The trillion dollar conundrum: complementarities and health information technology

Download 115.25 Kb.
Size115.25 Kb.
  1   2   3


David Dranove, Northwestern University

Chris Forman, Georgia Institute of Technology

Avi Goldfarb, University of Toronto

Shane Greenstein, Northwestern University

November 2012

We examine the heterogeneous relationship between the adoption of electronic medical records (EMR) and hospital operating costs at thousands of US hospitals between 1996 and 2009. Combining data from multiple sources, we first identify a puzzle that has been seen in prior studies: Adoption of EMR is generally associated with a slight increase in costs. We draw on the literature on information technology and productivity to analyze why this average effect arises, and explain why it masks important differences over time, across locations, and across hospitals. We find evidence consistent with this approach, namely, that: (1) EMR adoption is initially associated with a rise in costs; (2) EMR adoption at hospitals in favorable conditions – such as urban locations – leads to a decrease in costs after three years; and (3) Hospitals in unfavorable conditions experience a sharp increase in costs even after six years.
JEL Code: I10, L30

I. Introduction

More than a quarter century ago economists engaged in a vigorous debate about the benefits from investment in information technology (IT) in manufacturing and services. That debate was encapsulated in the Solow ‘Productivity Paradox’—“You can see the computer age everywhere but in the productivity statistics” (Solow 1987). That debate eventually faded from view, in part because the data began to reject it. Over time it was found that firms achieved productivity benefits from IT, just with a lag. Moreover, explanations for the lag emerged from considerable work on IT use in enterprises. The challenges to productivity benefits were due to the costly adaptations required for the successful implementation of new IT. In time, it was found that the firms realizing benefits from their IT investments were those that had made complementary investments in areas such as worker skills and organizational decision rights.1

A new manifestation of this debate has surfaced around the use of electronic medical records (EMR). A small sampling of research from the last half dozen years provides a sense of the uncertainty about the productivity benefits from these investments. The Congressional Budget Office states: “No aspect of health information technology (i.e., EMR) entails as much uncertainty as the magnitude of its potential benefits” (Congressional Budget Office 2008). A widely cited 2005 report by the RAND Corporation, published in the leading policy journal Health Affairs, estimates that widespread adoption of EMR by hospitals and doctors could reduce annual health spending by as much as $81 billion while simultaneously leading to better outcomes (Hillestad et al. 2005). Jaan Sidorov, a medical director with the Geisinger Health Plan, an early adopter of EMR, published a response to the RAND report in Health Affairs. Sidorov (2006) highlights the high cost of adoption and cites evidence that EMR leads to greater health spending and lower productivity. Other recent studies, cited below, also fail to find consistent evidence that EMR savings offset adoption costs.

What impact does EMR have on a key determinant of an existing organization’s productivity, such as its operating costs? We view EMR as a type of business process innovation, one that involves not only investments in IT but also changes in the operational practices within the adopting organization. If EMR is viewed in this way, how does that change the understanding of EMR’s impact? We argue that prior literature has missed important features of EMR by not building on prior research studying the adoption and productivity benefits of large scale enterprise IT and “insider econometrics” studies of IT adoption.2

Building on this prior research, we stress the complementary assets that reduce the costs associated with “co-invention,” which is the process of adapting an innovation to unique circumstances and turning the overall change into a net benefit to the enterprise. These complementary assets come from several sources. Local resources may be available as market services, such as expertise in implementing similar business process innovations, or widespread spillovers in how to use the IT. Resources available internally to the enterprise, such as expertise with other business processes inside the enterprise, may also help implement the business process innovation and often cannot be purchased from markets in the short run. To summarize, variance in local and internal resources provides an explanation for why the payoff from EMR may be delayed, and for why we observe variance in the returns to investments in enterprise IT across locations.

