William E. Encinosa, PhD; and Jaeyong Bae, MA
To encourage the adoption of electronic medical records (EMRs), in 2011 the Centers for Medicare & Medicaid Services (CMS) implemented an incentive system that rewards hospitals for implementation of the meaningful use regulations. This program was established by the Health Information Technology for Economic and Clinical Health (HITECH) Act of 2009. HITECH provides about $27 billion in incentives over a 10-year period for hospitals and physicians to adopt the meaningful use of EMRs.1
Prior to HITECH, meaningful use adoption was relatively low. Only 1.6% of hospitals in 2009 had EMRs that satisfy the HITECH rules for meaningful use.2
As of June 2013, more than 60% (n = 3481) of hospitals had received a HITECH incentive payment, for a total of $5.5 billion paid out to hospitals.
However, many doubts about the cost benefits of EMRs are now emerging. In a recent Wall Street Journal editorial, Soumerai and Koppel surmised that the United States will spend $1 trillion on health information technology (HIT) without much in cost savings.3
Much of this skepticism was exacerbated by a recent Agency for Healthcare Research and Quality (AHRQ) evidence report, Enabling Medication Management Through Health Information Technology, which examined almost 36,000 HIT medication studies and found 76 studies that focused on clinical outcomes. About 34% of those studies found a significant beneficial impact of HIT on outcomes. However, only 31 articles studied cost outcomes, and only 5 of those provided full economic analyses. As a result, the report concluded that “given the uncertainty that surrounds the costs and outcomes data, and limited study designs, it is difficult to reach any definitive conclusion as to whether the additional costs and benefits represent value for money.”4
A major limitation of this EMR literature, as the AHRQ evidence report concluded, is that most of the studies examined only 1 EMR medication function at a time. For example, 19 of the 31 cost studies examined only computer decision support functions. Only 2 studies examined the joint use of computer decision support functions and computerized physician order entry in hospitals.5
Of the 14 core measures that hospitals must satisfy for a Stage 1 meaningful use incentive payment, 5 in 2011-2013 deal with medication management functions in the EMR system. Yet no study has examined all 5 functions bundled together.
The reason for this lack of research is that most of the studies usually took place at 1 hospital and involved only a few thousand
patients. Such low power makes it almost impossible to study more than 1 EMR function. A simple power analysis prior to our research indicated that we would need 654,000 patients to detect a 10% change in adverse drug event (ADE) rates truly due to the 5 meaningful use medication management components. Such a large patient sample is beyond the scope of many studies, which may be the reason why the AHRQ evidence report found little evidence of the effect of EMRs on costs.
However, even if researchers have a large enough database to examine all 5 EMR functions, another problem is that an ADE measure that can be applied to large administrative data sets does not yet exist. In fact, because of this data measurement problem, Medicare has yet to establish a zero-payment policy for hospital-acquired ADEs among its list of never events, even though the National Quality Forum has mandated death orsevere disability due to medication errors as 1 of its 28 never events. Such an ADE measure for large databases is much needed under the Affordable Care Act, particularly since the Act launched the $1 billion Partnership for Patients initiative to reduce the incidence of 10 patient safety events (1 of which is ADEs) by 40% through the 26 Hospital Engagement Networks.
