ArticlePDF Available

Functional IT Complementarity and Hospital Performance in the U.S.: A Longitudinal Investigation

Authors:
Article

Functional IT Complementarity and Hospital Performance in the U.S.: A Longitudinal Investigation

Abstract and Figures

This paper examines complementarity between clinical health information technology (HIT) applications and their effects on three hospital-level performance measures: clinical quality, experiential quality, and healthcare cost. We emphasize two aspects of HIT use in hospitals. First, we focus on whether HIT applications are used to perform primary or support clinical functions. Contingent on whether the use of HIT applications is for performing only primary functions or both primary and support functions, we conceptualize symbiotic and pooled HIT complementarity, respectively. Second, we focus on whether HIT applications are implemented in the same time period or different time periods. Contingent on this temporal aspect, we conceptualize simultaneous and sequential HIT complementarity, respectively. We collected panel data on HIT implementation, clinical quality, experiential quality, and healthcare cost for 715 hospitals in the U.S. from four sources. Our results suggest that symbiotic, pooled, simultaneous, and sequential complementarity among HITs impact hospital quality and cost outcomes. Our results further indicate that these complementary effects differ across chronic and acute conditions. We also find that three-way complementarity has significant economic effects on quality and cost. In fact, post hoc analyses indicate that three-way sequential complementarity effects, which have not been previously examined, are particularly significant. This paper contributes to the literature by empirically examining different forms of HIT complementarity in hospitals. Our central message is that when assessing HIT value in hospitals, managers and researchers must pay attention to: 1) the clinical functions to which these technologies are applied; 2) the sequence in which these HITs are implemented; and 3) the prevalence of chronic versus acute patients admitted in the hospital.
Content may be subject to copyright.
Functional IT Complementarity and Hospital Performance in the U.S.: A Longitudinal Investigation
1
Abhay Nath Mishra
Debbie and Jerry Ivy College of Business
Information Systems & Business Analytics
Iowa State University
E-mail: abhay@iastate.edu
Mark Keil
J. Mack Robinson College of Business
Department of Computer Information Systems
Georgia State University
E-mail: mkeil@gsu.edu
Youyou Tao
College of Business Administration
Information Systems & Business Analytics
Loyola Marymount University
E-mail: youyou.tao@lmu.edu
Jeong-ha (Cath) Oh
E-mail: Jh.cath.oh@gmail.com
Abstract
This paper examines complementarity between clinical health information technology (HIT) applications and their effects
on three hospital-level performance measures: clinical quality, experiential quality, and healthcare cost. We emphasize
two aspects of HIT use in hospitals. First, we focus on whether HIT applications are used to perform primary or support
clinical functions. Contingent on whether the use of HIT applications is for performing only primary functions or both
primary and support functions, we conceptualize symbiotic and pooled HIT complementarity, respectively. Second, we
focus on whether HIT applications are implemented in the same time period or different time periods. Contingent on this
temporal aspect, we conceptualize simultaneous and sequential HIT complementarity, respectively. We collected panel
data on HIT implementation, clinical quality, experiential quality, and healthcare cost for 715 hospitals in the U.S. from
four sources. Our results suggest that symbiotic, pooled, simultaneous, and sequential complementarity among HITs
impact hospital quality and cost outcomes. Our results further indicate that these complementary effects differ across
chronic and acute conditions. We also find that three-way complementarity has significant economic effects on quality
and cost. In fact, post hoc analyses indicate that three-way sequential complementarity effects, which have not been
previously examined, are particularly significant. This paper contributes to the literature by empirically examining different
forms of HIT complementarity in hospitals. Our central message is that when assessing HIT value in hospitals, managers
and researchers must pay attention to: 1) the clinical functions to which these technologies are applied; 2) the sequence
in which these HITs are implemented; and 3) the prevalence of chronic versus acute patients admitted in the hospital.
Keywords: healthcare quality, healthcare cost, experiential quality, clinical quality, hospital performance, HIT
1
The authors wish to acknowledge helpful comments provided by seminar participants at Iowa State University, Loyola Marymount
University, Temple University, Workshop on Information Systems and Economics, and Conference on Information Systems and
Technology.
1
1. Introduction
As the literature in information systems (IS), economics, and management has established, organizations often employ
innovations in combinations (Barua et al. 1995; Bresnahan et al. 2002; Ennen and Richter 2010; Milgrom and Roberts
1990; Teece 2018). In recent years, the health care industry in the United States has spent almost $30 billion on health
information technology (HIT) innovations (Adler-Milstein and Jha 2017). Research in IS and health informatics has
examined HIT applications extensively, but contrary to the guidance from past research, which emphasizes the joint use
of innovations (Teece 2018), and actual practice in hospitals, extant literature has largely ignored interactive effects of
HITs (Hansen and Baroody 2020; Sharma et al. 2016). Recently, however, researchers have underscored the need to
study multiple HIT applications and their joint effects to obtain insight into the overall impact of technology on hospital-
level measures of quality and cost of care (Angst et al. 2012; Karahanna et al. 2019; Dranove et al. 2014).
Studying complementary effects of HIT applications is important to advance both theory and practice. From a
theoretical perspective, examining joint impacts of HITs enables theoretical development on the innovative ways in which
HIT applications can be leveraged and furthers our understanding of their overall impact at the hospital level.
Investigating individual HITs without considering interactions limits researchers’ knowledge of the joint impacts these
systems can have on hospitals. From a practical perspective, obtaining a more nuanced understanding of the joint
effects can lead to judicious and robust managerial actions regarding the use of and investment in HIT applications.
In this paper, we focus on two key nuances, which exemplify the need for contextualizing HIT applications,
studying them jointly, and expanding the conceptualization and empirical analysis of HIT complementarity. First,
hospitals routinely implement a diverse set of HIT applications to accomplish various clinical functions (Karahanna et al.
2019). These clinical functions pertain to the observation of ailing patients and provision of appropriate medical services
aimed at improving their health conditions. We conceptualize two broad categories of clinical functions primary and
support. Table 1 summarizes major differences between the two, including definitions, examples and key characteristics.
Table 1: Delineation of Primary and Support Clinical Functions
Clinical Functions
Examples
Definition
Key Characteristics
Primary clinical
functions
Patient
diagnosis, and
treatment2
Primary clinical functions are defined as those
clinical functions that are central to the health
condition diagnosis, care plan, and actual
1. Central to physicians’ work and
patient treatment.
2
https://www.ilo.org/public/english/bureau/stat/isco/docs/health.pdf
2
treatment of patients (Balogh et al. 2015; Del
Mar et al. 2006; Groopman 2008).
2. Physicians’ medical education
and training largely geared toward
these functions (Groopman 2008).
3. Patients are admitted to a
hospital primarily to get diagnosed
and treated (Balogh et al. 2015; Del
Mar et al. 2006).
Support clinical
functions
Clinical
documentation,
and reviewing
of results
Support clinical functions are defined as those
auxiliary clinical functions that enable care
providers to perform primary clinical functions
in a more comprehensive, accurate and
effective manner by augmenting relevant
patient information at the point of care.
1. Serve mainly to supplement the
primary clinical functions.
2. Require lower cognitive effort in
comparison (Bhargava and Mishra
2014).
Clinical functions significantly impact the treatment regimen and patient outcomes and constitute the largest
determinants of quality and cost in hospitals.
3
Accordingly, our focus in this research is on clinical HITs, which hospitals
use to accomplish clinical functions in patient care provision. Information complementarities facilitated by the joint use of
multiple HITs may have substantial synergistic impacts on clinical functions and hospital-level outcomes (e.g., quality
and cost of care). In order to gain a deeper understanding of these joint impacts, it is critical that we examine HITs
collectively rather than study each HIT individually.
Second, hospitals can implement multiple HIT applications concurrently or time-sequence them in different
orders, depending on their resource availability and clinical needs. Thus, primary and support clinical functions can be
enabled by HIT either simultaneously or in different sequences. To date, we know little about the impact of different HIT
implementation approaches on hospital-level quality and cost of care. However, because the sequence in which various
HIT applications are implemented affects the order in which clinical functions are enabled, it may be consequential from
the perspective of quality and cost of care.
Over the last two decades, researchers have sought to understand the joint impacts of information technology
(IT) applications, process transformations, and innovations on firm performance (Milgrom and Roberts 1990). A key
insight from this literature is that although the above-mentioned factors may be impactful in isolation, their joint effects
can exceed the sum of values generated by each element when they are applied in a mutually reinforcing manner - a
phenomenon known as complementarity. Complementarity improves organizational performance by streamlining
3
https://www.acponline.org/acp_policy/policies/controlling_healthcare_costs_2009.pdf
3
business processes, enabling information sharing and coordination among key entities, facilitating effective decision
making, expediting the execution of critical transactions, and enhancing customer satisfaction.
Although prior research has contributed to our understanding of IT complementarity, the use of this perspective
is largely lacking in the HIT literature (Dranove et al. 2014). We emphasize four resulting limitations, which inhibit our
knowledge of how HIT applications, directed toward different clinical functions and implemented in different sequences,
impact hospital-level cost and quality metrics. First, the HIT literature has paid little attention to whether HIT applications
are used for similar or different clinical functions. Because some HIT applications are geared toward primary clinical
functions and others toward support clinical functions, ways in which these applications interact with one another may
impact hospital performance. These aspects are understudied in the literature. Second, the HIT literature has ignored the
possibility that synergistic gains can occur when complementary factors interact simultaneously (i.e., in the same time
period) or sequentially (i.e., across time periods). Third, the HIT literature has not paid sufficient attention to empirically
estimating how disease characteristics, such as those found in patients with chronic and acute conditions, impact HIT
value (Adjerid et al. 2018; Bardhan et al. 2020). Fourth, the HIT literature suffers from empirical issues, such as the use
of data acquired from selected hospital systems (e.g., Aron et al. 2011; Bhargava and Mishra 2014; Hansen and
Baroody 2020; Hao et al. 2018), and limited analysis of second- and higher-order interactions among HIT applications.
In our digital world, complementary technologies are more important than ever, and it is vitally important to
orchestrate them together to maximize value (Teece 2018). An understanding of complementary technologies can
enable researchers to further theorize the nuanced ways in which organizations can create and appropriate value by
combining the strengths of multiple innovations. Complementarity considerations can be helpful in addressing
inconsistencies in the literature by sensitizing researchers to the fine-grained nature of the interactive effects of HIT.
Because evidence on the impacts of HIT on cost and quality has been mixed in the prior literature (Furukawa et al.
2010), it may be valuable to explore if the joint exploitation of technologies across multiple clinical functions might yield
better insights on the overall efficacy of HIT applications (Dranove et al. 2014).
Specifically, we address the following research question in this paper: how do pairwise and three-way
complementarities between different HIT applications, implemented to accomplish primary and support clinical functions,
impact hospital performance from the perspective of cost and quality of care? We attempt to address limitations in prior
4
research by examining HIT complementarity in a nuanced and staged manner. First, we conceptualize functional IT
complementarity and discuss its theoretical underpinnings. We focus on specific clinical functions HITs enable and their
interactions (Grandori and Furnari 2009). Subsequently, we add the time dimension and examine both simultaneous and
sequential interactions between HITs (Battisti et al. 2015). Next, we bring the use context into consideration and
empirically examine the differential impacts of HIT on patients with acute and chronic conditions. Acute conditions are
severe and sudden in onset, whereas chronic conditions last 12 months or longer, and this difference can potentially
impact HIT value. Finally, using panel data assembled from multiple sources and analyzed using fixed effects models,
we explore the impacts of HIT complementarity on the quality and cost of care at the hospital level. We examine three
hospital-level outcomes. The first, clinical quality, refers to the extent to which a hospital delivers clinical services
appropriate for the patient, resulting in desired health outcomes (Mainz 2003). The second, experiential quality, refers to
the patients’ perceptions of their inpatient experience with respect to the quality of their communication with hospital care
providers (Chandrasekaran et al. 2012; Senot et al. 2016). The third, healthcare cost, refers to the average cost of
patient treatment at a hospital. Our key contribution is at the intersection of HIT and complementarity, with concepts from
complementarity instantiated, operationalized, and analyzed to reflect the way in which hospitals implement and exploit
HIT over time.
2. Background Literature and Theoretical Development
We draw upon three literature streams. First, given our interest in hospital-wide HIT use and impacts, we draw upon
enterprise resource planning (ERP) systems literature to underscore that as with ERP systems, HIT use in hospitals can
have effects that cut across the entire organization and impact hospital performance. Second, we draw upon HIT value
literature that examines the impact of different HIT applications on hospital-level quality and cost metrics. Finally, we
draw upon complementarity literature to classify primary and support HITs, to conceptualize the nature of functional HIT
complementarity, and to assess how organizations leverage various innovations synergistically.
2.1 ERP Implementation and Organization-Level Impacts
A key insight from the ERP literature (see Nazemi et al. 2012; and Schlichter and Kraemmergaard 2010 for syntheses of
this literature) is that three major ERP features enhance organizational performance: capturing enterprise data and using
it for end-to-end planning and coordination; completing different transactions, and fulfilling orders efficiently; and enabling
5
enhanced decision making (Davenport 1998; Gattiker and Goodhue 2005; Holsapple and Sena 2005; Seddon et al.