We conduct an empirical examination of the impact of EMR adoption on hospital operating costs during the period 1996 to 2009, using the prior research on enterprise IT adoption to frame the analysis and results. The data come from several sources linking hospital costs to EMR adoption and the potential for complementarities. Our main analysis regresses logged operating costs on EMR adoption, hospital fixed effects, and a large number of controls. We focus on whether the impact of EMR is greater for hospitals that are positioned to exploit available complementarities. Thus, our key independent variable is the interaction between EMR adoption and the presence of local complements, as measured by the IT-intensity of local industry. Our key identification assumption is that EMR adoption is not correlated with unobservable cost factors that are differentially trending in hospitals with locally available complementary inputs relative to hospitals that lack these inputs. We explain below why we believe that this is a reasonable assumption; even so, we show robustness to instrumenting for EMR adoption using hospital proximity to EMR vendors and EMR adoption in alliance systems and geographically linked markets.

We find the evidence consistent with our reframing. The timing of cost savings is consistent with what we would expect given the literature on the productivity paradox in IT. For the average hospital, the gains from EMR adoption appear with some delay. Moreover, there is significant heterogeneity in the gains achieved that depend upon the local availability of complementary factors such IT workers. We focus on costs because much of the political discussion has emphasized cost savings, and because the multi-product nature of hospitals makes it easier to measure the implications of EMR for costs than for productivity. However, our focus on costs means that we cannot use our data to rule out the possibility that our results are paralleled by opposite results on clinical benefits, though the prior literature, including Agha (2012), Miller and Tucker (2011), and McCullough et al (2010), have found clinical benefits on average to be small.3

We find that hospitals that adopted EMR between 1996 and 2009 did not experience a statistically significant decrease in costs on average. In fact, under many specifications, costs rose after EMR adoption, particularly for the more advanced EMR systems. However, this effect is mediated by measures of the availability of technology skills in the local labor market. Specifically, in strong IT locations, costs can fall sharply after the first year of adoption to below pre-adoption levels. In weak IT locations, costs remain above pre-adoption levels indefinitely. Overall, hospitals in IT-intensive markets enjoyed a statistically significant 3.4 percent decrease in costs from three years after adoption of basic EMR and a marginally significant 2.2 percent decrease in costs from three years after adoption of advanced EMR. These results are significantly better than the up to 4 percent increase in costs after adoption by hospitals in other markets.

Figure 1 displays these general patterns in the raw data, comparing hospitals that adopt basic and advanced EMR before the adoption period, during the adoption period, and after the adoption period. For basic EMR, costs do not fall until three years after adoption. For non-IT-intensive locations, costs rise sharply in the year of adoption, and then fall back. For IT-intensive locations, costs fall with adoption, and are substantially lower three years after adoption. For advanced EMR, the patterns are similar: costs rise in the period of adoption for non-IT-intensive locations and fall over time for the other hospitals.

We provide evidence that the benefits of strong IT locations arise in part from an agglomeration of IT employment in (other) IT hospitals. Hospitals in locations with strong HIT employment enjoyed a statistically significant 6.9 percent decrease in costs from three years after adoption of basic EMR and a statistically significant 7.3 percent decrease in costs from three years after adoption of advanced EMR. However, concentration of IT employment in other industries is not associated with greater benefits from adopting basic or advanced EMR. Controlling for strong HIT employment, costs still fall more rapidly in strong IT locations than in weak ones. In short, one benefit of strong IT locations is a thicker labor market for HIT workers, though other benefits persist as well.

We also show results suggesting that complementary skills can be found internally in the hospital. For advanced EMR, the initial increase in costs is mitigated substantially if hospitals already have substantial software experience. Hospitals without experience are hurt in the short run for the most sophisticated technologies. We do find, however, that within a short time inexperienced hospitals can make up the difference. Specifically, the difference in costs after adoption for hospitals with and without internal expertise disappears within three years. This suggests that, in contrast to complementary assets that depend on a location with favorable agglomeration economies, some complementary assets to business process innovation can be acquired relatively quickly.

These findings have several implications. As annual U.S. healthcare expenditures climb towards $3 trillion and with spending forecast to exceed $4.5 trillion by 2020, many analysts hope that electronic medical records (EMR) can stem the tide (Centers for Medicare & Medicaid Services). For example, David Cutler and Melinda Beeuwkes Buntin make EMR the centerpiece of their “Two Trillion Dollar” solution for modernizing the health care system (Buntin and Cutler 2009). While some are confident in EMR, others remain cautious, especially due to EMR’s sluggish diffusion. As of 2009, only about 30 percent of America’s hospitals have adopted any advanced elements of EMR.4 This may have been due, in part, to the lack of consistent evidence of cost savings.