In this study, we attempted to overcome these major limitations. First, we examined all 5 meaningful use medication management components together. Second, instead of a few hospitals, we examined a wide array of hospitals in the real world, both advantaged and disadvantaged. We included all hospitals in Florida. Third, since we had all 2.6 million hospitalizations in Florida, we had the statistical power to detect the impact of full meaningful use medication management adoption on rare ADEs and their costs. We did this by developing one of the first measures of hospital-acquired ADEs that can be used on large-scale administrative data sets.METHODS
First, we used the 2010 Florida State Inpatient Database from the Agency for Healthcare Research and Quality’s Healthcare Cost and Utilization Project (HCUP).6 This database consists of all 2.64 million inpatient hospitalizations in Florida. We selected Florida because it has very accurately coded “present on admission” (POA) diagnosis data, especially for International Classification of Diseases, Ninth Revision E-codes (injury codes), compared with other states. Second, our source of EMR data linked to the Florida data was the American Hospital Association (AHA) Healthcare Information Technology (IT) Database Supplement to the AHA 2010 Annual Survey of Hospitals. (See Jha and colleagues7 for a description of this survey.) This survey first asked questions about meaningful use in 2010. Third, we linked the 2008 to 2010 Medicare Hospital Compare data to the Florida data in order to use Medicare’s hospital quality measures.Hospital-Acquired Adverse Drug Events
To identify ADEs, we use the algorithm developed by Lucado and colleagues8 for flagging ADEs in HCUP.8
However, many of the ADEs identified by the algorithm may have occurred outside of the hospitalization. To identify only those ADEs that occurred during that hospitalization, we redesigned the algorithm to flag ADEs that were coded as not being POA. The POA codes used in HCUP have been validated.9
Following Houchens and colleagues10
and the HCUP POA report,9
we screened out a few hospitals (5% of the hospitalizations) that had either more than 95% of their diagnoses coded as POA or more than 20% with POA missing.Meaningful Use
In the AHA IT supplement, we could identify hospitals that by 2010 had implemented the 5 Stage 1 meaningful use core measures that pertain to medication management:
Use computerized physician order entry for medication orders.
Implement decision support system for drug-drug and drug-allergy interaction checks.
Capability to exchange key clinical information (for example, problem list, medication list, medication allergies, and diagnostic test results) among providers electronically.
Maintain active medication list.
Maintain active medication allergy list.
We required functions 1, 2, and 5 above to be “fully implemented across all units” of the hospital. Unit implementation for functions 3 and 4 could not be ascertained from the AHA data. The adoption rates for these 5 measures are reported in Table 1
We used multivariate logistic regression analyses to examine 2 main relationships: (1) the relationship between each of the 5 core meaningful use medication management measures and the probability that a hospitalization will have an ADE and (2) the relationship between the number of the meaningful use medication management measures adopted (0, 1-2, 3-4, all 5) and the costs of a hospitalization. To further minimize endogeneity issues, all our regressions controlled for the following covariates. We included 26 collapsed chronic condition variables derived from 30 chronic conditions developed by Elixhauser and colleagues.11
We also included the following variables to control for potential confounding effects of patient severity: 6 age categories, sex, race, an indicator for emergency department admission, an indicator for surgery, and an indicator for whether the patient had more than 5 chronic conditions. We also controlled for the type of insurance (private, uninsured, Medicare, Medicaid) and for the median household income of the patient’s zip code. Finally, we controlled for hospital characteristics such as bed size, for-profit status, not-for-profit status, public hospital, teaching hospital, and the hospital’s percentage of missing POA codes. We also controlled for whether the hospital had a 20% or more quality improvement from 2008 to 2010 in terms of reduced 30-day Hospital Compare Medicare mortality rates for health attack, heart failure, and pneumonia. Finally, to control for the effects of other HIT components, we used an indicator of “advanced EMR” following Jha and colleagues’7
definition of comprehensive EMRs as a hospital adopting all 24 AHA EMR functionalities in all major clinical units. The means of all the covariates are found in Table 2
The HCUP hospital costs included all costs except physician and laboratory costs. The HCUP cost information was obtained from the hospital accounting reports collected by CMS. To estimate the hospital costs of an ADE, we regressed the log of costs on ADEs. We used the log of expenditures to control for the skewed distribution of expenditures. To avoid bias due to retransforming the log, we used a generalized linear model regression with a log link and a gamma distribution, following Manning and Mullahy.12
We included 100 diagnosis-related group fixed effects in the cost regressions.Selection Bias
PDF is available on the last page.