2010). Further, ERP systems enable complementarity through enterprise-level information integration. Like ERP
systems, HITs, such as electronic medical records, clinical decision support systems, and computerized provider order
entry systems, are used for enterprise-level information sharing, integration and care coordination. Thus, their joint
implementation has the potential to influence hospital-level performance.
2.2 HIT and Hospital-Level Implications
Considerable prior research has examined the impacts of HIT investments and use on the quality and cost of care. HIT
systems examined include those geared toward processing internal data for administration; enhancing clinical and
information workflows for patient care; and coordinating organizational activities, planning, and allocating resources.
Researchers have labeled these HITs administrative HIT, clinical HIT and strategic HIT, respectively (Angst et al. 2012;
Bardhan et al. 2015; Bhattacherjee et al. 2007).
Prior research has found mixed impacts of HIT on clinical quality. Some studies have reported that hospitals
with HIT have fewer patient complications and lower mortality rates (Amarasingham et al. 2009; Buntin et al. 2011; Lee
et al. 2013), while others have reported that HIT lowers mortality rates, but increases patient complications (Furukawa et
al. 2010; Lin et al. 2018), has little or no significant impact (Agha 2014), and has even adverse impacts (Ash et al. 2004;
Nebeker et al. 2005). These results may be a manifestation of a variety of quality measures used, resulting in
inconsistent comparisons. Prior research has also found mixed effects of HIT on experiential quality, with some studies
reporting improvements (Hsu et al. 2015; Queenan et al. 2011), and others reporting declines (Peng et al. 2020). Finally,
prior research examining HIT and cost has also found mixed results. For instance, Bardhan and Thouin (2013),
Ayabakan et al. (2017), and Ayer et al. (2019) found HIT to lower costs, but Agha (2014) found that HIT is associated
with a 1.3% increase in billed charges and reported no evidence of cost savings five years after adoption. Dranove et al.
(2014) found no significant decrease in costs after electronic medical records (EMR) implementation in hospitals. Others
have also found HIT to have no significant impacts on cost and efficiency (Furukawa et al. 2010; Sharma et al. 2016).
Researchers have offered several suggestions to resolve these inconsistent findings. First, researchers have
argued that the impact of HIT on quality and cost of care may be context-driven, and it may be useful to conduct further
research using panel data and to focus on the use context (Hydari et al. 2019; McCullough et al. 2010). Second,
6
scholars have advocated that quality and cost impacts of HIT be studied together (Agha 2014; Angst et al. 2011;
Dobrzykowski and Tarafdar 2015; Menon and Kohli 2013), as there may be trade-offs among hospital-level outcomes.
Because HIT has the potential to impact multiple outcomes, it is critical to examine these outcomes together to
thoroughly assess the effects of these applications. Further, studying individual rather than integrative impacts of HIT
applications may disguise their synergistic, joint effects on hospital performance, thus it is important to consider them
jointly (Bardhan and Thouin 2013; Bardhan et al. 2015; Menachemi et al. 2008; O’Connor et al. 2011).
To date, limited effort has been expended to examine the joint effects of clinical HITs. Ranji et al. (2014) studied
ordering and decision support applications and found that their joint use consistently reduces prescribing errors but does
not prevent adverse drug events. Scott et al. (2018) found that electronic documentation and decision support
applications are jointly effective in reducing the prescription of potentially unsuitable medications. Sharma et al. (2016)
used panel data to examine the joint effect of clinical IT and augmented clinical IT on quality and cost. They found that
the effect enhances experiential quality, but not cost outcomes. They also examined the interaction effect of clinical IT
and EMR and found it reduces cost. Sharma et al., however, did not consider use contexts, three-way interactions, or the
sequence of HIT implementation. In summary, joint effects of HIT applications remain underexplored in the literature,
hindering our understanding of the synergistic effects of these systems on hospital-level quality and cost metrics.
2.3 Complementarity of Innovations
The complementarity perspective posits that economic factors of production are complementary if the total value created
by combining two or more factors exceeds the sum of values generated through them in isolation (Milgrom and Roberts
1990). The extent and type of complementarity depend on interacting factors and how firms exploit them. Grandori and
Furnari (2009) suggest that these interacting factors can be similar or different and that firms can apply them to the same
or different application domains. Grandori and Furnari (2009) differentiate between symbiotic complementarity, which is
created when different factors are applied to the same application domain, and pooled complementarity, which is created
when similar factors are applied to different application domains.
IS scholars have shown that the pairwise interaction effects of IT innovations, process transformations, and
organizational changes can be synergistic (e.g., Barua et al. 1995; Bresnahan et al. 2002; Dewan and Kraemer 2000). A
recent advance in the literature has been the introduction of three-way complementarity analysis. The key idea in these
7
analyses is that correlations between any two factors are positive when the third factor is also present, but not
necessarily otherwise. Empirical work examining three-way complementarity is limited in the IS literature, with only two
published papers to our knowledge, neither of which examines HIT (Aral et al. 2012; Tambe et el. 2012).
2.4 Primary and Support HITs, and HIT Complementarity
The complementarity perspective provides an apt lens to investigate HIT’s impact on hospital performance, as it allows
us to assess joint effects of various HITs, over and above their individual effects. To develop our theoretical arguments,
we first conceptually delineate primary and support HITs. Subsequently, we apply the concepts developed in the prior
complementarity literature (e.g., Ennen and Richter 2010; Grandori and Furnari 2009) to advance our HIT
conceptualization and to elucidate how HIT applications complement one another.
2.4.1 Primary and Support HITs
Our conceptualization of primary and support HITs is closely related to that of the primary and support clinical functions
we defined earlier. We draw upon two sources to corroborate our conceptualization that: 1) the primary clinical functions
of physicians are diagnosing patients’ conditions and devising treatment plans for them, and prescribing medicines, tests
and procedures; and 2) the support functions are documenting patients’ medical history and viewing results of their prior
treatments and tests. The first source enables us to identify key clinical functions upon which to ground our notion of
complementarity. The second supports our categorization of these functions into primary and support functions.
Our first source is the venerable Institute of Medicine (IoM). A patient safety study conducted by the IoM
suggests that HIT applications enable four core clinical hospital functions (Tang 2003): (1) order entry/management,
which enables physicians to prescribe medicines and procedures, monitors duplicate orders and drug interactions,
eliminates errors caused by illegible handwriting, and reduces order processing time; (2) clinical decision, which provides
technology-assisted diagnosis to detect patients’ conditions and promotes the use of best clinical practices; (3) health
information and data, which captures patient information; and (4) results management, which electronically stores and
reports prior results, such as laboratory and radiology test results.
4
This classification has been employed by a number of
studies in the health informatics and IS literature (e.g., DesRoches et al. 2008; Hydari et al. 2019; Jha et al. 2009). We
4
The other four functionalities include electronic communication and connectivity, patient support, administrative processes, and
reporting and population health management. HIT applications that enable these functions have been categorized as administrative
and strategic HIT in the literature (Angst et al. 2012; Menachemi et al. 2007). Our focus in this study is on clinical HIT applications.
8
use it to explore four clinical functions which we label: (1) Order Entry and Management (OEM), (2) Decision Support
(DS), (3) Electronic Clinical Documentation (ECD), and (4) Results Viewing (RV).
5
Our second source is the literature on health informatics and health services research. The health informatics
literature identifies two fundamental clinical functions. (Riaño et al. 2012; Unertl et al. 2009; Walker and Carayon 2009).
The first function enables physicians to execute treatment plans by ordering medicines, clinical procedures, and lab tests
(i.e., the OEM function). The second enables them to diagnose patient conditions and provide evidence-based medical
guidance (i.e., the DS function). The health services research literature also supports the notion that patient diagnosis
and treatment are the primary functions performed by physicians (Balogh et al. 2015; Del Mar et al. 2006; Groopman
2008). This literature further suggests that these primary functions are aided by support functions, such as clinical
documentation and viewing of prior results. Access to patients’ medical information and how prior procedures and
medications impacted the patients helps doctors assess the efficacy of a treatment regimen. These latter functions
constitute support functions because their key purpose is to assist physicians in their primary clinical functions of patient
diagnosis and treatment. In summary, prior research supports our conceptualization that OEM and DS are primary
clinical functions and ECD and RV are support clinical functions.
Having established the significance of primary and support clinical functions, we now define primary and
support HITs as those information processing systems that enable care providers to accomplish primary and support
clinical functions, respectively. We conceptualize OEM HIT and DS HIT as primary HITs. Specifically, we conceptualize
OEM HIT as the level of computer provider order entry system implementation in a hospital. OEM HIT facilitates order
entry and management by enabling physicians to order medicines, procedures, and tests. We conceptualize DS HIT as
the level of clinical decision support system implementation in a hospital. DS HIT aids decision support by enabling
physicians to accurately diagnose patient conditions, follow latest guidelines, and provide patient-specific care. We
conceptualize ECD HIT as the level of electronic clinical documentation implementation in a hospital. Finally, we
5
The core clinical functions mentioned by Tang (2003) and the clinical functions conceptualized in this study have a one-to-one
correspondence. However, whereas Tang (2003) considers these four above-mentioned functions to be core, we provide a more
nuanced categorization of primary and support functions, based on actual clinical functions accomplished using these applications in
patient care provision.
9
conceptualize RV HIT as the level of results viewing implementation in a hospital. We further conceptualize support HIT
as the combined level of ECD HIT and RV HIT in a hospital.
6
2.4.2 Conceptualization of HIT Complementarity
The use of the complementarity perspective is limited in the HIT literature. To conceptualize the nature of interactions
between HIT applications, we build on Grandori and Furnari’s (2009) work. They posit that the extent and type of
complementarity depend on the interacting elements and the application domains in which they are employed. Here, we
draw a parallel between application domains and clinical functions. Just as Grandori and Furnari (2009) suggest that
interacting elements can be employed in the same or different application domain(s), we posit that OEM HIT, DS HIT,
and support HIT can be employed in various combinations. Functional HIT complementarity is the notion that the joint
application of these technologies in different primary and/or support clinical functions can generate synergies such that
the value produced from the joint application exceeds the sum of values produced through individual applications of the
same technologies. Our conceptualization of complementarity is more domain-specific due to our focus on HIT, and
more nuanced due to the delineation of primary and support clinical functions in our context. We posit that studies on
functional IT complementarity should pay careful attention to how the joint, rather than individual, application of
technologies enables organizations to accomplish various functions.
As primary HITs, OEM HIT and DS HIT are similar, but each focuses on different primary clinical functions.
Hence, any complementarity between these HITs is pooled, because similar factors are applied to different application
domains. OEM HIT and support HIT, as well as DS HIT and support HIT, when applied together, work synergistically
such that OEM HIT and DS HIT accomplish ordering and decision making respectively, and support HIT helps these
functions by enabling electronic documentation and results viewing. Thus, any complementarity between OEM and
support HIT and DS and support HIT is symbiotic, because different factors are applied to the same application domain.
Further, it is important to note that hospitals can implement OEM, DS, and support HIT simultaneously or sequentially in
any order in synergistic ways, thereby producing simultaneous and sequential complementarity, respectively. This
6
We combine ECD HIT and RV HIT into support HIT because our focus is on the primary clinical functions of decision support and
order entry and management, and the auxiliary role that support HIT plays in supporting primary functions.
10
nuance related to longitudinal aspects of complementarity is often ignored in the literature, with most studies focused on
cross-sectional analyses of simultaneous complementarity (Battisti et al. 2015; Brynjolfsson and Milgrom 2012).
2.5 Hypothesis Development
We develop our hypotheses in two stages. First, we focus on pairwise complementarity, establishing why the use of two
clinical HITs, which can be similar (e.g., both primary) or different (e.g., one primary and one support), and jointly applied
to similar or different clinical functions, can be complementary. We then integrate the temporal perspective and theorize
the effects that joint application of two HITs can have when they are implemented during the same time-period (i.e.,
simultaneous complementarity) and during different time-periods (i.e., sequential complementarity). In the second step,
we theorize the effects of three-way complementarities.
2.5.1 Pairwise HIT Complementarity: Functional and Longitudinal Perspectives
HITs enable hospitals to manage both the workflow and the information flow generated in clinical functions. The
implementation and integration of multiple HITs can enhance standardization of procedures and enable widespread use
of captured data for process improvements (Borzekowski 2009; Dranove et al. 2014; McCullough et al. 2016). When
primary and support HITs are jointly applied, the latter provides the information flow needed by the former to execute the
clinical workflow (Unertl et al. 2009). For example, imagine a physician who needs to prescribe treatment for a patient
after consulting prior medical records and results. Here, the primary function involves a physician completing a
prescription, and the support function entails a physician checking if this patient is allergic to the selected medicine
based on a review of prior records and test results. To complete the primary function, the physician needs to use OEM
HIT to select a class of drugs (e.g., antibiotics) and, within this class, select an individual drug (e.g., penicillin). To
complete the support function, the physician needs to check prior medication lists, physician/nursing notes and prior
results from support HIT to ensure the patient is not allergic to the selected medicine. This information supports the
workflow enabled by OEM HIT of prescribing medication, thus generating symbiotic complementarity.
The availability of patient medical history and the efficacy of prior treatments including medicines, tests, and
procedures through the use of support HIT enables physicians to prescribe treatments that are more likely to be clinically
effective for the patient (McCullough et al. 2010). As a result, the patient is less likely to suffer from adverse events,
which enhances care quality and reduces the need to provide uncompensated care to patients adversely impacted
11
because of preventable errors. Support HIT can also help prevent redundant tests and unnecessary procedures, thus
enhancing efficiency and reducing waste (Sharma et al. 2016). Finally, support HIT can improve coordination among
care providers, thus yielding additional benefits in terms of cost and quality of care (Romanow et al. 2018).