In order to spur EMR adoption, Congress in 2009 passed the Health Information Technology for Economic and Clinical Health Act (HITECH Act), which provides $20 billion in subsidies for providers who adopt EMR. Two thirds of hospitals said they planned to enroll in the first stage of HITECH subsidy programs by the end of 2012 (US Department of Health and Human Services 2011). The 2010 Patient Protection and Affordable Care Act also contains provisions promoting EMR adoption. Despite these legislative actions, many remain unconvinced of the benefits of EMR. Our findings also may help resolve the ongoing debate. Supporters and detractors both seem to treat EMR as if its economic impact is independent of other environmental factors, as if it either works or it doesn’t. This creates a conundrum for both sides. If EMR is going to save hundreds of billions of dollars or more, as its supporters claim, why isn’t it working in obvious ways? If it costs more than it saves, as the skeptics argue, why are policy makers so keen to expand adoption? Our results suggest that the debate about EMR should be reframed by drawing on the general literature on business adoption of IT, where it is very common for successful technology adoption to require complementary changes in business processes that rely on specific labor and information inputs. It is also common for new enterprise IT to be more productive when companies have access to these inputs in their local market. Using this experience, it is not surprising that EMR can simultaneously have the potential to generate substantial savings but demonstrate mixed results in practice.

We proceed as follows. Sections II and III describe the institutional setting for EMR, and some of the prior evidence about its effects on hospitals. This motivates a comparison in Section IV between EMR and the adoption of IT inside other organizations, which leads to a reframing of several key hypotheses. Sections V and VI present data and results. Section VII concludes.

II. What is EMR?

EMR is a catchall expression used to characterize a wide range of information technologies used by hospitals to keep track of utilization, costs, outcomes, and billings. In practice, EMR includes, but is not limited to:

  • A Clinical Data Repository (CDR) is a real time database that combines disparate information about patients into a single file. This information may include test results, drug utilization, pathology reports, patient demographics, and discharge summaries.

  • Clinical Decision Support Systems (CDSS) use clinical information to help providers diagnose patients and develop treatment plans.

  • Order Entry provides electronic forms to streamline hospital operations (replacing faxes and paper forms).

  • Computerized Provider Order Entry (CPOE) is a more sophisticated type of electronic order entry and involves physician entry of orders into the computer network to medical staff and to departments such as pharmacy or radiology. CPOE systems typically include patient information and clinical guidelines, and can flag potential adverse drug reactions.

  • Physician Documentation helps physicians use clinical information to generate diagnostic codes that are meaningful for other practitioners and valid for reimbursement

As this list shows, there is no single technology associated with EMR, and different EMR technologies may perform overlapping tasks.

Nearly all of the information collected by EMR already resides in hospital billing and medical records departments and in physicians’ offices. EMR automates the collection and reporting of this information, including all diagnostic information, test results, and services and medications received by the patient. EMR can also link this information to administrative data such as insurance information, billing, and basic demographics. EMR can reduce the costs and improve the accuracy of this data collection. Two components of EMR, Clinical Decision Support Systems and Computerized Provider Order Entry, use clinical data to support clinical decision making (Agha (2012) refers to this as a distinct category labeled Clinical Decision Support or CDS). If implemented in ideal conditions and executed according to the highest standards, EMR can reduce personnel costs while facilitating more accurate diagnoses, fewer unnecessary and duplicative tests, and superior outcomes with fewer costly complications.

Despite these potential savings, EMR adoption has been uneven. Table 1 reports hospital adoption rates for the five components of EMR described above. The data is taken from HIMSS Analytics, which we describe in more detail in Section V. Clinical Data Repository, Clinical Decision Support, and Order Entry are older technologies that were present in many hospitals in the 1990s. Even for these older technologies, adoption rates range from 75 to 85 percent in 2009. The remaining applications emerge in the early to mid-2000s. Adoption rates for these are below 25 percent.

