3
4
4
11
4
4
6
16
6
9
3
-0.30
-0.32
-0.33
-0.33
-0.33
-0.35
-0.35
-0.39
-0.40
-0.40
-0.43
-0.48
-0.49
-0.59
-1.03
1
This column is standardized by the number of hospitals in a
city as shown below:
Z=
Summer 2007 p. 4
Volume 2, Issue 2A
X −µ
σ
X = Number of Hospitals
µ = arithematic mean of distribution
σ = std . deviation of distribution
Conclusion and Implications
We demonstrate that contagion through LKS is a
measurable phenomenon as it relates to the adoption of
EMRs. We believe this provides an important
perspective on how innovations diffuse. As discussed,
we found that some cities emerged as highly
concentrated in EMRs and surprisingly, these are
typically not the cities that have been known to be
highly innovative. One possible explanation is that
some hospitals in specific geographic regions have been
more active in the movement to make their information
systems’ infrastructure highly interoperable so they can
exchange patient data. For example, in Salt Lake City,
Utah, Intermountain HealthCare and a network of
physician practices and other stakeholders have
collaborated to form a Regional Health Information
Organization (RHIO) as a means of sharing data
between sites. This data-share would not be possible
without EMRs so it stands to reason that there is a high
degree of contagion between sites – a fact that is
confirmed by our empirical analysis.
Because EMRs require the presence of existing
technological infrastructure, it is important to determine
which innovations are the most predictive of EMR
adoption. This study identifies that the concentration of
technologies is an important antecedent of the
likelihood of EMR adoption. Policy makers, hospital
executives, physicians, vendors and others are seeking
practical guidance on how to encourage the adoption
of EMRs. We identify specific factors important for
adoption and suggest that local influence may be one of
the most important means of spreading the innovation.
Thus, state-level, or even county-level initiatives may be
more important for rapid diffusion of EMRs than federallevel initiatives.
Suggested Citation:
Angst, C.M. and Agarwal, R., "Keeping Up with the Jones’: Influence
from Peer Hospitals on EMR Diffusion," CHIDS Research Briefing (2:2A)
Center for Health Information and Decision Systems, Robert H. Smith
School of Business, University of Maryland, Summer 2007, pp 1-4.
CHIDS Contact Information
Director – Ritu Agarwal, Professor and Robert H. Smith Dean's Chair of
Information Systems
Associate Director – Corey Angst, Research Assistant Professor
Center for Health Information and Decision Systems
Robert H. Smith School of Business
University of Maryland
Van Munching Hall
College Park, Maryland 20742
Ph: 301.405.0702
chids@rhsmith.umd.edu
www.smith.umd.edu/chids
CHIDS Membership Information
Become a corporate member of CHIDS and realize the benefits of
having access to a world-class team of researchers. All levels of
membership provide your organization with the following benefits:
- Early insights from research
- Access to top scholars
- Quarterly updates on key HIT activities
- Participation in bi-annual conferences
- Opportunity to influence CHIDS’ research agenda
- Priority access to research projects
CHIDS Corporate Members
Johnson & Johnson Health Care Systems, Inc.
Pfizer Inc
CapMed, a Division of Bio-Imaging Technologies, Inc.
CTIS Inc.
CHIDS • Van Munching Hall • College Park, MD • University of Maryland • www.smith.umd.edu/chids • chids@rhsmith.umd.edu • 301.405.0702
© 2007 Center for Health Information and Decision Systems CHIDS. CHIDS Research Briefings are published on a quarterly basis to update CHIDS members.