In a similar manner, a physician may need to determine a treatment plan based on the best available evidence
for a patient’s medical conditions and how (s)he has responded to prior treatments. It is important to check if the
prescribed drug interacts with other drugs the patient is taking. Here, the primary function involves the physician deciding
on a treatment to prescribe, while the support function involves the physician checking the patient’s medical condition
from prior records and prior results. To complete the primary function, the physician needs to use DS HIT to aid in
checking drug interaction alerts and the latest evidence to decide which medication will be most appropriate for the
patient. To complete the support function, the physician needs the patient’s medication-related information from support
HIT. This information supports the workflow enabled by DS HIT, again generating symbiotic complementarity.
As these examples show, symbiotic complementarity is possible because support HIT provides information on
the patient’s medical history and the efficacy of prior treatments, enabling physicians to devise more accurate, tailored
treatment plans based on the latest clinical guidelines and to prescribe more effective treatments (McCullough et al.
2010). The joint application of primary and support HITs creates opportunities for organizational learning and
coordination across different functions (Kremser and Schreyögg 2016), resulting in enhanced care delivered to patients.
We next examine pooled complementarity effects between two primary HITs applied to primary clinical
functions. DS HIT and OEM HIT can be applied together to facilitate workflows in decision-making and ordering. To
complete the former function, physicians use DS HIT to make suitable treatment decisions based on system-initiated and
patient-specific recommendations. To complete the latter function, physicians use OEM HIT to order the appropriate
treatment. The joint usage of two primary HIT applications in different functional domains clinical decision making and
order entry and management can create pooled complementarity, because while DS HIT can be used to provide
reference information and suggestions, OEM HIT can be used to execute medication orders and tests electronically.
Thus, when DS HIT and OEM HIT are implemented together, pooled complementarity is generated as these two primary
HITs working together generate synergies that can improve quality and reduce cost.
12
In the absence of support HIT, hospitals do not benefit from having patient histories or prior results, but the joint
use of OEM HIT and DS HIT can provide synergistic benefits. For example, hospitals can still benefit from joint effects of
two primary HITs, which enable physicians to prescribe based on best practices, while incorporating a level of decision
support that can protect patients from harm. These synergies improve clinical and experiential quality and lower cost by
decreasing adverse events and errors, enhancing the match between patient conditions and treatments provided, and
customizing care providers’ communication with patients (Romanow et al. 2018), leading to higher patient satisfaction.
It is worth noting that this automation manifests differently for primary and support functions. Technology-
enabled work practices created for support functions are more widely applicable because the informational support they
provide can shore up multiple primary functions. In contrast, work practices created for primary functions are specific to
them and may not complement other primary functions to the same extent. As discussed above, these combinations are
still valuable and can help reduce cost, enhance quality, and enhance patient satisfaction when used together.
Next, we shift our attention to exploring whether the HIT implementation sequence impacts how pairwise
symbiotic or pooled HIT complementarities enhance hospital performance. We posit that symbiotic complementarity
effects (i.e., when support HIT provides information to facilitate the workflow enabled by primary HIT) are more likely to
yield value when both primary and support HIT functions are implemented simultaneously or when support HIT is
implemented before primary HIT. In contrast, pooled complementarity effects are generated when primary HITs work
synergistically to support the workflow in both the ordering and decision-making application domains. Because one
primary HIT does not need another primary HIT to play a “supporting” role in the same application domain, synergistic
value can be generated when both primary HITs are implemented at the same time or in different periods in any
sequence. Thus, when considering automating two primary clinical functions, a dominant automation sequence may not
exist, in which case the order of implementation is immaterial. However, when one clinical function is supported by
another, it may be beneficial to have the two functions automated at the same time or to automate the support function
earlier. Based on the above theorizing, we offer the following two hypotheses:
Hypothesis 1A: Pairwise symbiotic complementarity effects are likely to enhance hospital-level clinical quality and
experiential quality, and reduce healthcare cost when primary and support HIT applications are implemented
simultaneously or sequentially such that support HIT is implemented before primary HIT.
Hypothesis 1B: Pairwise pooled complementarity effects are likely to enhance hospital-level clinical quality and
experiential quality, and reduce healthcare cost when two primary HIT applications are implemented simultaneously or
sequentially in any order.
13
2.5.2 Three-way HIT Complementarity: Functional and Longitudinal Perspectives
Joint application of the two primary HITs along with support HIT could potentially produce three-way complementarity
effects. Sharma et al. (2018) suggest that substantial benefits can accrue when organizations connect “separate islands
of automation” so information can flow from one island to another. When IT applications are integrated, information flows
are automatic - without the need to reenter information - thus reducing potential errors and costs, and facilitating faster
information sharing, care coordination, improved responsiveness and reduced rework. Due to this integration, hospitals
employing both primary and support HITs may benefit from a deeper and wider use of technology (Yu et al. 2009). In the
absence of such integration, information sharing could be difficult, leading to lower coordination and costly mistakes.
When hospitals use primary and support HIT simultaneously, a wide range of interlacing information flows and
workflows are generated, which are geared toward integrating different clinical functions. The joint application of HITs
can create practices that produce synergistic effects. One area where such effects can be realized is in the prevention of
medical errors. The Swiss Cheese Model is often used to explain how healthcare systems can serve as barriers that help
to prevent errors. These barriers (e.g., system warnings generated by DS HIT, reduction in prescription errors by virtue
of using OEM HIT, or the availability of patient history stored in support HIT), however, are imperfect and may be
vulnerable to unintended consequences or holes, just like Swiss cheese. For example, when using DS HIT alone, an
external drug classification system may wrongly suggest an antiplatelet drug for patients already taking one (Wright et al.
2016). When using OEM HIT alone, duplicate dose errors can be generated if the default setting is not changed. The
presence of holes in one barrier may not necessarily lead to an error, but when multiple holes happen to align, a medical
error can result that harms a patient (Perneger 2005). Implementing all technologies together adds more barriers,
reducing the chance of errors. For example, when a doctor prescribes orders with OEM HIT but accidentally fails to
change the default setting, DS HIT will generate a system alarm using patient history stored in support HIT. In this way,
OEM, DS, and support HIT complement each other to avoid preventable errors and reduce uncompensated medical
care. Safeguards engendered through the simultaneous use of OEM, DS and support HITs can be very effective in
enhancing clinical quality. Simultaneous application of three HITs can also lead to experiential quality improvement, as
patients’ medical information stored in support HIT complements ordering and decision-making functions. Drawing upon
information from both primary and support HIT, healthcare providers can work together better as a team, thus enabling
14
them to better meet each patient’s specific needs. Using OEM, DS, and support HITs together also enables physicians
to eliminate redundant medication orders, tests, and procedures, which reduces waste and enhances efficiency. Hence:
Hypothesis 2A: Three-way complementarity effects are likely to enhance hospital-level clinical quality and experiential
quality, and reduce healthcare cost when two primary HITs and support HIT are implemented simultaneously.
When hospitals implement HITs sequentially, the order of implementation may become important because
information flows necessary for executing clinical workflows may or may not exist. As mentioned earlier, when primary
HITs are reinforced by support HITs, it may be beneficial to have the former automated after the latter. If the support HIT
is implemented before primary HITs, for example, patient information and results of prior tests and medications can be
reviewed before ordering tests, treatments, and medications. This helps to ensure that prescribed medications and
treatments will be clinically effective for the patient, and the patient will be less likely to suffer from adverse drug events,
enhancing care quality and lowering the need to provide uncompensated care, arising because of preventable errors.
Further, this implementation sequence will enable physicians to consult support HIT and consider a patient’s prior history
and test results so they can tailor their communication according to each patient’s unique medical conditions. Finally, it
will also enable physicians to avoid ordering unnecessary tests, medications or procedures, leading to reduced cost and
enhanced efficiency (Sharma et al. 2016).Thus:
Hypothesis 2B: Three-way complementarity effects are likely to enhance hospital-level clinical quality and experiential
quality, and reduce healthcare cost when two primary HITs and support HIT are implemented sequentially such that the
latter is implemented before the former.
Three-way complementarity can also materialize when primary and support HITs are implemented such that at
least one primary HIT is implemented at the same time or before support HIT. We conjecture that this implementation
sequence is conducive to care providers’ communication with patients about their medical conditions. When
communicating with patients, it is vital for care providers to devote time to inform patients about the substantive issues
related to diagnosis, medications, and treatment. Although their medical history and test results certainly remain
important, patients are likely to attach more importance to communication and interpersonal interaction related to
diagnosis, medications, and treatment because of their direct relationship with recovery during and after a hospital stay.
This requires that healthcare providers be familiar with the new technologies and spend adequate time to communicate
with patients. Compared to support HITs, which may not be as disruptive to existing workflows, primary HITs may take
more time to integrate with the established flows. During the adjustment period, care providers may need to spend
15
additional time to learn the primary HITs and to adapt to the new clinical workflows. Thus, when one primary HIT and the
support HIT are implemented before the other primary HIT, care providers can first adapt to using the primary and
support HIT in a complementary manner before adjusting their practices to fit another primary HIT. With a shorter
adjustment period required for support HIT, care providers can dedicate more time to learn the primary HITs; this lets
them customize suggestions regarding medication, diagnosis, and treatment, leading to more effective communication,
better patient experience, and enhanced ratings. Similarly, when a hospital implements one or both primary HITs before
the support HIT, physicians get an opportunity to adjust their primary clinical functions related to treatment and
diagnosis. Subsequently, when the support HIT is implemented, they can leverage their experience and adjust their
support functions in a shorter time period, enabling them to communicate with patients about diagnosis, treatment,
health conditions, and prior results, leading to a better patient experience and higher ratings. Prioritizing at least one
primary HIT, can thus be beneficial from the perspective of experiential quality. Hence:
Hypothesis 2C: Three-way complementarity effects are likely to enhance hospital-level experiential quality when two
primary HITs and support HIT are implemented sequentially such that at least one primary HIT is implemented before or
at the same time as support HIT.
3. Data and Measurement
We constructed a longitudinal dataset spanning 2008-2013 from multiple archival data sources. Online Appendix A
provides details on variables, data sources, and correlations between variables. Our first data source is the Healthcare
Cost and Utilization Project’s state inpatient datasets (HCUP-SID), which provide information about clinical quality and
healthcare cost at the hospital discharge-level (see online appendix B, Part I for details). Our second data source is the
Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS) survey, which provides information
about experiential quality. This survey is administered to a random sample of adult inpatients after their discharge to
assess their hospital stay experience. Survey results are averaged to get experiential quality at the hospital-level (see
online appendix C, part I). We use the American Hospital Association (AHA) annual survey and its HIT supplement to
obtain hospital characteristics and HIT implementation data (see online appendices C, part II).
7
Our final sample
7
The annual survey provides hospital demographics and organization structure on more than 6,300 hospitals in the U.S. The HIT
supplement dataset contains HIT implementation information on more than 3,300 hospitals.
16
comprised data from hospitals in seven states: California, Florida, Maryland, New Jersey, New York, North Carolina, and
Washington. Hospitals with missing HIT information were dropped, resulting in 2,054 observations from 715 hospitals.
3.1 Operationalization of Hospital Performance: Clinical Quality, Experiential Quality and Healthcare Cost
We operationalize hospital performance with three measures which encapsulate different aspects of care delivery
clinical quality, experiential quality, and healthcare cost. Our first performance measure - clinical quality - is the inpatient
quality indicator (IQI) 91 composite quality. The IQI measures were developed by experts at leading research universities
and have been used extensively because these measures are rigorously-constructed, risk-adjusted, reliable, and face-
validated (Farquhar 2008; Mutter et al. 2008). IQI 91 composite quality measures consider mortality rates for six
conditions: Acute Myocardial Infarction, Heart Failure, Acute Stroke, Gastrointestinal Hemorrhage, Hip Fracture, and
Pneumonia. These rates vary substantially across hospitals, and evidence suggests that high mortality may be linked to
a deficiency in care provision, giving a direct indication of quality of clinical care patients receive in a hospital. The case
mix of hospitals, i.e., the risk profile of patients, is explicitly considered in these measures to ensure that hospitals
admitting sicker patients are not unduly penalized (Menachemi et al. 2007; Menachemi et al. 2008).
In order to leverage the healthcare context and provide a more nuanced perspective, we draw upon recent
research that has underscored that chronic and acute conditions have different characteristics and that their treatments
require different approaches (Bardhan et al. 2020). Because chronic and acute patients differ in their clinical care needs,
physicians may use different workflows and information flows when treating them (Chaudhry et al. 2006). Accordingly,
we measure clinical quality separately for chronic and acute conditions. The measure we use for clinical quality - IQI 91
composite quality - considers mortalities for both chronic and acute conditions. To evaluate clinical quality for chronic and
acute conditions separately, we mapped the IQI 91 conditions against the chronic indicator variable from the HCUP
dataset. We found that Acute Myocardial Infarction, Heart Failure, and Acute Stroke are chronic diseases, hence,
mortality rates for these three conditions constitute our clinical quality measure for chronic conditions. We also found that
HCUP data classifies 96.8% of Gastrointestinal Hemorrhage, Hip Fracture, and Pneumonia admissions as acute; we
used only this data, discarding the remaining admissions data for these conditions. Mortality rates calculated from the
remaining data for these three conditions constitute our clinical quality measure for acute conditions. Part II of online
Appendix B provides detailed information about the calculation of these measures. Following prior research, and to meet
17
distributional assumptions needed for regression, we applied logit transformation on the clinical quality measures for both
chronic and acute conditions (Collett 2003).