While informative, Table 1 lacks several crucial pieces of information. It lacks comparable data on physician adoption of EMR, for example, which is much lower than hospital adoption (Callaway and Ghosal 2012). Our data do not tell us about intensity of use by physicians and staff within hospitals, about the details of the installation, or on how close operations come to ideal conditions. Interviews with hospital administrators suggest that adoption can be uneven within hospitals, with some departments enthusiastically embracing change while others do not. Although beyond the scope of this study, compatibility issues may shape the success of EMR at a regional level, and this too is missing from the table. There are many different EMR vendors and their systems do not easily interoperate. As a result, independent providers cannot always exchange information, which defeats some of the purpose of EMR adoption (Miller and Tucker 2009).

III. Evidence on the Potential Savings from EMR

Has adoption of EMR reduced hospital costs? This section reviews prior evidence, stressing the absence of work focusing on operational savings, lack of emphasis on complementarities with the labor market, and the absence of accounting for the functional heterogeneity of EMR’s components. This discussion will motivate our concerns and our approach to framing the study of EMR’s impact on productivity using past research that emphasized enterprise IT as a business process innovation.

Every EMR study remarks on the expense. One prominent estimate, from the Congressional Budget Office (CBO 2008), estimates that the cost of adopting EMR for office-based physicians is between $25,000 and $45,000 per physician, with annual maintenance costs of $3000 to $9000. For a typical urban hospital, these figures range from $3-$9 million for adoption and $700,000-$1.35 million for maintenance. These costs are quite significant: If the adoption costs are amortized over ten years, EMR can account for about 1 percent of total provider costs. It would be no surprise, therefore, if research suggested that EMR does not pay for itself, let alone generate hundreds of millions of dollars in savings.

In their review of 257 studies of EMR effectiveness, Chaudry et al. (2006) note that few studies focus on cost savings, providing, at best, indirect evidence of productivity gains. Most of the studies they review focus on quality of care. Ten studies examine the effects of EMR on utilization of various services. Eight studies show significant reductions of 8.5-24 percent, mainly in laboratory and radiology testing. While fifteen studies contained some data on costs, none offered reliable estimates of cost savings.

Hillestad et al. (2005; the widely cited RAND study mentioned in our introduction) uses results from prior studies of EMR and medical utilization and extrapolates the potential cost savings net of adoption costs. They estimate that if 90 percent of U.S. hospitals were to adopt EMR, total savings in the first year would equal $41.8 billion, rising to $77.4 billion after fifteen years. They also predict that EMR adoption could eliminate several million adverse drug events annually, and save tens of thousands of lives through improved chronic disease management.

Sidorov (2006) challenges these findings, arguing that the projected savings are based on unrealistic assumptions. For example, the RAND study appears to assume that EMR would entirely replace a physician’s clerical staff. Sidorov argues that providers who adopt EMR tend to reassign staff rather than replace them. To take another example, EMR is supposed to eliminate duplicate tests, while it is just as likely that EMR may allow providers to justify ordering additional tests.5 Buntin et al. (2011) review 73 studies of the impact of EMR on medical utilization. EMR is associated with a significant reduction in utilization in 51 (70 percent) of these studies. They do not identify any studies of EMR and costs.

Indeed, we have identified only three focused cost studies. Borzekowski (2009) uses fixed effects regression to examine whether early versions of financial and clinical information technology systems generated significant savings between 1987 and 1994. He finds that hospitals adopting the most thoroughly automated versions of EMR realize up to 5 percent savings within five years of adoption. He also finds that hospitals that adopt less automated versions of EMR experience an increase in costs. His conclusions mirror the popular discussion: there appears to be the potential for savings but there is little understanding of the drivers of the heterogeneity across hospitals. Second, Furukawa, Raghu, and Shao (2010) study the effect of EMR adoption on overall costs among hospitals in California for the period 1998-2007. Also using fixed effects regression, they find that EMR adoption is associated with 6-10 percent higher costs per discharge in medical-surgical acute units, in large part because nursing hours per patient day increased by 15-26 percent. This is plausible because nurse use of EMR can be very time consuming. Third, Agha (2012) uses variation in hospitals’ adoption status over time, analyzing 2.5 million inpatient admissions across 3900 hospitals between the years 1998-2005. Health IT is associated with an initial 1.3 percent increase in billed charges. She finds no evidence of cost savings, even five years after adoption. Additionally, adoption appears to have little impact on the quality of care, measured by patient mortality, medical complication rates, adverse drug events, and readmission rates. While not directly about costs, Lee, McCullough, and Town (2012) document small positive effects of hospital IT on productivity.

None of the studies frame EMR in the context of the prior literature on enterprise IT. In other words, there is no examination of factors that shape availability of complementary components such as local expertise or prior experience with related technology. This may be due to a lack of familiarity with the theoretical frameworks that would suggest such differential effects. In the next section, we offer such a framework, based on research on the productivity of large scale IT projects in enterprises, and develop some specific implications for the deployment of EMR.

IV. Information Technology and Complementarities

The existing literature on effective implementation of IT within businesses has emphasized the view of IT as a business process innovation.6 Such innovations alter organizational practices, generally with the intent of improving services, reducing operational costs, and taking advantage of new opportunities to match new services to new operational practices. Typically this type of innovation involves changes in the discretion given to employees, changes to the knowledge and information that employees are expected to retain and employ, and changes to the patterns of communications between employees and administrators within an organization.

Because important innovation in enterprise IT occurs on a large scale, it typically involves a range of investments, both in computing hardware and software, and in communications hardware and software. It also involves retraining employees and redesigning organizational architecture, such as hierarchy, lines of control, compensation patterns, and oversight norms. In the discussion below, we draw on a wide literature to explain a number of common misunderstandings about business process innovations.

For example, there is a misperception that new IT hardware or software yields the vast majority of productivity gains without the need for adaptation by the firm. Prior research has shown that each generation of IT is not readily interchangeable with older products or processes, meaning that the initial investment often does not generate a substantial productivity gain until after complementary investments, adaptations, and organizational changes (e.g. Bresnahan and Greenstein 1996; Bresnahan, Brynjolfsson, and Hitt 2002; Bartel, Ichniowski, and Shaw 2007; Bloom, Sadun, and Van Reenen 2012). Many of these necessary changes are made long after the initial adoption. Hence, it is common for IT investments to have no or negative returns in the short run before yielding positive returns. Among the functions mentioned in EMR, for example, CPOE generates many changes to routine processes. These changes often take time to make, and their productivity gains can come long after the initial rollout.

Thus, business process innovation is not equivalent to installing shrink-wrap software for a PC that works instantly, or merely after training of staff. Instead, prior studies stress the importance of co-invention, the post-adoption invention of complementary business processes and adaptations aimed at making IT adoption useful (Bresnahan and Greenstein 1996). The initial investment in IT is not sufficient for ensuring productivity gains. Those gains depends on whether the employees of the adopting organization–in the case of hospitals, administrative staff, doctors, and nurses–find new uses to take advantage of the new capabilities, and/or invent new processes for many unanticipated problems. For example, at one ophthalmology unit at a teaching hospital, the physicians could not find a way to put their traditional “hand-drawings” into the new formats. They found that the new electronic formats sometimes reduced the richness of the information they could record.

This relates to another common misunderstanding: expectations that the entire cost of investment is incurred as monetary expense. Non-monetary costs comprise a substantial risk from installing a business process innovation. Prior studies emphasize the cost of delays, for example. Delays can arise from non-convexities in investment (e.g., all the wiring must be installed before the communications routines can be tested), the technical necessity to invest in one stage of a project only after another is completed (e.g., the client cannot be modified until the servers work as designed), lack of interoperability during upgrades (which some software handles better than others), and cognitive limits (e.g., staff does not anticipate idiosyncratic issues until a new process is at their fingertips). Moreover, interruptions to ongoing operations generate large opportunity costs in foregone services that can be substantially mediated with internal resources (e.g., development of middleware by in-house IT staff) for which there may be no market price or, for that matter, no potential for resale.7

Thus, planning is another common difficulty in IT adoption. Though the installation of any substantial business process innovation requires planning – i.e., administrative effort by the enterprise in advance of installation – such planning alone rarely ends the administrative tasks required to generate productivity gains. Administrative effort does not cease after installation, and continues throughout implementation. Hiring and training personnel generates use of new hardware, software, and procedures. New users in new settings then notice unanticipated problems, which generates new insight about unexpected issues.