Our second performance measure is experiential quality, which we operationalize using two variables:
communication score and rating score (see online Appendix C, Part I for HCAHPS survey items and calculations). The
communication and rating scores are calculated by using the equations below, with i as the individual hospitals
measured in year t, Q as the average communication score measured by four items, and PR as the rating percentage.
Following recent research (Senot et al. 2016; Sharma et al. 2016) and based on reasons discussed above, we applied
logit transformation on these computed scores (Collett 2003).
  
  

  
 

Our third performance measure is healthcare cost. As discussed earlier, there are inherent differences in
chronic and acute conditions (Bardhan et al. 2020), because of which HIT may have differential effects on hospital-level
healthcare costs for chronic and acute patients. Thus, we examine the impact of HIT on healthcare costs separately for
chronic and acute conditions. We calculate hospital-level healthcare cost for chronic and acute conditions in two steps.
First, we focus on patients who were admitted to the hospital with one of the six IQI 91 health conditions mentioned
earlier. For these patients, we multiply the summarized charge amount for each discharge from the charge file in the
HCUP-SID dataset by the cost-to-charge ratio. Second, we averaged the costs calculated in the previous step over
chronic and acute conditions. We label these costs IQI 91 chronic costs and IQI 91 acute costs, respectively. Following
recent research, these cost measures are log transformed for further analysis (Senot et al. 2016).
3.2 Operationalization of Explanatory Variables
The main explanatory variables in our model corresponding to HIT implementation include DS HIT, OEM HIT, ECD HIT,
and RV HIT. DS and OEM are primary HITs and ECD and RV jointly constitute support HIT. HIT implementation is
measured using a six-point scale, where 1 indicates “fully implemented across all units,” and 6 indicates “not in place and
not considering implementing.” The AHA HIT survey items comprising these measures are in online Appendices C and
D. To calculate HIT implementation levels, we recoded the original data (see online Appendix C, Part II). We constructed
three HIT variables DS, OEM, and support HIT by counting the number of technologies fully implemented at a
18
hospital for each HIT (Angst et al. 2012; Menachemi et al. 2008). We then standardized variables to remove potential
multicollinearity in interaction terms. Items measuring HIT implementation are presented in online Appendix D.
3.3 Operationalization of Control Variables
Hospital size, profit, and teaching status were included as control variables. Further, because a hospital’s HIT use can
be influenced by other hospitals in the market (Miller and Tucker 2009), we included IT network effects and market
competition in the hospital referral region (HRR) as control variables. Network effect was measured by averaging HIT
levels across hospitals in the HRR, excluding the focal hospital. Competition was measured by Herfindahl-Hirschman
Index (HHI). Please see online Appendix A, Part I for details on all the control variables.
4. Data Analysis and Results
Prior empirical research has used two approaches to establish complementarity (Aral et al. 2012; Brynjolfsson and
Milgrom 2012; Tambe et al. 2012). The first approach is used when a performance outcome variable is either not of
interest or unavailable. Under this approach, complementarity is established if different practices or elements are found
to cluster more significantly than they would by random chance. The second approach is used when a reliable
performance variable is available and of interest. Under this approach, complementarity is established if joint application
of practices or elements is found to lead to better performance than the summation of impacts of individual practices or
elements. To analyze complementarity thoroughly, we provide results from both approaches.
4.1 Assessment of HIT Complementarity via Partial Correlation Test
Using the first approach, we conducted a partial correlation test to examine the association among HIT applications over
time. We found that after controlling for hospital characteristics, IT network effect, market competition, state, and year,
correlations between DS and OEM HIT, DS and support HIT, and OEM and support HIT are 0.45, 0.53 and 0.36,
respectively. These correlations are significant at p< 0.001. We also examined how implementation levels of one HIT
change over time when other HIT applications are both high (one standard deviation above the mean), or both low (one
standard deviation below the mean), or mismatched. Our findings suggest that the implementation levels for a HIT with
two other HITs matched at a high level grow at a faster rate over time compared to those with HITs mismatched or
matched at a low level. These results are shown in Figures 1, 2, and 3 in online Appendix E. These tests provide
suggestive evidence of complementarity among HIT applications (Brynjolfsson and Milgrom 2012).
19
4.2 Assessment of HIT Complementarity via Panel Data Analysis
Using the second approach, we conducted panel data analysis to assess HIT complementarity. Based on the outcome of
the Durbin-Wu-Hausman test, we use the fixed effect (FE) model as our main empirical method. We estimated the
following econometric model.
Performance (i, t) = β0 + HITitβ + Controlitη + αi +
t +εit
The dependent variables include different quality and cost measures for a hospital in year t; β0 is the constant, HITit are
HIT variables, including lagged individual IT variables (e.g., DSi, t-1 , OEMi, t-1 , Support i, t-1), and pairwise and three-way
interactions between lagged IT variables (e.g., DS i, t-1Support i, t-1, OEM i, t-1Support i, t-1, DS i, t-1OEM i, t-1, DS i, t-
1OEM i, t-1Support i, t-1); Controlit are the time-varying control variables, including hospital characteristics variables
(e.g., Hospital Size (log) i,t, Teaching i, t, and Not-For-Profit i, t) lagged IT network effect variables (e.g., IT Network Effect i,
t-1), market competition effect (HHI i, t ); αi is the hospital fixed effect that absorbs time-invariant unobserved hospital-
specific effects;
t is year fixed effect that captures year-specific effects shared by all hospitals; and εit is the time-varying
unobserved hospital effect. Controlling for unobserved hospital-specific effects allows us to control for other unobserved
time-constant hospital factors that may affect outcomes. By adding year dummies to control for the year-specific effect,
we can rule out the time trend explanation on hospital outcome changes.
Identification in our analysis comes from time series variation in HIT implementation levels by hospitals. Given
that we are interested in examining the causal effect of HIT on hospital performance, we now address potential validity
threats, such as selection effect, reverse causality, simultaneity, and unobserved hospital heterogeneity. Since not every
hospital completes the AHA’s IT supplement survey, we performed the Heckman selection test to examine the potential
selection bias. Our results indicate that selection bias is not a significant concern in our data (see Table F1 in online
Appendix F). To assess reverse causality, we estimated several models with one-year lagged performance indicators as
explanatory variables and HIT implementation levels as outcomes. Our results show that reverse causality is not a
significant concern (see Tables F2-F4 in online Appendix F). We can rule out simultaneity based on the extensive
evidence in prior literature for a considerable time gap between IT implementation and organization-level impacts.
Finally, the FE model allows us to account for the unobserved hospital heterogeneity that could impact performance.
20
Having addressed potential validity threats, we proceeded to estimate FE models with both pairwise and three-
way HIT interactions. In these analyses, we lag the HITs by one year and two years because the effect of lags on
outcomes fades after two years. We present our results in Tables 2-5. We first present results for pairwise simultaneous
complementarity, lagging three HITs together by either one year or two years. Subsequently, we present results for
pairwise sequential complementarity with some HIT variables time-lagged one year and others two years. This lag
structure is needed to create sequential complementarity terms. Finally, we present results for three-way
complementarity. We summarize our results and provide hypothesis tests in Tables 6 and 8.
4.2.1 Pairwise HIT Complementarity Results
Table 2 presents the effects of simultaneous HIT complementarity on hospital performance measures clinical quality,
experiential quality and healthcare cost when each HIT is lagged either one-year (see Part I) or two-years (see Part
II).
8
Table 2: Simultaneous Complementarity Models
Clinical Quality
Experiential Quality
Healthcare Cost
(1)
(2)
(3)
(4)
(5)
(6)
Chronic
Acute
Communication
Rating
Chronic
Acute
Part I. Models for DSt-1, OEMt-1, and Supportt-1 (all HITs time-lagged one year)
Symbiotic
Complementarity
DS t-1Support t-1
-0.0045
0.0059
0.0022
0.0086
-0.0001
-0.0057
(0.012)
(0.006)
(0.005)
(0.010)
(0.005)
(0.004)
OEM t-1Support t-1
-0.0045
0.0012
0.0064
0.0095
-0.0066
-0.0003
(0.016)
(0.008)
(0.006)
(0.010)
(0.006)
(0.006)
Pooled
Complementarity
DS t-1OEM t-1
0.0031
-0.0075
0.0006
0.0006
-0.0074
0.0009
(0.013)
(0.008)
(0.005)
(0.009)
(0.006)
(0.006)
Three-way
Complementarity
DS t-1OEM t-1Support t-1
-0.0226+
-0.0078
0.0063
0.0141
0.0052
-0.0011
(0.013)
(0.007)
(0.005)
(0.010)
(0.005)
(0.004)
# Observations
2,054
2,054
2,049
2,049
2,054
2,054
# hospital
715
715
713
713
715
715
R2
0.044
0.045
0.373
0.119
0.131
0.125
Part II. Models for DSt-2, OEMt-2, and Supportt-2 (all HITs time-lagged two year)
Symbiotic
Complementarity
DS t-2 Support t-2
0.0142
0.0015
0.0050
0.0177+
0.0021
0.0000
(0.015)
(0.008)
(0.005)
(0.010)
(0.005)
(0.005)
OEM t-2 Support t-2
-0.0477**
-0.0055
0.0029
-0.0010
-0.0187*
-0.0187+
(0.018)
(0.011)
(0.005)
(0.011)
(0.009)
(0.010)
Pooled
Complementarity
DS t-2 OEM t-2
0.0259+
0.0052
-0.0047
-0.0120
-0.0004
0.0004
(0.015)
(0.008)
(0.004)
(0.009)
(0.009)
(0.010)
Three-way
Complementarity
DS t-2 OEM t-2 Support t-2
-0.0115
0.0051
0.0099*
0.0113
-0.0026
-0.0060
(0.016)
(0.009)
(0.004)
(0.010)
(0.008)
(0.009)
# Observations
1,548
1,548
1,541
1,541
1,548
1,548
# hospital
646
646
644
644
646
646
R2
0.043
0.042
0.328
0.093
0.124
0.119
(1) Robust standard errors in parentheses (2) *** p<0.001, ** p<0.01, * p<0.05, + p<0.1 (3) Direct IT effect is included (4) All control variables
including hospital-level variables (e.g., size, profit status, and teaching status), IT network effect, market competition effect, state and year effect are
included (5) Constant is included
8
In the interest of space and focus, we do not report experiential quality separately for chronic and acute patients.
21
Next, we present three sets of analyses to assess pairwise sequential HIT complementarity. Part I in Table 3
provides impacts of interaction effects when support HIT is lagged two periods while DS and OEM HITs are lagged one
period. Part II in Table 3 shows performance impacts of interactions when support HIT is lagged one period while DS
and OEM HITs are lagged two periods.
Table 3: Sequential Complementarity Models of Support HIT
Clinical Quality
Experiential Quality
Healthcare Cost
(1)
(2)
(3)
(4)
(5)
(6)
Chronic
Acute
Communication
Rating
Chronic
Acute
Part I. Models for DSt-1, OEMt-1, and Supportt-2
Symbiotic
Complementarity
DS t-1Support t-2
0.0082
0.0005
-0.0039
0.0090
-0.0104
-0.0034
(0.016)
(0.010)
(0.006)
(0.015)
(0.006)
(0.006)
OEM t-1Support t-2
-0.0024
-0.0017
0.0125+
-0.0195
0.0017
-0.0009
(0.021)
(0.013)
(0.007)
(0.018)
(0.008)
(0.008)
Pooled
Complementarity
DS t-1OEM t-1
0.0087
-0.0176+
0.0043
0.0117
-0.0024
0.0043
(0.014)
(0.009)
(0.006)
(0.013)
(0.006)
(0.006)
Three-way
Complementarity
DS t-1OEM t-1Support t-2
-0.0300+
-0.0025
-0.0006
0.0255
-0.0074
-0.0126+
(0.016)
(0.011)
(0.006)
(0.018)
(0.008)
(0.007)
# Observations
1,191
1,191
1,188
1,188
1,191
1,191
# hospital
499
499
499
499
499
499
R2
0.041
0.040
0.320
0.105
0.109
0.109
Part II. Models for DSt-2, OEMt-2, and Supportt-1
Symbiotic
Complementarity
DS t-2 Support t-1
0.0005
-0.0000
0.0061
0.0068
0.0063
0.0039
(0.016)
(0.008)
(0.005)
(0.011)
(0.007)
(0.007)
OEM t-2 Support t-1
-0.0051
0.0145
-0.0124*
-0.0077
-0.0021
0.0022
(0.018)
(0.009)
(0.006)
(0.012)
(0.007)
(0.008)
Pooled
Complementarity
DS t-2 OEM t-2
0.0031
0.0086
0.0015
-0.0008
-0.0103
-0.0100
(0.014)
(0.007)
(0.004)
(0.010)
(0.008)
(0.008)
Three-way
Complementarity
DS t-2 OEM t-2 Support t-1
0.0225
-0.0115
0.0130**
0.0073
-0.0034
-0.0066
(0.016)
(0.008)
(0.004)
(0.014)
(0.007)
(0.007)
# Observations
1,191
1,191
1,188
1,188
1,191
1,191
# hospital
499
499
499
499
499
499
R2
0.046
0.040
0.325
0.093
0.110
0.109
(1) Robust standard errors in parentheses (2) *** p<0.001, ** p<0.01, * p<0.05, + p<0.1 (3) Direct IT effect is included (4) All control variables
including hospital-level variables (e.g., size, profit status, and teaching status), IT network effect, market competition effect, state and year effect are
included (5) Constant is included
Part I in Table 4 depicts performance impacts of interactions when OEM HIT is lagged two periods while DS
and support HITs are lagged one period. Part II shows the impacts of interactions when OEM HIT is lagged one period
and DS and support HITs are lagged two periods.