As an example of the necessary adjustments and co-invention required for successful EMR investment, consider one large teaching hospital that supported a diverse and geographically dispersed affiliate network.8 The IT staff configured the records for patients to suit the needs of physicians treating severe medical issues. The central hospital saw many severely ill patients, since it was a major trauma center for its region. After rolling out this new system, the doctors at the satellite campuses complained of wasting time “cleaning up the records” because no responsibility was assigned for updating the records after an urgent event. The doctors at satellite campuses also frequently found themselves wading through many screens when routine issues did not require it, and while the patient was present, diminishing the patient experience. It also lengthened the physician’s day, as they spent time updating records. In this simple example, gaining the maximal productivity gains required tailoring the software to the specific types of users and the specific setting, as well as implementing procedures to keep it updated.

Two key empirical implications arise from this discussion. First, given that there was considerable heterogeneity across US locations in the availability of complementary factors, such as skilled labor and knowledge spillovers (Forman, Goldfarb, and Greenstein 2005, 2012), third-party software support and service (Arora and Forman 2007), and infrastructure (Greenstein 2005, Greenstein and McDevitt 2011), there should have been a visible relationship between investment in health IT and local conditions in a limited metropolitan geographic area. Large cities had thicker labor markets for complementary services or for specialized skills. We expect that thicker markets lowered the (quality-adjusted) price of obtaining IT services such as contract programming and of hiring workers to develop in-house functions.9 Such locations may also have had better availability of complementary information technology infrastructure, such as broadband services. Increases in each of these factors may have increased the (net) benefits of adopting complex technologies in some cities and not others, other things being equal. Non-IT employees in advantaged locations also may have adapted more easily to EMR. Overall, the presence of thicker labor markets for technical talent, greater input sharing of complex IT processes, and greater knowledge spillovers in cities should have increased the benefits to adoption of frontier technologies in big cities relative to other locations (Henderson 2003; Forman, Goldfarb, and Greenstein 2008).

A few examples help to illustrate. One example is El Camino Hospital in Mountain View, California (e.g., near Silicon Valley).  This hospital is an otherwise small community hospital that would normally be a laggard in advanced IT, but not due to its location. It could hire plenty of sophisticated administrators to implement the components of EMR, and IT was "in the air", so the hospital was able to gain worker acceptance and adapt to its advanced EMR system. This type of spillover is not unique to Northern California. It is also related to hospitals in the Milwaukee area, which is not a location normally regarded as one with a thick market for IT talent. However, the largest EMR provider in the country, Epic, has its headquarters in Madison, Wisconsin, less than a 90 minute drive from Milwaukee. This proximity resulted in early and extensive support for hospitals in the Milwaukee area, giving them more experience and, hence, greater success with advanced EMR services than otherwise would be expected.10

This framework has a second implication, which is less surprising, as it mimics long-standing economic work on learning curves inside organizations. Enterprises with existing IT facilities should expect lower co-invention costs than establishments without extensive operations, and that should shape costs around the time of adoption. Prior IT projects may reduce development costs if on-staff programmers are able to transfer lessons learned from one project to another.11 Prior work on other IT projects may create learning economies and spillovers that decrease the costs of adapting general purpose IT to organizational needs, reducing the importance of external consultants and local spillovers. For example, many major medical centers in the US – such as Duke, Vanderbilt, Hopkins, UPMC, Yale, or Washington University in St. Louis – invested in advanced IT in order to remain competitive, and those centers initially built their EMR with in-house software instead of packages, using internal expertise during every additional investment. That internal expertise proved valuable when the hospitals later adopted packages and customized them to their organizations.

In summary, if the productivity impact of EMR follows patterns seen with other types of IT, then it should come with a lag. Furthermore, the productivity impact of EMR should depend on factors that shape the supply conditions for complements, such as the experience of a hospital’s IT staff, as well as the local labor market for skilled labor and third-party software and support.

Share with your friends:
  1   2   3

The database is protected by copyright © 2019
send message

    Main page