Table 4: Sequential Complementarity Models of OEM HIT
Clinical Quality
Experiential Quality
Healthcare Cost
(1)
(2)
(3)
(4)
(5)
(6)
Chronic
Acute
Communication
Rating
Chronic
Acute
Part I. Models for DSt-1, OEMt-2, and Supportt-1
Symbiotic
Complementarity
DS t-1Support t-1
-0.0048
-0.0005
0.0009
0.0089
0.0024
0.0032
(0.015)
(0.008)
(0.005)
(0.011)
(0.006)
(0.006)
OEM t-2Support t-1
0.0152
0.0172
-0.0062
-0.0056
-0.0044
-0.0019
(0.017)
(0.011)
(0.006)
(0.012)
(0.008)
(0.009)
Pooled
Complementarity
DS t-1OEM t-2
-0.0169
-0.0072
-0.0007
-0.0033
0.0010
0.0011
(0.013)
(0.010)
(0.006)
(0.012)
(0.008)
(0.008)
Three-way
Complementarity
DS t-1OEM t-2Support t-1
0.0137
-0.0026
0.0161***
0.0225
-0.0041
-0.0076
(0.013)
(0.008)
(0.004)
(0.014)
(0.008)
(0.008)
# Observations
1,191
1,191
1,188
1,188
1,191
1,191
# hospital
499
499
499
499
499
499
R2
0.044
0.036
0.324
0.096
0.103
0.104
22
Part II. Models for DSt-2, OEMt-1, and Supportt-2
Symbiotic
Complementarity
DS t-2 Support t-2
0.0071
0.0011
0.0066
0.0178
-0.0040
-0.0073
(0.017)
(0.008)
(0.005)
(0.011)
(0.007)
(0.006)
OEM t-1 Support t-2
-0.0192
-0.0051
0.0073
-0.0013
-0.0026
-0.0043
(0.018)
(0.010)
(0.005)
(0.009)
(0.008)
(0.008)
Pooled
Complementarity
DS t-2 OEM t-1
0.0101
0.0008
-0.0039
-0.0140
-0.0048
-0.0017
(0.015)
(0.008)
(0.004)
(0.009)
(0.007)
(0.008)
Three-way
Complementarity
DS t-2 OEM t-1 Support t-2
-0.0218
-0.0024
0.0155***
0.0195+
-0.0047
-0.0049
(0.016)
(0.009)
(0.004)
(0.011)
(0.007)
(0.007)
# Observations
1,191
1,191
1,188
1,188
1,191
1,191
# hospital
499
499
499
499
499
499
R2
0.042
0.036
0.336
0.109
0.109
0.108
(1) Robust standard errors in parentheses (2) *** p<0.001, ** p<0.01, * p<0.05, + p<0.1 (3) Direct IT effect is included (4) All control variables
including hospital-level variables (e.g., size, profit status, and teaching status), IT network effect, market competition effect, state and year effect are
included (5) Constant is included
Part I in Table 5 illustrates performance impacts of interactions when DS is lagged two periods while support
HIT and OEM are lagged one period. Part II presents impacts of interactions when DS is lagged one period while OEM
and support HIT are lagged two periods.
Table 5: Sequential Complementarity Models of DS HIT
IQI 91 Clinical Quality
Experiential Quality
IQI 91 Healthcare Cost
(1)
(2)
(3)
(4)
(5)
(6)
Chronic
Acute
Communication
Rating
Chronic
Acute
Part I. Models for DSt-2, OEMt-1, and Supportt-1
Symbiotic
Complementarity
DS t-2Support t-1
0.0059
0.0072
0.0014
0.0083
0.0081
0.0053
(0.015)
(0.007)
(0.005)
(0.011)
(0.006)
(0.006)
OEM t-1Support t-1
-0.0104
0.0062
0.0036
0.0124
0.0041
0.0053
(0.020)
(0.009)
(0.006)
(0.012)
(0.006)
(0.006)
Pooled
Complementarity
DS t-2OEM t-1
-0.0019
-0.0001
-0.0022
-0.0217*
-0.0110*
-0.0068
(0.016)
(0.008)
(0.005)
(0.011)
(0.005)
(0.005)
Three-way
Complementarity
DS t-2OEM t-1Support t-1
-0.0030
-0.0074
0.0105*
0.0232+
0.0009
-0.0016
(0.016)
(0.008)
(0.005)
(0.012)
(0.005)
(0.005)
# Observations
1,191
1,191
1,188
1,188
1,191
1,191
# hospital
499
499
499
499
499
499
R2
0.045
0.039
0.323
0.096
0.112
0.106
Part II. Models for DSt-1, OEMt-2, and Supportt-2
Symbiotic
Complementarity
DS t-1 Support t-2
0.0079
-0.0041
-0.0001
-0.0004
-0.0065
-0.0013
(0.017)
(0.010)
(0.006)
(0.013)
(0.006)
(0.006)
OEM t-2 Support t-2
-0.0288
-0.0022
0.0122
0.0138
-0.0262*
-0.0251*
(0.024)
(0.015)
(0.008)
(0.018)
(0.011)
(0.012)
Pooled
Complementarity
DS t-1 OEM t-2
0.0059
0.0048
-0.0058
-0.0035
0.0058
0.0069
(0.014)
(0.009)
(0.005)
(0.010)
(0.008)
(0.009)
Three-way
Complementarity
DS t-1 OEM t-2 Support t-2
-0.0142
-0.0098
0.0109+
0.0053
0.0022
-0.0046
(0.017)
(0.011)
(0.006)
(0.018)
(0.009)
(0.009)
# Observations
1,191
1,191
1,188
1,188
1,191
1,191
# hospital
499
499
499
499
499
499
R2
0.039
0.033
0.325
0.102
0.123
0.123
(1) Robust standard errors in parentheses (2) *** p<0.001, ** p<0.01, * p<0.05, + p<0.1 (3) Direct IT effect is included (4) All control variables
including hospital-level variables (e.g., size, profit status, and teaching status), IT network effect, market competition effect, state and year effect are
included (5) Constant is included
Table 6 summarizes our hypotheses and results on pairwise HIT complementarity. The results collectively provide
strong support for the two pairwise HIT complementarity hypotheses. Both the hypotheses in the paper are supported.
Table 6: Summary of Pairwise Complementarity Results
Type of
Complementarity
Hypothesis
Findings
Hypothesis
Support
23
Pairwise symbiotic
complementarity
Hypothesis 1A
Pairwise symbiotic complementarity effects enhance clinical quality for
treating chronic conditions (p<.01) and reduce healthcare cost (p<.05 and
p<0.1) only when primary and support HITs are implemented simultaneously
for two years (see Table 2, Part II, and Table 5, Part II). We also find that
pairwise symbiotic complementarity effects enhance rating score (p<.1) only
when primary and support HIT are implemented simultaneously for two years
(see Table 2, Part II) and enhance communication score (p<.1) only when
support HIT is implemented one year before primary HIT (see Table 3, Part
I). We do not find evidence of pairwise symbiotic complementarity in other
HIT combinations.
Supported.
Pairwise pooled
complementarity
Hypothesis 1B
Pairwise pooled complementarity effects enhance clinical quality for treating
acute conditions (p<.1) when both primary HITs are implemented
simultaneously (see Table 3, Part I) and reduce healthcare cost for treating
chronic conditions (p<.05) when primary HITs are implemented in different
time periods (see Table 5, Part I).
Supported.
4.2.2 Three-Way HIT Complementarity
Following prior literature, we assess three-way complementarity in two steps (Aral et al. 2012; Brynjolfsson and Milgrom
2012; Tambe et al. 2012). We first examine which three-way interactions are significant. Subsequently, we analyze
which of these significant three-way interactions are complementary. This is done by performing a system test of
complementarity to assess if each pair of HIT variables is complementary and then if all three HIT variables are
complementary together (Aral et al. 2012; Tambe et al. 2012).
Significant three-way interactions among primary and support HIT variables are reported in tables 2-5. In order
to assess complementarity for each significant three-way interaction, we first focus on the three pairs of variables
contained in each such interaction. For each pair of variables, we calculate two differences: 1) when two variables are at
high level, what the performance difference is when the third variable is high vs. when it is low; and 2) when two
variables are at low level, what the performance difference is when the third variable is high vs. when it is low. If the
former performance differential is larger than the latter for experiential quality, or lower for costs and mortality for these
variable combinations, then there is evidence for three-way complementarity. The final test is a system test considering
all three pairs of variables within each three-way interaction simultaneously (see online appendix G for details).
In Table 7, we present the system test of complementarity for all significant three-way interactions in Tables 2-5
(complete details are in tables G1-G8 in online Appendix G). Of the 11 significant three-way interactions, three-way
complementarity is strongly supported for nine and moderately supported for two.
Table 7: System Tests of Complementarities
System Complementarity
Test
Outcome
Significance
DS t-1OEM t-1Support t-1
[F(1,1,1) - F(0,1,1) + F(1,1,1) - F(1,0,1) + F(1,1,1) - F(1,1,0)] -
[F(1,0,0) - F(0,0,0) + F(0,1,0) - F(0,0,0) + F(0,0,1) - F(0,0,0)] < 0
Clinical Quality
(Chronic)
p=0.032*
24
DS t-2OEM t-2Support t-2
[F(1,1,1) - F(0,1,1) + F(1,1,1) - F(1,0,1) + F(1,1,1) - F(1,1,0)] -
[F(1,0,0) - F(0,0,0) + F(0,1,0) - F(0,0,0) + F(0,0,1) - F(0,0,0)] > 0
Communication
Score
p=0.003**
DS t-1OEM t-1Support t-2
[F(1,1,1) - F(0,1,1) + F(1,1,1) - F(1,0,1) + F(1,1,1) - F(1,1,0)] -
[F(1,0,0) - F(0,0,0) + F(0,1,0) - F(0,0,0) + F(0,0,1) - F(0,0,0)] < 0
Clinical Quality
(Chronic)
p=0.081+
DS t-1OEM t-1Support t-2
[F(1,1,1) - F(0,1,1) + F(1,1,1) - F(1,0,1) + F(1,1,1) - F(1,1,0)] -
[F(1,0,0) - F(0,0,0) + F(0,1,0) - F(0,0,0) + F(0,0,1) - F(0,0,0)] < 0
Healthcare
Cost (Acute)
p=0.019*
DS t-2OEM t-2Support t-1
[F(1,1,1) - F(0,1,1) + F(1,1,1) - F(1,0,1) + F(1,1,1) - F(1,1,0)] -
[F(1,0,0) - F(0,0,0) + F(0,1,0) - F(0,0,0) + F(0,0,1) - F(0,0,0)] > 0
Communication
Score
p=0.030*
DS t-1OEM t-2Support t-1
[F(1,1,1) - F(0,1,1) + F(1,1,1) - F(1,0,1) + F(1,1,1) - F(1,1,0)] -
[F(1,0,0) - F(0,0,0) + F(0,1,0) - F(0,0,0) + F(0,0,1) - F(0,0,0)] > 0
Communication
Score
p=0.011*
DS t-2OEM t-1Support t-2
[F(1,1,1) - F(0,1,1) + F(1,1,1) - F(1,0,1) + F(1,1,1) - F(1,1,0)] -
[F(1,0,0) - F(0,0,0) + F(0,1,0) - F(0,0,0) + F(0,0,1) - F(0,0,0)] > 0
Communication
Score
p=0.000***
DS t-2OEM t-1Support t-2
[F(1,1,1) - F(0,1,1) + F(1,1,1) - F(1,0,1) + F(1,1,1) - F(1,1,0)] -
[F(1,0,0) - F(0,0,0) + F(0,1,0) - F(0,0,0) + F(0,0,1) - F(0,0,0)] > 0
Rating Score
p=0.025*
DS t-2OEM t-1Support t-1
[F(1,1,1) - F(0,1,1) + F(1,1,1) - F(1,0,1) + F(1,1,1) - F(1,1,0)] -
[F(1,0,0) - F(0,0,0) + F(0,1,0) - F(0,0,0) + F(0,0,1) - F(0,0,0)] > 0
Communication
Score
p=0.013*
DS t-2OEM t-1Support t-1
[F(1,1,1) - F(0,1,1) + F(1,1,1) - F(1,0,1) + F(1,1,1) - F(1,1,0)] -
[F(1,0,0) - F(0,0,0) + F(0,1,0) - F(0,0,0) + F(0,0,1) - F(0,0,0)] > 0
Rating Score
p=0.057+
DS t-1OEM t-2Support t-2
[F(1,1,1) - F(0,1,1) + F(1,1,1) - F(1,0,1) + F(1,1,1) - F(1,1,0)] -
[F(1,0,0) - F(0,0,0) + F(0,1,0) - F(0,0,0) + F(0,0,1) - F(0,0,0)] > 0
Communication
Score
p=0.010*
*** p<0.001, ** p<0.01, * p<0.05, + p<0.1
Table 8 summarizes our hypotheses and results on three-way HIT complementarity. The results collectively
provide strong support for the hypotheses. H2C is completely supported, and H2A and H2B are partially supported. As
these results emphasize, different combinations of HITs have disparate impacts on outcomes, therefore, a nuanced
perspective is necessary when examining the theoretical and practical implications of HIT complementarity.
Table 8: Summary of Three-Way Complementarity Results
Type of
Complementarity
Hypothesis
Findings
Hypothesis
Support
Three-way
simultaneous
complementarity
Hypothesis 2A
Three-way complementarity effects significantly enhance clinical quality for
treating chronic conditions when two primary HITs and support HIT are
implemented simultaneously for one year. We also find that three-way
complementarity effects significantly enhance communication score when
two primary HITs and support HIT are implemented simultaneously for two
years (see Table 7). We do not find significant impacts on clinical quality for
treating acute conditions or rating score.
Partially
supported.
Three-way sequential
complementarity
Hypothesis 2B
Three-way complementarity effects enhance clinical quality for treating
chronic conditions (p<.1) and reduce cost for treating acute conditions
(p<.05) when two primary HITs and support HIT are implemented
sequentially such that the latter is implemented one year before the former
(see Table 7). The impact on chronic clinical quality is weakly significant. We
do not find significant impacts on acute clinical quality or chronic cost.
Partially
supported.
Three-way sequential
complementarity
Hypothesis 2C
Three-way complementarity effects significantly enhance communication
score and rating score when two primary HITs and support HIT are
implemented sequentially such that at least one primary HIT is implemented
before or at the same time as support HIT (see Table 7). All the effects are
significant at the p<.05 level and only one is significant at the p<.1 level.
Supported.
4.3 Post-hoc Analyses
We calculate the economic effects of three-way complementarities using coefficients and standard errors reported in
Tables 2-5. These results are summarized in Table 9. We note that economic effects of three-way complementarity
translate to improvements of up to 2.15%, 0.73% and 1.20% at the hospital-level, respectively, in chronic quality,
25
communication score and rating score, and a reduction of $301 per discharge in acute cost. In aggregate terms three-
way complementarity results in reduced mortality rate, with 215 deaths prevented out of 10,000 admitted patients with
Acute Myocardial Infarction, Heart Failure, and Acute Stroke; and for every 10,000 admitted patients, 73 additional
patients who respond positively on communication score and 120 additional patients who respond positively on rating
score. These effects are economically substantial and were uncovered only after considering simultaneous and
sequential three-way complementarity. These findings have not been previously reported in the literature.
Table 9: Economic Effects of Three-way Complementarities
Impact on Clinical Quality (Chronic) [L = Low; H=High]
DS t-1OEM t-1Support t-1
(Simultaneous)
At high DS and Support HIT, chronic mortality rate decreases 0.75% as OEM increases from L to H.
At high OEM and Support HIT, chronic mortality rate decreases 0.67% as DS increases from L to H.
At high DS and OEM, chronic mortality rate decreases 2.15% as Support HIT increases from L to H.
DS t-1OEM t-1Support t-2
(Sequential)
At high DS and Support HIT, chronic mortality rate decreases 1.79 % as OEM increases from L to H.
At high OEM and Support HIT, chronic mortality rate decreases 0.54 % as DS increases from L to H.
At high DS and OEM, chronic mortality rate increases 0.27% as Support HIT increases from L to H.
Impact on Experiential Quality (Communication Score) [L = Low; H=High]
DS t-2OEM t-2Support t-2
(Simultaneous)
At high DS and Support HIT, the communication score increases 0.18% as OEM increases from L to H.
At high OEM and Support HIT, the communication score increases 0.27% as DS increases from L to H.
At high DS and OEM, the communication score increases 0.43% as Support HIT increases from L to H.
DS t-2OEM t-2Support t-1
(Sequential)
At high DS and Support HIT, the communication score increases 0.20% as OEM increases from L to H.
At high OEM and Support HIT, the communication score increases 0.64% as DS increases from L to H.
At high DS and OEM, the communication score increases 0.05% as Support HIT increases from L to H.
DS t-1OEM t-2Support t-1
(Sequential)
At high DS and Support HIT, the communication score increases 0.31% as OEM increases from L to H.
At high OEM and Support HIT, the communication score increases 0.38% as DS increases from L to H.
At high DS and OEM, the communication score increases 0.14% as Support HIT increases from L to H.
DS t-2OEM t-1Support t-2
(Sequential)
At high DS and Support HIT, the communication score increases 0.46% as OEM increases from L to H.
At high OEM and Support HIT, the communication score increases 0.56% as DS increases from L to H.
At high DS and OEM, the communication score increases 0.73% as Support HIT increases from L to H.
DS t-2OEM t-1Support t-1
(Sequential)
At high DS and Support HIT, the communication score increases 0.40% as OEM increases from L to H.
At high OEM and Support HIT, the communication score increases 0.40% as DS increases from L to H.
At high DS and OEM, the communication score increases 0.27% as Support HIT increases from L to H.
DS t-1OEM t-2Support t-2
(Sequential)
At high DS and Support HIT, the communication score increases 0.43% as OEM increases from L to H.
At high OEM and Support HIT, the communication score increases 0.08% as DS increases from L to H.
At high DS and OEM, the communication score increases 0.67% as Support HIT increases from L to H.
Impact on Experiential Quality (Rating Score) [L = Low; H=High]
DS t-2OEM t-1Support t-2
(Sequential)
At high DS and Support HIT, the rating score increases 0.27% as OEM increases from L to H.
At high OEM and Support HIT, the rating score increases 0.48% as DS increases from L to H.
At high DS and OEM, the rating score increases 1.20% as Support HIT increases from L to H.
DS t-2OEM t-1Support t-1
(Sequential)
At high DS and Support HIT, the rating score increases 0.37% as OEM increases from L to H.
At high OEM and Support HIT, the rating score increases 0.37% as DS increases from L to H.
At high DS and OEM, the rating score increases 0.72% as Support HIT increases from L to H.
Impact on Healthcare Cost (Acute) [L = Low; H=High]
DS t-1OEM t-1Support t-2
(Sequential)
At high DS and Support HIT, acute healthcare cost decreases by $228 as OEM increases from L to H.
At high OEM and Support HIT, acute healthcare cost decreases by $301 as DS increases from L to H.
At high DS and OEM, acute healthcare cost decreases by $241 as Support HIT increases from L to H.
4.4 Robustness Checks
We performed a series of tests to assess the robustness of our results. We estimated models with other specifications,
used alternative operationalizations and included additional controls in our model. We found that our results are robust to
26
these changes. We tested for contemporaneous cross-equation error correlations due to potential trade-offs between
quality and cost. We estimated a system of seemingly unrelated regressions (SUR) to allow correlation among the error
terms in different models. The results largely agree with the baseline model, indicating that our results are robust. Next,
we operationalized HIT variables using the Saidin Index, which is an alternative operationalization of HIT (Sharma et al.
2016). The index is a weighted sum that gives higher weights to rare HITs compared to those widely used by other
hospitals. Results obtained from analyses using the Saidin Index are largely consistent with those from our baseline
models. Finally, we included additional hospital characteristics, such as female patient percentage, percentages of
Medicare and Medicaid patients, and health system affiliation, to assuage concerns that our analysis may be missing
potentially important variables. These analyses provided similar results.
5. Discussion
The current literature pays insufficient attention to the fact that HIT applications can complement one another in multiple
ways that impact quality and cost of care delivered in hospitals. Agarwal et al. (2010) posit that a coherent understanding
of the relationship between HIT and its impact on quality and cost has not yet been reached. Our study contributes to
this research stream by adopting a nuanced perspective, examining the complementarity between primary and support
HITs used in hospitals. Our results indicate that different combinations of HIT applications, applied toward different
clinical functions, implemented simultaneously and temporally-sequenced, impact hospital performance metrics (i.e.,
clinical quality, experiential quality, and healthcare cost) differently and the effects differ for chronic and acute patients.
We contribute to the literature by extending the complementarity thinking to the HIT domain. Our findings suggest that
one combination and sequencing of HIT applications may not serve the needs of all hospitals.
We next discuss the overall impact of HIT on clinical quality, experiential quality, and cost. Our results show that
pairwise complementarity between OEM and support HIT improves clinical quality performance for chronic patients when
both are lagged two years. However, we find that pairwise complementarity between DS and support HIT has no
significant impact on clinical quality for either acute or chronic patients. We also notice that pairwise complementarity
between DSt-1 and OEM t-1 results in enhanced clinical quality for acute patients. Finally, we find that two three-way
interactions between DS, OEM and support HIT, when all three HIT applications are lagged one year and when support
HIT is implemented one time period before OEM and DS reduces chronic mortality, and thus enhances clinical quality.
27
Our results suggest that significant pairwise interaction effects between DS and support HIT, OEM and support
HIT, and DS and OEM may sometimes lead to decreased experiential quality, but the significant three-way interaction
effects always lead to experiential quality enhancement. Three-way complementarity between DS, OEM, and support
HIT potentially produces synergistic effects in three aspects of care - the advice (achieved through DS HIT), prescription
(achieved through OEM HIT), and patient document and result viewing, therefore enhancing experiential quality.
Finally, examining the impact of HIT on cost, we find that although the direct effects of DS, OEM, and support
HIT may increase and decrease costs (direct effects are not presented in Tables 2-5), significant two- and three-way
interactions always result in lower costs. It is important to note that examining impacts of HIT individually can mask their
overall effects. This result may explain why prior studies, which examine HIT effects in isolation, find it to be associated
with both higher and lower costs (Agha 2014). When investigated in combination, HITs improve cost performance.
5.1 Limitations of the Research
Our study suffers from several limitations. First, we used HCUP data from only seven states in the U.S. Although these
states account for approximately 35 percent of the U.S. population, other researchers may wish to consider collecting
data from more states. Second, some of our hypotheses receive partial support, so a priori we cannot predict which
combination of HIT variables will be impactful on cost and quality for all hospitals. Third, although our results strongly
suggest that researchers should consider the joint effects of HIT, this paper does not completely eliminate the
inconsistencies in the literature regarding the impact of HIT on various measures of hospital performance. Fourth, we do
not hypothesize the impact of every combination of HIT complementarity on each hospital-level outcome examined (e.g.,
see H2C). Finally, this study does not examine how various HITs are used, only if a hospital has fully implemented a HIT
across all units or not. Additional research is needed to examine how hospitals use HITs to institute clinical, operational,
and strategic changes.
5.2 Research Contributions and Implications
This research provides contextualized instantiation of and rigorous empirical findings on HIT complementarity. Previous
HIT and IT complementarity literature has proceeded largely in isolation, with complementarity concepts mostly ignored
in HIT research. Our study provides contextual instantiation and empirical operationalizations of these concepts in the
context of HIT. To our knowledge, this research is the first attempt to provide a fine-grained perspective on and
28
instantiate HIT complementarity in the IS literature. We extend prior literature by studying both simultaneous and
sequential complementarity to explain performance. Researchers may wish to experiment with different lag structures to
examine sequential complementarity. Technological innovations may create synergies with other innovations that are
implemented in the same or a different time period. We propose, and our results strongly suggest, that researchers need
to incorporate sequential complementarity in future research. Previous research indicates that changing a subset of
organizational practices may lead to lower performance and recommends changing entire practices in a system
simultaneously for maximum benefits (Brynjolfsson and Milgrom 2012; Milgrom and Roberts 1990). Yet, in practice,
coordination problems can prevent simultaneous implementation of all the changes. Our study shows that
complementarity can enhance performance under both simultaneous and temporal regimes. It adds to the previous
research that has found evidence that simultaneous complementarities improve performance but has ignored
complementarity from a temporal perspective and in a health care context. The nuanced contextual instantiation enables
us to reveal some of the conditions when HIT complementarity impacts performance. Future research can undertake
rigorous theorizing on the various complementarity aspects examined contextually and empirically in this paper and
further extend the complementarity framework for HIT and other contexts.
This research underscores the importance of context in assessing complementarity in two ways. First, in
contrast to much extant work on complementarity, which examines IT at a generic level (Aral et al. 2012), this study
adopts a functional approach in studying various technologies. In doing so, it focuses on clinical HIT applications and
advances complementarity theorizing by conceptualizing functional aspects to emphasize which hospital clinical
functions these HITs enable and whether these functions are primary or support. Second, our results demonstrate that
despite some similarities, clinical HIT applications impact care quality and costs differently for chronic and acute patients.
Additionally, our results indicate that more HIT combinations impact quality and cost for chronic than acute patients. To
our knowledge, these results have not been demonstrated in the HIT literature using large-sample panel data. We
believe these are strong results and signify important contributions to the HIT and IT complementarity literature because
although some extant IS research focuses on the role of IT on self-care and monitoring for chronic patients, our research
shows that even when patients are in the hospital, HIT may be more beneficial for chronic patients. These results
illustrate the need for researchers to consider the impact of HIT on chronic and acute conditions separately.
29
5.3 Implications for Practice
For practitioners, one of the most challenging problems is understanding how to implement the various available HITs in
a way that will yield the most positive impacts on quality and cost of care. This study informs top managers that there is
no single combination of HIT applications that uniformly improves clinical and experiential quality and reduces cost for all
hospitals. Thus, managers must be cognizant of which HIT interactions improve which performance metric under which
conditions. Although, the concept of “best practices” is prevalent in the practitioner literature, when organizations try to
implement them, they rarely have as much success as the exemplars because best practices devoid of context can be
misleading (Brynjolfsson and Milgrom 2012).
This study offers managers several pointers for effective investment in and use of HITs for hospital-level
impacts. Our results suggest that pairwise symbiotic and pooled complementarity likely have the best performance
effects when both HIT functions are implemented simultaneously. Pairwise symbiotic complementarity effects are also
likely when support HIT is implemented before primary HIT, but pairwise pooled complementarity effects are likely
irrespective of the order in which the primary HITs are implemented. In other words, when considering pairwise
complementarity, managers may want to implement primary and support HIT simultaneously or implement support HIT
before primary HIT. Finally, this research informs us that pairwise effects of different HITs impact cost and quality
outcomes related to chronic and acute diseases differently.
With three-way complementarity, hospitals are likely to benefit from enhanced clinical and experiential quality
when they implement primary HITs and the support HIT simultaneously. When sequencing HIT implementations,
managers should recognize that implementing the support HIT before primary HITs is likely to reduce cost and enhance
quality, while implementing at least one primary HIT before the support HIT is likely to enhance only experiential quality.
Our key message to top managers such as hospital chief information officers and chief operations officers is to
prioritize different combinations of HIT contingent on the performance variables they are targeting and the stock of
resources for their hospitals, but also to realize that technology may not impact all outcomes. Additionally, decision
makers must consider how systems of HIT integrate into different disease workflows when they allocate resources. Our
results indicate that HIT applications impact performance metrics, but these impacts need to be further theorized and
analyzed in a fine-grained manner with multiple years of longitudinal data.
30
References
Adjerid I, Adler-Milstein J, Angst C (2018) Reducing Medicare spending through electronic health information exchange:
The role of incentives and exchange maturity. Information Systems Research. 29(2):341-361.
Adler-Milstein J, Jha AK (2017) HITECH Act drove large gains in hospital electronic health record adoption. Health
Affairs. 36(8):1416-1422.
Agarwal R, Gao G, DesRoches C, Jha AK (2010) Research commentary-The digital transformation of healthcare:
Current status and the road ahead. Information Systems Research. 21(4):796-809.
Agha L. (2014) The effects of health information technology on the costs and quality of medical care. Journal of Health
Economics. 34(C):19-30.
Amarasingham R, Plantinga L, Diener-West M, Gaskin DJ, Powe NR (2009) Clinical information technologies and
inpatient outcomes: A multiple hospital study. Archives of Internal Medicine. 169(2):108-114.
Angst CM, Devaraj S, D'Arcy J (2012) Dual role of IT-assisted communication in patient care: A validated structure-
process-outcome framework. Journal of Management Information Systems. 29(2):257-292.
Angst CM, Devaraj S, Queenan C, Greenwood B (2011) Performance effects related to the sequence of implementation
of health information technologies. Production and Operations Management. 20(3):319-333.
Aral S, Brynjolfsson E, Wu L (2012) Three-way complementarities: Performance pay, human resource analytics, and
information technology. Management Science. 58(5):913-931.
Aron R, Dutta S, Janakiraman R, Pathak PA (2011) The impact of automation of systems on medical errors: Evidence
from field research. Information Systems Research. 22(3):429-446.
Ash JS, Berg M, Coiera E (2004) Some unintended consequences of information technology in health care: The nature
of patient care information system-related errors. Journal of the American Medical Informatics Association.
11(2):104-112.
Ayabakan S, Bardhan I, Zheng ZE, Kirksey K (2017) The impact of health information sharing on duplicate testing. MIS
Quarterly. 41(4):1083-1103.
Ayer T, Ayvaci M, Karaca Z, Vlachy J (2019) The impact of health information exchanges on emergency department
length of stay. Production and Operations Management. 28(3):740-758.
Balogh EP, Miller BT, Ball JR (2015) Improving Diagnosis in Health Care. National Academic Press, Washington, DC.
Bardhan I, Chen H, Karahanna E (2020) Connecting systems, data, and people: A multidisciplinary research roadmap
for chronic disease management. MIS Quarterly. 44(1):185-200.
Bardhan I, Oh J-hC, Zheng ZE, Kirksey K (2015) Predictive analytics for readmission of patients with congestive heart
failure. Information Systems Research. 26(1):19-39.
Bardhan I, Thouin MF (2013) Health information technology and its impact on the quality and cost of healthcare delivery.
Decision Support Systems. 55(2):438-449.
Barua A, Kriebel CH, Mukhopadhyay T (1995) Information technologies and business value: An analytic and empirical
investigation. Information Systems Research. 6(1): 3-23.
Battisti G, Colombo MG, Rabbiosi L (2015) Simultaneous versus sequential complementarity in the adoption of
technological and organizational innovations: The case of innovations in the design sphere. Industrial and
Corporate Change. 24(2):345-382.
Bhargava H K, Mishra AN (2014) Electronic medical records and physician productivity: Evidence from panel data
analysis. Management Science. 60(10):2543-2562.
Bhattacherjee A., Hikmet N, Menachemi N, Kayhan VO, Brooks RG (2007) The differential performance effects of
healthcare information technology adoption. Information Systems Management. 24(1):5-14.
Borzekowski R (2009) Measuring the cost impact of hospital information systems: 19871994. Journal of Health
Economics. 28(5):938-949.
31
Bresnahan TF, Brynjolfsson E, Hitt LM (2002) Information technology, workplace organization, and the demand for
skilled labor: Firm-level evidence. The Quarterly Journal of Economics. 117(1):339-376.
Brynjolfsson E, Milgrom P (2012) Complementarities in organizations. Gibbons R, Roberts J, eds. Handbook of
Organizational Economics. Princeton University Press, Princeton, NJ. 11-55.
Buntin MB, Burke MF, Hoaglin MC, Blumenthal D (2011). The benefits of health information technology: A review of the
recent literature shows predominantly positive results. Health Affairs. 30(3): 464-471.
Chandrasekaran A, Senot C, Boyer KK (2012) Process management impact on clinical and experiential quality:
Managing tensions between safe and patient-centered healthcare. Manufacturing & Service Operations
Management. 14(4):548-566.
Chaudhry B, Wang J, Wu S, Maglione M, Mojica W, Roth E, Morton SC, Shekelle PG (2006) Systematic review: impact
of health information technology on quality, efficiency, and costs of medical care. Annals of Internal Medicine.
144(10):742-752.
Collett D (2003) Texts in Statistical Science. Modelling Binary Data. Chapman & Hall, London.
Davenport T (1998) Putting the enterprise into the enterprise system. Harvard Business Review. 76(4):121-131.
Del Mar C, Doust J, Glasziou P (2006). Clinical Thinking. Oxford: Blackwell Publishing Ltd., BMJ Books, Oxford, United
Kingdom.
DesRoches CM, Campbell EG, Rao SR, Donelan K, Ferris TG, Jha A, Kaushal R, Levy DE, Rosenbaum S, Shields AE,
Blumenthal D (2008) Electronic health records in ambulatory carea national survey of physicians. New
England Journal of Medicine. 359(1):50-60.
Dewan S, Kraemer KL (2000) Information technology and productivity: Evidence from country-level data. Management
Science. 46(4):548-562.
Dobrzykowski DD, Tarafdar M (2015). Understanding information exchange in healthcare operations: Evidence from
hospitals and patients. Journal of Operations Management. 36(1):201-214.
Dranove D, Forman C, Goldfarb A, Greenstein S (2014) The trillion dollar conundrum: Complementarities and health
information technology. American Economic Journal: Economic Policy. 6(4):239-270.
Ennen E, Richter A (2010) The whole is more than the sum of its parts-or is it? A review of the empirical literature on
complementarities in organizations. Journal of Management. 36(1):207-233.
Farquhar M (2008) AHRQ quality indicators. In: Hughes RG, ed. Patient Safety and Quality: An Evidence-Based
Handbook for Nurses. Rockville, MD: Agency for Healthcare Research and Quality.
Furukawa MF, Raghu T, Shao B (2010) Electronic medical records, nurse staffing, and nursesensitive patient
outcomes: Evidence from California hospitals, 19982007. Health Services Research. 45(4):941-962.
Gattiker TF, Goodhue DL (2005) What happens after ERP implementation: Understanding the impact of inter-
dependence and differentiation on plant-level outcomes. MIS Quarterly. 29(3):559-585.
Grandori A, Furnari S (2009) Types of complementarity, combinative organization forms and structural heterogeneity:
Beyond discrete structural alternatives. In M. Morroni (ed.) Corporate Governance, Organization and the Firm:
Co-operation and Outsourcing in the Global Economy. Edward Elgar Publishers, London. 63-86
Groopman J (2008) How Doctors Think. Houghton Mifflin Company, Boston, MA.
Hansen S, Baroody AJ (2020) Electronic health records and logics of care: Complementarity and conflict in the U.S.
healthcare system. Information Systems Research. 31(1): 57-75.
Hao H, Padman R, Sun B, Telang R (2018) Quantifying the impact of social influence on the information technology
implementation process by physicians: A hierarchical Bayesian learning approach. Information Systems
Research. 29(1):25-41.
Holsapple CW, Sena MP (2005) ERP plans and decision-support benefits. Decision Support Systems. 38(4):575-590.
32
Hsu J, Huang J, Fung V, Robertson N, Jimison H, Frankel R (2005) Health information technology and physician-patient
interactions: Impact of computers on communication during outpatient primary care visits. Journal of the
American Medical Informatics Association. 12(4):474-480.
Hydari M, Telang R, Marella WM (2019) Saving patient RyanCan advanced electronic medical records make patient
care safer? Management Science 65(5):2041-2059
Jha AK, Orav EJ, Epstein AM (2009) Preventing readmissions with hospital discharge planning. New England Journal of
Medicine. 361(27):2637-2645.
Karahanna E, Chen A, Liu QB, Serrano C (2019) Capitalizing on health information technology to enable digital
advantage in U. S. hospitals. MIS Quarterly. 43(1):113-140.
Kremser W, Schreyögg G (2016) The dynamics of interrelated routines: Introducing the cluster level. Organization
Science. 27(3):698-721.
Lee J, McCullough JS, Town RJ (2013) The impact of health information technology on hospital productivity. The RAND
Journal of Economics. 44(3):545-568.
Lin SC, Jha AK, Adler-Milstein J (2018) Electronic health records associated with lower hospital mortality after systems
have time to mature. Health Affairs. 37(7):1128-1135.
Mainz J (2003) Defining and classifying clinical indicators for quality improvement. International Journal of Quality in
Health Care. 15(6):523-530.
McCullough JS, Casey M, Moscovice I, Prasad S (2010) The effect of health information technology on quality in US
hospitals. Health Affairs. 29(4):647-654.
McCullough JS, Parente ST, Town R (2016) Health information technology and patient outcomes: The role of information
and labor coordination. RAND Journal of Economics. 47(1):207-236.
Menachemi N, Saunders C, Chukmaitov A, Matthews MC, Brooks RG (2007) Hospital adoption of information
technologies and improved patient safety: A study of 98 hospitals in Florida. Journal of Healthcare
Management. 52(6):398-409.
Menachemi N, Chukmaitov A, Saunders C, Brooks RG (2008) Hospital quality of care: does information technology
matter? The relationship between information technology adoption and quality of care. Health Care
Management Review. 33(1):51-59.
Menon NM, Kohli R (2013) Blunting Damocles' sword: A longitudinal model of healthcare IT impact on malpractice
insurance premium and quality of patient care. Information Systems Research. 24(4): 918-932.
Milgrom P, Roberts J (1990) The economics of modern manufacturing: Technology, strategy, and organization. The
American Economic Review. 80(3):511-528.
Miller AR, Tucker C (2009) Privacy protection and technology diffusion: The case of electronic medical records.
Management Science. 55(7):1077-1093.
Mutter RL, Wong HS, Goldfarb MG (2008) The effects of hospital competition on inpatient quality of care. INQUIRY: The
Journal of Health Care Organization, Provision, and Financing. 45(3):263-279.
Nazemi E, Tarokh MJ, Djavanshir GR (2012) ERP: A literature survey. The International Journal of Advanced
Manufacturing Technology. 61:999-1018.
Nebeker JR, Hoffman JM, Weir CR, Bennett CL, Hurdle JF (2005) High rates of adverse drug events in a highly
computerized hospital. Archives of Internal Medicine. 165(10):1111-1116.
O’Connor PJ, Sperl-Hillen JM, Rush WA, Johnson PE, Amundson GH, Asche SE, Ekstrom HL, Gilmer TP (2011) Impact
of electronic health record clinical decision support on diabetes care: A randomized trial. The Annals of Family
Medicine. 9(1):12-21.
Perneger TV (2005) The Swiss cheese model of safety incidents: Are there holes in the metaphor? BMC Health Services
Research. 5(1):1-7, https://doi.org/10.1186/1472-6963-5-71.
33
Peng DX, Ye Y, Feng B, Ding DX, Heim GR (2020) Impacts of hospital complexity on experiential quality: Mitigating roles
of information technology. Decision Sciences. 51(3):500-541.
Queenan CC, Angst CM, Devaraj S (2011) "Doctors’ orders––If they’re electronic, do they improve patient satisfaction?
A complements/substitutes Perspective. Journal of Operations Management. 29(7-8):639-649.
Ranji SR, Rennke S, Eachter RM (2014) Computerised provider order entry combined with clinical decision support to
improve medication safety: A narrative review. BMJ Quality & Safety. 23:773-780.
http://dx.doi.org/10.1136/bmjqs-2013-002165
Riaño D, Real F, López-Vallverdú JA, Campana F, Ercolani S, Mecocci P, Annicchiarico R, Caltagirone C (2012) An
ontology-based personalization of health-care knowledge to support clinical decisions for chronically ill patients.
Journal of Biomedical Informatics. 45(3):429-446.
Romanow D, Rai A, Keil M (2018) CPOE-enabled coordination: Appropriation for deep structure use and impacts on
patient outcomes. MIS Quarterly. 42(1):189-212.
Schlichter BR, Kraemmergaard P (2010) A comprehensive literature review of the ERP research field over a decade.
Management. 23(4):486-520.
Scott IA, Pillans PI, Barras M, Morris C (2018) Using EMR-enabled computerized decision support systems to reduce
prescribing of potentially inappropriate medications: A narrative review. Therapeutic Advances in Drug Safety.
9(9):559-573.
Seddon PB, Calvert C, Yang S (2010) A multi-project model of key factors affecting organizational benefits from
enterprise systems. MIS Quarterly. 34(2):305-328.
Senot C, Chandrasekaran A, Ward PT, Tucker AL, Moffatt-Bruce SD (2016) The impact of combining conformance and
experiential quality on hospitals’ readmissions and cost performance. Management Science. 62(3):829-848.
Sharma L, Chandrasekaran A, Boyer KK, McDermott CM (2016) The impact of Health Information Technology bundles
on Hospital performance: An econometric study. Journal of Operations Management. 41(1):25-41.
Tambe P, Hitt LM, Brynjolfsson E 2012. The extroverted firm: How external information practices affect innovation and
productivity. Management Science. 58(5):843-859.
Tang P (2003) Key capabilities of electronic health record system: Letter report. National Academic Press, Washington
DC.
Teece DJ (2018) Profting from innovations in the digital economy: Enabling technology standards, and licensing models
in the wireless world. Research Policy. 47(8):1367-1387.
Unertl KM, Weinger MB, Johnson KB, Lorenzi NM (2009) Describing and modeling workflow and information flow in
chronic disease care. Journal of the American Medical Informatics Association. 16(6):826-836.
Yu S, Mishra AN, Gopal A, Mukhopadhyay T (2009) IT infusion and its performance impacts: An empirical analysis of
eProcurement in the service industy. ICIS Proceedings 121, http://aisel.aisnet.org/icis2009/121.
Walker JM, Carayon P (2009) From tasks to processes: The case for changing health information technology to improve
health care. Health Affairs. 28(2):467-477.
Wright A, Hickman TT, McEvoy D, Aaron S, Ai A, Andersen JM, Hussain S, Ramoni R, Fiskio J, Sittig DF, Bates DW
(2016) Analysis of clinical decision support system malfunctions: A case series and survey. Journal of American
Medical Informatics. 23(6):1068-1076.
Article
Using a large-scale panel of U.S. hospitals across health referral regions (HRRs), we empirically examine how a hospital's and its neighboring hospitals' adoption of the Clinical Decision Support System (CDSS) (and their meaningful use of CDSS) affects the hospital's quality of care. We find that CDSS adoption significantly reduces a hospital's readmission rate of heart failure, acute myocardial infarction, and pneumonia patients. We also reveal the regional spillover effects of adoption, where neighboring hospitals' CDSS adoption in the same HRR also reduces a hospital's readmission rates. Such spillover effects become more significant with electronic data interchange, facilitating seamless and errorless communications among hospitals. We also show that the spillover effects may depend on HRRs' different market structures and hospitals' meaningful-use status. Our findings offer theoretical and managerial insights for both healthcare researchers and practitioners.
Article
Full-text available
Health information exchanges (HIEs) are entities that have emerged in healthcare delivery markets across the United States. By providing an interorganizational information system (IOIS) and governance over use of this system and the information exchanged through it, HIEs enable more routine and efficient electronic sharing of patient information between disparate and fragmented healthcare providers. This should result in improved quality and efficiency of care. However, significant questions persist about the extent to which HIEs produce these benefits in practice, particularly in terms of reducing healthcare spending. We use transaction cost economics (TCE) to theorize that HIEs establish a quasi-hierarchy that decreases frictions associated with information sharing in ways that reduce healthcare spending, and that the magnitude of reductions is greater when (1) insurer and provider incentives align, and (2) HIE capabilities mature. We can test these conjectures because HIEs, unlike other efforts that provide IOIS, are typically confined to regional markets and develop heterogeneously between these markets, introducing variation in insurer-provider incentive alignment and HIE maturity. Leveraging a unique national panel data set, we evaluate whether HIEs reduce spending for the largest insurer in the United States, i.e., Medicare, and whether incentives and HIE maturity modify the magnitude of reductions. We find significant spending reductions in healthcare markets that have established operational HIEs, with an average reduction of $139 per Medicare beneficiary per year (1.4% decrease) or a $3.12 billion annual reduction in spending if HIEs were nationally implemented in 2015.We also find that these reductions occur disproportionately in healthcare markets where providers have financial incentives to use an HIE to reduce spending and when HIEs are more mature. Our results inform an important open empirical question in the healthcare domain related to the value of HIEs, while also joining perspectives from TCE with the IOIS literature to understand the factors that may be relevant to IOIS value creation more generally.
Article
The U.S. healthcare system has experienced rapid growth in the adoption and use of clinical health information technology, such as electronic health record (EHR) systems. In a field as complex and pluralistic as healthcare, the introduction of these sweeping information technology platforms is transforming the practices, roles, interdependencies, and communication mechanisms that connect a heterogeneous mix of stakeholders. In this study, we assess the impact of EHR adoption and use on the practices of U.S. healthcare stakeholders and the institutional logics that they instantiate. We identify four prominent logics around the adoption and use of EHR systems—medical professionalism, private sector managerialism, technical design, and regulatory oversight. Our analysis reveals the ways in which practices enacted through EHR systems interact to engender a confluence of the institutional logics that mark the field. Specifically, we theorize a mechanism by which the reticulation, or intertwining, of practices mediated by an EHR system variously fosters complementarity and conflict between institutional logics based on the commensurability of the organizing principles invoked. This proposed framework offers a novel approach to the role of information system use in the institutional dynamics of the healthcare system. In addition, this research offers a range of practical insights regarding the ways in which the diverse institutional logics of the U.S. healthcare system may be impacted by the growth of EHR system use.
Article
This study measures the impact of information technology (IT) use on hospital operating costs during the late 1980's and early 1990's. Using a proprietary eight-year panel dataset (1987-1994) that catalogues application-level automation for the complete census of the 3,000 U.S. hospitals with more than 100 beds, this study finds that both financial/administrative and clinical IT systems at the most thoroughly automated hospitals are associated with declining costs three and five years after adoption. At the application level, declining costs are associated with the adoption of some of the newest technologies, including systems designed for cost management, the administration of managed care contracts, and for both financial and clinical decision support. The association of cost declines with lagged IT as well as the cost patterns at the less automated hospitals both provide evidence of learning effects.
Article
Hospitals today are posed with many requirements arising from government regulations and financial incentives to improve patient experiential quality. An administrative challenge thus results from hospital complexity, which can circumvent efforts to enhance experiential quality. This study examines associations of two dimensions of hospital complexity (i.e., hospital safety and patient case mix) with patient‐level experiential quality. The study also investigates how health information technology (IT), via clinical IT and administrative IT, differently mitigates the two dimensions of complexity. Using unique proprietary secondary data sources, we assemble a patient‐level experiential quality dataset. The dataset enables us to simultaneously examine hospital and individual patient factors that may affect experiential quality of individual patients. The findings indicate that a hospital's service variety and case mix are both negatively associated with experiential quality. Although not directly affecting experiential quality, administrative IT mitigates the negative associations of both types of complexity with experiential quality. Post hoc analyses reveal nuanced relationships between hospital complexity, IT, and disaggregated experiential quality dimensions.
Article
Health information exchanges (HIEs) are expected to improve poor information coordination in emergency departments (EDs); however, whether and when HIEs are associated with better operational outcomes remains poorly understood. In this work, we study HIE and length of stay (LOS) relationship using a large dataset from the Healthcare Cost and Utilization Project consisting of about 7.4 million treat‐and‐release visits made to 63 EDs in Massachusetts. Overall, we find that HIE adoption is associated with a 10.2% reduction in LOS and the percentage reduction increases to 14.8% when the hospital is part of an integrated health system or to 21.0% when a patient has a previous visit to an HIE‐carrying hospital. We further find that 1) teaching hospitals benefit more from HIE adoption compared with non‐teaching hospitals, 2) patients with severe or multiple comorbid conditions spend less time in the ED under HIE presence. Together, these results imply that 1) HIE adoption reduces overall ED LOS, 2) wider HIE adoption would scale up the benefits for individual hospitals, 3) magnitude of the association between HIE and LOS is higher when financial incentives for HIE adoption are stronger (e.g., integrated health systems), and 4) the size of the reduction depends on certain contextual moderating factors. Given that HIEs are a key component of healthcare delivery and ongoing reforms, we believe that our findings have important implications and may inform policymakers regarding the nationwide HIE adoption. This article is protected by copyright. All rights reserved.
Article
Prescribing of potentially inappropriate medications (PIMs) that pose more risk than benefit in older patients is a common occurrence across all healthcare settings. Reducing such prescribing has been challenging despite multiple interventions, including educational campaigns, audits and feedback, geriatrician assessment and formulary restrictions. With the increasing uptake of electronic medical records (EMRs) across hospitals, clinics and residential aged care facilities (RACFs), integrated with computerized physician order entry (CPOE) and e-prescribing, opportunities exist for incorporating clinical decision support systems (CDSS) into EMR at the point of care. This narrative review assessed the process and outcomes of using EMR-enabled CDSS to reduce the prescribing of PIMs. We searched PubMed for relevant articles published up to January 2018 and focused on those that described EMR-enabled CDSS that assisted prescribers to make changes at the time of ordering PIMs in adults. Computerized systems offering only medication reconciliation, dose checks, monitoring for medication errors, or basic formulary information were not included. In addition to outcome measures of medication-related processes and adverse drug events, qualitative data relating to factors that influence effectiveness of EMR-enabled CDSS were also gathered from selected studies. We analysed 20 studies comprising 10 randomized trials and 10 observational studies performed in hospitals (n = 8), ambulatory care clinics (n = 9) and RACFs (n = 3). Studies varied in patient populations (although most involved older patients), type of CDSS, method of linkage with EMR, study designs and outcome measures. However, assuming little publication bias, the totality of evidence favoured EMR-enabled CDSS as being effective in reducing the prescribing of PIMs in hospitals, although results were more mixed for ambulatory care settings and RACFs. While absolute effects in most positive studies were modest, they suggest EMR-enabled CDSS are feasible and acceptable to clinicians, and if certain design features are adhered to, there is potential for even greater impact.
Article
Evidence linking electronic health record (EHR) adoption to better care is mixed. More nuanced measures of adoption, particularly those that capture the common incremental approach of adding functions over time in US hospitals, could help elucidate the relationship between adoption and outcomes. We used data for the period 2008-13 to assess the relationship between EHR adoption and thirty-day mortality rates. We found that baseline adoption was associated with a 0.11-percentage-point higher rate per function. Over time, maturation of the baseline functions was associated with a 0.09-percentage-point reduction in mortality rate per year per function. Each new function adopted in the study period was associated with a 0.21-percentage-point reduction in mortality rate per year per function. We observed effect modification based on size and teaching status, with small and nonteaching hospitals realizing greater gains. These findings suggest that national investment in hospital EHRs should yield improvements in mortality rates, but achieving them will take time.
Article
The value-capture problem for innovators in the digital economy involves some different challenges from those in the industrial economy. It inevitably requires understanding the dynamics of platforms and ecosystems. These challenges are amplified for enabling technologies, which are the central focus of this article. The innovator of an enabling technology has a special business model challenge because the applicability to many downstream verticals forecloses, as a practical matter, ownership of all the relevant complements. Complementary assets (vertical and lateral) in the digital context are no longer just potential value-capture mechanisms (through asset price appreciation or through preventing exposure to monopolistic bottleneck pricing by others); they may well be needed simply for the technology to function. Technological and innovational complementors present both coordination and market design challenges to the innovator that generally lead to market failure in the form of an excess of social over private returns. The low private return leads to socially sub-optimal underinvestment in future R&D that can be addressed to some extent by better strategic decision-making by the innovator and/or by far-sighted policies from government and the judiciary. The default value-capture mechanism for many enabling technologies is the licensing of trade secrets and/or patents. Licensing is shown to be a difficult business model to implement from a value-capture perspective. When injunctions for intellectual property infringement are hard to win, or even to be considered, the incentives for free riding by potential licensees are considerable. Licensing is further complicated if it involves standard essential patents, as both courts and policy makers may fail to understand that development of a standard involves components of both interoperability and technology development. If a technology standard is not treated as the embodiment of significant R&D efforts enabling substantial new downstream economic activity, then rewards are likely to be calibrated too low to support appropriate levels of future innovation.